Hi friends 👋,

Happy Friday and welcome back to our 204th Weekly Dose of Optimism!

The pace of all of this good stuff continues to blow my mind. We sent the Dose last Friday morning. It was jam-packed. “How can things get any better than this?” I asked myself.

And then the big tech companies came out in favor of open-weight models out of pure self-interest, which I wrote about in AI is Oil, Not God.

And then… the Sixers signed Lebron James!

I guess the real lesson, no matter which particular stories we cover each week, is to Trust the Process.

This week’s theme is physical AI - robots, robot brains, drones, and AVs, with a little robot regulation in the Extra Doses.

Let’s get to it.

(1) Enigma Comes Out of Stealth with $71M & 100 Robot Online Control

A company called Enigma is a little less enigmatic after emerging from stealth on Monday with a $71 million seed round led by Index Ventures and Ribbit Capital, plus its decision to open up more than 100 real AI-powered robots available online for anyone to control in real-time. You can go play with them right now, and tell them what to paint, how to lightsaber, to please defuse that bomb, or even which experiments to run.

The cute little robot arms you’re controlling are actual physical machines in facilities in Israel and California, taking your instructions through the browser.

Enigma is building robot-agnostic software, foundation models for robotics, and user interfaces that make robots easier to command across different hardware and tasks. Its bet is that studying how humans engage with robots produces both better interfaces and possibly a different kind of robotic brain.

The public robots.online experiment lets users try text, audio, demonstrations, and other interaction patterns while the company observes how people naturally attempt to get machines to do things. If the race in robotics is about accumulating real training data fast and cost-efficiently, this is a novel and orthogonal approach to that which doubles as a unique marketing stunt.

Part of the reason the company is able to do something so different is that its founders aren’t robotics people. Jonathan Jacobi and Gal Niv met in teenage hacking competitions and served together in Israel’s Unit 8200. Neither is a roboticist. That might be a good thing - in deep tech, people can often get stuck in whichever path they focused on for their PhD. No PhD, no stuck.

Will it work? Who knows, I’m not a robotics PhD. But it looks fun and different.

(2) Gemini Robotics 2 Brings Whole-Body Intelligence to Robots

BAH GOD THAT’S GOOGLE DEEPMIND’S MUSIC!

The frontier lab exploring the most frontiers, and the one that has done by far the most to advance the state of robots and world models, is back with Gemini Robotics 2, its next-generation robotics model family for controlling robots across whole-body movement, dexterous manipulation, embodied reasoning, and multi-robot collaboration.

The release includes three models: Gemini Robotics 2, a vision-language-action (VLA) model that maps vision and language into motor control; Gemini Robotics ER 2, an embodied reasoning (ER) model for planning longer tasks and coordinating robots; and Gemini Robotics On-Device 2, a local model that can adapt to new robot bodies with a few hours of data, typically fewer than 200 examples.

With these models, GDM is moving beyond tabletop manipulation into full-body control. The model can control Apptronik’s Apollo humanoid as it walks, crouches, reaches, and manipulates objects in cluttered environments. It can also operate different end effectors, including five-fingered hands and two-finger grippers, across tasks like tying a knot, sealing a ziplock bag, packing tools, and placing objects. DeepMind says the same model checkpoint controlled multiple embodiments, including Apollo variants and a Franka Duo setup.

Humanoid robots haven’t been solved yet. Travis Kalanick is still right, for now. GDM notes that movement speed still needs work, and the benchmarks show uneven performance on multi-finger dexterity. But it is a meaningful shift in the robotics stack, with one model family handling perception, planning, control, adaptation, safety, and collaboration across different bodies.

Ironically, after the team by the name got subsumed into Google DeepMind, robots are finally getting a … Google Brain.

(3) DoorDash Announces DoorDash Air Drones

So we have robots that anyone with a computer can control, and models that can control any robot that walks, and now DoorDash is making its own flying robots, aka drones.

DoorDash announced on Wednesday that DoorDash Labs earned FAA Part 135 air carrier certification and launched DoorDash Air, its in-house drone delivery program. Part 135 lets DoorDash operate as an air carrier, and the company says it is only the eighth drone operator to earn it.

In its tweet, DoorDash said, “Drones are just one delivery option DoorDash Labs built to scale delivery for merchants, alongside Dashers, Dot, and our autonomous delivery partners. Everything from the aircraft to the merchant handoff system have been designed to get your orders to you quickly and safely.”

The company’s strategy and bet is clear here: they believer that it already owns the most important part of the delivery ecosystem - the network of customers and merchants and the technology to coordinate it all - and that it can commoditize the ways that it gets things from one party to the other. Whatever is at the right spot on the cheap/fast Pareto Frontier, they can choose.

Someone smarter on drones than I am said they’re making their drones less as a standalone drone play and more as a template to build standards and encourage competition in the space. Like Nvidia and Microsoft signing the letter, DoorDash is trying to commoditize its complements.

It’s a super interesting bet, like Uber trying to make its own self-driving cars. It’s interesting to note that Uber is no longer doing that, but that it is providing a demand cannon to dedicated AV makers. In any case, I want drone delivery ASAP, and I’m glad DoorDash is making the push to speed the whole thing up.

And look, I think it’s cool, but the most-liked reply to DoorDash’ tweet is:

So we have a little ways to go before people get used to a burrito landing in their yard.

(4) Zoox Wins 1st US Approval for Robotaxis w/o Steering Wheels

Everyone is super excited about computer AI and I get it but I continue to think that the way cooler and way more underappreciated thing is what’s going to happen in the physical world, and in particular the way that we move people and things.

While DoorDash is flying our burritos overhead, Amazon’s Zoox will be zooking passengers around the streets sans human driver, sans, even, the potential for a human driver.

Per Reuters, “Zoox unit won ​U.S. approval on Wednesday for limited commercial deployment of its novel steering-wheel-free robotaxis, a first for the autonomous ride industry, ‌the U.S. auto safety agency said Thursday.”

This isn’t the first self-driving car, obviously. Waymo and Tesla are operational in a number of American markets, and they are incredibly cool. But it’s the first vehicle designed from the ground-up for a self-driving world that will be commercially available in the US, the Adam or Eve of future generations of vehicles that never even contemplated that a human might drive it.

In Zoox’s case, that means no driver’s seat or steering wheel, and “two rows of inward-facing ​seats” so passengers can spend their ride hanging out more easily. I would imagine that in the future, cars will be designed more and more like moving rooms. Hop in, work or enjoy a movie or get a spa treatment or whatever, and get out at your destination.

I can’t wait to see the horrendous things people do to and in these cars.

(5) OpenAI Gives 10k Researchers Free Access to Frontier Models

There’s almost nothing I’ve found as fascinating this year as Sam Altman / OpenAI’s comms transformation, and there’s nothing that’s captivated me over the past week as much as market incentives aligning almost everyone against the market leader, Anthropic, and towards things that are great for everyone else.

I tweeted about the former after Sam’s Invest Like the Best appearance, and wrote about the latter after the open letter here.

This latest announcement by OpenAI, and Sam’s commentary, combines both.

Basically, OpenAI’s new message since bringing the TBPN guys on (correlation or causation, who knows) has been that they’re building tools to empower people to do amazing things. Replacing people is weird. Trying to do everything themselves is weird. Thinking they can predict the future is weird. Just make tools, give them to people, and see what happens.

This is both a response to how little people want the AI labs to tell them what their future is going to be, and to Anthropic’s more opinionated views. I’m not passing a value judgment here - I believe that Dario fully believes what he’s saying, and that he’s doing a principled thing by speaking out about what he sees to be the risks to his product even if it hurts him or the company. I just happen to strongly disagree with that view (he’s a million times smarter than me and in the center of this, so grain of salt) and it also gives OpenAI a wide-open lane for counterpositioning.

The ILTB conversation was a great example of the comms side of this - Sam was very on-message about AI being a tool. I especially liked the point that it’s not even that out of line with historical trends in progress, that everyone always just wants to feel that they’re in a special role in the special part of history. The whole thing is designed to deflate fears that AI will replace us and to make Anthropic look weird.

But those are words, as is signing the open letter, and at the same time, OpenAI is likely lobbying behind the scenes for regulatory capture, too.

What’s amazed me most this week, and made me appreciate capitalism, is that even if you don’t trust Sam or OpenAI or big tech, companies are incentivized to do things that are good for the public, too.

In this case, counter-positioning against Anthropic trying to go more vertical in certain categories like biotech means that OpenAI is giving free frontier models to tens of thousands of researchers because “the best way to do this is for us to empower scientists, not to try to figure out everything ourselves.”

A jab, to be sure, but if the result is that more researchers get more powerful tools with which to benefit us all, jab away.

EXTRA DOSES: HT Conversation, FCC Covers Foreign Robots, Wall-Penetrating Camera, Pangram, No Science Breakthroughs This Week

(Extra 1) My Conversation with General Catalyst CEO Hemant Taneja

Throwing this one above the paywall as an extra little tweet. There is a ton going on in the private markets, a ton of it driven by AI, and there are very few firms deploying more capital in more different ways to AI in the private markets than General Catalyst. In conjunction with the release of his Q2 Letter, I sat down with GC’s CEO Hemant Taneja for a conversation on the markets and AI generally, how GC is approaching it specifically — including owning a healthcare system and making Anthropic its biggest position — and what the role of a venture capitalist is today.

This was a fun one because there are definitely some things we disagree on, and we went into them, but we are both fundamentally optimistic about technology.

You can also find the conversation on Spotify here.

Read more

Happy Saturday! The past couple of days have been a technology business strategy nerd’s Super Bowl plus a Pro-Progress Person’s Super Bowl. Whenever you combine commoditizing complements with fighting the precautionary principle, I’m not going to be able to stop myself from writing something.

This is lightly edited and is just my quick thoughts, so I’m throwing a bunch of it behind the paywall for the real ones (not boring world subscribers).

Tell our enemies that they may take our lives, but they'll never take our freedom!

Let’s get to it.

Yesterday, Satya Nadella and Jensen Huang, among other leaders in tech, signed and shared a letter in support of open-weight models. Jensen even set up a twitter account to share it.

I loved it, and tweeted that this was Capitalism at its best.

From a business perspective, this is classic tech strategy stuff.

“Smart companies try to commoditize their products’ complements.”

– Joel Spolsky, Strategy Letter V

All of the signatories either make open-weight models models or are complementary to them. Models run on Nvidia chips. They power Microsoft products. a16z and Y Combinator back companies that support, use, and compete with models. Palantir’s Alex Karp has been banging this drum because he wants to disintermediate models in the enterprise. Meta is more open source than the other similar American labs. Hugging Face is where open models live. Dell sells the machines they run on. CrowdStrike benefits if defenders have access to the same intelligence as attackers. Box, ServiceNow, Replit and Perplexity all get better when capable models become cheaper, more plentiful and easier to customize. Google, a new signatory, got off of its slow, bureaucratic ass and signed because Google will be fine either way, and maybe open source models will allow it to serve search ads more cheaply at some point. Mariana Minerals needs to give its comms person a raise, or send a big thank you to a16z.

The more competition there is at the model layer, the less Anthropic and OpenAI are able to run away with the frontier and build a duopoly, the more each of the signatories benefits.

Signing this letter is a selfish act that also happens to align with the interests of consumers and millions of businesses.

And that’s great! That’s capitalism, baby!

Their self-interests are aligned with our interests as consumers and businesses using these models and living in a world in which these models (and the leaders of the labs that create them) exist.

Not everyone agrees with me or the signatories. There are people who believe that it’s better to lock these models down because they might be dangerous.

I think it comes down to whether you think AI is more Oil or God, more “an economically useful commodity that can be scaled and refined to act as a multiplier on everything we do” or “supremely intelligent and powerful being that’s going to wake up and make humanity subservient (if we're lucky).”

I am strongly on the Oil side, as I wrote in The Goldilocks Zone in June 2024. My belief is unchanged since.

In fact, I think it’s super interesting that Satya and Jensen, both of whom are at the very bleeding edge of this technology, wrote in support of open-weight models. Surely, if they saw something that made them believe that these models were going to become conscious and take over the world, they wouldn’t be as openly pro-diffusion, even if they were short-term incentivized to be. I think they know that AI is Oil, too.

But aren’t they going to be dangerous?

My friend Matt Kaufman at Collaborative Fund, who is a lot smarter than me, emailed me after seeing my tweet with just that concern:

Just saw your tweet on the Jensen note and I’ve seen a few others supporting open-weight model proliferation. I’m on the fence here and if you have 2 min to type out a quick reply, I’d be very curious how you’re thinking about it.

I’d definitely prefer a more open ecosystem were it not for risks like models being fine-tuned to remove bio / cyber safeguards, and then resulting in harm. This seems trivially easy and perhaps isn’t a big issue now, but likely will be once 1) the open ones are much more powerful and 2) dangerous actors outside our tech bubble catch on to the capabilities here.

I think there are many domains where LLMs more favor offense rather than defense so this feels like a scary reality and I therefore sympathize with arguments that they should be tightly regulated like weapons.

I quickly banged out a response and I wanted to share it with you, even though it was dashed off in a few minutes and a little messy, because I’ve been meaning to write something and this is close. I’ll add some footnotes on things I’ve thought more about since hitting send.

Read more

Hi friends 👋,

Happy Friday and welcome to the 203rd Weekly Dose of Optimism.

At least here in New York City, the weather is 10/10, it’s a summer Friday, and you probably want to spend some time outside. So let’s cut this intro short and here’s an audio version (my AI voice, so please excuse the pronunciation of “a16z”) if you want to pop it in your ears and walk around.

Peep the Extra Doses for some wild science, an excellent essay on the Frontier, and an absolutely stellar 50-year land art project.

Let’s get to it.

(1) Travis is Back (with $1.7B from a16z)

In Dose #185 in March, we talked about Travis Kalanick’s emergence from the stealthiest stealth in startup history with his new/old company, Atoms, which plans to do everything.

Well Travis is back again.

On Wednesday, Atoms announced a $1.7 billion equity investment led by a16z, with Ben Horowitz joining the board, which is a very rare occurrence these days, to make up for past mistakes on both sides, per Travis:

I’ve known Marc and Ben for a long time, and we got oh-so-close to partnering up at Uber in 2011. I blame Marc, Ben blames himself but let’s just say it was on all of us and in 2017 Uber suffered the consequences of not having Marc on the board - iykyk.

Notably, Uber is also participating. Benchmark is not.

“On many levels, this round is a bit of unfinished business,” Kalanick wrote. In 2011, a16z came very close to leading Uber’s Series B before the deal fell apart and Menlo Ventures led the $32 million round instead. a16z, which is assembling the mid-2010 founder Infinity Stones with Kalanick and Adam Neumann, doesn’t make the same mistake twice.

Not that it really matters when it’s Travis Kalanick, but what is a16z investing in?

After leaving Uber heartbroken in 2017, Kalanick bought City Storage Systems, turned CloudKitchens into its best-known business, and spent eight years building in near-total silence. Employees couldn’t put the company on their LinkedIn, for example.

Meanwhile, the company hired thousands of people, expanded across 30 countries, bought and developed real estate, wrote restaurant software, built robots, and learned what it takes to operate complicated physical systems every day. As part of this round, those businesses have been merged into one Atoms equity structure.

Atoms looks less like a startup than a conglomerate: food, mining, and transport all under one corporate umbrella.

Kalanick’s unifying idea for Atoms (and Uber before it) is that physical industries can be built like computers. In Kalanick’s model, manufacturing is the CPU that transforms atoms, real estate is the storage layer, and transportation is the network that moves atoms from one place to another.

Food is the furthest along version of Kalanick’s computer today:

• CloudKitchens supplies delivery-first real estate.

• Otter is the restaurant OS.

• ProFood builds large-scale food infrastructure.

• Picnic handles office lunch delivery.

• Lab37 builds the robots.

Lab37’s Bowl Builder takes an order from a restaurant’s point-of-sale system, dispenses and weighs the ingredients, applies sauce, lids the bowl, labels it, puts it in a bag with utensils, and closes the bag. The machine can assemble up to 300 bowls per hour (200 if it has to bag them, which Kalanick said on TBPN is surprisingly hard), and Lab37 says it can reduce assembly labor by more than 40%. It is already operating in restaurants.

On the Mining side, Atoms acquired Pronto, the autonomous-haulage company founded by former Uber and Google autonomy pioneer Anthony Levandowski, and made it the technology engine of Atoms Mining.

Pronto retrofits existing Caterpillar, Komatsu, and other haul trucks instead of requiring a mine to buy an entirely new fleet. At Heidelberg Materials’ Lake Bridgeport quarry in Texas, a mixed fleet running Pronto’s system autonomously hauled more than two million tons of limestone in under eight months. Pronto now has an agreement to deploy on more than 100 trucks across over a dozen Heidelberg sites, is expanding to two more quarries this year, and signed a partnership with Hitachi Construction Machinery last week. In his latest TBPN interview, Kalanick says autonomy can improve mine productivity by 30% to 40%.

Transport, the third Atoms division, is still mostly hidden. Kalanick describes it as a wheelbase for robots. TBD.

In total, Atoms has assembled seven operating companies and shown credible proofs in food and mining. It is yet to be seen whether he can turn all of those pieces into one really big computer, if he ends up with a disconnected menagerie of good businesses, or if the whole thing crumbles under its own weight.

But this Travis Kalanick waging war on some of civilizations’s biggest challenges, all at once, with potentially the biggest chip on any billionaire’s shoulder and a desire to make Benchmark miss out on trillions of dollars in returns. If you want to bet against him, that’s on you. I’m happy cheering on The Return of the King.

(2) Claude (and Levent Alpöge) Disprove the Jacobian Conjecture

During the World Cup final on Sunday night, Anthropic mathematician Levent Alpöge posted the finale of one of algebraic geometry’s most famous open problems: ‘hello there the jacobian conjecture is false.’ He thanked his friend Akhil for asking about it and his other ‘close friend,’ Claude Fable 5, for working on it during the game.

The Jacobian conjecture goes like this. Take a polynomial map from n-dimensional complex space back to itself. If its Jacobian determinant (the local measure of how the map stretches and squishes space) is always the same nonzero number, the map is locally reversible everywhere. The conjecture, generalized by Ott-Heinrich Keller in 1939, says that it must therefore be globally reversible, with a polynomial inverse.

Alpöge’s three-variable map has a Jacobian determinant that is always -2, satisfying the conjecture’s premise. But it sends three different points—(0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2)—to the exact same point, (-1/4, 0, 0). A map that turns three inputs into one output cannot have an inverse.

Look… I don’t know what I’m talking about here. Better to play with Simon Rezchikov’s visualization and explainer here if you’re really curious.

The bigger point is that these models, pointed by human mathematicians, are going on a conjecture disproving SPREE, to the point that Joscha Bach (jokingly?) proposed a moratorium on disproving until they’re good enough to prove.

That comment was in the replies to another disproval:

Adios Dinitz-Garg-Goemans! It looks like the Jacobian disproval was the 4-minute mile of disproving longstanding conjectures. Now everyone knows it’s possible, and they’re burning tokens on disproving their favorites. They’ll probably all fall soon (unless Anthropic keeps throttling (which I reproduced)).

At which point: 1) we’ll see how that translates into applications, and 2) neti neti.

(3) India’s First Private Rocket - Skyroot Vikram-1 - Reaches Orbit

Gopika Gopakumar and Nivedita Bhattacharjee for Reuters

Indian space startup Skyroot has reached orbit with Vikram-1, the country’s first privately developed orbital rocket.

My friend Christian Keil at a16z described Skyroot’s Vikram-1 like this: “Think of it as the Indian Electron. Roughly equivalent in size to the RocketLab rocket, also non-reusable.” I want to be clear that he described it that way. Apparently a lot of his “new Indian haters” didn’t like him comparing it to the “workhorse rocket for a $40 billion company (RocketLab).”

So just the facts: Vikram-1 is 22 meters tall and carries up to 350 kilograms. It has three solid stages and a liquid orbital-adjustment module powered by a 3D-printed engine. On its maiden flight, it delivered customer payloads to a 450-kilometer orbit. The whole trip took about fifteen minutes.

Skyroot’s speed is impressive. India opened its space sector to private companies in 2020. Six years later, a startup founded by former Indian Space Research Organization engineers (ISRO is India’s NASA) has joined the very small club of private companies that have built a rocket and put useful payloads into orbit.

India joins an even smaller club: countries whose private companies have achieved orbit. The other members are the United States, China, and kind of New Zealand. Rocket Lab is Kiwi by birth but was headquartered in the US by the time Electron first reached orbit from New Zealand in 2018. I say we count it.

Russia, amazingly, hasn’t joined the club, despite its decorated state space program. The Soviet Union put Sputnik into orbit in 1957, but seventy years later, no private Russian company has done the same. Communism fall hard.

Most of the world is happy for India, aside from a few MAGA anon accounts who totally would have launched a rocket to orbit if the Indians hadn’t taken the job.

Chak De, India! 🇮🇳

(4) Science: A New Golden Age

Director Michael Kratsios

On Tuesday, White House science advisor Michael Kratsios released Science: A New Golden Age, a 123-page attempt to do for American science in 2026 what Vannevar Bush’s Science: The Endless Frontier did in 1945.

Bush’s report established the federal government’s role in supporting basic research, helped lead to the creation of the National Science Foundation, and designed the system that has funded American science for the past eighty years.

Kratsios’ argument is that Bush’s system has worked!… so well that it calcified.

The federal government distributes roughly $200 billion in research and development funding every year, largely through a model in which scientists submit detailed project proposals to panels of other scientists, who attempt to reach consensus on which proposals deserve funding. It is pretty good at separating plausible science from nonsense. It is less good at distinguishing potentially revolutionary science from merely solid science, because potentially revolutionary science often looks like nonsense before it works. It is not good at Taking Weird Ideas Seriously.

The process has also become suffocatingly slow. Researchers spend nearly half of their time on administrative work. From 1991 through January 2025, the federal government added at least 270 new requirements to research grants. Some grants take as long as twenty months to move from submission to award, which the report helpfully points out is almost as long as it took Boeing to move the 747 from the drawing board into production.

Science: A New Golden Age proposes treating that $200 billion more like a portfolio, with different funding mechanisms designed for different kinds of scientific work.

Some researchers would receive long-horizon funding based on their talent and track record instead of being forced to invent a new project every three years. Howard Hughes Medical Institute provides its investigators with roughly $10 million over seven years, with minimal reporting requirements and tolerance for early failure. Research cited in the report found that those scientists produced high-impact work at nearly twice the rate of similarly accomplished federally funded peers. So why not do that at the Federal level?

Smaller, exploratory projects could receive “fast grants,” a la the Collisons during Covid, based on applications just a few pages long, with decisions made in under a month instead of six to nine (nice). Reviewers could receive “golden tickets” that allow each of them to champion one unconventional proposal that the rest of the committee dislikes. Agencies could use more prizes, advance market commitments, and ambitious challenges that pay for results instead of reimbursing inputs.

The report proposes regranting programs that give scientists close to a field small pools of money to distribute to promising people and projects before consensus forms around them. It calls for more DARPA-like program managers who can construct a thesis, recruit teams, and make a coordinated portfolio of bets. And it wants every major science agency to create a metascience unit that runs controlled experiments on the funding process itself, testing whether fast grants, golden tickets, partial lotteries, or other mechanisms actually produce better science. I love all of that.

If you have followed the Progress Studies world for the past few years, the report reads like a greatest-hits album. It cites Heidi Williams on rebuilding the NIH, Patrick Collison’s Fast Grants, Sam Rodriques and Adam Marblestone on Focused Research Organizations, Caleb Watney on X-Labs, Ben Reinhardt on unbundling the university, the Vesuvius Challenge, ARPA-E, HHMI, and the Institute for Advanced Study.

The report is not an executive order and does not immediately change how NIH or NSF awards grants. But it was released alongside an FY2028 R&D Priorities Memorandum directing federal agencies to incorporate the recommendations into their next budget proposals and submit implementation plans within ninety days.

There is a big risk here, too. The report argues that federal science funding needs to be more “politically accountable.” In the constitutional sense—that elected governments decide how public money is spent—that is obviously true. In practice, giving political appointees more discretion over scientific grants creates an obvious path from “fund high-risk work that consensus panels overlook” to “Eric Trump receives $17 billion to disprove string theory once and for all.”

A separate proposed OMB rule would give political appointees the power to block grants before they are issued and terminate existing awards that no longer align with administration priorities. That is not the same thing as golden tickets, fast grants, or empowered technical program managers, and it would be a mistake to smuggle political control in under the banner of metascience.

The best version of this plan would create more independent sources of scientific judgment, not replace one centralized consensus with another. It would combine traditional peer review, long-term bets on individual scientists, fast grants, prizes, lotteries, regranting, and DARPA-style portfolios, then publicly measure which approaches work for which kinds of problems. Program managers should have more freedom to make unusual bets, but their selections, conflicts, milestones, and long-term track records should be legible.

There is a ton of great stuff in here, stuff I would have fallen over if you’d told me was coming from the White House just a few years ago. Now, the challenge will be turning those ideas into practice for the benefit of all mankind.

(5) Anduril and Archer Bring the Thunder

A reader commented last week that they were unsubscribing because they thought I was celebrating war by highlighting autonomous boat startup Saronic’s Very Big Week.

To be very clear, I don’t celebrate war and I think very few people building for the DoW do either. It’s a dumb, blunt way to resolve conflicts that we couldn’t figure out how to resolve in a better way. Every life lost during a war is a tragedy.

But war is unfortunately a fact of life, and if that’s the case, I want America to have a strong enough arsenal to deter it, and ideally an autonomous enough one to put fewer human lives at risk. As Anduril put it, “The modern battlefield is a robotic kill zone.”

To that end, Anduril and Archer’s announcement of Thunder, “a Group 5 autonomous attack rotorcraft,” is a welcome one. The idea is that these are wingmen for human-flown Apache helicopters, pushing out ahead to scout, eliminate threats, and take fire otherwise intended for the people if necessary.

In the anime launch video (see: The Great Differentiation), Anduril shows the autonomous Thunder flying ahead and wiping out drones and missiles that threatened the human pilots’ helicopters. Turns out, they’re father and son, and they’re able to reunite after the mission that might otherwise have killed them.

It’s a bit on the nose, but I think that points to the way that Anduril views the program, as a way to put autonomous mass in harms way instead of humans.

Hopefully, one day, war will just be autonomous machines fighting autonomous machines, wasting both sides’ money but keeping the people out of it until it becomes too much of a simulacrum and we just call the whole thing off. This is another small step in that direction.

In nearer-term optimistic news, Archer is introducing the same platform for commercial use as Archer Halo.

It will focus on “uncrewed commercial missions across search & rescue, logistics, offshore energy, humanitarian & medical cargo, maritime operations and more.”

EXTRA DOSES: Science Breakthroughs (including a potential aging breakthrough in mice, Doudna CRISPR work, and a pancreatic cancer vaccine), Chris Paik on the Frontier, Star Axis

Read more

Hi friends 👋 ,

Happy Friday and welcome to our 202nd Weekly Dose of Optimism!

As always, a big week here at the Dose. This week, we had a lot of good news come in pairs: two Saronic stories, two Austin stories, two robotics stories, and two open-weight model stories. So I cheated and combined them to squeeze a little extra optimism into your Dose.

Let’s get to it.

Today’s Weekly Dose is brought to you by… Not Boring

At the bottom of this and every Weekly Dose, we share some extra goodies, including Science Breakthroughs from Ulkar Aghayeva, who I first found through her work with collaborators ranking 2025’s papers and breakthroughs for their potential impact. Now, you get those glimpses into the future as they happen each week.

That’s one of many goodies we’re bringing to paid not boring subscribers, from the full version of Riding the Leopard to co-written essays with founders on the cutting edge of robotics, AI, manufacturing, and more. We’ll keep bringing you more, and I hope you’ll join us.

Subscribe now

(1) Saronic’s Very Big Week

We last covered Saronic, the Austin-based autonomous shipbuilding startup, in Dose #197, when its Corsair drone boats rescued a downed Apache crew near the Strait of Hormuz. A little over a month later, the company has had an even more absurd week.

On Monday, three of its 24-foot Corsair autonomous boats were used in combat (not rescue) for the first time, striking an Iranian submarine and ship-maintenance facility at Bandar Abbas. The military has not publicly assessed the damage. A system built by a four-year-old startup went from prototype to a live combat mission, a month after another Corsair performed the first known rescue of downed pilots by an autonomous boat. Seems to be working.

Then yesterday, Saronic announced that it had chosen Brownsville, Texas, for Port Alpha, a more than $3 billion greenfield shipyard. The first phase will cover 835 acres, with room to expand to nearly 4,400. Construction is supposed to begin this year, operations in 2028, and the company expects to create as many as 10,000 direct jobs over the next decade.

The initial yard will be able to build ships up to 850 feet long. Fully built out, Saronic says it could reach roughly 2 million gross tons of annual shipbuilding capacity. The plan is to build crewed and uncrewed military ships, container ships, roll-on/roll-off vessels, tankers, and icebreakers, using the same autonomy and advanced-manufacturing stack as the drone boats.

Proving that a startup can build autonomous boats that are useful in combat is huge, but proving that they can build a great American shipyard again might be even bigger.

If you’ve been reading the Dose for a while, you might be thinking … wait, wasn’t that shipyard supposed to be a part of California Forever? Good memory. It was. California fumbled it, because local and state government failed to pass legislation around a proposed permitting framework in time.

“We hope this missed opportunity serves as a wake-up call that inaction and political gridlock have real costs for all Californians,” the California Forever-led labor and business coalition that pushed for the legislation wrote. Until then, in the words of Tim Riggins, it’s Texas Forever.

(2) Big Week for Austin, Actually: Base Launches in Austin + TerraFirma

Speaking of Austin companies doing big things… IT’S COMING HOME.

No, not the World Cup. Jude Bellingham shit-talked Messi, so Messi decided to end England’s run in the semis.

I’m talking about Base Power Company, which after more than two years operating out of Austin, TX is finally operating in Austin, TX.

Base is beginning to serve homeowners inside Austin Energy territory as part of a 40-megawatt residential battery agreement with the city-owned utility. Base will install and maintain the batteries. Austin Energy will be able to dispatch the fleet when demand and wholesale power prices spike. Homeowners get whole-home backup when the grid goes down. Win-Win-Win.

In the very same week, another Austin company, TerraFirma, which is “Building giant robots to build a brighter future,” announced a $100M Kleiner Perkins-led Series A to make construction robots. In the SpaceX-vets’ words, “We’re a new type of company, a robotic construction company that builds the full technology stack needed to deliver an order-of-magnitude improvement in one of the world’s oldest, largest, most important, but least efficient industries.”

It seems like the TerraFirma guys also got every great vertical integrator co-founder/exec on the cap table, including Anduril’s Matt Grimm, Hadrian’s Chris Power, and their Austin neighbor, Base’s Justin Lopas.

It’s time to build. Hook ‘em. Texas Forever.

(3) Don’t Forget About the Humanoids: Walden and Sunday

Regular-sized robots can raise money, too, and it seems like they can do more than that.

On Wednesday, Walden Robotics came out of stealth with $300 million in seed funding at a $1.1 billion valuation co-led by Deviation Capital and a trio of Toyota investors (Toyota Motor Corp, Toyota Invention Partners, and Toyota Ventures). Its robots are already doing useful work in a North American Toyota factory, too.

The Toyota connection makes sense, because Walden spun out of Toyota Research Institute in January. CEO Russ Tedrake, an MIT professor who led TRI’s robotics and machine learning team for nearly a decade, brought over many of the people behind Diffusion Policy, Large Behavior Models, OpenVLA, and the Drake robotics simulator. By February, the new company had robots in production - on the factory floor in two months!

The robots are built as a full stack: hardware, software, models, and an application layer for deploying them into real workflows, a la Standard Bots. They learn tasks through demonstrations, then improve through practice, starting with the annoying jobs in manufacturing and logistics that have remained hard to automate because they change too often or require a little too much judgment for traditional fixed automation.

Most of us don’t live in factories, though, we live in homes. And there seems to be good news for us home-dwellers, too.

Yesterday, Sunday Robotics introduced ACT-2 Preview, its new robotics model for Memo, the company’s home robot. Sunday says ACT-2 can learn a new behavior from a single fine-tuning example, generalize it zero-shot to real homes it has never seen, and perform with a 99% success rate.

It’s still a preview, but they got ahead of the slick demo criticism by introducing something called Solves. “Progress in robotics is difficult to measure because demos vary by setup. Demo ≠ Solved,” CEO Tony Zhao tweeted. “A Solve declares two boundaries: scope and adaptation cost. Without both, 99% has no context.”

They found a “general recipe for Solves: scale pretraining, then hill-climb with minimal in-house data. For the first time, one fine-tuning example can teach a new behavior that generalizes.”

Blah blah blah that’s a lot of robot words. What can it actually do for me?

And now for your Moment of Zen (3 hours of Memo folding laundry)…

(4) FDA Approves New Merck Pill to Slash Cholesterol Levels

Sorry the picture for this one isn’t that exciting. Merck doesn’t do fancy pictures for their FDA-approval announcements. What they do do is make drugs that save lives, which, if you have to choose, is what you want.

Yesterday, the FDA approved Merck’s Lipfendra, the first oral PCSK9 inhibitor. It’s a once-daily pill that lowers LDL cholesterol about as much as the injectable PCSK9 drugs: a placebo-adjusted 56% in a broad hypercholesterolemia trial and 59% in patients with the inherited form.

It’s a bad time to be LDL Cholesterol. In Dose #195, we talked about Eli Lily’s new PCSK9-targeting gene therapy. Which means it’s a good time to be a heart. LDL cholesterol is one of the leading causes of heart attacks. PCSK9 is a protein that destroys LDL receptors on liver cells. If you can block it, more receptors survive to pull bad cholesterol out of the blood.

This isn’t the first PCSK9-targeting drug. There’s the upcoming gene therapy, and there are injected antibodies like Repatha and Praluent that do this very well. They are also injections, which turns out to be a meaningful barrier when you are asking millions of people to take something for years.

Luckily, the other thing they do uniquely ahead of the curve is macrocyclic peptides. They signed a $220M biobucks deal with not boring capital portfolio company Unnatural Products in 2024 to work on macrocyclic peptide, which it describes as “the next wave of drug discovery.”

What macrocyclic peptides allow you to do is to take larger molecules that normally require injections (they generally get chewed up in the gut) and deliver them orally. Lipfendra is a macrocyclic peptide, folded into a shape that helps it make the journey from mouth to target intact.

This isn’t a new capability, but it’s a new delivery mechanism: heart attacks can now be attacked back with antibody-like potency in a $10.50 daily pill. Eat your heart out.

(5) Kimi and Thinky Open Up the Frontier

Thinking Machines and Moonshot AI

It feels like just weeks ago that the US Government shut down Anthropic’s Fable 5 for being too dangerous while the USMNT squashed its opponent in a World Cup Match. Oh how things change.

In bad news for people who think that AI models are super super dangerous and might even kill us all on behalf of their owners like the Genie in Aladdin when Jafar has the lamp, and good news for normal people, a Chinese open source model has basically caught up to Fable 5 (and OpenAI’s GPT-5.6 Sol).

Yesterday, Beijing-based Moonshot AI released Kimi K3, the first open model in the 3-trillion-parameter class. K3 has 2.8 trillion total parameters, a one-million-token context window, and native vision. It is live now in Kimi’s apps and API, with the full weights scheduled for release by July 27th.

The big model is sparse, activating just 16 of 896 experts at a time. Moonshot says its new Kimi Delta Attention makes decoding up to 6.3x faster in million-token contexts, while a second architectural change called Attention Residuals delivers roughly 25% higher training efficiency at less than 2% additional cost. Taken together with the larger mixture of experts and new training recipes, the company claims about a 2.5x improvement in scaling efficiency over Kimi K2.

Moonshot says K3 still trails Claude Fable 5 and GPT-5.6 Sol overall. But on its published evaluations, it is in the same band and wins individual agentic and coding tests: 91.2 on BrowseComp versus Fable’s 88.0 and GPT-5.6 Sol’s 90.4, for example, and 88.3 on Terminal Bench 2.1 versus Fable and Opus 4.8 at 84.6. Those are company-run results and will need independent replication, but it looks like an open model is competing with the big boys.

This comes on the heels of Mira Murati’s Thinking Machines Lab releasing its first model, Inkling, on Wednesday, with the weights available to download under an Apache 2.0 license. Inkling is a 975-billion-parameter mixture-of-experts model with 41 billion active at a time. It accepts text, images, and audio, has a context window up to one million tokens, and can be fine-tuned through Thinking Machines’ Tinker platform.

Inkling is not the best model in the world, either, by Thinking Machines’ own admission. Its bet is that it is a strong, broad model that other people can turn into the best model for one particular thing.

To make the point, the team had Inkling write and run its own fine-tuning job, evaluate the new weights, and switch itself over to a version that never uses the letter “e.” Then, they opened it up to everyone, in the hopes that someone has an even better idea than that, which is how free markets work.

Open models competing is great because competition is great, because it means that more value will accrue to the many companies that make up the economy versus a handful of closed labs, because it lessens the odds that Dario is in charge of the world, and because open models, theoretically, compound on each other.

This seems to be happening here. Inkling’s architecture draws on China’s DeepSeek-V3, and Thinking Machines used data generated by Moonshot’s Kimi K2.5 during post-training.

Open-weight is not the same thing as fully open source. Neither lab is handing over every piece of training data and code, and almost nobody is going to run a multi-trillion-parameter model on a gaming PC. But weights are the valuable part if you want to inspect a model, customize it, improve it, or build on it permissionlessly.

Plus, it’ll give Ramp’s new AI Token Spend Management tool more ammo in the fight against bad ROT.

EXTRA DOSES: Science Breakthroughs, Space Tissues, Senra

Read more

Subscribe now

Hi friends 👋,

Happy Thursday!

Just about three years ago, I was having coffee with another VC, someone whose portfolio I really like, and I asked him which one he was most excited about. He asked, “Do you know what wire harnesses are?” and I said something like “like bundles of wires?” and he said “kinda” and then told me about Jordan Black and Senra Systems.

Ever since then, it’s been a company that I’ve wanted to understand better, and as robotics has begun to take off and reindustrialization is underway and aerospace and defense are booming and the future looks more and more electric, it’s become even more relevant. I keep hearing about it from people I respect, too.

The reason wire harnesses are interesting, aside from the fact that they’re in everything electronic, from toys to Jeeps to satellites to F-35s, is that it’s devilishly hard to automate their manufacture. They’re high-mix, low volume, 3D, bendy, and floppy, among other tricky attributes. They’re often the component holding up the production of our most advanced machines, because you need them to turn everything else on, and once they’re in, a faulty harness can precipitate recalls or worse.

And because skilled assembly, and especially skilled assembly of wire harnesses, is one of the most challenging parts of the manufacturing process to automate, figuring out how to help workers build more of them, faster and more reliably, even as many of the skilled workers who know how to make them retire, may set the pace of American manufacturing.

Plus, I’m a sucker for giving humans superpowers instead of replacing them, and Jordan argues that “the more the easy stuff gets automated, the more valuable skilled humans will become.”

When Alexa Liautaud at GC introduced me to Jordan, I thought it would be a great opportunity to learn everything I can about wire harnesses and skilled assembly and what the future of manufacturing in America looks like from him, Senra’s Head of Strategic Finance Greg Brainard, and one of their investors, Sam Rohr at JAWS. And of course, to write something with them to share what I learned with you.

So to celebrate Senra’s $65M Series B from Lowercarbon & Interlagos with Sequoia, Founders Fund, General Catalyst, Andreessen Horowitz, Founders Fund, Dylan Field, CIV, 8VC, The Friedkin Group, JAWS, Sozo Ventures, and Alumni Ventures…

Let’s get to it.

Senra Systems: Harnessing Human Skill

A Co-Written Essay with Senra Systems CEO Jordan Black

One of the most notable things about Senra Systems is that so many people who don’t have to work another day in their lives are choosing to work there. What makes that even stranger, to the casual observer, is what they’re choosing to work on: wire harnesses.

Some of you will know what a wire harness is. It is the nervous system of a machine, hundreds of wires, bundled and crimped by hand, routing power and signal to every component that needs to turn on in any modern electrical system, from a satellite to a coffee maker to your car. But many of you won’t, which is fine, and even many of those of you who do won’t understand what makes them important or challenging, and an even smaller number of you will understand why so many people who have made tens of millions of dollars, or more, building reusable rockets in order to make life multiplanetary have decided to dedicate their next chapter to building them instead of angel investing from a beach.

One difference between them and the rest of us is that they have been bottlenecked by wire harnesses. They’ve done the ridiculously hard work of, say, building chopsticks that can catch a Statue of Liberty-sized rocket as it stumbles back to Earth, only to have to wait for the wires to turn the whole thing on. They’ve undoubtedly attempted to just do it themselves, because that’s their whole thing, only to find out firsthand what makes it so tricky to do well.

So when two of their own went and started a wire harness company, they’d seen enough to look a little more, and the more they peeled, the more they liked.

In the US alone, annual spend on wire harnessing for just aerospace and defense is something like $25-50 billion. Maybe the money itself isn’t that important anymore, but it says that the problem is big. It’s also getting bigger. This is one of those classic stories of an industry whose demand is ripping just as the old skilled hands who were trained during America’s manufacturing golden age are aging out. So the problem is big, and it’s urgent. And the answer is obvious, on the surface: automate!

And that’s where Senra really gets ya, if you’re the kind of person who likes solving really hard problems, because automating skilled assembly is probably the hardest thing you can do in manufacturing, especially when you’re talking something that’s high mix, low volume, flexible, and very complex, like aerospace and defense wire harnessing.

If you view America’s ability to manufacture as one big engineering challenge, and your goal, ultimately, is to automate as much of it as you can, you arrive pretty quickly at skilled assembly, manufacturing work that requires trained human judgment and dexterity instead of simple repetitive motion, as the place the machine gets stuck. Within skilled assembly, wire harnesses are perhaps the trickiest to get right, or at least the trickiest that we need a lot of, the trickiest without which nothing else turns on. So if you’ve been trained to heatseek bottlenecks at the system level and throw everything you’ve got at them until they’re knocked out and you can move on to the next one, then wire harnessing is the logical place to start.

And it turns out you can’t automate it, yet. In Elon’s five-step design algorithm - make requirements less dumb, delete the part or process, simplify or optimize the design, accelerate cycle time, automate - automation comes last. You can’t automate until you can configure.

What you need to do with skilled assembly generally and wire harnessing in particular, and what Senra is doing, is to make everything around the skilled assembler better, so that the skilled assembler can deliver wire harnesses at 4x the speed of traditional vendors while maintaining exceptional quality and accuracy.

Senra gives skilled workers superpowers and trains up tons of them so that they can make wire harnesses faster and with fewer defects so that everyone else can make whatever it is that they make faster and cheaper so that America might be able to manufacture competitively again.

We’d like to tell you a little bit about how, and why.

Why, er, harnesses?

Wiring harnesses are like referees - most people only notice them when they mess up.

President Kennedy promised that America would go to the Moon and that we would be first, not “first but, first and, first if, but first period.” The Apollo Program was the vehicle through which this promise would be made real over eleven years, through the toil of 400,000 engineers and workers, and roughly $25 billion (nearly $200 billion in today’s dollars) of concentrated national will directed at a single goal.

What we remember is that America posthumously proved him right, when Apollo 11 astronauts Neil Armstrong and Buzz Aldrin walked on the moon as Michael Collins orbited above. What we forget is that the very first Apollo mission, Apollo 1, ended in tragedy.

In January 1967, three NASA astronauts, Gus Grissom, Ed White, and Roger Chaffee climbed into the Apollo 1 command module for what was supposed to be a routine pre-launch dress rehearsal on the launchpad at Cape Canaveral.

As the three men sat in the oxygen-rich capsule, behind a hatch that was designed to open inwards in a process that took 90 seconds, a fire broke out. It killed them within a minute. Investigators later determined the fire was sparked by a wiring fault. The review board cited poor workmanship and vulnerable wiring design as underlying causes. It was a bad wire harness that took the lives of those three astronauts.

That was a long time ago. Surely we’ve gotten better at bundling a bunch of wires together since?

On June 9th, Jeep recalled more than 1 million Wranglers and Gladiators, warning that they may catch on fire and advising owners to park them outside and away from other vehicles or structures.

The issue, as you might have guessed, was a wiring harness, specifically an electrical connection issue in the power-steering-pump wiring harness, something like this:

Drivers are advised to “inspect and, if necessary, repair or replace the wiring harness.”

Wiring harnesses can kill people when done wrong, but even in business-as-usual situations, they can kill production timelines.

Few people in the world know this better than Jordan Black.

Since working as a host at a burger restaurant at 16, Jordan has mixed service and engineering. He’s manned pizza joints, barbacked, bounced, and bartended, and even worked as a librarian. In each, he was obsessed with making customers happy and in each, he learned how to work with people. He loved solving hard physical problems, too. Back in his freshman year of college, high school, Jordan began working as a technician fixing roller coasters at the Santa Monica Pier. He parlayed that experience into internships at Enerpac and then The Ford Motor Company, which is where he expected to return after graduating from University of Wisconsin-Madison until SpaceX called and hired him as an Avionics Manufacturing Development Engineer. What that meant in practice was a lot of wire harnesses, wire harnesses for Starship and Dragon Spacesuits and Radio Frequency Cables and more.

Over the next five years at SpaceX, Jordan worked his way into the center of advanced hardware manufacturing, eventually running an Avionics R&D team with visibility into nearly every component that went into rockets and satellites. He saw firsthand how SpaceX would bring capital-intensive manufacturing capabilities in-house to increase iteration speed. He was often the one doing it. But highly skilled assembly work, the craftsmanship layer underlying nearly every advanced hardware system, remained dependent on fragmented suppliers operating with outdated systems and limited scalability, even at SpaceX.

A wire harness is in everything, nothing turns on without them, and everybody hates them, for two reasons.

One, to design it, it’s all in Excel spreadsheets and PowerPoint slides.

Two, to manufacture it, it’s all done by hand, arts and crafts style.

“It’s a really screwed up supply chain from start to finish.”

Because the wire harness supply chain is screwed up, everything else is. The wire harness is one of the last designed but first needed parts of any modern aerospace and defense system. You can’t know where your wire harnessing should go in a missile until you’ve figured out how you’re going to build the entire missile. You don’t know how long the missile is going to be, where the sensors are going to go, what needs to connect to what, where, until you’ve actually figured all of that out. Then and only then can you design the wire harnessing that connects it all. But then you have to design and produce custom wire harnessing as quickly as possible, because you won’t be able to turn the thing on in order to demonstrate it to the customer until you have it.

The wire harness is kind of like the fartlek of manufacturing; when it’s your turn, you have to sprint from the back of the back to the front as fast as you can. “You can get as far ahead on everything else as possible,” Jordan puts it, “but you can’t make a grilled cheese sandwich without the cheese.” (Wire harnesses are both cheese and fartlek here, for those keeping metaphor score at home.)

Sourcing high-quality harnesses is harder than you’d expect, too, because it sounds so simple. “Just give me a cable that plugs this thing into this thing.” But some of these designs end up being really complex - you need a cable that has 45 different offshoots, and they all need to be built to spec, so they’re highly custom products.

And for customers like SpaceX they’re really high mix, which makes them almost impossible to automate. One SpaceX wire harness will look totally different from this other SpaceX wire harness, and they might need five of this one kind but 100 of another kind, and so maybe it’s not surprising when you read it, but it was surprising to Jordan when he just needed wire harnesses pronto, that smaller wire harness shops would quote four to six month lead times on new programs.

Jordan learned this the hard way, going around to suppliers when he needed harnesses to make whatever he was working on work, and they’d come back with “Oh, cool, we can do it in four months,” but he’d need it way faster than that, because think about it; as an engineer, whether you’re building rockets or stringing a Christmas tree with lights, the first thing you actually need to do is to make sure that when you plug it in, the thing turns on.

A longstanding theme of Not Boring is that if a company is a bad customer or supplier, they might make a very good competitor.

So Jordan decided to leave SpaceX and start a wire harnessing company, teaming up with Ben Shanahan, a neuroscientist turned SpaceX software engineer who worked on WarpDrive, the company’s ERP / manufacturing execution system (MES), the software that runs the factory that is the product.

In March 2023, they founded Senra Systems.

Senra’s Strategy

In its early years, Senra has focused primarily on manufacturing wire harnesses for aerospace and defense (A&D) customers.

A&D is the highest-margin segment, the most fragmented slice of the market, and it’s ITAR-protected, meaning that customers are effectively required to source American. It is also the most challenging part of the market, and the least amenable to automation in the near-term, given the high mix and low volume compared to, say, a Toyota Camry. It is where America’s skilled workforce crunch is felt most acutely, because the work can neither be automated nor outsourced.

Put simply, the American wire harness workforce is on the decline while the demand for wire harnesses for A&D is taking off. There are fewer technicians to make more harnesses.

There is no government statistical category for “wire harness technician,” but the closest public workforce category - electrical, electronic, and electromechanical assemblers - has a median age of 47.2. More than a quarter of the workforce is 55 or older, approaching retirement age, while only 8% is under 25, stepping in to backfill the roles. Even if there were more people excited about taking the roles you still wouldn’t be able to scale up in time because that training period takes 18 months. In a typical shop, you’ll have a year and a month apprenticeship, and it’s like, “There’s this guy Doug, he’s been building Boeing harnesses for 25 years. Just go follow him around and see if you can learn something.” The industry isn’t training replacements at anything close to the rate at which it is losing experienced hands. The IPC, the electronics manufacturing trade association, calls the shortage of adequately skilled workers “chronic.”

Even if demand were flat, that would be a problem. Demand, however, is not flat. America sent 3,708 objects into space in 2025, up from 2,261 in 2024, and those numbers should look tiny in a decade.

Meanwhile, the fiscal year 2026 national defense budget request crossed $1 trillion for the first time, including $153.3 billion for procurement and $142 billion for research & development. And the systems that the DoW will be purchasing and developing will require a lot of wire harnesses. As we wrote in Electromagnetism Secretly Runs the World, “Electronics comprise 35% of the cost of the F35 Lightning II, more than the cost of the engine itself, and 15% of the Pratt & Whitney F135 engine, which costs $20 million. By the 2030s, when it’s projected that defense contractors will be building the F-47, they’ll be spending over 40% of the $300 million airframe on electronics.” The more electronic our vehicles and weapons go, the more wire harnesses they’ll need.

The government knows its suppliers are not prepared. In a 2024 review of the commercial aviation supply chain, the Government Accountability Office (GAO) found that 15 of the 17 manufacturers it interviewed had struggled to hire enough skilled workers to meet demand. Nine of 15 suppliers said they had also struggled to fill orders as demand rebounded. There’s no reason to believe this is any better for the Defense sides of these Aerospace businesses.

The roughly 800 American harness shops that make up the supply base are largely small, owner-operated businesses. Their ability to grow is constrained by the amount of veterans they have to train apprentices, but the veterans are retiring, training takes years, they’re swamped with the growing demand, and apprentices aren’t signing up anyway. Worse, when they retire, their knowledge is gone with them, because they aren’t spending precious time writing it down.

So as much as Senra is manufacturing wire harnesses, its more important role may be “manufacturing” wire harness technicians to make all of the wire harnesses and increasingly electronic world needs.

This supply/demand imbalance also creates the market foothold for Senra. Because declining supply is buckling under growing demand, A&D is also where there is the most room for improvement: lead times on a new program are 4-6 months, only 25-50% of wire harnesses pass on the first try (first-pass yield), and quality is so inconsistent that many customers have to employ in-house technicians whose job it is to fix the wiring they receive from their wiring suppliers.

All of those characteristics make wire harnessing for A&D particularly amenable to Senra’s approach, which is to get configuration right long before worrying about automation.

Automation can be thought of as replacing humans with machines. Configuration is creating a system with any combination of humans, machines, software, materials, and processes to turn a design into a product. It is the layer above automation, and counts automation as one of its tools.

You can’t automate until you can configure, and long before getting to automation, there’s a lot that you can do to improve wire harness manufacturing by redesigning the system.

In a traditional shop, practically everything runs on tribal knowledge locked in veterans’ heads. This is the same challenge that Hadrian identified in precision machine factories; to fix the problem, it’s “turning tribal knowledge into scalable software and processes.” It also rhymes with what Poetic does; instead of just magically throwing AI at a problem, “We need people to get out there into Minnesota to be like, what the hell do you guys do all day?”

Relying on veterans’ head-knowledge works decently well as long as it’s the veterans doing the work, but people-as-operating-system doesn’t scale. Take training. In a typical shop, you’ll have an 18-month apprenticeship, and it’s like, “There’s this guy Doug, he’s been building Boeing harnesses for 25 years. Just go follow him around and see if you can learn something.” And that’s really it. There’s no trade school for skilled assembly work.

Jordan compares wire harnessing today to a Cheesecake Factory recipe where skilled workers can just look at the menu and kind of figure out how to make everything that’s on there. The hope is that other metaphorical chefs will watch them do that enough that they, too, can look at the menu and reverse engineer the recipe.

That doesn’t work that well now, let alone when the skilled vets retire, or when demand continues to grow and manufacturers need to bring on a lot more people to meet it.

So you have this situation where the incumbents actually do a pretty poor job…

… that is not fully automatable yet, but where there is an absolute ton of low-hanging fruit for a player with scale ambitions to make the investments into getting the configuration right.

Which, at its simplest, is what Senra does.

You know what, I’m just going to hand the mic to Jordan for the rest of this piece to explain.

Standardize → Optimize → Automate

The way we see it, you need to standardize, then optimize, then automate, in that order.

What Senra is doing is taking that Cheesecake Factory menu item, then reverse engineering, in software, what the recipe is, then training cooks over a few weeks to be able to cook to that recipe, reliably. That’s standardization.

To end the food metaphor, what Senra does today is take a wire harness design and turn it into work instructions, which break down the component parts of building a wire harness and serve it up on a silver platter to a technician.

Get everything as step-by-step as possible. This is how you cut the wire for this specific harness, this is how you strip it, this is how far back you strip the rubber that encloses the metal, this is how you connect it to this connector, and this is how you lay everything out on the form board. Check check check check check, until you have a working wire harness.

Today, Senra takes in a wire harness drawing from a customer, and then have a team of engineers look at the drawing and convert it to work instructions. Right now, the process is like 20% automated, which is fine, because even having work instructions is a step-up. Over time, it can use those human-crafted work instructions to train models that can get to somewhere like 90% of the way from drawing to work instructions, with an engineer doing the final check-off, which would speed up the process and increase accuracy. This is optimization.

Even before then, though, there’s a lot that you can do with really good work instructions, each of which feeds on the others.

To start, you can train people up much more quickly and effectively. The question is how you can grab someone who has never seen a wire harness, certify them through Senra’s apprentice program in a month instead of a year or two, and get them to work meeting customer demand. The key is that because workers follow a recipe for each harness, they don’t need to keep all that tacit knowledge in their head.

Senra hires people who have never touched a wire harness, and after the four-week program, they’re building harnesses on the production floor for SpaceX and Anduril.

Those same explicit steps mean we know which parts are needed where when, so we can optimize our supply chain and inventory. It would surprise you how big a source of error just giving technicians the wrong components is. With work instructions in software, we can make sure that the instructions on how to kit components for each job, including what goes in the kit, are accurate. All of that can tie back to our inventory, to purchase orders. Then our inventory tracking system goes from being a kind of crazy startup system to fully optimized and ready to scale.

If inventory runs smoothly, then it becomes even easier to get someone trained up and productive more quickly, because they have everything they need at their fingertips. The machine runs more smoothly through a bunch of mutually reinforcing improvements.

We are basically creating a digital twin of the manufacturing process, capturing and digitizing data at every point so that we can optimize each and every part. Then we build custom machines that integrate with our software.

For something like cutting and stripping a wire, since our software knows what’s in each kit and what it’s being used for, our technician can scan a barcode from the machine and input the wire, and the machines cut and strip exactly to the specification we need.

The line between optimization and automation gets a little blurry here.

Something like cutting and stripping is relatively easy to build a machine for; that machine is a tool for the human technician, who drives the process.

Crimping – taking a piece of metal, putting a wire in it, and squeezing down with a tool - we can make a machine for that next. Something like connector insertion is such a high-dexterity operation that we probably won’t automate it for a very long time.

Our intent is to train our technicians, optimize everything around them, give them custom machines, give them better tools, and remove any of the non-value-add work from their plates.

We think we can make our technicians 3x more productive than industry average simply by optimizing everything else around them. We care a lot more about that than we do about full lights-out automation.

Meanwhile, again because we can track the process step-by-step, we embed quality control throughout the process, so we catch things when they go wrong. Inspection during the process is something we’ve already begun to automate, and thanks to a combination of everything we just described and that automated inspection, we’re at a 99% first-pass yield, versus 75% for our competitors.

And we’re tracking and collecting data at every part in the process, from inventory to quality control, in our Amp platform, which is like our operating system, Senra’s version of SpaceX’s WarpDrive. It spans the full thread: Design (ITAR-compliant cloud harness design tool) → Quoting & BOM (AI reads messy aerospace RFQs) → Procurement & Warehousing (kitting) → Manufacturing Planning (work-instruction generation) → Build (a 100% digital thread for every crimp, connection, and serial) → Quality (real-time enforcement + traceability).

The operating system, training environment, data infrastructure, and years’ worth of data give our workers superpowers, and the results have been strong. We are higher-quality, higher-reliability, and faster-turnaround than the incumbents. We can deliver in four weeks something better than they’d be able to turn around in four months. And we can bid on jobs at a 20% margin to win the business with the comfort that we’ll be able to optimize the process to well north of 50% over the life of the contract. And of course, the more we do, the more we learn, the faster we optimize, and the better we get.

In the near-term, our focus is on scaling wire harness capacity while improving efficiency and maintaining quality in order to meet the tremendous demand that is currently underserved.

For a sense of scale, individual customers of ours are doing hundreds of millions of dollars a year in wire harnesses, and they are impatient for us to expand capacity to take on more of that business. Our customers don’t like having to employ people to fix their wire harness suppliers’ mistakes, nor do they like waiting four months to turn on their new product.

If we can continue to execute, we can continue to grow our share of the market even as the market expands, and we hope that by offering a better product, faster, to our customers, we’ll be able to speed up American manufacturing on the whole. Our job is to eliminate this bottleneck for the industry.

For now, the fact that this is really hard to do well is our moat, but as we scale, we dig traditional 7 Powers moats in the form of Scale Economies and Process Power. We should be able to offer a better, faster product than the fragmented incumbents, or than any would-be venture-backed challenger if we continue to execute.

More than that, we want to be the company our customers turn to whenever they have any wire harness needs because we know how to help them from the very beginning, the same way that the expert at Blockbuster could point you to the right horror movie or classic Western or “I’m going through a breakup and I need a great romcom,” the same way that I could Norm from Cheers a patron with a fresh drink and open ear when I was a bartender. The practical way we do this is that we’re on contract with our customers to provide design and engineering feedback on their harnesses, even before they get to the manufacturing stage.

All of that said, this opportunity is way bigger than wire harnesses for aerospace and defense.

Optimizing All Skilled Assembly

We started in wire harnesses for A&D because it’s what we know best, and because it’s the most challenging, highest-willingness-to-pay portion of the skilled assembly market. At $25-50 billion annually, it is a big enough market to keep us busy for a while.

But we want to fix all of skilled assembly. That’s why we’ve been able to attract the team we have, to solve the biggest challenge in manufacturing. As we get good at the hardest part, adjacencies are already opening up. In the past month alone, we’ve been hit up by companies making chillers for data centers and mail trucks for the US Postal Service, and a lot in between. Both the AI data center buildout and the electrification of everything means that there’s going to be a lot more wiring everywhere.

The fun part, and the part that we debate internally, is that expanding into automotive and industrials requires a different skillset than A&D. One satellite company inside of one Prime builds something like 8,000 unique wire harnesses every year. That electric truck company has just 25 that they put on every single truck, over and over again. This would move us into low-mix, high-volume, which can be more (although not fully) automated today. Doing this work wouldn’t build the same hard-to-replicated skilled workforce bench that high-mix A&D work does, but if we can do it at a positive (and expanding) margin, it allows us to get to scale more quickly and establish a foothold on the commercial side.

It’s a matter of when, not if, because we want to own as much of the $200 billion total wire harness market as we can eventually. We want to be in the data centers, the trucks, the nuclear reactors, the aircrafts… in everything, because wire harnessing is in everything, and we think we can offer a better wire harnessing product across the stack.

Moving into lower-mix, higher-volume wire harnessing also serves as a bridge to the rest of the skilled assembly market. Things like sensors, electromechanical assembly and testing are skilled assembly jobs adjacent to what we do now, and demanded by the same customers. If we do a great job on wire harnesses for our customers, we can say, “Oh you have a problem with sensors, too? Let us give it a shot.”

As we figure out how to standardize, optimize, and automate as much of wire harness manufacturing as possible, everything else becomes much easier. Because we tinker at the configuration level, we can copy-and-paste a lot of the processes and software from wire harnessing into sensors and electromechanical assembly, and tweak specifics like which machines and components we need. We built Amp to generalize; it’s a design and manufacturing software platform, not a wire harness-only tool.

Wire harness manufacturing is really a wedge into our broader ambition: how do you make the products of skilled assembly available more like Amazon than like manufacturing?

You should be able to go to our website, design your harness or your sensor or whatever you need, click buy, and get it two weeks later.

A lot needs to happen for that vision to come to life. We probably need to pony up for senra.com, for one thing. We need to offer better design software; the stuff people use today sucks. It’s really fragmented and not standardized, and there’s an opportunity for us to do it better, and turnaround outputs faster when customers design to our standards. We need to standardize and optimize across a growing number of products, and turn designs into work instructions and kits of parts smoothly. We need our people to be able to turn those work instructions into anything quickly and at high-quality. We need to inspect it all in real-time to maintain quality. We need to make new machines when it makes the work better, and look for opportunities to automate once we’ve standardized and optimized all we can.

Ultimately, we need to build up massive capacity so customers can get whatever they need really, really quickly, so that they can build, iterate, and scale their main products really, really quickly.

It’s that opportunity to unblock American manufacturing that’s why so many people who don’t need to work anymore are choosing to work at Senra.

Gathering the Grinders

In the past month, we’ve made two ex-SpaceX hires that we’re really excited about.

Ken Venner was the CIO at SpaceX, where he architected WarpDrive, which we’ve modeled Amp after. He’s a legend. He’s joining us as CTPO (Chief Technology and Product Officer) because, as he told us during the process, “This is the most excited I’ve been about a startup since SpaceX.”

Jessica Chavarria started her career in contract manufacturing, moved to SpaceX as a Requirements Planner in 2011, and worked her way up to Supply Chain Manager over eight years at the company. She has worked increasingly senior supply chain roles at a number of A&D companies, and just joined us in June to go back to her contract manufacturing roots as our Head of Supply Chain.

We’ve noticed something with the SpaceX alumnae. There’s a group that will make like $3-5 million and take it easy. Then there are the people who have been there long enough, and in important enough roles, to make hundreds of millions of dollars who have negative desire to retire and sit on a beach sipping pina coladas. These are people who worked for over a decade at a crazy company and loved the rush. They want to do it all over again. You can’t take the grinder out of them.

We’ve created a place where the grinders can grind on a challenge that’s endlessly deep and deeply important. We see it as validation of the opportunity that the people who’ve been at SpaceX long enough to know where all the bodies are buried are saying “If we could solve this, this is the most important thing we could work on.” Now it’s up to all of us to actually go solve it.

If you’re reading this, you’re probably bored of hearing that the US forgot how to build. That we sent the jobs overseas in search of profits, and hollowed out the golden goose. That you’ve heard it a lot doesn’t make it any less true, but the interesting question isn’t what happened in the past, but what we’re going to do to fix it.

We think that skilled assembly is the fulcrum point. It’s a really fucking big problem waiting to happen; you can’t necessarily feel it today, but you’ll feel it in the next two to five years. Our skilled workforce is retiring, just as the demand for their skills is surging. We’ve realized that we need to build big things here again, but all of those big things are made up of a lot of little, complex, custom things that a shrinking group of people know how to make.

A lazy solution to this coming crisis is that we’ll just let the robots take care of it. As you understand by now, most of our industrial base’s skilled assembly isn’t standardized or optimized, and until you standardize and optimize, you can’t automate.

When the time comes - when the brains are smart enough and the hands are dexterous enough - we believe that robotics adoption in manufacturing will be constrained less by hardware and more by the absence of structured, well-configured operational environments. We’re solving that problem first, by encoding manufacturing into executable, measurable, and machine-readable systems.

Far before automating skilled assembly becomes possible, we will reindustrialize by giving humans better work environments, environments that work for them. That’s little things – our attrition is comically low because we offer air conditioning and a pleasant workplace - and big things, like making sure they have the tools, components, and instructions they need at all times.

All of this is really hard. It’s not as clean or as sci-fi as throwing robots and ASI at the problem. But we love that. We’ve been out here screaming into the ether: humans matter, we need to figure out how to scale them.

Ours is a belief that the more the easy stuff gets automated, the more valuable skilled humans will become. We agree with ex-Uber-now-Atoms founder Travis Kalanick, who predicted on TBPN that skilled workers will get paid like LeBron James because, “Humans [are going to] become more and more valuable because they will be the long pole in the tent to progress - and that progress is going to accelerate and get faster and more robust.”

The future, like the past, is human. It’s just that we can give humans superpowers now.

If you’re out there and you hear our scream, come join us.

Thanks to Jordan, Greg, and Sam for working with me on this piece, and to Alexa for the introduction!

That’s all for today. We’ll be back in your inbox tomorrow with our Weekly Dose.

Thanks for reading,

Packy

Hi friends 👋,

Happy Friday and welcome back to our 201st Weekly Dose of Optimism.

We’re jam packed again, and you don’t come here for the intro, so…

Let’s get to it.

Today’s Weekly Dose is brought to you by… Mercury

Most founders understand their finances in spreadsheets, and that’s a problem.

Your bank has the data. But to actually use it — check your cash position, reconcile transactions, understand your burn — you export it, paste it somewhere else, and do the work outside your account.

Mercury Command closes that loop. It’s AI built directly into Mercury* that surfaces insights and takes action from your live account data. Ask what you need to know, then act on it instantly — follow up on an outstanding invoice, set a limit on a card, categorize a transaction — all in the same conversation. You approve every step. Command executes.

Try Mercury Command →

*Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column N.A., Members FDIC.

(1) Aalo Goes Critical on the 4th of July

On July 4th at 12:20am, Aalo Atomics went critical, becoming the fourth advanced nuclear company to accomplish the mission by the July 4th, 2026 deadline, and the third whose founder we had on Age of Miracles back in 2023. I actually wrote a small personal check into Aalo before the podcast, back when I didn’t think nuclear was fund investable but that it was incredibly important. I was right on the latter but very wrong on the former.

Aalo joins Antares Nuclear, Valar Atomics, and Deployable Energy in hitting the target, with a number of others including Radiant, Oklo, X-Energy, Deep Fission, and Last Energy expected to hit criticality later this year or next, and deploy commercially in the months and couple years ahead. As Abundance Institute’s Chris Koopman put it, “In just over 13 months, the United States of America went from a handful of executive orders to FOUR DIFFERENT advanced nuclear reactors reaching criticality.”

This is an insane state of affairs given where nuclear was just three years ago. Getting any one advanced reactor to criticality by July 4, 2026 would have been considered a minor miracle, but getting four with more to come means that there’s going to be competition. Whoever can deploy most quickly, with a product that customers want, at the best price, will win. The field shifts from science to manufacturing. I think that good ol’ competition is what it’s going to take to make nuclear cheap and abundant.

One of nuclear’s challenges has been that utilities are terrible customers for the product. Fortunately, they are not the main customers for these smaller advanced reactors. Data Centers are, and they will fund whoever gets them something that works fast (and safe, cheap, clean, and eventually affordable).

(2) Data centres are a crucial test of US industrial resolve

Josh Zoffer for The Financial Times

This is the argument that I made in Thank God for Data Centers, and part of the argument that my friend Josh Zoffer made in the Financial Times this week:

The data centre bonanza is a special kind of opportunity: one where deep-pocketed buyers like Silicon Valley’s hyperscalers are willing to open their wallets to pay for fast, reliable access to industrial technology. It offers a chance for the US to get ahead in the next key technologies and to build domestic supply chains based on demand rather than subsidies and tariffs.

I’m glad to see Josh making the case for U.S. data centers (we use er here, FT) in front of the policy and business leaders who read the FT, and making it so eloquently. Data centers are useful in their own right, but they’re also a turbocharged way to turn on that big, beautiful American Demand engine and fund the technology that will carry us into the Age of Miracles.

(3) American Turbine Comes Out of Stealth

John McElhone

Including gas turbines. Lots and lots of gas turbines. If you can make a gas turbine, or retrofit a jet engine to work as a gas turbine, the data centers would like to buy it from you. And this week, American Turbine came out of stealth to sell it to them.

What’s interesting about AT’s approach is that instead of very large, complex turbines that take a long time and a lot of scarce expertise to make, they’re making small, highly manufacturable turbines that they can get to customers quickly. If you need a lot of MW, they’ll just sell you a lot of turbines.

For now, the goal is speed, and buyers are happy to pay for speed in lost efficiency. Over time, though, whether data center demand slows or the supply chain gets worked out and time-to-power is no longer as big an issue, turbines will still have a ton of value.

The beautiful thing about turbines in the American context is that America has bountiful natural gas resources. We have an advantage there, if we have the turbines to turn it into power.

Part of American competitiveness in the century ahead will be beating China at things China currently dominates, but an equally important part will be winning things that we can do uniquely well. Turning natural gas into power to turn ideas into economic activity is one of those things.

(4) Vigils Uses Colors That Are Grown, Not Applied

Neri Oxman for OXMAN

Not everything in the good future will be manufactured or petrochemicals-based, though. Some of it will be grown naturally.

Neri Oxman, the ex-MIT designer, engineer, scientist, and artist working on computational design, synthetic biology and digital fabrication whose 2023 Lex Fridman appearance was one of my favorites, announced a new line of capes from her company, OXMAN, called Vigils, “whose color is not applied but grown. We engineer bacteria to produce indigo and melanin pigments, then let them work directly on the surface of a 3D-knitted silk textile. When the process is complete, the cells are washed away, leaving only the pigment they have grown into the fiber.”

“For a century,” Oxman (the person, not the company) tweeted,

One day, OXMAN might even apply this technology to garments people actually wear.

(5) 1X Gives Neo Alien Hands

Giving robots hands that work like human hands is one of the hardest challenges in robotics. If these videos are real, it looks like 1X may have solved it.

Yesterday, they announced their “25 Degree of Freedom (DOF), tendon-driven hands for the NEO humanoid platform– achieving near human-level dexterity, strength, safety, and reliability.”

Watch the video. Or this one.

1X argues that with these hands, the hardware problem is basically solved and what’s left is gathering the data that drives the robots capabilities. It’s a bit of a self-serving argument, because 1X is selling and deploying tele-operated robots into homes in order to collect data, but any good company argument should be self-serving. And that is, once again, an absurdly impressive hand.

One thing I particularly like is that they call the hand an “API to the Physical World,” which is a catchy way to express that our hands are how we interact with the world. Our legs can walk us up to a door, but our hands push it open. Our eyes can locate a glass in space, but our hand picks it up.

The new 1X hand can pick up a wine glass with ease. It can do LEGOs. It can even screw in a lightbulb. If you want to learn more about what it does, and how it works, check out their blog post.

Over the past couple of days, OpenAI released a new voice model that talks more like a human and 1X released hands that are more human, and in both cases, unexpectedly, it feels that the more human-like features we give the machine, the clearer it is that they are not becoming human but will be very useful to us.

Long humans.

EXTRA DOSES: Science Breakthroughs, Aleph Speech, Levin, Quantum

I know you’re probably overdosed on optimism at this point, but I do think the Science Breakthroughs are worth reading as an Ulkar-vetted glimpse into the future. This is the stuff that’s not on X yet. Check ‘em out.

Read more

Hi friends 👋,

Happy Friday and welcome to our 200th Weekly Dose of Optimism.

A full 200 weeks after Dan and I started writing the Weekly Dose at the depth of the despair market in the July 2022, the optimism is at an all-time high around these parts. We write this 200th Dose to you on the eve of America’s 250th birthday, just two days after the USMNT beat Bosnia and Herzegovina, a man down, to advance to the round of 16 and the Sixers (finally) fleeced the Celtics in a trade. Welcome to Philly Jaylen Brown. Come join him Lebron. America feels like it’s on fire, because it’s way too hot outside for one, but more importantly because the energy around the World Cup has been exquisite, we’ve played the role of host to the world well, the markets are soaring (if scary), and everyone’s a little happier in the summer.

Since we started the Dose four years ago, obesity has been cured, cars have begun driving themselves commercially, AI has become very smart and useful, there has been a renaissance of atoms companies in America, and multiple new advanced nuclear reactors have gone critical. That was just a quick list off the dome; if you’ve been following along, you know that practically every week is that jam-packed.

And this week is no different, so…

Let’s get to it.

Today’s Not Boring is brought to you by… Framer

Launch fast. Design beautifully. Build your company’s website on Framer.

Framer helps teams design, build, and launch their marketing sites lightning fast. With the ability to publish hundreds of CMS pages in a single click, operate at a global scale with seamless localization, and even host unified content across multiple domains, teams have never been able to ship faster. Companies like Miro, Bilt, and Perplexity trust Framer to achieve:

• Speed without chaos: ship pages and updates faster without turning the site into a fragile set of one-off hacks.

• Reduced dependency: shift routine brand and marketing work out of product engineering queues.

• Production-grade foundation: Run real marketing systems (CMS, SEO, performance optimization) with governance and collaboration.

👉 Build your company’s site on Framer today

(1) Conception Bio Generates First Early Human Eggs from Stem Cells

Imagine telling George Washington that in four generations, scientists in the nation he founded would turn blood cells into egg cells, which might become human.

The team at Conception, a startup in Berkeley, is doing just that. The company says it has grown the first early human egg cells, called primary oocytes, from stem cells. To do it, they drew blood, reprogrammed the blood cells into induced pluripotent stem cells, then coaxed those into miniature human ovaries that in turn grow the eggs inside them. Their early eggs went through meiosis, and support cells wrapped each one in a follicle, recreating the defining structure of a real developing ovary. The whole thing is absolutely insane.

In vitro gametogenesis has worked in mice since 2016, when Conception’s collaborator Katsuhiko Hayashi made mouse eggs that grew into healthy, fertile pups. Humans have been far harder, which is why this took Matt Krisiloff, an OpenAI founding-team alum who’s talked openly about wanting a child biologically related to both him and a partner, eight years of trying. The urge to reproduce is one of nature’s strongest, and Matt is putting his to work on behalf of all humankind.

The caveats, for now, are that these are early-stage oocytes, not mature eggs ready for IVF, and clinical use is years away. But what it points to is a world in which we can make as many healthy eggs as we need from a single drop of blood. Elizabeth Holmes weeps.

There will be a lot of debate over the ethical use of these eggs, I’m sure, in the coming years. They will allow for two biological fathers, improve the efficacy of embryo selection (more eggs to choose from), extend a woman’s fertile life (and restore it in cancer patients), compress generations of breeding (no gestation period, theoretically), and de-extinct more animals.

To my mind, it passes the “increase the range and depth of experience in the universe” test.

(2) This Cell Feeds, Grows and Reproduces. And It’s Manmade.

Carl Zimmer and Marco Hernandez for The New York Times

Big week for tiny little cells.

Where Conception took human stem cells and turned them into an egg, scientists at the University of Minnesota created the first synthetic cell with a complete life cycle assembled entirely from non-living chemicals. They call it SpudCell.

For as long as we’ve had biology, making a living cell required a living cell. Cells divide, life reproduces sexually and asexually, thanks to Yamanaka, we can turn specialized cells into iPSCs and then into whatever kind of cell we want, etc…

SpudCell is built different, created in a lab and able to do things like “feed, grow, reproduce and compete with one another for food.”

I would try to compress an explanation of how they did it, but honestly, the NYT does an excellent and visual job and you should just read their version or the 190-pager the lab put out.

Kate Adamala, the lead researcher behind SpudCell, argues that the core functions of life “do not need some magical spark.” With SpudCell, we may be able to begin engineering life from scratch, instead of starting with an E.coli.

Drew Endy, whose belief that “Atoms are Local” and humans should be able to grow anything Elliot has written about a bunch, is teaming up with Adamela on a non-profit research organization, Biotic, that will open source and develop the work with hundreds of scientists. To the New York Times, Endy compared “SpudCells to a biological version of the Wright flyer, the crude plane that the Wright Brothers used to make the first sustained controlled flight in 1903, ushering in the age of airplanes.” Let’s fly.

(3) Meta Releases v2 of Its Mind-Reading Model

Relative to the other two, this is pretty ho-hum, but Meta just released a new version of its model that can read minds from outside the brain and translate brain activity into words. Whatever. Yawn.

(4) Nuclear Race Update

There’s so much happening in nuclear around the July 4th criticality target date that we’re giving it a whole section with the highlights:

• Radiant receives 100 pounds of (not boring capital portfolio company) Standard Nuclear’s medium enriched uranium TRISO fuel for its first reactor

• Aalo loads fuel bundle into its first reactors and begins splitting atoms pre-criticality

• Valar Atomics powers NVIDIA Spark with nuclear power

• Deployable Energy achieves criticality in Unity demonstration reactor

• It looks like Aalo is going to attempt to go critical today.

And it’s only Friday morning… let’s see what nuclear fireworks this weekend brings.

(5) America’s Next 250

By Scott Nolan

I don’t normally include the week’s essay as one of the five main entries in the Dose, but I can’t think of a better way to head into this monumental 4th of July Weekend than by reading Scott’s on how much progress America has made in its first 250 years and what it will take to blow that out of the water in the next 250.

EXTRA DOSES: Science Breakthroughs, RocketLab, Etched

Read more

Subscribe now

Hi friends 👋,

Happy Thursday, Happy 4th of July, and Happy 250th Birthday, America 🇺🇲

The United States Men’s National Team is advancing to the Sweet 16 after beating Bosnia in what is now called Soccer. The sun is shining incredibly hard on America. We are just two days away from the Fourth of July. And this particular 4th of July marks the 250th birthday of the United States of America.

I figured we should do something special to celebrate.

In May, I had the chance to moderate a panel with Oklo’s Jake DeWitte and General Matter’s Scott Nolan at Utah’s Operation Gigawatt Summit. After getting into the nitty gritty about what it would take to scale nuclear now that it’s both regulatorily possible and popular, I asked a final question: “What do you want the next 250 years in America to look like?”

I really liked Scott’s answer, which looked back on everything we’d achieved and forward to everything we might… if we get a few important things right in the present.

Scott is someone with a front row seat to America’s potential. He was a very early engineer at SpaceX, has spent over a decade as a partner at Founders Fund, and recently launched General Matter to make sure that nuclear energy has the fuel it needs to power America’s growth.

So I asked him to share what he thinks America’s next 250 will look like, and what we need to do to make that good future happen.

Let’s get to it.

America’s Next 250

By Scott Nolan

To imagine what America might look like 250 years from now, a good place to start is what life was like in the colonies that would become the United States of America 250 years ago, on July 4th, 1776.

In 1776, we did not have running water, light bulbs, electricity, or gas stoves in our homes. We had no insulin, no penicillin, and no x-rays. We did not have cars, and we certainly didn’t have planes. We didn’t have trains or even horse-drawn streetcars.

In 1776, there was no anesthesia. Instead, you were held down and fully conscious during surgery. Surgeons didn’t realize that they needed to wash their hands and the average American was most likely to die from an infectious disease. We didn’t understand blood types, and blood transfusions could not be safely performed. Infant and maternal mortality were extraordinarily high.

There was no synthetic fertilizer with which to grow food — we used animal manure — and no refrigeration to store it once harvested. Food supply was limited by season and geography, stretched with salt or smoke or ice.

There was no telegraph. Paul Revere used a horse to warn that the British were coming. Storms arrived with even less warning; there was no weather forecasting. Because there was no real-time coordination across even modest distances, there were no standardized time zones. Every town kept its own time by the sun.

Pause on this one for a second. Today, it is trivial to communicate with anyone anywhere in the world, at any time. We inhabit a shared present. That experience did not exist in 1776. Information could move only as fast as the horse, ship, or person carrying it. A New Yorker could imagine that something was happening in Virginia, but they couldn’t know what. By the time the news arrived, it was history.

We could continue listing things that did not exist then but seem eternal now — the U.S. dollar, modern steel, plastics, stainless steel, concrete skyscrapers — but what’s even more fascinating is what the absence of those things meant for the experience of being human.

When Thomas Jefferson drafted the Declaration of Independence in 1776, he did so with a quill pen on a homemade portable desk. The digital world did not exist. No photographs or recorded sound meant that nobody alive in 1776 could preserve exactly what a person looked or sounded like. A gone person stayed gone in all but memory and art.

Americans alive at the signing of the Declaration of Independence worked not for fulfillment but to prevent their families’ hunger. They lived in a world of wood: downed trees were both America’s dominant energy source, and its main building material. We had very little mastery over the natural world. We were input-takers, not input-makers.

250 years ago, there was a lot we didn’t know, and a lot we didn’t have.

But we had capitalism, safe and bountiful geographical blessings, and waves of intrepid new Americans in search of a better life, and a frontier spirit. Thanks to the constant combination and recombination of essentially just those attributes, the last 250 years have taken us from something pre-industrial to an industrial, digital age.

Two hundred and fifty years seems like a long time, but it’s just three peoples’ lifetimes at current averages. It would have been six lifetimes back then, except the average American’s life expectancy has doubled over the past 250 years.

Even in one generation, a lot has changed. I spent my formative years growing up outside of St. Louis, Missouri in the 1980s. My sense of Americana was formed in Mark Twain’s America. I remember riding my bike down our suburb’s sidewalks, past the BMX half pipes in neighbors’ driveways and crumpled Busch cans along the road. I remember eating cheeseburgers at a McDonald’s that was on a Mississippi riverboat, which felt every bit as cool to a kid as you might imagine.

And I remember floating in our humble above-ground pool, under a tree, exploring TIME magazine’s July 1988 Onward to Mars issue and looking at pictures of the Mars Rover concept, wondering about space.

It’s grounding to think about how much has changed in less than one lifetime, a fraction of 250 years. In many ways, the world is as different between my childhood and today as it was between 1776 and my childhood.

When I was a kid in the 1980s, the Internet was still ARPANET, a military and academic network. Tim Berners-Lee wouldn’t propose the World Wide Web until 1989, and Marc Andreessen wouldn’t popularize the browser until 1993 when I was entering middle school.

Today, 96% of Americans use the internet to connect, work, learn, entertain ourselves, and increasingly, use AI models.

Americans today primarily do these things from our mobile phones, which 98% of Americans now own. All but the least connected Americans carry a device in their pockets, giving them powers Jefferson could only have dreamed of.

The Human Genome Project, the ambitious quest to map the blueprint of life, kicked off in 1990, and was completed thirteen years later, at a cost of $3 billion. Since then, the cost to sequence a human genome has fallen faster than Moore’s Law, from $100 million at the turn of the new millennium a quarter century ago to a few hundred dollars today, with the newest machines pushing $100.

As a result, the genome has gone from a singular scientific monument to a routine source of data. Researchers can sequence millions of people, compare their genetic differences, identify the mutations that cause disease, trace cancers as they evolve, diagnose rare conditions, and design drugs for increasingly precise groups of patients. With impressive frequency, they are using these tools and others, such as CRISPR and CAR-T, to create therapies that fight back against the leading causes of death, like heart disease and cancer.

When I was growing up in the 1980s, America did have electric cars, but they were much closer to golf carts than a modern Tesla. The Comuta-Car was a small lead-acid electric city car produced until 1982 that boasted about 40 miles of range. Excluding the Cybertruck, which I greatly respect, the aesthetics of modern EVs have changed as well.

Now, not only are there millions of Teslas on the road, but they can drive themselves. Motor vehicle crashes remain one of the nation’s largest causes of preventable death, but those, too, may be on the decline. Over the past three years, autonomous vehicles have evolved from a novelty to a reliable part of everyday commutes, from Tesla FSD to Waymo, both about 10x safer per mile than manual driving.

At the founding of America, there were no cars. Now the cars drive themselves, and do it more safely than humans.

We have made a tremendous amount of progress in science and in the digital world in the two and a half centuries since our nation’s founding, much of it since I was born, and more of that in the second half of my life than the first. The past five years alone saw the introduction and widespread adoption of those self-driving cars, large language models like GPT and Claude, and GLP-1s.

American progress has been world historic, and it is accelerating.

Then there are the things that we have not made real progress on, or have even backslid on until very recently.

One of these is nuclear power, where we made practically no progress on nuclear energy in the past quarter century, despite nuclear having fewer emissions than solar and being statistically safer than wind.

In 2001, the United States generated 768 million MWh of nuclear power. In 2023, the United States generated 775 million MWh of nuclear power. That’s less than a 1% increase in two decades.

In fact, on a per capita basis we’ve made barely any progress on energy at all. We fell off the “Henry Adams Curve,” on which America steadily grew usable energy by 7% per year, in the 1970s and never recovered.

Have we just gotten more efficient with the energy we use, making us all richer? Unfortunately, no. If you look at the average personal income on a real-dollar basis post-1970, when there was the oil shock and regulatory and cultural institutions making it harder to build, there hasn’t been real wage growth per capita.

Real wage growth is a measure of our prosperity in the physical world, and we haven’t made progress in our ability to physically sculpt our environment in over half a century. America has lived one-fifth of its young life in what Tyler Cowen calls The Great Stagnation.

That is changing.

America’s 250th Birthday is an opportunity not only to reflect and predict, but to commit to making the progress of the next 250 years dwarf the past 250.

How do we do it?

When asked what form of government the Constitutional Convention had created, Ben Franklin famously quipped, “A republic, if you can keep it.” What he was saying was that the Constitution could create a republic, but it would be up to engaged citizens to work to maintain it.

The same is true for growth, prosperity, and abundance within the republic that we’ve managed to keep for a quarter millennium. The American project is of the people, by the people, and for the people. There is no monarch or dictator promising to provide. If we want the growth, prosperity, and abundance of the next 250 years to exceed those of the last 250, it’s up to every one of us to keep it in the best way we see fit.

For my part, that’s helping support energy abundance, because there is no such thing as a prosperous low-energy country.

I have been fortunate to live out the dream that began while reading TIME in our modest pool. At Cornell, I joined the rocket club and built propulsion systems with my friends. When a young company called SpaceX was looking for interns in 2003, I signed up. I came back full time when I graduated in 2004, just in time to contribute to the original Merlin propulsion systems and Dragon capsule.

One of the things you learn at SpaceX is to identify bottlenecks and to attack them. That applies to both micro engineering topics and macro civilizational issues. It’s been clear for a long time that the bottleneck to civilizational abundance and individual prosperity is energy, and it’s been even clearer that nuclear power is the best way to generate it. It is clean, safe, energy-dense, reliable, and scalable in any geography.

At Founders Fund, where I’ve been a Partner for about 15 years, we searched for transformational nuclear investments for over a decade. In 2023, we invested in Radiant, a microreactor startup founded by ex-SpaceX engineer Doug Bernauer. Doug started Radiant to tackle a bottleneck of his own: if we wanted to power a Mars civilization, we’d need to send nuclear reactors up there. Here on Earth, clusters of stranded energy demand, like remote Alaskan villages and military bases, needed power as well and were willing to pay. Both could be powered by containerized reactors.

Over the past three years, as baseload power demand expanded dramatically, partly driven by AI, and regulation began modernizing, a number of ambitious new nuclear companies like Radiant were founded. With more advanced reactors coming online, the bottleneck in nuclear energy shifted to something that the U.S. had once produced but stopped: enriched uranium. This meant America relied on foreign companies, including from our adversaries, for over 20% of the enriched uranium traditional reactors run on, and 100% of the fuel consumed by advanced reactors. After nearly a year looking for an American company to invest in that would enrich uranium at scale, I realized none existed.

So at the end of 2023, I started a company called General Matter. We are enriching uranium in America to fuel the nuclear reactors that will power America into the 23rd Century and beyond.

The formula is simple. To unlock economic growth, generate power. To generate power, build nuclear reactors. To fuel those nuclear reactors, enrich uranium. Enrichment was the most important bottleneck I felt I could help move the needle on. Today, almost three years later, our team at General Matter is restoring U.S. capability at our facilities from California to Kentucky, with the support of the U.S. Department of Energy.

America will progress over the next 250 years to the extent that our citizens continue to find and eliminate bottlenecks and come up with entirely new capabilities. All of these capabilities will require more energy.

If we get energy production right, what will the next 250 years look like? Let’s start with the next 25 and build from there.

We will have a base on the Moon. As my Founders Fund colleague Mike Solana will tell you, Moon Should Be a State.

We will have direct-to-cell communication anywhere on Earth, connected by an ever-expanding fleet of telecommunications satellites. An ever-expanding network of sensors subtly blanketing the globe will give us real-time visibility into the status of the home planet. Although our ancestors couldn’t know what was happening in Virginia from New York, we will know everything that is happening everywhere.

When we want to see something for ourselves, we will zip around the globe supersonically. We will build infrastructure to reclaim roads and parking spaces for plazas and green spaces. And when we do need to drive, our cars will be fully electric and fully autonomous. There will be no more meaningful car emissions, and there will be far, far fewer traffic fatalities.

Cars won’t kill us, and neither will today’s most deadly diseases. We are already experiencing medical breakthroughs that make our lives longer and healthier, from GLP-1s to CAR-T cell therapies to early detection. Just last month, Midjourney announced a new way to scan the body and Aleph previewed a new way to scan the brain, both using ultrasound, which now comes on a chip. These will accelerate, as human scientists devise new ways to fix and enhance our bodies, from epigenetic reprogramming to bioelectric therapies, and use AI to turn over every stone.

Twenty-five years from now, we will look back on this current era as AI’s awkward adolescent years. Having figured out the right relationship between humans and our intelligent assistants, we will have AIs helping us everywhere in our daily lives. The term “artificial intelligence” will be as outdated as saying “color TV” today. We will have integrated it into how we live, work, and play. Robots will be a routine part of life. Every person who can today afford a car might tomorrow own a personal chef, a massage therapist, a physical trainer, and the staff of a five-star hotel… all in one robot.

Provided there’s enough energy to power today’s tech revolution, it’s hard to comprehend how right things could go. In just 25 years, the amount of time since I was first in the rocket club in college, we will be living in a future that looks like sci-fi compared to today.

Just as our lives today are dramatically better than the Founding Fathers could have imagined, we’ll have lives that are better than we could possibly imagine today in the next 25 years… if we allow ourselves to.

On this trajectory, life in 250 years will be practically incomprehensible from today’s perspective. Sci-fi does the best job imagining, but even sci-fi failed to anticipate the world we now live in. For the same reason that free market capitalism beats centrally-planned communism, it is impossible to predict exactly what the future will be like, or just how great it can be.

In 250 years, most transformatively, people will be living indefinite periods of time, effectively as long as we want to. We will have full command over the Earth and its oceans — to travel wherever we’d like, whenever we want, to terraform, and to grow anything. Malthus will have been proven completely wrong. Our world will look very much like Star Trek.

Inevitably, this will include expanding off of the home planet. We will have the technology to physically support hundreds of billions of people or more on Earth, but people are born to dream and expand. To support those fundamental human desires, we will need to spread.

We will expand our civilization to other planets. I hope that we export our founding American values, including the freedom to experiment with new systems based on the will of the people living there.

All of this could be the inheritance we pass to future generations of Americans, if — like Franklin said — we can keep it. These next 10 to 20 years will determine the course of the following centuries.

None of this will just magically happen. It’s going to take hard effort and tough calls to reach a future that will look like today’s science fiction.

If we can’t find the will to build data centers, and the energy to support them while giving every citizen more energy, for less… if we can’t fix and expand the grid to carry all of the electricity that we’ll need, even though we have the technology to do so… if we can’t continue the momentum around nuclear until it shows up in thousands of powered and grid-connected reactors… if we can’t build that base on the Moon and use it to start pushing towards other planets… in 250 years, it won’t mean these things were impossible. It will mean something has gone horribly wrong.

That something might be cultural. Maybe hatred towards data centers and a modern Luddite movement. Maybe we’ve faced something worse, like another World War.

All of these fears have the same cure: abundance. Conflicts are based on scarcity or the concern about future scarcity. But if America re-accelerates our output, and if both us and China realize we’ll be living in a post-scarcity world anyway, there’s far less reason for conflict.

The task on which the next American quarter-millennium will be built is to dramatically increase clean, safe, baseload energy, so that we can use it to power everything else, unlocking a world of abundance.

The time to do all this is now. On a timescale as long as 250 years, getting things a little bit more right now has an enormous compounding effect on our trajectory.

There is perhaps no better example of that principle than the document whose anniversary we celebrate this weekend: the Declaration of Independence.

“We hold these truths to be self-evident, that all men are created equal, that they are endowed by their Creator with certain unalienable Rights, that among these are Life, Liberty and the pursuit of Happiness.”

The America of 250 years ago didn’t always live up to those promises. But there’s no doubt Jefferson’s quill set us on our path: a 250-year long pursuit of the ideals captured in those well-crafted words. What started as the American Experiment has become a republic worth keeping.

The Pursuit of Happiness is the magical part. Correctly interpreted, those words speak to the opportunity for any American to build a good life for themselves and their family, not simply to joy or comfort. Joy, in the American context, is often found in the exhaustion of a job well done.

America has always granted its citizens the right to be industrious, the right to go do things, build, and create prosperity. These rights are fundamental to who we are.

What has made America’s first 250 years great is that it has been this place where people knew they could come and have the space and freedom to build the future they believed would be good. The U.S. has always been the place for that, a place of freedom to, not just freedom from.

The success or failure of the next 250 years will depend on these fundamentals. If the U.S. continues to be the place where we celebrate ambition — and if we commit ourselves to our best visions for the future — we will thrive, and the rest of the world will thrive with us.

America should be the birthplace of superintelligence, the nuclear renaissance, the spacecraft that takes us to Mars, and the first people to populate the moon. This is the frontier country, wherever the frontier happens to be.

We will have another 250 years of growth, prosperity, and abundance if each of us chooses to continue to work for them. The most beautiful thing about the American project is that there is no one to blame, and no one coming to save us, other than ourselves.

Thanks to Scott, Avery, and Sally for working with me on this piece, and to the Founding Fathers for America.

That’s all for today. We’ll be back in your inbox tomorrow with our Weekly Dose.

Thanks for reading,

Packy

Hi friends 👋 ,

Happy Friday and welcome to our 199th Weekly Dose of Optimism! For some reason, maybe it’s the fantastic summer vibes, this one was particularly fun to write, so without further ado…

Let’s get to it.

This Week’s Weekly Dose is brought to you by… Mercury

Heads up: This is how banking* works now

You’ve seen what AI can do when it’s built into the right system. Mercury Command is that for your finances.

Say what you need and the work gets done — payments, forecasting, categorization, invoices — across all of Mercury. You approve every action, and every answer is generated from your Mercury data with full account context and a traceable record.

No dashboards. No exports. No friction. Just total command.

Try Mercury Command →

*Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column N.A., Members FDIC.

(1) Stripe Launches Intercept to End Respiratory Infections

for Stripe

I have had a cold all week. You forget how little fun a cold is! It sounds innocent, if annoying, a little cold. You don’t want to be the guy who complains about a cold. Rub some dirt in it, get back to work. But I’ve been operating at like ~40%. It would be way better if I didn’t have a cold…

And thanks to a new effort from Stripe, in the future, I (and everyone else) may not have to. Nan Ransohoff, who runs Public Goods at Stripe, announced that they’re launching Intercept: “a $500M philanthropic initiative to make respiratory infections, like the common cold and flu, a thing of the past.”

As Nan wrote:

We treat respiratory infections as a minor nuisance, but that’s really not the case. Most of us will spend 5% of our lives (!) sick from these viruses, they kill 1M people a year, cost $600B annually in productivity, and periodically threaten civilization through pandemics.

But respiratory infections are technically challenging and they’ve been underfunded, so we haven’t made much progress against them. Every year, we just accept that part of our year is going to suck (more time if we have kids, and suck more if it’s a particularly bad virus that kills us).

Intercept is solving the underfunded part, with money from Stripe, Anthropic, The OpenAI Foundation, Flu Lab, and people from Jane Street. It believes it can solve the technical parts, with a mix of broad-spectrum preventatives and air cleaning technologies.

Obviously, I hope they pull it off, and I think they will, but in any case, bravo to Nan and Stripe for continuing to rethink what modern institutions can to against the big problems facing humanity, and to the funders for taking a big, concentrated swing. This is what our billionaires should be doing, and I hope as many people copy Stripe here as copy the design of their homepage. Magnificenza!

(2) Congress Passes 21st Century Road to Housing Act

Ronda Kaysen for The New York Times

On Tuesday, the House voted 358-to-32 in favor of a bill, the 21st Century ROAD to Housing Act, in a touchingly bipartisan agreement that we need to build more housing in this country.

Vacancies are too low, mortgage rates are elevated, home prices keep rising, and the rents are too damn high! The bill, negotiated over many months, aims to bring down housing costs using a number of tools:

It loosens federal regulations, making it easier, faster and cheaper to build; eases lending rules; rewards communities that build; delivers aid to communities reeling from disasters; and, in a policy that proved to be one of the biggest flash points but was favored by Mr. Trump, sets new limits on the role institutional investors can play in the market.

Hear hear. We at the Dose are big believers in making it much much easier to build new housing, through regulatory and technological means. Not Boring Capital invested in Cuby to tackle the technical part, and now the government is doing its part to clear the way for new homes.

This is an obvious one. Voters, whether R or D, want to be able to afford a place to live. It’s clear that many rules in place get in the way of that happening.

No one in their right mind would oppose this thing…

To that end, President Trump is currently refusing to sign the bill unless Congress passes his SAVE America Act (his priority elections/voter ID legislation, which he called a “National Emergency”), and has doubled down since, telling House Speaker Mike Johnson “nobody gives a f*ck about housing” and telling the press that he’s not signing it because he doesn’t want to lower home prices for those who own them.

As of this morning, Speaker Johnson said that he’s formally sending the bill to the White House after a productive conversation with President Trump.

So, optimistic because a healthy majority from both sides understands that a lot of Americans do give a f*ck about housing and are finally doing something about it, and because the majority is so strong that it’s veto-proof, but also a good reminder that these NIMBY f*cks can be really dangerous.

(3) U.S. Bets Billions in Low-Cost Loans Can Revive Nuclear Power

Jennifer Hiller for WSJ

Another big week for nuclear!

On Tuesday, the Department of Energy’s Loan Program Office (of which the Dose and the broader Not Boring universe have long been fans, shout Julie Kozeracki) conditionally commited $17.5 billion in funding to support the construction of 10 Westinghouse AP1000 reactors.

These are the big boys, the 1GW guys that you’re familiar with, the gigareactors that can power whole towns by themselves indefinitely, but that can take over a decade and over $10 billion each to build.

These loans are meant to ease some of the financing and timeline fears that utilities have had when considering embarking on AP1000 builds.

Per the DOE:

EDF financing will support up to five loans, each loan supporting two reactors at a project site. Westinghouse will partner with up to five eligible utilities and energy companies nationwide to procure the long-lead items at a fixed price. Each project will be jointly owned by Westinghouse and a utility or energy company partner. Both Westinghouse and the partner are required to fully commit their project equity, $500 million each ($1 billion total per project), upfront prior to accessing DOE loan funds. Purchasing for each project will be staggered based on the timing of equity commitments and other relevant factors. Westinghouse has signed letters of intent with seven potential partners, each with identified project sites.

When Westinghouse and its utility partners commit their combined $1 billion in equity upfront, the DOE will chip in $3.5 billion in low-cost loans early, so that the project can start to purchase long lead-time items that typically delay nuclear projects, like containment vessels and turbines.

It’s a smart way for the government to aim its balance sheet at a high-leverage point in the process. Big win for the big reactors, while the small ones keep racking up wins too…

Yesterday, Aalo Atomics announced that DOE Secretary Chris Wright had “just signed the approval to turn on our first nuclear reactor at the Aalo-X site.” If successful in turning it on, Aalo would be the third advanced reactor company to go critical before America’s 250th Birthday on July 4th, meeting President Trump’s goal, and it would be the biggest to do so.

Meanwhile, Valar Atomics, whose criticality we covered last week, has gone on to generate power and then demonstrated 24 hours of continuous operations at 100% power.

Between all of this reactor progress and the USMNT’s performance at the World Cup, this summer is giving Nuclear Football a whole new, and much better, meaning (although it’s called Soccer now).

(4) Aleph Neuro Obtains Highest Resolution Extracranial Brain Image

Ultrasound is on an absolute tear. Prophetic’s dreaming headband. Last week’s Midjourney Scanner. And now a startup, Aleph Neuro, announced that it has “obtained the highest-resolution 3D images of the human brain ever taken from outside the skull” using “ultrasound transmitted through the skull, with a contrast agent.”

Best I can tell, the co-founders Marley and Lev injected contrast microbubbles into the bloodstream of their own brains, fired ultrafast ultrasound through their own skulls, tracked individual bubbles moving through their own brain vessels, and computationally combined thousands of frames into a super-resolution 3D map of their own brains.

Per Marley:

I could never convey the feeling of this discovery: it’s a random Tuesday and, all of a sudden, we saw a branching, treelike structure.

Moreover, it’s my own brain. A brain was looking at itself, while also knowing that it was the first time anyone in the world had seen this kind of brain image; this was divine poetry.

The tech is super cool. Aleph wants to make telepathy happen, and long time readers will know how I feel about telepathy. The cooler part of this story, though, is the good ol’ fashioned scientist-as-subject research the founders did on themselves to get to this breakthrough.

Marley put out a thread on the whole journey that I hope inspires you to find something that you know little about but want to, or a thing that you want to exist but don’t know how, and then to go out and learn whatever it takes to learn and do that thing.

~*In the age of AI*~ this feels like a pretty great skillset to have.

(5) Analogue Group Funds Soviet Papers Translation

Friend of Not Boring Aishwarya Khanduja’s Analogue Group announced the inaugural cohort of its Revival Fund: “an experimental fund dedicated to restoring neglected, illegible, or prematurely dismissed research to active circulation.”

Recipients are doing things like “reviving molecular inference by building DNA Learning Systems for programmable biology - sequences that learn, adapt, and act through chemistry” and “reviving brain-in-vat research with modern perfusion technology - bringing back forgotten work on sustaining living brains outside the body” and “reconstructing China's engagement with anomalous bodily phenomena from 1979-1999” that might not otherwise have a home but are important, and do now.

The one I’m most excited about is one that I talked to Aish about last December: translating old Soviet research papers. I’d had enough conversations with founders who told me that actually the tech they’re building is based off a Soviet research paper from like 1957 that I figured there was probably a lot there.

Now, Seconds is working on it at SovietRxiv, so we will find out soon enough! If you’ve been wanting to start a sci-fi company but have just been short on ideas, get in there and start digging for золото.

Poetically, the same day that Aish announced that they were translating the Soviet papers, Mike Grace announced that Longshot Space had raised $20M.

A conversation with Mike, in which he mentioned that the idea for his launcher came from an old Soviet paper, was one of that small handful that led me to believe that there was gold in them hills.

EXTRA DOSES: Science Breakthroughs, Space Wars, Impulse, General Intuition

Read more

Hi friends 👋,

Happy Friday and welcome back to our 198th Weekly Dose of Optimism.

I spent the week at the third annual Reindustrialize conference, which has the highest density of people we’ve covered in the Dose and in not boring more generally than any other event I know of, so the optimism is flowing particularly smoothly through my veins.

Let’s get to it.

Today’s Weekly Dose is brought to you by… Mercury

The hidden cost of banking* isn’t fees. It’s navigating all your tabs and dashboards.

Every finance task takes multiple steps. Check runway? Find the right report, cross-reference your burn, interpret the graph. Pay a contractor? Track down the invoice, navigate to transfers, fill in the details. Freeze a card? Dig through settings.

Mercury Command eliminates the switching tax. It’s AI built directly into your Mercury account that understands natural language and executes from one place — payments, invoices, categorization, team management. You review and approve every action. Command does the rest.

Try Mercury Command →

*Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column N.A., Members FDIC.

(1) Midjourney Announces Midjourney Medical + Scanner

When the AI image and video company Midjourney said that they were going to be hosting a demo of their first hardware product this week, most people, understandably, assumed that they would unveil something related to AI image and video. A digital canvas, perhaps. Glasses. VR. Art-immersed physical environments, even.

Wrong. All wrong. Thinking way too small.

“Today we're gonna announce something a little weird and a little crazy,” the company wrote, “but also spectacular and filled with hope.”

When I wrote a Deep Dive on Ezra a couple years back, I started the piece with a vision of the future:

In the year 2050, each time you step in front of the mirror in your bedroom, it runs a quick and dirty MRI on you. The MRI won’t be as high-resolution as something you would have gotten in a tube in 2023, but it doesn’t need to be. AI fills in the gaps and pulls signal from noise.

Most days – 99.9% of days – you don’t even remember you’re being scanned. There’s nothing to report, so your mirror reports nothing.

Very occasionally, though, it might notice enough of a change in your cells that it recommends a higher-resolution, localized scan on, say, your prostate, moves some things around on your calendar, books you an appointment, and summons your car to whisk you away to an imaging facility for a closer look.

That is basically the weird, crazy, spectacular, hopeful product that Midjourney announced, the first from the newly-formed Midjourney Medical: the Midjourney Scanner, “something as powerful as MRI, and as casual as a trip to the spa.”

It’s not an MRI, but the idea is to use lots of ultrasound and compute to get to the same place. From Midjourney’s (very well-done) announcement post:

It starts by stepping into a shallow pool of golden light. You then begin to descend into the water. Your body passes through a ring of underwater sensors, each acting like a dolphin, using its echolocation. The sensors send ultrasonic sound waves through your body from every angle. With enough waves, and enough angles, we form an image of what’s happening inside your body.

The goal is for this process to take no more than 60 seconds.

You go into the water, you come out of the water, and you’re done.

While you’re in there, ultrasonic waves hit your body and ripple back millions of times per second, producing terabytes of data. Then, computers turn all of those waves into images to form a 3D map of your body down to the millimeter.

Midjourney calls the process Ultrasonic CT (computer tomography), a whole body ultrasound scanner. They explain it in more detail here:

Of course, this is just an announcement, and there’s a lot to build and a lot to prove. Some people are unsurprisingly more skeptical of the Scanner, and others are more measured about what it is and isn’t. I really like Friend of Not Boring and Ezra co-founder Emi Gal’s breakdown.

Having said that… a few thoughts beyond the technology itself:

1. It’s very cool to see a company branch out so aggressively. It’s imaging-adjacent, and it requires a lot of computation, but the company admits that “It’s not related to anything you’ve seen from us so far.” But, they write: “However, we feel an obligation as people standing on the frontier to look at the foundations of the human experience and ask: ‘What do we want to be different?’ ‘How do we want to be different?’ and ‘What do we want to become?’” That’s a very cool way to think about it, and I hope more companies get creative with what they can do given a bunch of smart people, money, and attention, even if many fail in the process of unfocusing.

2. Midjourney hasn’t raised VC and is profitable, which gives them the leeway to do this (or whatever they want).

3. The reason they’re able to do this so quickly is because they’re partnering with Butterfly Network, which is a fascinating company that makes ultrasound-on-chip. Interestingly, Matthew de Jonge, the founder and CEO of hearing aid company Fortell, which we covered in Dose #172, is the ex-VP of Product at Butterfly. You can do magical things with hardware when you’re willing to go all the way down to the chip.

4. If you look at David Holz’s bio, this isn’t as crazy as it looks. He’s built hardware as the co-founder and CTO of Leap Motion, worked on planetary imaging at NASA, and researched neuroimaging algorithms at Max Planck Florida to map an “entire rat brain” to the cell level.

   

5. Holz said the Scanner is the first of eight major announcements it plans to make in 2026. MOAR.

6. This is a new entry into The Great Differentiation canon, and it’ll be interesting to see how the other seven fit in (or don’t). Different products as differentiation would be fun.

(2) Making the Invisible Visible with the Laser Phase Plate

CZ Biohub & UC Berkeley

Bigggg week for seeing litttttle things.

Zuck’s CZ Biohub, in partnership with UC Berkeley, announced their Laser Phase Plate last week, which looks and sounds cool, but which honestly I didn’t understand until I saw a lot of very smart people tweet something to the effect of “most people won’t appreciate how huge this is but I do because I’m very smart” so I looked into it.

Basically, scientists can’t see the proteins inside of a cell in real-time in their natural environment. They can image ~10% of the proteins once they’re pulled out and purified, using cryo-electron microscopy (cryo-EM), or < 1% when they’re actually working inside a cell, but most of our cell’s proteins operate in the shadows. They stay too faint to see because current imaging techniques can’t create enough contrast.

In light microscopes, the normal ones, scientists have understood how to solve the contrast problem since Fritz Zernike won a Nobel for it in 1942. With an electron microscope, though, it’s way harder, because anything you place in an electron beam gets zapped and fried. So fifteen years ago, Berkeley physicist Holger Müller proposed swapping the physical part for a laser. Cool idea, but “because light barely interacts with electrons, the laser would have to be extraordinarily intense,” so intense as to be impractical.

Until now! The CZ Biohub and UC Berkeley researchers (led by Müller himself) published two papers (Science and Nature Communications) and a preprint explaining their laser phase plate, including Müller’s solution to the laser intensity problem:

Müller’s solution was a so-called Fabry-Perot cavity: a pair of mirrors that bounce a laser beam back and forth roughly 10,000 times. Each bounce amplifies the intensity until the light focused between those mirrors reaches roughly 100 million times the intensity of the surface of the Sun — the brightest continuous-wave laser in the world.

See the results for yourself. The image on the left is with the beam off, and the right is with the beam on (you can drag it on the Biohub blog).

Interestingly, better contrast also unlocks cryo-electron tomography, or cryo-ET, which “builds three-dimensional reconstructions of structures inside intact cells — revealing not just protein shapes, but how proteins interact with their neighbors,” and which I bring up because how often do I get to cover tomography in back-to-back entries?

The hope is that the better we can see, the better we can fix. We down with LPP.

(3) Valar Atomics Goes Critical

Out in Emery County, Utah, Valar Atomics’ Ward 250 reactor went critical. This makes it the second new advanced reactor to go critical before America’s 250th Birthday on July 4, 2026.

What makes this one particularly sweet is how much doubt Valar founder and CEO Isaiah Taylor faced when he initially announced the company a little over two years ago. He didn’t have a traditional background - no college degree, let alone nuclear physics PhD. The way that he talked about building a lot of reactors really fast, and even using them to produce hydrocarbons if that’s where the first market was, was just Not How Things Are Done Here.

His reputation preceded him. And it was totally wrong.

When I got to speak with him for Age of Miracles, it was obvious that Isaiah was a rare actual example of that thing that everyone claims to be: a first-principles thinker, which is something it’s easier to be when you come from outside the industry.

He did things like start with the market and work back to the specific technology, make systems-level trade-offs for speed over efficiency, and compare the cost of machines of different sizes (like buses and gas plants) to nuclear reactors instead of just taking for granted that nuclear reactors are expensive. He became one of our go-to people for Age of Miracles because he was so clear in his explanations and novel in his approaches.

Anyway, it’s cool to see him just do the thing in such a short amount of time. Now, he and the Valar team have to fulfill the rest of the mission: first, power operations by July 4th, then, print tens of thousands of reactors across the globe to make energy 10x cheaper than it is today with mass-scale nuclear fission.

Being Critical: 0

Going Critical: 1

(4) Anduril Wins Production Contract for U.S. Air Force CCA Program

Wayyyy back in Dose #91, in April 2024, we led with Anduril’s selection for the U.S. Air Force Collaborative Combat Aircraft Program. In that early phase, for which it beat out a number of incumbent primes, “Anduril will design, manufacture, and test production-representative CCAs.” Then, we wrote, “Ultimately, the USAF will make a final, multibillion-dollar production decision in two years from now — it said that it wants ~1,000 craft — with the goal of having fully operational crafts by the end of the decade.”

Two years is up ladies and gentlemen, and … drum roll please… we have a winner.

This week, the Air Force picked Anduril’s FQ-44 “Fury” as one of two jets going into production for the CCA fleet: uncrewed fighter jets designed to fly alongside human pilots as autonomous wingmen.

Anduril went from prototype award in 2024 to a production contract this June, which is the fastest any fighter has gone from prototype to production in more than half a century. The decision landed four months ahead of schedule, so good on the Air Force, too. It’s also the first time a new company has won a fighter aircraft program since the 1970s, before the F-16 existed.

The Air Force wants more than 150 FQ-44s by the end of the decade.

Anduril didn’t get the W solo. General Atomics won an identical contract for its own jet. But the Air Force resolicited Boeing, Lockheed, and Northrop, the incumbents who’ve owned fighter programs for fifty years, and it passed over them in favor of Anduril once again.

It’s time to produce now, lots to prove, but it’s a great step on the way towards “saving western civilization by saving taxpayers hundreds of billions of dollars a year as we make tens of billions of dollars a year.”

(5) Notes from Reindustrialize

This week, I went on my annual pilgrimage to the Motor City for Reindustrialize. This was my third, and in my opinion, the best yet.

The thing that’s stayed consistent is the density of great people. They save me like 10 discrete trips by pulling so many good people to Detroit.

Where else can you eat Westmag hot dogs with your favorite founders, investors, and policy people while watching a drone show in a metal factory?

What’s changed is that where Reindustrialize 1.0 focused on arguing that we should build things in America again, Reindustrialize 3.0 was packed with examples of companies actually building in America. It was much more focused on execution details than before, because Reindustrialization is happening.

I got to moderate two panels - one with Micah Springut of Monumental Labs and Alexis Ohanian of 776 on AI and the Workforce (and how to build beautiful things), and one with Sam D’Amico (Impulse), Justin Lopas (Base), and Adam Warmoth (Chariot Defense) on The Electric Slide. Both got very specific - the right balance of automation and craft, how to decide where to integrate and where to buy, etc…

It’s wild how much can change in two years. At Reindustrialize 1.0, I moderated a panel making the case for nuclear. To my left were Valar Atomics’ Isaiah Taylor and Decisive Point’s Tommy Hendrix.

This week, just this week, Isaiah’s company went critical, and the company that Tommy founded after Reindustrialize 1.0, Standard Nuclear, filed its S-1 to go public. Not Boring Capital is an investor in Standard Nuclear because of Reindustrialize.

There’s a ton of momentum, capital, and talent going into making big things again. I can’t wait to see what this year’s crop announces during Reindustrialize week two years from now.

EXTRA DOSES: Science Breakthroughs, SpaceX & the Sentient Sun, Dean W. Ball

Read more

Hi friends 👋 ,

Happy Friday, welcome to the 197th Weekly Dose of Optimism, and a very Happy SpaceX IPO Day to those who celebrate!

It’s been a big week over here at Not Boring HQ. On Monday, we published a co-written essay on flying cars with Tsung Xu. On Wednesday, we published one on tokenminning with Markie Wagner. That night, I got to go to the craziest basketball game I’ve ever seen. Yesterday, the World Cup started right here in America.

And today, we have wall-to-wall, handcrafted optimism for you, brought to you by our new friends at Pangram, who are allies in the fight against AI-written slop.

We have a church that has taken lifetimes to build, and a trial of a drug that may help extend our lifetimes. We have Doudna back at it with a cancer-shredding CRISPR. We have drone boats saving soldiers, and autonomous planes that fly right over the water. We have money for robots and Bezos has money for manufacturing. And we have Science Breakthroughs, a view of America through Freddy’s eyes, nuclear batteries, and even more. What a week for the optimists.

Let’s get to it.

Today’s Weekly Dose is brought to you by… Pangram

Pangram is the most accurate AI detector on the market with a 1-in-10,000 false positive rate and segment-level scoring your team can act on. They are hosting a live session for compliance, fraud and security teams on how AI detection fits in a security and compliance stack.

If your team reviews filings, claims or vendor deliverables every day, this session is for you.

Reserve your spot.

(1) Pope Blesses La Sagrada Família’s Tower of Jesus Christ

This week, to mark the 100th anniversary of the architect Antoni Gaudí's death, Pope Leo XIV blessed the Tower of Jesus Christ at Barcelona’s Sagrada Família basilica. It LSF’s tallest tower, making it the tallest Catholic church in the world, and its central one. There’s more construction to be done on the rest of the church, but the ceremony was beautiful enough to lead off with, given how much I love the story of La Sagrada Família.

I wrote about it in I, Exponential back in August 2023. You can read the LSF section here.

In 1883, a small group of Catholic devotees of Saint Joseph (Josephites) entrusted a young Catalan architect, Antoni Gaudí, to build them a church. As architect Mark Foster Gage writes in a piece for CNN:

Gaudí’s vision of the church was so complex and detailed from the start that at no point could it be physically drawn by hand using the typical scale drawings so common to almost all architectural projects. Instead, it was almost entirely constructed through the making of large plaster models to communicate Gaudí’s desires to the army of stonemasons slowly liberating its form from blocks of local Montjuïc sandstone.

Then Gaudí was hit by a tram car and died in 1926, with the church only 10-15% completed. For the past century, teams of builders and architects and technologists have worked to figure out exactly what he was going for, and then to build it.

The reason I love the story, other than LSF’s beauty and the fact that Barcelona is one of my favorite cities in the world, is that the reason we’ve been able to get it to this point over the past half-century is that the technology finally caught up to what was in the architect’s head.

In 1979, a 22-year-old Kiwi Cambridge grad student, Mark Burry, visited Sagrada Família and interviewed some of Gaudí’s former apprentices. They showed him the boxes of broken model fragments, and offered him an internship. Burry got to work trying to reconstruct the mind of Antoni Gaudí in order to construct the church that lived within.

He tried to hand-draw the “complex intersection of weird shapes, including things like conoids and hyperbolic paraboloids” but he realized that the tool wasn’t up to the task. The computer saved the day.

Burry brought in software used to design airplanes to solve the otherwise-impossible problem of translating Gaudí’s sketches of bone columns into 3D models.

In order to actually construct the building, Burry and team hooked their computers up to a relatively new invention, CNC (computer numerically controlled) machines. CNC machines were themselves a product of a number of technological advances in computing power, data storage, electronics, motors, material science, user interfaces, networking and connectivity, control systems, and software designs. All of those curves converged in time for Burry to feed his 3D models into CNC machines that could precisely carve their designs out of stone.

Today, the team working on Sagrada Família uses a full arsenal of modern technology, from 3D printers to Lidar laser scans, from sensors to VR headsets.

And now, this great tech-enabled architectural wonder of the world has its central tower completed, is nearing full completion, and has the blessing of the Pope.

As impressive as the building itself is, the Pope blessed both it and his Nova Knicks in the same week. He’s the real MVP.

(2) World-first: therapy to make cells young again trialled in a person

Heidi Ledford for Nature

Five years ago, David Sinclair's lab at Harvard made old, blind mice see again by turning back the biological clock in their cells. They coaxed aged optic-nerve neurons to behave young again, and regrow.

This week, for the first time, they did it to a person.

Life Biosciences, the Boston company built on Sinclair's work, announced it has treated the first participant in the world’s first clinical trial of partial cellular reprogramming. The trial is targeting glaucoma, a disease that slowly kills the neurons of the optic nerve, which don’t typically regenerate. They’re betting that the therapy will make the cells young enough to try.

The work is based on Japanese Nobel Laureate Shinya Yamanaka’s eponymous Yakanama Factors, four genes that can rewind any adult cell all the way back to a stem-cell state. The problem is, a reset retina cell forgets it's a retina cell. So Life Bio uses three of the four factors (dropping c-Myc, the one most tied to cancer) and nudges the cells only partway back: younger, but still themselves. In Sinclair's 2020 mouse study, that partial nudge regenerated neurons and reversed vision loss in elderly and glaucomatous mice.

The approach - partial epigentic reprogramming - is similar to the NewLimit approach to curing mouse (and eventually human) hangovers and liver damage that we covered in last week’s Dose.

We flagged the potential for this trial when the FDA cleared it back in January. Now, there's a human being walking around with partially reprogrammed cells in their eye.

Before we get too excited, this is a safety trial, and it’s only in one patient so far. There are real concerns. Push cells too far back, or in the wrong tissue, and you risk tipping them cancerous. Longevity scientist Matt Kaeberlein put it plainly: the upside is big if it can be done safely, but the tech is early and the downside risk is severe.

That’s the only place to start, though, and the potential here is enormous. Two weeks in a row, we’ve covered credible teams pursuing trials to reverse aging, that horrible disease that ultimately kills every human on earth and degrades our quality of life in the process. The sooner we get these therapies, the longer we live younger.

(3) Doudna's Lab Build a CRISPR to Selectively Shred Cancer Cells

Andy Murdock for Innovative Genomics

Speaking of cells… this week Jennifer Doudna's Innovative Genomics Institute published a paper in Nature describing a CRISPR system that does the opposite of what it usually does. Instead of fixing a broken gene, it finds the cells carrying one and destroys them.

Most cancer drugs are inhibitors that tamp down an overactive gene. But the most common cancer driver is the reverse: a tumor-suppressor switch that's been snapped off. p53, the gene that normally keeps cells from turning cancerous, is mutated in roughly half of all cancers and up to 70–90% of the nastiest ones: ovarian, pancreatic, non-small-cell lung. You can't restore a lost function by inhibiting something, which is why, after 35 years of trying, there's still no p53 drug.

So first author Jingkun Zeng went the other way. He engineered CRISPR-Cas12a2 to watch for the RNA signature only a p53-mutant cell produces, and when it sees that signature, the enzyme shreds all the genetic material inside that one cell, killing it while leaving healthy cells alone. In a dish of cells differing by a single DNA letter, it wiped out the mutants and left the normal ones almost untouched. Because it's programmable, a new mutation just means writing a new guide RNA.

The caveat is that this is still cell culture, and as Zeng flags herself, the hard part will be delivery, first in animals, then humans, to get a big genome-cutting enzyme into every target cell in a living body.

But still! People have been trying to target p53 for 35 years. Doudna & Co have figured out a way to just shred it. Say it with me… get fucked, cancer.

(4) Saronic Drone Boat Rescues Downed Apache Crew

Nicholas Kulish and Eric Schmitt for The New York Times

On Monday, Iran downed an American Apache helicopter near the Strait of Hormuz, sending two crew members into the water. That’s bad news, not the stuff of the Dose.

The Dose-worthy part is that those two crew members were rescued by a drone boat, a Saronic Corsair. “It was the first U.S. rescue carried out by an autonomous surface vessel, remotely piloted by a human operator, the Central Command spokesman, Capt. Tim Hawkins, said on Tuesday.”

Per The New York Times, “The vessel carried the Apache’s pilot and gunner to another location, where they were picked up by a helicopter to complete the rescue.”

Again, two American military members were stranded in the water in hostile territory, and they were rescued by drone boats built by an American startup that we’ve covered previously in the Dose, just one week after the same company launched its much larger Marauder, which it built in an American shipyard (America is supposed to be bad at shipbuilding) in under a year.

If that’s what American Dynamism looks like, then all aboard. (jk they’re autonomous)

(5) Standard Bots Raises $200 Million at $1 Billion

Here at the Weekly Dose, we love when good things happen to our friends. A controversial stance, perhaps, but we stand by it.

So we were very happy to see our friend Evan Beard raise $200 million at $1 billion for Standard Bots, America’s largest manufacture of AI-native industrial robots. Evan said that the company is on track to deploy 10% of America’s industrial robots by 2027. That’s not saying as much as it should: last year, China installed 9x more robots than the US.

With the money, Standard Bots plans to manufacture robots - from metal in to robots out - and deploy them in manufacturers across the country. More robots, Evan argues, more competitive American manufacturing.

If you want to learn more about what Standard Bots is building and why their Many Small Steps approach might be the way to actually make robotics happen, check out the essay Evan and I co-wrote in January:

EXTRA DOSES: Science Breakthroughs, Freddy, Poseidon, Prometheus, Zeno Power, Criminals Hate Flock Safety

Read more

Subscribe now

Hi friends 👋,

Happy Wednesday and welcome back!

A couple of months ago, my friends Adam and Ben at Genius Ventures asked if they could introduce me to one of their favorite founders, Markie Wagner.

The Markie Wagner? The Choose Good Quests Markie Wagner? The drop an all-timer then go quiet for years, cooking up something spoken of in hushed tones Markie Wagner? The grew up inside of a computer and dreamed as a young girl in Southern California, seriously, of making computers do the work that humans shouldn’t have to Markie Wagner?

Of course I wanted to meet Markie Wagner.

So we met a month ago at Soho Diner and I ordered a milkshake and she asked them to cut up a bowl of fruit. She asked for my lore, which was boring, and I asked for hers, which she weaved non-stop for the next hour, landing so naturally on why she’s building what she’s building that it seemed almost pre-destined.

She also told me, before everyone else came to the same conclusion, that tokenmaxxing was bullshit, because behind closed doors, the Fortune 500 CEOs she works with were all saying some version of “We committed to all this token spend and I have no idea what we’re getting out of it.”

She was right, I think she’s going to be right again, she’s backed by Founders Fund, Kleiner Perkins, Genius Ventures, and OpenAI to go prove it, and now she’s explaining her logic publicly in her first written piece since Good Quests.

So this is where we are heading, according to Markie Wagner.

Let’s get to it.

Today’s Not Boring is brought to you by… Deel

New to global hiring? Start here.

Hiring internationally is complex. Learn what an Employer of Record is and how startups use EORs to hire global talent compliantly.

Get the Guide

Return on Tokens (ROT)

Co-Written with Markie Wagner

The promise of AI is that it will turn businesses into software so that they can evolve over millions of tiny iterations. Beautiful, ideal, complex things can only emerge as the result of tremendous trial and error over time. You cannot build perfection, only discover it.

Capitalism is organizational evolution. Millions of businesses compete in the marketplace with offerings that they think customers will want. Some thrive and grow. Others die. Each company evolves, too. People come and go. An experiment becomes a process, a process becomes a web of tacit knowledge. Products are introduced, and products are retired.

This constant evolution is why we enjoy the standard of living we enjoy today, and why ours will look primitive to future generations. Accelerating it is my Good Quest, because if every business can evolve to its ideal form, it will create trillions of dollars of value and unblock all of the other Good Quests.

I dropped out of the research world because it felt like the wrong hill to climb, and I went out into America to just do work, so that I could figure out how to make computers do the work, so that humans could direct the computers in evolving the work.

What I suspected before and learned in my travels is that the way that the market has implemented AI thus far is the wrong way. It’s not endgame. It is too wasteful, too forgetful, and too imprecise. I’ve been in the fucking Sahara Desert out here fighting demons to learn this wisdom.

Tokenmaxxing Clearly Isn’t It

Tokenmaxxing - literally maximizing the amount of tokens you or your organization spends, tracked in leaderboards and rewarded with trinkets - was a mass delusion, something like a commercial form of AI psychosis.

Tokenmaxxing was a lab-grown supermeme that worked better than the labs could have hoped.

Picture this. Anthropic and OpenAI release a product, Agents, in the form of Claude Code/Cowork and Codex, respectively, that are basically lab employees working inside of customers’ companies and are given company credit cards with no spending limit (tokenmaxxing suggests the more they spend the better they’re doing) to spend on behalf of their real employer, the lab. Anthropic ships a bunch of Agents into, say, KPMG, which commits to a certain spend in exchange for discounts (token commits), KPMG’s employees are encouraged to use Agents to do everything they can possibly think of (lots of dashboards), and then these Agents, which again you can think of as digital Anthropic employees with no-limit KPMG credit cards that they can use to spend on Anthropic, run up token bills to their heart’s content. Employees who direct their Agents to use the most tokens are recognized as AI Innovators.

Certainly, some people recognized that it was a delusion. They would ask questions like, “But are the Agents doing anything useful? Aren’t they just building dashboards? Please can someone show me something useful they’ve built with an Agent?” but those sane few were met with the killer retort: “Skill Issue.”

Some people, they were told, were building immensely valuable things with Agents, the same way that some people had a super hot girlfriend at summer camp but you’ve never met her. If you couldn’t figure out how to do the same, well, welcome to the Permanent Underclass.

Everyone fell for it, for a while. The market incentivized companies to spend tokens, so boards incentivized leaders to spend tokens, so leaders incentivized managers to spend tokens, so managers incentivized employees to spend tokens. Nobody had an incentive to say that the tokens aren’t doing useful stuff.

I talk to these people all the time and every company has some version of the same conversation. Someone who’s running the AI team goes, “We’ve made a ton of progress this quarter. We spent $50 million on tokens.” And everyone nods and claps. “Usage is up. We’ve built 3,000 Agents. We shipped 10 million lines of code.” And you’re like… what? And then I’d ask, “Hey did you measure accuracy for the fraud Agent?” And they’d go, “Yeah… it’s about 50%.” People are at 99%!

But the models improve and everyone’s tokenmaxxing and you don’t want to be the luddite, so you keep lazily throwing Agents at everything and hoping they learn.

All of this happened, by the way, right as the labs switched from subscription-based to consumption-based revenue models, so companies had no time to prepare.

It is no wonder token usage, and therefore lab revenue, went parabolic.

It took Uber, that last era’s poster child of VC-subsidized demand, to break the spell. Its CTO said that the company had burned through its 2026 Claude Code token budget by April. In May, its COO said that the company was having a harder time justifying its AI spend, because the link between AI consumption and shipped features “is not there yet.”

What followed was like that scene in Mean Girls where Tina Fey asks the students to raise their hands if they feel personally victimized by Regina George, and one hand goes up, then all of the rest of the hands go up.

There was the consultant saying his client accidentally burned half a billion dollars on Claude Code. Amazon shut down its AI leaderboard. Legora CTO Jacob Lauritzen told Harry Stebbings that token leaderboards “lead to tokenmaxxing, which is people just burn tokens just to look good. That’s a really stupid way to do anything.” Ramp’s Veeral Patel called it the Token Casino: “useful software wrapped in mechanics that make spend feel like progress. It starts with the oldest trick in the book: abstract the money.” Palantir CEO Alex Karp told the TBPN boys that tokenmaxxing is like “a porn addiction.”

Even Sam Altman, a prominent token vendor himself, admitted on CNBC that “You hear companies saying, ‘I am spending a ton of money on AI, and I know some great stuff is happening, but I know there’s a ton of waste, and you know, when… how long do I have to wait for it to really show up in revenue, and how long do I have to wait to really get the costs under control?’” It had become, he admitted, a “huge issue.”

The issue is the companies have focused on maximizing tokens, assuming that tokens = value.

Every cycle has its dumb metric. In the mid-nineteenth century, the market wanted miles of railroad track as a proxy for future monopoly and the benefits thereof, and so railroads raced to lay miles, often along the same routes as competitors. At the turn of the 21st century, the market wanted eyeballs, and so dot coms attracted eyeballs and served them up on a platter. In the 2010s, the market wanted top-line gross revenue, and so companies like WeWork delivered top line gross revenue.

This cycle has tokenmaxxing.

Which is not to say that tokens can’t be valuable. Cornelius Vanderbilt’s New York Central ended up becoming very valuable, as did the Pennsylvania Railroad. Google and Facebook have converted eyeballs to cashflow better than anyone has ever converted anything to cashflow. Uber ended up turning top line growth into market dominance and turning that into $10 billion in 2025 free cash flow.

The question is always: can the thing generate returns?

For tokens, the question is: what is your Return on Tokens (ROT)?

Return on Tokens

When you invest in a new machine, you expect it to generate a return. When you hire an employee, you expect them to generate a return. Business is the process of making investments big and small and expecting them to create more value than they cost.

Tokens need to be held to the same standard.

Return on Tokens = (Value of Output - Cost of Tokens) / Cost of Tokens x 100

There are two ways, then, to increase your ROT. You can create more valuable things with them, or you can spend less on them. Ideally, you spend less to create more value.

The first thing that companies are focused on, because it is easier to measure than output value, is spending less.

Now that the spell has been broken, cooler heads are proudly discussing “routing” as a means to lowering the cost. Use Anthropic and OpenAI’s best models for the really big brain stuff, but do most of the work with cheap Chinese open source models. Coinbase CEO Brian Armstrong’s recent tweet is a good example of this logic:

You can see this in the OpenRouter AI Model Rankings. The move to Chinese models actually showed up in lockstep with consumption-based pricing, although this is a self-selecting group of users that were already thinking about routing tasks to the right models. The rest are scrambling to do the same now.

It’s a good start, but Agents spending tokens, American or Chinese, to figure everything out from scratch is not endgame, either.

Because you know what’s cheaper than Chinese models? Code.

Code, good old fashioned deterministic code, is not only cheap, it is a better fit for most economically valuable work. We have learned this lesson.

In the past, companies hired humans to do all manner of repetitive tasks. Before “computers” were digital, they were humans.

About half a century ago, we began the process of taking repetitive tasks that humans did, like calculating the trajectory of a missile or the profits of a business, and handing them over to software. Code ran more reliably than even the most reliable human computer. It made no mistakes. It answered instantly. Enter the same numbers and same formulae in the same cells in an Excel spreadsheet anywhere in the world, at any time, and it spit out the same number.

Then we got Agents, and we forgot the lesson. We decided that we needed to throw these pseudo-humans at everything, because everyone else was. Agents are great at some things, but they’re not the right shape for a lot of others.

It’s no wonder companies aren’t getting a positive ROT on their tokens. All the dashboards have been dashboarded, and now they’re sending Agents to do software’s job.

There is an argument to be made that a lot of companies aren’t getting a positive Return on Tokens because they don’t know how to use them yet, or because they haven’t re-architected their companies to be AI-native yet. This is one of the reasons, perhaps, that both Anthropic and OpenAI have launched consulting subsidiaries to help companies better deploy tokens. And there are certainly examples of startups built during the AI era that seem to be using tokens to great effect, which shows up in their own supersonic revenue growth. If the Old Economy can’t generate a ROT, well, this is creative destruction baby.

And while it’s certainly true that not all companies are deploying AI equally well, we believe that there are more fundamental structural reasons for the negative ROT: Agents are the wrong architecture for most work.

Agents Aren’t It (For Most Work) Either

There are three structural reasons for Agents’ negative ROT:

The Agentic architecture can’t do long-running work at the nines of quality that real economic work requires.

Agents improvise. They’re spawned fresh onto repetitive tasks like every day is their first day on the job, which hurts consistent accuracy. For new features, prototypes, or dashboards, 80% accuracy is fine. For the real repetitive work on which the economy runs, like fraud detection or underwriting decisions, 80% accuracy is 0% usable.

Engineers don’t know what to build because they don’t do the work.

Most of the process-driven work we’re describing exists as a combination of written rules, which Agents can ingest, and then like 3,000 tacit rules and sub-rules that live in people’s heads, in offices around the country, far away from the engineers’ San Francisco desks. AI can only evolve what it can touch, which is why it’s been great at coding but has largely failed to do useful things in the enterprise.

The original sin is that there are no goals.

If people have no goals then the Agent has no goals, and then the thing achieves no end. Without a goal to hill-climb against, code (whether written by humans or generated by Agents) decays into slop in the limit because there’s no purifying force to evaluate what’s good and bad.

One of the beautiful things about Agents, from a laziness perspective, particularly when you are being encouraged to spend a lot of tokens, is that you can just set them loose without knowing exactly what it is that you’re solving for. They can go spin on a vague instruction for a while, bring something back that’s decent but not perfect, and then go out and spin some more.

This process drives more token spend without delivering any value, which is a fast track to negative ROT.

People are searching for new things for Agents to do assuming that AI will do for everything else what it’s done for code. But it doesn’t have to.

There is a surprising amount of work that is best done with plain old code. The challenge has been that, until recently, there were not enough engineers to turn everything every business does into code, and then update it as things changed. There are now. AI makes writing code trivial, so if we can get the knowledge out of people’s heads, we can turn businesses into code.

AI is a Compiler, Not a Runtime

Basically, software works in two steps, thinking and doing.

First, thinking: you take the goals and requirements for what a piece of software should do and compile it into code that a computer can run.

Then, doing (and doing and doing and doing): every time it needs to do the thing, the code runs cheaply and predictably (or deterministically).

While computer science has a precise definition of a compiler, you can also think of a software company or a software engineer as a compiler. They take the goals and requirements and turn it into code. Then customers buy the code they built, and run it over and over again. This is the beauty of the zero marginal cost software business, and it’s why companies can sell software that took millions of dollars to develop for $20 per month and still generate mouthwatering margins.

The way that most people think about (and use) Agents today is that they replace both the software company and the software. That is the wrong way to think about it.

Agents should replace the software company; they should take goals and requirements in English (or whatever language), and turn it into code that runs over and over and over, deterministically.

Thinking is expensive but happens rarely. Doing is cheap and happens forever.

Agents should do the thinking, code should do the doing.

For most economic work, you want to use humans to figure out the rules, use AI to turn the rules into code, and then run that code forever at near-zero token cost, only bringing the AI back in when the rules change.

Why would you use a prompt to add two numbers? Just write a line of Python, dawg.

The current Thinking-Doing Ratio (TDR) in AI implementations is roughly 1000:1, which is not surprising. San Francisco is a Thinking town. Anthropic’s hats say, simply, “thinking.”

Silicon Valley built AI assuming work is mostly thinking, but work is mostly doing.

Chat is the rare exception where you genuinely don’t know what comes next. So maybe customer support chat continues to churn through thinking tokens (although even customer support Agents kick complex problems to humans). Almost nothing else in a business looks like constant improvisation.

So we use Agents for Doing, but Agents are the BlackBerry of doing. They are not where most work will get done inside of companies in five years’ time. It will get done in the deterministic code that they write.

The Agents’ role is in compiling into code, not into running and doing work day to day. Which means that it’s more like CapEx than OpEx. Everything you think is gonna be AI running is just going to be code running.

Everyone thinks the thing that is going to change in the world is that AI is going to become a person, but the real change is that a business is going to become a piece of software.

That’s the world we’re building at Poetic.

Turning Businesses Into Software That Evolves

We are building the antidote to tokenmaxxing: software that tokenminns itself.

Poetic is a new class of software: adaptive like AI, reliable like code.

We use AI as the compiler. We learn everything that a business does by taking in all of the processes that are written down, then going on-site in Nebraska or Providence or wherever the work is done, sitting on people’s shoulders, and asking “What did you just do?” “Why did you do that?” hundreds of times to learn the thousands of hidden tacit rules on which every company runs. Then we turn it into code.

The code is the runtime. When the world stays the same, this code runs the exact same steps every time. When the world changes, it learns, regenerates, tests itself against the objective, and then runs the new code until the world changes again.

The result is 100x less token usage and nines of accuracy on complex tasks. Put differently, each token you spend does 100x more, and it does it right.

The value of the output increases, because Poetic does something that your business actually needs to do, over and over. And the cost of tokens is lower, because Poetic only uses tokens when the world changes. Combined, Poetic generates a clear, measurable Return on Tokens.

We are doing it today, for companies like AIG, SoFi, and Chime. AIG CEO Peter Zaffino said that Poetic has already “achieved 99%+ quality outcomes on multi-hour processes - delivering real enterprise value.”

These companies’ leaders believe what we believe: that every business will have to be re-founded as a software business. The story of the next decade is the beginning of those new businesses. Some will be truly new, built from scratch. Others will be businesses that have existed for hundreds of years, brave enough to reinvent themselves.

Everyone talks about the fact that it took reconfiguring factories around electricity to benefit from electricity, and follows that with the AI equivalent of “so the new businesses that are built to throw a ton of electricity at the problem will win.” What you really need to do is refactor the businesses into code.

Doing that takes a ground game, going deep into the guts of companies, wherever they are, to understand how they work and migrate their logic into programs. We need people to get out there into Minnesota to be like, what the hell do you guys do all day?

Most of our team are engineers, a lot of them ex-Palantir, who spend weeks at a time on-site with customers, learning from them, getting into the nittiest of gritty details.

The term gets a bad rap, but relative to engineers who spend all day at a desk prompting Claude, they are the most Social Engineers. Engineers who understand people, business, and AI will rule the world. If that sounds like you, come join us at Poetic.

It’s hard work, but the biggest mistake of the AI era so far has been believing that anything worth doing could be easy.

This is worth doing. It’s what I’ve wanted to do for as long as I can remember, because all of the institutions that run our world, every business, every government, does so much stuff that doesn’t make sense, operates way more slowly than it should, and gums up the works.

Our goal is to discover the perfect process for every business - the plan, the set of steps that is ideal for achieving your goals. This will evolve as the world evolves.

The role of AI is not “Agentic,” improvising as it goes. It is an evolutionary force: changing, testing, evaluating what plan is most successful and sticking to it until a better one emerges.

In the end, this is what a business is. A piece of living, evolving software reconfiguring, testing, evaluating itself, hurtling towards ideal form at the fastest clip imaginable. Humans exist to define what good looks like, not how to get there. Shaping behavior, directing behavior.

Through running in production, the system gains a record of what happened. What happened, step by step, for every dispute, underwriting case, insurance claim. Thus, every process change becomes testable. The answer to “what if” is known after minutes of backtesting. Run both scenarios in shadow, compare outcomes, decide which is better.

When impact is entirely known, there is little risk - you know exactly what would have happened. Change simply becomes a choice. Then the choice becomes: which outcome is ideal? The process lead, the person responsible for making sure the process achieves the goals above it, simply makes choices.

To make a change, you have to know what the impact of the change is. It’s easy to generate code, hard to know what happens if you run it. After months of running, you’ll be able to ask questions like “What if we approved every dispute under $25?” and know in minutes.

The hill-climb towards the most beautiful process will then begin. Experts experiment, asking what-if questions. Now that humans are no longer bottlenecks, they can begin searching.

This is endgame.

Every business is not just a piece of software; it’s a piece of software constantly editing, testing, evaluating changes. Evolving at the highest frame-rate possible, climbing towards its most correct form. All energy is spent evolving, figuring out the ideal form of the rules.

We don’t use tokens to run the business. We use tokens to turn the business into code and evolve it. We tokenminn to ROTmaxx.

Over billions of years, we have evolved from ocean slime, through trial and error, into fish, lizards, voles, monkeys, and humans.

We don’t want to have to wait billions of years for businesses to evolve into their diverse and ideal forms, and Agents won’t build them. You cannot build the butterfly.

Beautiful, ideal, complex things can only emerge through evolution. I want to speed it up and see how far we can go.

Thanks to Markie for dropping her knowledge and to Adam and Ben for introducing us!

That’s all for today! We’ll be back in your inbox on Friday with a Weekly Dose.

Thanks for reading,

Packy

Subscribe now

Hi friends 👋,

Happy Monday!

J. Storrs Hall first asked “Where is my flying car?” in 2005. Sixteen years later, in 2021, Stripe Press was able to republish his cult classic because the question remained open. Five years later still, in 2026, a year in which computers write most of the code themselves and medicines seem poised to cure everything that ails us, Hall is still waiting along with the rest of the world.

There are many reasons we don’t yet have flying cars, but there are even more to suggest that now may be the time to build them.

About a year and a half ago, I started to notice Tsung Xu tweeting about a VTOL (vertical takeoff and landing) plane he was building from scratch. Then he kept making progress.

I knew and respected Tsung from his writing on energy and materials - I linked to his work way back in September 2022 - and thought it was cool that he was playing around with drones, but I didn’t think much more of it, until he reached out to ask if I wanted to learn more about the company he was building, Vight.

On that call, we nerded out about the myriad why nows for flying cars, what a practical flying car might look like, and what it would take to make them affordable enough to impact humans’ daily life. I angel invested in Vight, and have gotten a front-row seat to the progress Tsung and team have been making towards delivering Hall a satisfactory answer.

Then, when I wrote that “I have a hunch that drones, EVs, and EVTOLs should expand local frontiers” in April’s Scarce Assets, I asked Tsung - the rare entrepreneur who I knew as a writer first - to write us an essay expanding on how he’s planning to expand the radius of daily life.

The goal for these essays is to understand the bets that people building at the frontier are making with their careers. They are one person’s view, by definition, because I think that the beliefs that makes talented people go all-in on something are the best raw material I can give you for shaping your own beliefs on the future. And the future that Tsung is betting on is one that I want: one where we finally get flying cars.

Let’s get to it.

Today’s Not Boring is brought to you by… Framer

Launch fast. Design beautifully. Build your company’s website on Framer.

Framer helps teams design, build, and launch their marketing sites lightning fast. With the ability to publish hundreds of CMS pages in a single click, operate at a global scale with seamless localization, and even host unified content across multiple domains, teams have never been able to ship faster. Companies like Miro, Bilt, and Perplexity trust Framer to achieve:

• Speed without chaos: ship pages and updates faster without turning the site into a fragile set of one-off hacks.

• Reduced dependency: shift routine brand and marketing work out of product engineering queues.

• Production-grade foundation: Run real marketing systems (CMS, SEO, performance optimization) with governance and collaboration.

👉 Build your company’s site on Framer today

Expanding the Radius of Daily Life

It’s hard to fully imagine how drastically flying cars will change our lives, because we don’t have flying cars yet.

When you think of flying cars, you might think of cars with wings that let you fly right over the traffic from your current home to your current office, or of air taxis that shuttle you between Manhattan and JFK. You think about traveling faster between the places you already go.

That undersells the promise of flying cars. What flying cars do is expand the radius of daily life.

In his 1934 tome Technics and Civilization, Lewis Mumford credits the mathematician Bertrand Russell for discovering what would come to be known as Marchetti’s Constant:

Mr. Bertrand Russell has noted that each improvement in locomotion has increased the area over which people are compelled to move: so that a person who would have had to spend half an hour to walk to work a century ago must still spend half an hour to reach his destination, because the contrivance that would have enabled him to save time had he remained in his original situation now—by driving him to a more distant residential area—effectually cancels out the gain.

As Cesare Marchetti detailed in his 1994 paper Anthropological Invariants in Travel Behavior, “humans are ‘naked apes’ in many dimensions of their behavior involving territory and personal contacts.” It is an anthropological invariant rooted in territorial behavior that when humans access a faster mode of transportation, they use it to travel further instead of traveling less. We are born to spread.

We like to commute about 30 minutes each way, or about an hour a day. Give us a faster vehicle, and we don’t pocket all the time we save. We spend it to reach more of the map.

The ability to fly point-to-point further in 30 minutes than you can drive today promises more optionality: more places where you can live, work, and visit, more often. This might mean living where you can have more land, or by the water, or in a community that fits your values, and still working where the opportunities are. At scale, it might even mean reshaping where economic opportunity exists. In the future, the choice will be yours, and it will happen sooner than you might think.

That is the future I’m working on at Vight. For a number of reasons that we’ll discuss, I believe that now is the time to build the flying car we were promised.

To get there, the experience we build needs Tesla supervised-FSD levels of intuitiveness, flying you the vast majority of the time. It needs to be safe. It needs to be mass manufactured, eventually, like a Model T.

Most importantly, it needs to be fast and offer freedom, to be able to fly to most places cars can drive to, and some places that cars can not.

That is the point of technology, to push out the efficient frontier and eliminate trade-offs. And if we want to explore the physical world nearly as easily, and much more magically, than we explore the digital one, we will need to eliminate the trade-off between speed and freedom.

Speed or freedom: we don’t have both (yet)

For all of our cumulative efforts, we have yet to devise a mode of transportation that is freer than walking. You start wherever you are, and stop wherever you want. You can duck into alleys or buildings, climb up or down stairs, or simply stop and wonder, wherever you’d like.

Walking, though, is slow, so humans have invented myriad modes of transportation that can get us from point A to point B in less time, at the expense of point-to-point freedom.

Bicycles are faster, but they can’t reach quite everywhere you can on foot. A car is faster than either, but they can only take you as far as the edge (or endpoint) of the road network.

There’s another class of vehicles, hub-to-hub vehicles, that offer less point-to-point freedom, but require less work from the traveler and, in the case of commercial airplanes, go much faster in transit. Each requires you to get somewhere, at an appointed time. Riding a bus requires first going to a bus stop, a train leaves from the train station, and an airplane takes off from an airport after some unknown but now-almost-guaranteed delay. Using any of these means real friction getting to and from the hub, and once at your destination airport, you still need another mode of transportation to cover the “last mile” to your destination.

Still, these are the marvels we know. The car has been the best form of point-to-point freedom for the last 120 years in America and other wealthier countries. Aircraft win on raw speed over any surface, land or water. But cars and aircraft each solve only one axis of the problem.

You can see the split in the history of transportation: one lineage optimized for point-to-point freedom, the other for speed between hubs.

The future we want requires both: aircraft speed with car-like point-to-point freedom.

To understand why flying cars are the answer to this particular trade-off, we will traverse the history and near-future of transportation technologies. We are in the midst of a revolution in these technologies across many fronts - from autonomous vehicles to supersonic planes to air taxis - each of which will change the way that we move ourselves and our things. We’ll also find that while each is awesome at certain things, they don’t eliminate the specific trade-off we’re after in the way that flying cars uniquely do.

Flying cars, of course, come with their own challenges, which we’ll discuss, but we believe that expanding peoples’ worlds is worth the fight.

Because where we’re going, we don’t need roads.

The limitations of cars and roads

Thanks to the incredible progress made by Waymo and Tesla, people are starting to get rightly excited about self-driving cars. “Always ten years away,” they’re here now. Just last week, Tesla opened up its unsupervised Robotaxis to serve the entire Austin Metro area.

Self-driving cars will serve a growing portion of our driving miles in the decades to come, and their growth will save many lives and give us time to do other things. I’m personally excited for when they start redesigning self-driving cars as moving lounges instead of traditional cars. But until and unless they replace almost all of the human-driven cars on the road and learn to communicate losslessly with each other, they don’t solve the speed-freedom trade-off, because they’re still cars, and they still need roads.

Cars have been incredibly useful in helping us get to places further away and faster, while protecting us from the elements, but they’re fundamentally a 120-year-old technology that depends on road infrastructure. Roads present chokepoints, are jammed with traffic, and slow us down.

Despite the fact that our cars are much more capable than the Ford Model T, cars in 2026 don’t take us from point A to B much faster in urban or suburban areas than Tin Lizzie did. During rush hour, cars crawl at 17-24 mph in LA and 11-15 mph in New York City. A smaller city like Austin does not fare much better at 14-33 mph.

Adding more lanes to the highway famously doesn’t help much. Traffic is the canonical example of induced demand: add more lanes, and you encourage more people to drive, and the traffic is just as bad. Worse, even, because it’s not the highway itself that’s a chokepoint, but the on and off ramps. Off the highway, traffic lights gum up the works. Even if Autobahn-like unlimited highway speeds were approved in your city, you’re still idling at the next traffic light once you get off during rush hour, and the highway itself will still block up as a result. All that assumes the highway will be built, which is probably too much to ask if you’re living in California, for example.

There’s also the physics problem. Drag increases with the square of velocity. By increasing your speed from 70 mph to 200 mph, you are effectively increasing your drag by ~8x. Power scales with drag x velocity, so about 23x more power is needed at 200 mph vs 70 mph. To go really fast, the range of your car, whether gas or electric, is going to take a massive hit. Driving that fast will massively increase the wear and tear on your tires and other components. Making 200 mph normal, efficient, safe road travel is a completely different problem than making cars drive that fast.

The capabilities of the infrastructure gate the capabilities of the vehicle. You could be driving a Corolla or a Bugatti and it would not matter in cities during peak hour.

That said, roads and later highways did expand the capabilities of cars. The infrastructure answered the question “Where can cars go?”. Carmakers could then invest a century in making them easier to drive and driving down costs to make them usable by more people. Automating the ignition, automatic transmission, and manual then adaptive cruise control were all examples of this. With self-driving today, the easiest to use car is the one that does not need to be driven.

A car’s fundamental job is to get you from A to B. By that standard, a Model Y with supervised FSD or Waymo does that job better than a car that requires you to still drive it. In fact, many people I know primarily drive a Tesla with HW 4 because of FSD. It’s a functional utility, but such a step change in ease of use that they can’t live without it. The best version of a car is no longer the most expensive purist vehicle but the one that gets the most people from A to B with the least effort.

Still, AVs make cars easier to use but don’t solve the challenges with the limitations of roads.

AVs are really convenient, but in the context of our speed-freedom matrix, they’re not any faster. In fact, hailing Waymo is slower than just getting in your car and driving. This is especially true in rush hour. Fundamentally, they don’t expand the radius of daily life.

No matter how much intelligence the vehicle has, the infrastructure layer is the limiting factor in point-to-point speed.

So… go above the infrastructure, right?

The limitations of today’s personal aircraft

Unfortunately, personal aircraft like the Cessna 172 have not changed much in the last 50 years, either.

The biggest upgrade on base models was shifting from “six pack” steam gauges to “glass cockpits,” which effectively offer digitized flight displays for pilots and passengers. The 172 is still one of the best selling GA aircraft by units. Newer 172s give better situational awareness for pilots, but they still require a human to be in-the-loop flying the aircraft with a control stick, throttle and rudder pedals. At $400,000, a new 172 is not even close to affordable, let alone easy to fly.

Personal aircraft are also limited by runway dependence and operating friction. There will only ever be a limited number of people who will purchase a vehicle that is usually stored far away from where their home is.

To fly, the pilot needs to drive to the GA airport, go through a manual pre-flight checklist that could take 15-30 minutes, and crank the engine, perhaps multiple times. Many engines still use a carburetor to mix air and fuel for combustion, components phased out by the automotive industry since the late 1980s.

Personal aircraft never could offer the same point-to-point freedom as cars did. The open skies were gated by sparse airports. Aircraft had speed, but airports and runways answered “Where can it go?” with “To and from an airport.” So the industry optimized for capability without reducing pilot workload; it could not optimize for everyday usability. The result was better aircraft, but not aircraft that more people could use.

Imagine how many people would drive cars if they were not parked at home, but in parking garages miles away. This is the challenge of any hub-to-hub vehicle, minus the scheduling issues in this case, because you can take off and land when you want, and it is the fundamental limitation of personal aircraft.

GA aircraft sales: booms, busts and runway dependence

That limitation is one of the reasons sales of general aviation (GA) aircraft sales have crashed since the 1970s. There are many, but if people were clamoring to fly 172s, I suspect the market would have figured out the rest.

Americans did use to buy GA aircraft, back when they were more affordable. Higher volumes enabled lower prices, and lower prices enabled lower volumes. In the late 1970s, US shipments of GA aircraft were close to 20,000 units per year:

Without looking hard, you can see that shipments fell off a cliff beginning in the late 1970s, falling nearly 90% to 2,000 units in 1985. To understand why “this time might be different” for flying cars, we should understand what’s in the black box on GA aircraft.

First, as always, money. The second 1970s oil shock and the early 1980s recession, combined with high interest rates, made new purchases, especially financed ones, unappealing. At the same time, in an echo of the nuclear industry around the same time, aircraft OEMs had expected that the boom times would continue, and so they overproduced. By the 1980s, technology improvements compared to the decade prior were minimal, there was a glut of used supply practically as capable as new vehicles, and tax incentives pushed people to buy used.

Normally, people blame the lawyers. J. Storrs Hall, too, assigns them a good deal of the blame:

The 1970s brought an increase in product liability, a major social change that was nominally aimed at safety. Scare stories worked on juries just as well as on the reading public, and juries would vote enormous awards for accidents that didn’t have any reasonable connection to malfeasance on the part of a manufacturer . . . This led directly to the collapse of the general aviation industry. Over the course of the 1970s and 80s, it was strangled by the explosion of product-liability lawsuits.

Again, as with nuclear and regulators, that read confuses the timing. Product liability lawsuits locked in the collapse in piston engine aircraft sales that was already underway. Brian Potter wrote a great analysis on this, Planes, Claims, and Automobiles, which includes this quote from John Baker, the former president of the Aircraft Owners and Pilots Association (AOPA):

Product liability judgments are not the cause of the new aircraft shortage in which we find ourselves. It is merely a symptom. The cause of the problem was clearly some unbelievable bad business decisions by the manufacturers 15 to 20 years ago, which is compounded by some lousy products. If the industry was annually producing the 20–25,000 quality products at an affordable price that the marketplace would absorb, then the per-unit product liability insurance costs would not be significantly greater than they were in the mid-70s.

John Baker, 1988 letter, emphasis mine.

So fix demand. Simple, right?

It’s worth a quick look at the very left of the chart above. The peak shipments ever were actually not in the 1970s, but in 1946, at 35,000 units. As pilots retired, there was a subsidy for flight training and most of the flight hours were from instructional flights. When the new pilots were trained and subsidies went away, the demand did, too.

Market gluts are not sustainable drivers of demand. For that, you need new technologies that actually improve the product in material ways. For that, though, you need demand. It’s a vicious circle when it works against you.

The lack of technological improvements was a major reason the GA aircraft industry never recovered after the 1970s. New aircraft engine programs cost around $100 million, and no one wanted to take the risk of investing in better technology in such a small market. Even when companies promised new compelling concepts in GA, the aircraft still ran into the same problems of runway dependence and difficulty to fly.

What about GA helicopters? They can take off and land vertically (VTOL), but lose out on ease of flying and safety perceptions. There is a high level of pilot skill required to learn to fly a helicopter safely, as well as the ongoing maintenance of those skills. In order to remove much of that pilot workload, you’d need to purchase stability augmentation and autopilot systems. But your autopilot system will still cost around $50k+ on top of an already expensive $500k helicopter like an R44. Because they’re complex machines and prone to mechanical failure, helicopters also place a high maintenance and operational burden with manual pre-flight and post-flight checks to ensure safety.

(Packy note: Kobe Bryant died in a helicopter crash on my birthday and Puja and I once watched a helicopter crash into New York’s East River, so I don’t plan to ever fly in a helicopter.)

Helicopters are getting upgraded, though. Skyryse takes existing aircraft and removes all mechanical controls, installs a digital fly-by-wire system, and replaces the cyclic, collective, and tail rotor pedals with a single control inceptor (like a specialized joystick) and a pair of intuitive touchscreens. It makes helicopters much easier to fly, and adds autonomy to the mix. At the moment, however, Skyryse’s system costs ~$400k more than a SAAS upgrade on an R44, and is being used on more expensive craft, like the Blackhawk Helicopter.

Helicopters are cool, but they, too, don’t technologically solve the trade-off for regular people.

Today, though, we finally have new technology: eVTOLs.

What about Air Taxis?

Going from a metro airport to downtown, air taxis can be faster than taking a car, but they are still hub-based. They have about as much point-to-point freedom as taking a bus, in that you still need another mode of transportation to get to and from the vertiports they plan on using. Once you’re there, though, air taxis go further, faster than buses can.

Air taxis use VTOL technology, which stands for vertical takeoff and landing. As the mouthful of a name suggests, this technology makes precise takeoff and landing possible. Air taxis demonstrated that electric VTOLs (eVTOLs) can fly, but they are limited by the constraints of the takeoff and landing hubs they must operate from.

These hubs, the takeoff and landing infrastructure for air taxis, are called vertiports. With commercial services, they are in many ways like a train station. People need a place to wait for their ride. Trains are awesome, and they expand the world too, so we’re excited for the possibilities that air taxis open up. Adding new options on the hub-to-hub frontier is good, because places that are currently inaccessible for urban workers will become accessible.

It’s just that we don’t think they’re the end game, and they face their own challenges.

Charging and maintenance logistics are tricky, for example. Where should the vehicles be kept when charging? There’s a lot of complexity on the ground to be managed. Plus, it’s challenging to get permits for vertiports, especially when sited in dense urban areas, and it becomes more challenging when you need to establish a network of them for any of them to be useful.

Air taxi companies have also been constrained by the regulatory pathways available to them. The regulatory burden is a large reason why US-based air taxi companies have been in development for 8 to over 15 years but are not yet in commercial operations, although that is showing signs of thawing slowly.

The air taxi model likely works better in dense cities, and particularly in places like China, where it’s much easier to stand up urban vertiports. Even with pilotless operations, though, Chinese air taxi company EHang’s cost structure is high. Focused on tourism as their first wedge, 298 RMB ($45) per person would “basically cover flight costs”for two people for up to ~20 minute flights, as per a recent earnings call.

These costs should come down with scale. Part of the challenge is that EHang’s fleet is still fewer than 10 aircraft. But ground ops, low utilization, and vertiport overhead add layers of economic difficulty. The extra friction associated with putting landing pads on buildings in crowded cities - even in China, let alone America, makes it much harder for air taxis to come down the learning curve.

That said, American companies like Joby and Archer are making real progress.

Joby has logged over a thousand test flights across three countries, completed Stage 4 of FAA Type Certification (the furthest any eVTOL has gotten in the US) and is now in Stage 5, the final phase, with a type certificate targeted for late 2026. They’ve demonstrated JFK-to-Manhattan routes and been selected for the FAA’s eVTOL Integration Pilot Program across 13 states.

Archer closed Phase 3 of its own certification process in April, is flight testing Midnight nearly every day at Hawthorne Airport in LA, and is building out a Stellantis-backed manufacturing facility in Georgia targeting 650 units per year by 2030.

Both companies have proven that eVTOL aircraft can fly reliably in real-world conditions, navigate complex certification processes, and attract serious capital and partnerships: Delta and Toyota for Joby, United and Stellantis for Archer. That work has de-risked the core technology for the entire category, including us. And in the short-term, they will offer transit options that we don’t and unlock routes that we won’t. We are targeting different initial markets, and we want both to succeed.

In the future, cities will be dotted with vertiports, but I believe that can only happen meaningfully once the unit economics close and operational friction can be removed. Then, demand will merit the investment necessary to make vertiports work.

Air taxis will replace and improve on some hub-to-hub transport modes. They will not deliver full point-to-point freedom, though, which is what we are attempting to achieve. We want to give people their own flying cars.

In that pursuit, there are several key lessons we’ve learned from air taxi companies:

• Regulations are destiny and constrain the GTM pathway,

• Slow certification pathways drag iteration loops, and

• eVTOL flight dynamics are hard but have been proven solvable.

Many people conflate air taxis and eVTOLs, whereas air taxis are just one application of electric VTOL technology.

That’s why I use “flying cars” in this essay, even though the aerospace industry says eVTOL. eVTOL describes the technology. Flying car describes the category we’re focused on: aircraft speed with car-like point-to-point freedom. Calling this an eVTOL is a little like calling an automobile a horseless carriage. Technically understandable, but not how normal people think about the thing once it becomes truly useful.

Extreme speeds are coming, but not for daily travel

It’s worth noting some other ways we’ll move faster and further in the future. There are a few archetypes that are being developed today that focus on extreme speed, hub-to-hub.

In ten to twenty years we could be taking supersonic or hypersonic aircraft, leveraging the very low drag in the upper atmosphere to fly at Mach 2 or higher across the Pacific, across the Atlantic, or anywhere on the planet.

Suborbital rockets like Starship would make these trips less comfortable, but would take you from any launch site to any landing site at up to Mach 25. Again, these are hub-to-hub options, and the hub, for the foreseeable future, is a launch pad somewhere far enough from people to launch rockets from. They optimize for speed but at the cost of point-to-point freedom of the vehicle.

As a result, you need very powerful and large vehicles that don’t make sense for point-to-point use. These are better suited for moving groups of passengers very quickly over very large distances. It’s a logical but radical extension of the hub model.

The future of long-distance travel will be supersonic and maybe suborbital, and we can’t wait, but these journeys will fall outside of Marchetti’s constant.

For fast, point-to-point travel that expands where we can go in 30 minutes or an hour, there’s one option left.

Flying cars offer speed and point-to-point freedom

It should be pretty clear by now that no widely used or anticipated vehicle category offers both speed and point-to-point freedom. The tradeoff we’ve faced for as long as we’ve built tools to augment how far and fast we can travel still exists.

Flying cars will break the trade-off.

Let’s define a flying car. Most people think of a flying car as a car that flies and can travel on roads. I think it’s simpler than that.

A flying car is really a vehicle that flies point-to-point with car-like convenience.

People will disagree and say, “That’s not a flying car. It can’t use roads.” As we talked about earlier, roads are an infrastructure constraint. Flying cars will not need to be roadable to be useful.

They have VTOL capabilities, allowing for takeoff and landing anywhere there’s a small pad. That means vertiports, eventually, but it also means right in your backyard.

At Vight, our first vehicle platform will cruise at about 110 mph, and future platforms will target close to 300 mph. Plus (and this is also very important), they will dramatically reduce the workload required to pilot an aircraft. This trifecta has never been possible before.

Flying cars fill the empty quadrant in our 2x2:

VTOL changes the answer to “Where can it go?”. Software-defined flight changes the answer to “How hard is it to operate?”. Put them together, and personal aviation can finally make flying dramatically easier and more accessible. Cars spent a hundred years turning a vehicle that already had endpoint freedom into something easier, cheaper, safer, and eventually intelligent. Flying cars start with the benefit of that entire software and autonomy stack being mature.

That is why we can now target a step change in the ease of flying a personal aircraft. For most of aviation history, that was not possible. You could add better instruments, better autopilots, or better training, but the aircraft was still fundamentally a pilot-workload machine tied to runways and airports. A practical flying car will use intelligence to absorb more of the workload, while also removing the runway as the core endpoint constraint.

So why not just make the runway shorter? Couldn’t STOLs (Short Take Off and Landing) be the answer for faster point-to-point travel?

After all, short takeoffs with an STOL bush plane means you can take off with as little as 50 feet of runway.

The problem is they are still dependent on a runway. Yes, runways can be shorter, but STOLs still need clear approaches, clearance over obstacles like trees, dependence on wind directions and maintenance of the surfaces. If you put all of these together, that ends up requiring about ~5x more land relative to a personal-use eVTOL. In addition, pilots need to be more skilled and have less margin for error on takeoff and landing.

Flying cars, done right, can offer a different experience.

The pilot experience for a flying car

You tap your phone, set your destination, and start the automated preflight checklist. The battery starts pre-conditioning and the onboard system runs safety and other checks for you, all in the background. You walk outside to your Vight aircraft on your pad. You take a quick walk around the vehicle, hop in, press a button on the single control stick, and take off.

Once off the ground, you will be able to use the advanced autonomous assist features. Tap to confirm transitioning from hover to forward flight, and when approaching your destination, tap again to confirm the landing zone. At any time during the flight, you can direct the aircraft using the control stick. It should be a nearly pedestrian, boring experience. When flying with the autonomous capabilities we will have for our first-gen aircraft, it will be as simple as that. The experience is far closer to that of just getting in your Tesla and driving with FSD than it is to that of piloting a personal aircraft.

We can only leverage this breakthrough if safety and redundancy are built into the design from day one. Our aircraft will use redundant motors, battery packs, flight computers, sensors, and other safety-critical components, with the system designed so that no single failure should lead to loss of the aircraft. It will intelligently handle situations like a motor or single battery pack failure as a pilot-assist feature, while keeping the pilot in control. We’re sizing the propulsion system to handle weather like wind gusts at up to 10,000 ft above sea level. And in the extremely unlikely case of a complete loss of power, the vehicle will deploy a whole-aircraft parachute.

Just as critically, we’re designing to dramatically reduce pilot workload, one of the major contributors to the pilot-related causes that dominate GA aircraft accidents. In AOPA’s 2023 non-commercial accident data, pilot-related causes accounted for two thirds of aircraft fatal accidents and almost 80% of helicopter fatal accidents. Simplifying controls, having guardrails during flight, and intelligent pilot-assist features help avoid the primary cause of accidents from happening.

The design breakthrough with flying cars is that it can operate like both an eVTOL and a fixed wing aircraft, depending on which phase of the flight you’re in.

The takeoff and landing phases are vertical or near vertical, which gives the precise endpoint capabilities we’ve been talking about. During the transition phase, the wings tilt forward, shifting the lift production from the rotors to the wings. Once the aircraft transitions, it climbs on its wings, at a forward airspeed of over 100 mph and much higher efficiency than in hover. It then stays wingborne for the vast majority of the flight, like an aircraft, before transitioning back to hover for landing.

For eVTOLs, cruise is several times more efficient than hover, so maximizing time flying like an aircraft is important. Not to scale.

This is possible because a multicopter’s core flight control system is already software-defined. The pilot doesn’t have to directly handle aircraft stability, whether the vehicle is a 250 gram drone or a 1200 kg eVTOL. The onboard flight controller does that continuously, adjusting RPM for each rotor hundreds of times per second to maintain stable hover. Without that stabilization layer, a multirotor would be effectively unflyable by a human using a stick alone.

Our platform’s autonomy and VTOL capabilities enable dramatically lower pilot workload, while still maintaining control that mimics how a great pilot would fly a helicopter. A helicopter is very hard to fly, and especially to hover, without simplified flight controls, and the flight controls in helicopters today are not simple.

This lack of easy-to-use controls was one of the reasons GA aircraft and helicopters weren’t able to reach the mass market. Flying is very skill-intensive.

With supervised autonomy, however, we can have guard rails through the whole flight envelope, as well as auto-takeoff, auto-landing, and intelligent sensing. On our first generation platform, these are all sophisticated pilot-assist functionality so the pilot is always in control. That’s only possible when a pilot-defined system becomes software defined.

Beyond that, the Vight flying experience will simplify things like radio and air traffic control (ATC), or getting updated altimeter readings (i.e. for barometric altitude) while coming into landing. Today, even small cockpit tasks are manual for piloting a small aircraft. A pilot may still get the local altimeter setting by listening to an ATIS/AWOS broadcast, copying down the pressure value, and setting it in the barometric altimeter before departure or arrival. It’s not hard, but it is an example of many GA-specific microtasks that have to happen correctly. This is why pilots lean on checklists, but to make flight as simple and normal as driving, they should not have to.

For Vight, our customers will still need a private pilot’s license (PPL) that is required to fly any GA aircraft. That said, we’ll standardize and accelerate this experience, thanks to MOSAIC, which we’ll discuss more shortly. Customers will do either a week-long intensive course or finish over a few weekends. They will learn on site about the nuances of our simplified approach to aviating, navigating and communicating (A-N-C). Because of the pilot workload required to A-N-C, attaining a PPL has traditionally taken 60 to 80 hours. With simplified controls, we will target the 40 hours minimum of flight time required to attain a PPL. This is similar to the amount of hours required to become a competent driver, with states like CA, NY and FL all requiring 50 hours of driving experience for teenagers before granting a license.

We’ll give our pilots the confidence to know what to do in a low-probability failure event, like a motor failure or battery pack failure. They’ll know how to land safely and how to deploy the whole vehicle parachute in a failsafe scenario.

That experience sounds simple, as it needs to be. But the technology to deliver this experience wasn’t feasible until now.

Why flying cars now?

To have car-like convenience, we need to build VTOL aircraft that are easy to fly, with supervised autonomous capabilities from the start. They will also need to be fast, have useful range and be low cost to purchase, operate and maintain.

This requires the vehicle have several key underlying technologies and principles:

• Software-defined systems and autonomous capabilities

• Electric powertrains

• Designed for higher volumes and standardized manufacturing

• A regulatory path

Let’s walk through each of these.

Software-defined flight closes the sim-to-real loop

The most important technology for a flying robot is not actually the vehicle itself. It’s the development process for closing the gap from simulation to reality for how the vehicle flies and performs.

That loop has historically been slow for novel aircraft. High-fidelity models were expensive, specialized software was locked behind costly licenses, and compute was slow and expensive. Closing the gap between simulation and flight test took time.

The best hardware companies turned this loop into an advantage. In their first version of their Roadster, Tesla used a Matlab simulation to model the whole vehicle. In the early days of NVIDIA, when running out of cash, they bet on emulation software to test the RIVA 128 GPU design virtually. This allowed the GPU to be designed in only six months instead of 18 months. SpaceX made reusable rockets possible through tight integration of simulation, flight software, sensors, actuators, and control loops.

Flying cars need the same type of iterative cycle. The faster we close the sim-to-real gap, the faster the aircraft becomes safer, easier to fly, and more capable.

Agentic engineering accelerates this further. Agents in the loop in experiments maintain requirements and do other grunt work. Engineers still make the design decisions and integrate learnings from results to close the loop. We’ve been agentic-native from our inception at Vight. All engineering, compliance and ops are in repos that give the latest frontier models access to do much of the grunt work.

Autonomy reduces and eventually removes pilot workload

Aerial autonomy is not easy, but it has a different shape than road autonomy. A car in a city has to reason about pedestrians, cyclists, lane markings, traffic lights, construction, and other vehicles. Once an aircraft is in the air, the environment is more structured and less cluttered.

The enabling stack has also gotten much better. Cameras, inertial sensors, GPS, LiDAR, onboard compute, and vision models have all improved dramatically while costs have fallen. Tesla’s FSD system is one example of how far camera-based perception has come, but aviation will likely use a more redundant sensor stack. Zipline has made similar strides in their platform’s autonomous capabilities. The components needed for supervised autonomy are mature and standardized and the autonomy problems are hard but solvable.

We will also be able to use synthetic training data before we have large-scale fleet data. Today, Tesla trains a world model on fleet video, generates future camera and sensor data from a given scene and action, and then creates adversarial corner cases that are hard to collect naturally. We aim to apply this to Vight. Rare events, sensor failures, and emergency procedures can be simulated repeatedly before they happen in the real world.

The electric stack makes VTOL practical

We’ve seen remarkable improvements in many standardized components, especially in the last few decades. Performance continues to improve as technologies keep coming down their cost curves.

Batteries

As I’ve written about previously, lithium ion cell and pack costs per kilowatt-hour have come down dramatically since the first commercial cells that Sony used in 1991. They’ve become almost 99% cheaper at a compounding 12.5% annual rate, as deployed volumes have risen about a million fold.

Specific energy and power (both per unit mass) have both increased substantially in that time. Energy density is the biggest lever on range. They are up 3x-5x since 1991 depending on the cell. Improved packaging has increased pack level energy. Power density is actually just as important for VTOL capabilities, as it’s the higher power discharge rates in hover and takeoff that are critical for our use cases. There, 2.5 kW/kg is commercially available at volume and at a low premium over lower power cells.

Batteries are the best baseline energy source for flying cars. They simplify their development and operations, requiring only a standard port, like the SAE J3400 in the US, for up to ~20 kW home and destination charging. We will extend the range on future platforms, using a series hybrid architecture with a piston or turbine generator. In a hybrid architecture, engines can be sized for efficiency in forward flight, and provide additional power on takeoff and landing.

Even in a hybrid powertrain, batteries are still required to provide fast power transients. Hovering in wind, maintaining control margins, and responding to motor commands require near-instantaneous power changes. Engines and generators would take seconds at best to respond, and would be too slow to handle those rapid transients.

Electric motors and power electronics

Electric motors have also gotten dramatically better and cheaper.

Specific power and torque have been improving consistently for permanent magnet synchronous motors (PMSM) for decades now. They are highly efficient at a wide range of rpm, and can be direct drive for less moving parts and complexity. High end commercial PMSM motors like Joby’s have more power per kg than all but the most exotic turboshaft engines.

DC electric motors like PMSM have surged in specific power, from hundreds of W/kg to now high thousands of W/kg, with most of the gains coming in the past decade or so.

High torque at low rpm is also a driving constraint for direct drive motor performance. Fast adjustments in motor rpm for maintaining hover stability require high torque motors to spin up and down rapidly. That’s no longer a problem, as the best commercial PMSM electric motors today are extremely high torque, with specific torque around 50 Nm / kg.

There’s still a ton of headroom here to design more optimized motors, partly because there have been so few motors built specifically for VTOL aircraft. Few iterations have yet happened on designs to continue improving them. There is even a motor designer I know who has used workstations in his garage (and more recently on AWS) to design motors in simulation that match the performance of best-in-class commercial motors, at a fraction of the cost. Expect motors to continue getting better while continuing to get cheaper.

Power electronics have improved alongside the motors. Inverters convert battery DC into precisely controlled three-phase current for the motors, which means thrust can be commanded almost instantly and very precisely. Silicon carbide power electronics in inverters enable high voltage and efficiency in smaller form factors. That means inverters are becoming an integrated part of the propulsion system rather than a mass penalty. Plus, they continue to get cheaper as they scale.

Embedded systems

The whole electric flight stack has been rapidly maturing. Sensors have become cheap and precise enough that drones and flying cars can measure their attitude, position and health in real-time. Embedded computers have become powerful enough to make that understanding actionable for control and navigation. Both the sensors and computer have also become much lighter, even as their performance has improved. This is the reason an iPhone or drone has better sensing and computation than aircraft had access to for most of aviation history.

This might surprise most, but the embedded compute supply chain is one layer of the Electric Stack not dependent on China. OEMs like ST Micro and TI operate global manufacturing networks. That said, commodity parts can’t simply be dropped into a certified aircraft, but the playing field has certainly changed. Redundant and reliable embedded systems that are low cost and lightweight are now practical.

The small aircraft cost premium can collapse by an order of magnitude

No laws of physics say that flying cars should be much more expensive than automotive at scale. But today, there’s more than an order of magnitude delta between the cost structure of a Cessna vs that of a Tesla.

Low volumes explain much of that difference. The chart below (log scale on both axes) shows vehicle price adjusted by weight against annual unit volumes. It’s a good-enough proxy for cost structure differences between the vehicles. Robinson sells around 100 R22 helicopters per year, Cessna around 200 172s. These aircraft designs are fundamentally unchanged except for optional glass cockpits, touchscreen flight displays. Contrast that with a Ferrari 296 GTB with far higher performance engines, and they are still about 2x the cost per pound.

Look at the clear pattern of the falling cost/lb for different types of vehicles. Performance cars at low volumes cost more than a Model 3 or Corolla at much higher volumes. The Nighthawk is a low-volume, high-cost premium boat, vs the Bayliner M17. The pattern holds across structural materials as well, including composites. For example, see the Ferrari 296 vs BMW i3 cost/lb and unit volumes, with the i3 having a composite monocoque structure but at a much lower cost/lb.

It’s very, very hard to make things at low cost when only a hundred are made per year.

Product value is upstream of manufacturing volume. As we discussed above, the job for a car of going from A to B is often best done by one that drives itself. Software-defined intelligence is the logical conclusion for cars that are easy to use. Once a vehicle clears the baseline for acceleration, highway speed, range, and safety, reducing the driver’s burden becomes one of the most important forms of capability. Runway dependence and high pilot workload kept aircraft optimized around skilled performance. Ease of use, and the technology that unlocks it, could now become the primary dimension of aircraft design.

So the question is, why do we make so few personal aircraft?

We sprinkled clues throughout, but it’s effectively a high Idiot Index, Elon’s phrase for ratio of finished product cost over input material cost. Personal aircraft have had a high Idiot Index because of the skill required to fly them, their dependence on runways, and a high certification burden. It’s also a result of the low demand and glacial speed of technological progress in most of the GA industry for the last 50 years. You can’t approach the cost of materials if you’re not iterating.

The opportunity is to shrink the cost delta to only a double digit premium over cars. We can do that with two seat electric aircraft designed from the start for high-rate manufacturing, integrating low cost components into redundant systems, and low cost fabrication and assembly.

Composites have gotten cheaper and more reliable at scale. The higher-rate composite processes are becoming increasingly mature, though not widely used in aerospace. The problem for GA aircraft has been that they don’t have enough volume to justify processes like light RTM.

With volume, we can justify building out the flying car supply chain. The industrial base underpinning the tech stack for flying cars has not been decided yet.

The critical components like the high-performance electric motors and scaled-volume composite fabrication that we need are not concentrated in one country. This was an important reason for starting Vight now. We want to do final assembly, and vertically integrate increasingly more aspects of the full technology stack over time in the US: the software, compute, powertrain, structure and more.

The hard part is now execution, not invention

Building flying cars is definitely not easy. But we’re no longer waiting for miracle leaps in technology like batteries. The biggest risk is now integrating a system that offers a compelling product that is easy to use.

The first wave of eVTOL companies forced a lot of the unknowns into the open. Learnings like how to handle complex transition aerodynamic interactions, redundancy of distributed electric propulsion, flight-test discipline. Those are still hard problems. But they are now engineering problems. They do not require a new battery chemistry, a new motor physics, or a new theory of flight.

One non-technical thing that has blocked the future of flying cars has been the regulatory pathway. For flying cars to be feasible today, even with all of the improvements on the tech side, we’d need a new one.

Is there now a regulatory pathway for such a product to exist? Yes indeed.

MOSAIC, the new regulatory path

For decades, personal flying cars were trapped between two regulatory paths: too small to be useful, or too expensive and slow to certify. The FAA’s MOSAIC ruling opens up another, better path.

MOSAIC stands for Modernization of Special Airworthiness Certification. It’s a regulatory path that is a performance-based approach that is designed to enhance safety without overburdening the engineering and testing required. It also opens up more useful VTOL electric aircraft for private use.

MOSAIC is a huge upgrade to the Light Sport Aircraft (LSA) category that had initially been introduced in 2004 to make safe personal flying more accessible and lower the certification burden. Since then, the FAA has seen that light-sport aircraft (LSA) have become as safe as other general aviation aircraft in recent years, giving them the confidence to expand the type and capabilities of LSAs, while intending to make them even safer. The FAA also wanted to encourage people to fly fewer amateur-built aircraft and more factory-built LSAs under MOSAIC.

For the future of flying cars, that unlocks all these capabilities that we’ve been talking about.

MOSAIC makes critical changes to unlocking the high point-to-point freedom future at high speeds. It allows, for the first time, VTOL-capable general aviation aircraft to be certified without having to go down the very expensive 21.17(b) path that air taxis had to follow, or be constrained into a small form factor like a Jetson or Pivotal ultralight.

It unlocks building VTOL-capable platforms, initially with two seats with up to 250 knot (287 mph) cruise speed, that can be flown in any airspace you can fly a GA personal aircraft today. There’s no restriction on max takeoff weight.

So why don’t air taxis just try to do this? Because their aircraft, companies, and certification programs were built around a different product. A 21.17(b) air taxi is a commercial aircraft for an operator network. A MOSAIC aircraft has to be designed from the start around the personal-aircraft pathway to be viable.

This is why the first market really matters, and why we’re taking a different approach.

The initial opportunity

Flying cars won’t start on downtown rooftops at first. The better initial opportunity is with owners of private properties who already pay for vehicles that expand what they can reach. Examples of such vehicles include utility terrain vehicles (UTVs) and powerboats.

An acreage sized property becomes a home base from which our customers can take off and land. A second property, a trailhead, winery, golf course, or resort become potential destinations. Vight aircraft will be intuitive and easy to fly, right from your backyard.

Many of our early backers want the vehicle for personal use. They’re tech-forward, excited by a vehicle that expands their possibilities for leisure and work at hard-to-reach locations. In addition, helicopter pilots already fly routes like this, albeit with significant workload. With the owner’s permission and the right local approvals, they can take off and land without needing runways.

To have real point-to-point freedom, our vehicles need to operate near where customers live, and eventually work. That makes noise a core design requirement. The first wave of eVTOL companies has already shown good progress, with aircraft that are dramatically quieter than helicopters in both hover and cruise. They got there by optimizing the aircraft as an integrated system, and the propeller blade design in particular. This is one of the places where Vight can leverage the hard-won learnings of what came before.

It’s easy to forget the leisure vehicle market is already large. Boats and RVs are not the same product, but they prove that people spend serious money on leisure-focused vehicles. The US powerboat market is about a 200,000-unit-per-year market, and about $15 billion. The RV market is also tens of billions of dollars and ships over 300,000 units per year. In the US alone, there are more than 10 million registered powerboats and RVs. Those numbers are more than twice as large globally, but the US is by far the biggest market for these vehicles.

This is how large the leisure-only use case for human-piloted VTOLs is already, without considering the flying or autonomous capabilities of our platform on day one and as we continue executing.

Another compelling early market is for flight clubs and schools. Today, flight clubs are member-run non profit organizations that let pilots share the cost of owning and using aircraft. The limiting factor is they are runway-dependent. Runway-independent aircraft allow for flying experiences to move closer to the places people already travel for leisure, work, and community.

This is a small market compared to our vision, to be sure, but a necessary one. We will take whatever ravenous demand exists in the early days, because scale will be how we win. Once we’ve ramped up to about a thousand units per year for the first platform, we’ll have the ability to scale rapidly down the cost curve.

We are targeting a purchase price of $200k-$250k at those volumes. We’ll continue to obsess about simplifying and automating the experience, improving speed and range, and shrinking the on-ground footprint. Of course, we don’t intend to keep the purchase price out of reach for most people for too long. By our fourth generation platform, we are targeting a sub $100k price point, and in partnership with the FAA, we aim to make the flying experience as effortless as taking a Waymo is today.

The master plan for new software-defined hardware has been codified for a while. The Tesla Roadster is the classic vehicle example of this, with about 1,000 units of the Roadster per year at similar inflation-adjusted prices to what we’re targeting. Henry Ford shipped multiple vehicles before the Model T became a breakout success, even more so in rural than urban areas.

As personal transport became more capable, people spent more on it. Cars became a better investment as highways were built out and vehicles became safer and easier to use. The Model T entered production in 1908 and over the next century, transportation rose from a negligible share of household spending to over 15%.

I saw this pattern up close. My dad immigrated to Australia when I was three. In his first year, my dad would walk for hours looking for work. His first car was a beat-up $500 manual early 80s Ford Meteor. His second was a $2,000 Holden Camira, an automatic but it still ran on leaded gas. He washed that car proudly. A year later, though, he bought a brand-new 1994 Toyota Camry.

That is what better transportation does. It stops being just an expense and starts becoming a bigger part of daily life.

With each vehicle generation, we’ll leverage our learnings to build more capable, more intelligent platforms that cost less and unlock significantly more demand. Then, we’ll continue down the cost curve, opening up even more demand.

This is the use case that’s eluded aviation forever. But once we have vehicles shipped to paying customers, Vight can then convert learnings from those early flights into an operating flywheel.

The operating flywheel

We have a long way to go to get from where we are today to our full vision, but our bet is that operating from private property compounds.

Takeoff and landing sites are owned or controlled by our customers, avoiding the need for public vertiports with high buildout cost and permitting friction. Even more importantly, private properties unlock high point-to-point freedom, enabling flights to happen where our customers will need them. More flights happen, across more geographies and uses.

Every aircraft and every flight teaches us how the vehicle behaves and how customers actually use it. It also helps speed up iterations on improving maintenance, site operations, and what makes the experience feel safe and simple.

Those lessons feed five loops at once:

1. Aircraft model loop

2. Customer experience loop

3. Endpoint/destination loop

4. Cost loop

5. Autonomy and services loop

1. Aircraft model loop

For development, the impact implications are massive as we can feed that back into both our simulation model, closing the gap on sim-to-real and better predicting how our platforms will fly before they are in the air.

2. Customer experience loop

For customers, it means we can improve the flying and maintenance experience through the vehicle life cycle. This includes over-the-air updates for non-safety critical features, including autonomous pilot-assist, and predictive maintenance, such as swapping battery packs near end of usable life.

3. Endpoint/destination loop

For endpoints (i.e. destinations), each site gives us a clearer operational playbook for what works. Property and flight club owners can use that playbook to activate more sites, especially at places like resorts, clubs, private communities, and industrial parks.

Indeed, fly-in communities already exist where residents can access by plane or helicopter. There are a few hundred in the US, and about 30,000 people live in them. The largest, Spruce Creek Fly-In, has about 5,000 residents in about 1,500 homes spread over about 1,300 acres. Others are spread all over the country, with even some that occupy small islands like North Captiva Island in Florida. Some are suburban, like Lakeway in Austin, and others are remote, like Alpine Park in Wyoming or Mountain Air in North Carolina.

Over time we’ll see shared endpoints emerge for approved users. These will not need the infrastructure required for an urban vertiport or runways. We’ll also be able to piggyback on any existing vertiport build-out by other companies, as we won’t gain an advantage by owning and building out this infrastructure. That said, we will help with standardized operating playbooks to make setting up infrastructure as easy as possible for property owners. We expect many land owners to be motivated to set up infrastructure that enhances the usage of their properties, and want to make this as seamless and profitable for them as possible.

4. Cost loop

We’ll also come down the cost curve by making better platforms at lower cost. Cruise speed, range, intelligence, noise reduction, and hybrid powertrains will all improve vehicle capabilities, increasing demand to keep pulling costs down. The goal is to drive down the unit costs toward a car-like cost structure, with only a double-digit premium at scale, as we discussed in the cost premium section above.

5. Autonomy/services loop

Over time, more pilots utilize the network. This accelerates the operating flywheel and lets us double down on future platforms and more advanced autonomy.

As our fleet compounds flight hours and we continue to engage with the FAA towards end-to-end unsupervised autonomy. We’ll standardize and automate maintenance, servicing and site ops, accelerating the advantage that will deepen with more flight hours. Our advantage will be the real world data lead that accelerates our ability to validate our autonomy models in sim. Simulation will help scale that evidence, handling every edge case required for safe, reliable truly autonomous flight and streamlined operations, starting with smaller pilot programs for fully autonomous operations. Once proven, that enables a network of autonomous flying robotaxis that do not require vehicle ownership.

When people picture flying cars at scale, they usually imagine cities filled with aircraft moving between rooftops. It’s true that they will increasingly fly to business campuses, industrial parks, and eventually more urban corridors, and this will drive real demand. But I don’t think that is the biggest unlock.

The biggest unlock is actually expanding the map: making places that are currently too far, too remote, or too inconvenient feel close enough to become part of daily life.

Expanding the radius of daily life

The biggest impact of flying cars will be expanding the radius of daily life. The promise is optionality: more places where you can live, work and visit.

No one can predict the future, but also, this future will not happen without highly motivated people aligned on the goal. We think this future could happen in three phases:

• Existing acreages and remote destinations become more valuable.

• More people move into areas that most benefit from flying cars.

• Cities lose their monopoly on proximity (but also benefit in this future).

Being able to travel point-to-point at 120 mph expands where we can go on a daily basis. There’s no limitation from traffic or roads that have to follow the contours of the land. What was once a weekend trip becomes a quick hop. As we upgrade cruise speeds to approach 300 mph and beyond, our daily radius will continue to expand:

Marchetti’s Constant has proven so reliable throughout human history and in so many regions that he called it an “invariant.” The hard part is expanding where people can get within 30-60 minutes. If you pull that off, it seems inevitable that people will take advantage of the speed to spread out, and when they do, we can expect some interesting second- and third-order effects.

Existing acreages and remote destinations become more valuable. Over several product generations, when you can compress a 200 mile / 3-4 hour drive into a 45 minute flight, the land inside the larger circles becomes a compelling real estate play (not investment advice). Packy was one of the first people I spoke to who understood how this expands the local frontier, as he wrote about in Scarce Assets.

More people move into areas that most benefit from flying cars. Before cars and highways, suburbia was not a thing. Suburbs shaped where many people live today, defined by cars as the primary mode of transport in almost every American city. They made access to downtown areas closer for people living further away from them.

Autonomous and fast long-range eVTOLs will also enable a new pattern of living. It will be easier to live on a ranch, a forest cabin, or new developments like CA Forever. Mountain or island homes that were too disconnected to be practical will become normal places to live, work, and build. That’s only possible when the people, activities and amenities take a fraction of the time to reach.

People in remote areas already have access to Starlink. They have battery backup and can be off-grid with solar if they choose to. Small modular electrolyzers and direct air-capture will enable on-site synthetic hydrocarbon production. Drone delivery services will be normal. Autonomous trucks will move heavier and bulkier supplies to wherever people want to live.

The hinterlands become more valuable, more compelling, more usable. Regional economies can become more connected without needing to route everything through the same city center. Flying cars will decentralize economic opportunity.

Cities lose their monopoly on proximity (but also benefit in this future). Cities will not just disappear. They benefit too, mostly through services and shared endpoints. Flying car garages in cities might look like the automated car parking warehouses that have existed for decades. Ground robots or automated tugs will move aircraft between landing pads, charging bays, maintenance areas, and dense storage slots, the same way automated parking systems already move cars through compact garages.

But cities stop being the default hubs for opportunity. As more people, goods, energy, data and eventually manufacturing organize around new endpoints, the frontier of daily life moves outward. The suburbs give way to more distributed communities and an incredibly exciting future on our planet.

If this all sounds too utopian for you, this vision will take decades and multiple product generations. The flying car layer of this future is very hard. We’ll need to make the experience safe, affordable, quiet enough, and easy enough to operate. We’ll need to earn the trust of our customers, property owners, insurers, local authorities, and the FAA. Future platforms with unsupervised autonomous capabilities will require many iterations of our operational flywheel, and ongoing engagement with regulators. But it’s a good quest and, as we’ve shown, it’s possible. Now we just need to do it.

This is just the beginning

It is too narrow to think of flying cars only as vehicles. They are autonomous robotics platforms operating in open physical space. Many robotics categories will automate tasks inside the world we already inhabit: homes, warehouses, factories, sidewalks.

Flying cars do something different. They change the shape of the world people can inhabit. That is why the better analogy is the car and the highway network. They changed where people lived, what land was valuable, how regions grew, and what freedom on the open road felt like.

Flying cars will help usher in the next paradigm for lived environments. Cars and highways enabled suburban life for tens, if not hundreds, of millions of families in developed countries. Flying cars will unlock new ways of living in new locations for just as many, if not more, people in the decades ahead. They will unlock freedom in the open skies.

Growing up, each new transportation mode changed the size of my world. My first bike gave me suburban reach in Adelaide, Australia, letting me visit my friends on weekends without needing a parent driver. My first car expanded that reach. Both made more places feel reachable and made my life feel less bounded. While the specific places are different, this growing sense of freedom is a universal experience.

That freedom, the expansion of what is possible in daily life, is the promise of flying cars.

Nothing worth doing is easy. Making life incredible on Earth will be extraordinarily hard. But if we’re locked in on the mission and we execute well, the path is becoming clear. We can do this.

If you want to help make this future real, come build it with us.

Thanks to Tsung for writing and teaching me about flying cars.

That’s all for today. We’ll be back in your inbox this week.

Thanks for reading,

Packy

Hi friends 👋 ,

Happy Friday and welcome back to our 196th Weekly Dose of Optimism!

A new American nuclear reactor has gone critical. Dan and I are doing a Hyrox in a few hours. New York City is hot and buzzing. What a week for the optimists.

Let’s get to it.

Today’s Weekly Dose is brought to you by… Guru

When it comes to implementing AI within your company, most orgs struggle with a knowledge problem. You can plug in all the AI tools you want, but if your internal info is scattered, outdated, or contradictory, you just get the wrong answers faster.

Guru fixes that.

It’s a central, governed knowledge layer that sits underneath your tools and actually makes them useful. Teams like Spotify and Brex use Guru to:

• Keep company knowledge in one place (and actually maintained)

• Verify information so it stays accurate over time

• Power AI tools with trusted, up-to-date context

• Deliver answers with sources, not guesses

Fix the input once with Guru, and it improves every output, across every tool your team uses. If you’re investing in AI, Guru can help make it actually work.

Try Guru Today

(1) Antares Goes Critical

Big day for the U S of A, and for our friends at Antares.

Last night, Antares announced that its Mark-0 low power reactor was brought to criticality at Idaho National Lab with a self-sustaining fission reaction. In doing so, it became the first novel reactor design to undergo a fueled test in over 50 years.

“We are now the first reactor to meet the intent of President Trump’s May 2025 EO 14301,” CEO Jordan Bramble said, “which calls for three reactors to meet this milestone before America’s 250th birthday on July 4, 2026.”

This is a massive milestone, and one that seemed implausible even a year ago. It’s also just the beginning - more reactors will go critical in the coming month, and Antares itself is moving from testing to producing electricity. Or as Jordan put it, “We’ve made neutrons. Next up: electrons.”

Two and a half years ago, Julia DeWahl and I started a podcast called Age of Miracles to try to understand why America no longer made this miracle technology, and what it might take to change that. We talked to most of the founders in the space, some of whom are racing to criticality alongside Antares. Turns out, Julia, the co-founder of Antares, was the one who pulled it off, alongside Jordan and the rest of the team. It is an incredible accomplishment to go from “Why aren’t we doing this?” to doing it - going critical - in under three years. I’m very proud of Julia for making it happen, and excited for the country now that we’re producing new nuclear in America again. 🇺🇸

(2) New Limit Raises $435M to Bring First Aging Reprogramming Medicine into Human Trials & Cure Hangovers

for

There are no approved treatments for alcohol-related liver disease. It kills 30,000 Americans a year, about 40% of the population has some degree of fatty liver, and nothing.

Until now. NewLimit, the epigenetic reprogramming company co-founded by Coinbase CEO Brian Armstrong and computational biologist Jacob Kimmel, just closed a $435 million Series C led by Founders Fund with Thrive, Greenoaks, and others at a $3.1 billion valuation, more than triple what the company was worth a year ago. For good reason - NewLimit is going into human trials with the first-ever test of an age-reprogramming medicine in human subjects, starting in 2027 in Australia, after success in mice.

NewLimit fed old mice alcohol as their sole calorie source for 11 days and then gave them a binge dose. The untreated old mice flipped onto their backs and stayed sedated for 8 to 12 hours, like me if I have a couple drinks now. Young mice on the same protocol just popped up like it was nothing. But when old mice got a single dose of NewLimit’s therapy, RNA encoding specific transcription factors, delivered via lipid nanoparticles, they stopped passing out. They were as good as young.

As Kimmel told Ashlee Vance on Core Memory, “It is just binary. They’re either passed out or they’re not.” The therapy didn’t help them metabolize alcohol faster. It restored the liver cells’ intrinsic ability to withstand stress and regenerate, as if they were young again.

The discovery engine behind those transcription factor combos is an AI model called Ambrosia that ingests published gene-function literature, protein sequences, and DNA binding motifs, then trains on roughly 10,000 real lab experiments. It explains more than half the variation in results and can be run in reverse: you specify the cell state you want, and Ambrosia proposes the combination most likely to produce it. As frontier LLMs and protein-folding models improve, the embeddings Ambrosia feeds on get better for free.

NewLimit initially expected a decade-plus timeline to reach the clinic. They’re now five years ahead of schedule. The first indication will be steatotic liver disease delivered via a 20-minute IV infusion, with the long-term goal of a subcutaneous pen for broader GLP-1-like use.

Beyond the liver, they’re engineering nanoparticles to target blood vessel lining (with a focus on chronic kidney disease), T-cells (autoimmune conditions), and eventually the blood-brain barrier.

The sooner they get through their roadmap, the longer we all live.

(3) Top AI CEOs Call for Law Protecting Against Biological Weapons

Amrith Ramkumar for WSJ

AI for bio will be used for a lot of good. We need to prevent it from being used for bad.

The leaders of the big AI labs don’t agree on much these days. Over the past week, for example, OpenAI CEO Sam Altman said that he doesn’t like Anthropic’s telling everyone they’re going to lose their jobs in one interview and that AI budgeting has become a huge issue for some companies, which would hurt both OpenAI and Anthropic but which I think he’s probably OK saying out loud because Anthropic is the one that just filed an S-1. “Sam Altman. You could parachute him into an island full of cannibals and come back in 5 years and he’d be the king.”

One thing they do agree on, though, is that people shouldn’t be able to use their products to make bioweapons. Sam, Anthropic CEO Dario Amodei, and Google DeepMind CEO Demis Hassabis are among the signatories on a new letter urging Congress to pass laws that would require companies that sell synthetic DNA and RNA to screen their customers and block any combinations that could be dangerous.

Per the WSJ, “Trump previously revoked a Biden-era executive order that resulted in a gene synthesis screening framework. The White House last year said it would replace the Biden framework with its own screening guidelines but hasn’t yet published a replacement policy.”

Congress should probably go ahead and just do this. “AI is going to kill us all” is better left an Anthropic marketing tactic than a real scenario.

(4) Helion Raises $465M to Accelerate Commercial Fusion Development

On Age of Miracles, one of the people Julia and I interviewed was Helion CEO David Kirtley. Coming in, we’d heard from people in fusion that Helion was brash, and more “Silicon Valley” than others. Its SpaceX-like cadence of building new generations while still testing older ones was aggressive.

We both came out of the interview incredibly impressed, and believing he might just pull it off.

This week, Helion got one step closer. It announced a $465 million Series G led by Thrive Capital (big week), bringing its total funding to $1.5 billion and valuing the company at $15.5 billion.

In February, its Polaris became the first privately developed fusion machine to demonstrate measurable deuterium-tritium fusion and hit plasma temperatures of 150 million degrees Celsius, or ten times the heat of the core of the sun. It’s also the first private fusion machine to operate with D-T fuel, after becoming the first company to receive regulatory approval to possess and use tritium for fusion energy production. Meanwhile, Orion, Helion’s 50-megawatt facility in Malaga, Washington and first commercial fusion power plant, is already under construction, with a contract to sell electricity to Microsoft starting in 2028.

Helion will use the money to scale manufacturing from Polaris, which proved the physics, to Orion, which is an actual power plant that aims to produce electricity that customers actually use using the same energy-generation process used by the sun.

It’s fusion, and it’s really hard, and there’s lots to prove, but the funding will help the company take a swing at the goal Kirtley laid our in our conversation: to make “generators per day rather than generators every few years” by 2030. Ab sole.

(5) 80-year-old Alzheimer’s Patient Showed Recovery with Mushrooms

Frontiers in Neuroscience

An octogenarian Japanese-American woman with a 10-year history of Alzheimer's disease, including five years of near-total functional decline, monosyllabic speech, chronic urinary incontinence, flat affect, and dependence in virtually all daily activities, was given a single 5-gram dose of psilocybin mushrooms.

Nineteen hours later, after fits of intense sweating, she woke up and “the patient spontaneously initiated autobiographical conversation lasting several hours.” Over the following days and weeks, her urinary continence came back after five-plus years, and she began walking independently, dressing herself, sustaining eye contact, retrieving contextual memory contextual, and engaging emotionally.

A second session a month later produced spontaneous humor, vivid emotional imagery, and increased agility. She told her caregivers, unprompted: “It is pleasant to come here.”

This was a single case and not a clinical trial. The authors are careful to note that causality can’t be established, and the improvements were transient. Whether or not it was the mushrooms, what’s incredible is that, even after 5 years, her capacity for all of those things was still in there. As the researchers put it, “Residual functional capacity may persist in advanced Alzheimer's disease and may become transiently accessible under specific neuromodulatory conditions.”

My grandmother had Alzheimer’s, and me and my loved ones getting it is one of my biggest fears, so any good news about this terrible disease is welcome, and I want to believe. If taking some mushrooms helps fight Alzheimer’s, sign me up for the clinical trials.

EXTRA DOSES: No Science Breakthroughs this week (back next), but we have a lecture on the grid & batteries, MAFIA, Hoffman, and Resonant Computing below…

Read more

Subscribe now

Hi friends 👋,

Happy Tuesday! Welcome to a Deep Dive I’ve been excited to write since The Electric Slide, about a company I invested in to build one layer of the Electric Stack - motors and actuators - in America.

If you’ve heard of Westmag, it might be because they make really great hats. Everyone loves the Westmag hats. I’m wearing one right now, which co-founder David Hansen gave me right off his head. This hat, I think, in a decade or two, will be a collector’s item.

I think that because over the past year, in stealth, Westmag has been making a lot more than hats. It’s been making motors and actuators, and it’s been making them to scale, so that in a decade or two, when we sit on the grass looking up at drone-decorated skies while robots do our chores, those drones and robots will be driven by Westmag motors and actuators.

Let’s get to it.

Today’s Not Boring is brought to you by… Framer

Launch fast. Design beautifully. Build your company’s website on Framer.

Framer helps teams design, build, and launch their marketing sites lightning fast. With the ability to publish hundreds of CMS pages in a single click, operate at a global scale with seamless localization, and even host unified content across multiple domains, teams have never been able to ship faster. Companies like Miro, Bilt, and Perplexity trust Framer to achieve:

• Speed without chaos: ship pages and updates faster without turning the site into a fragile set of one-off hacks.

• Reduced dependency: shift routine brand and marketing work out of product engineering queues.

• Production-grade foundation: Run real marketing systems (CMS, SEO, performance optimization) with governance and collaboration.

👉 Build your company’s site on Framer today

America Spins on Westmag

If you drive out to South San Francisco, past and away from the city’s “Stop Hiring Humans” billboards and into the industrial part, you will find a low-slung facility from which David Hansen and Jordan Sanders plan to sling American-made motors and actuators to the American companies making drones, robots, and eventually, anything that moves under the power of electrons.

It is called Westmag, a portmanteau of Western and Magnetics, which pretty much sums up the mission. Electric motors are spinning magnets, and if America is to participate in the coming electric revolution, we are going to need to make them in the West.

This is the kind of statement you read on Twitter, at varying degrees of jingoism, as in “Get those Chinese motors out of my yard.” But it isn’t self-evident that we need Western motors, or, therefore, Westmag. The promise of globalization was cheaply-made Chinese motors powering expensively-designed American products for the benefit of all mankind.

Certainly, we want to design our own robots and drones. Intellectual Property is what America does best. We probably want to manufacture them here, or in a friendly country, too. Even if you don’t believe the China hawks’ predictions that there will be a hot Great Power Conflict within the decade, it is better to err on the side of not putting a heavy, software-controlled, and easily-bugged Chinese hunk of metal inside every American home, or, eventually, putting every American human inside a flying, software-controlled, and easily-bugged Chinese hunk of metal to fly through the sky.

But where do you draw the line?

Should you make every magnet that goes inside every motor in the USA? Should you mine the rare earths that make the magnets here, mine them elsewhere and refine them here, or import them? The rare earths are commodities, after all, and impossible to bug or remotely control with known technology. And even assuming that you want to make everything here, is there a company to be built doing so? Who in their right mind would want to compete with China’s massive scale and subsidy advantages?

American motors seem to violate both David Ricardo’s concept of Comparative Advantage and Michael Porter’s Five Forces. And yet, Westmag exists, and I invested in the company, alongside a16z, Founders Fund, Lux Capital, NFDG, Menlo Ventures, and my Electric Slide co-author, Sam D’Amico.

In that essay, I wrote, “I even just invested in a stealth company making electric motors.” Westmag is that company. We also shared the reason America wants a company like Westmag, and many similar ones in all areas of the Electric Stack, to exist in America:

Manufacturing and design are inextricably linked. When you make things, you learn how to make them better. You learn which parts of the underlying stack need to be improved, improve them, and make better products. This is a theme that comes up over and over again in our Electric Stack story.

In the Electric Era, maintaining design leadership without manufacturing leadership is not a coherent strategic position, and one that gets less coherent the better you believe AI will get.

Elon has solved this by building practically everything inside of his own companies. There is a parallel American ecosystem emerging to serve the growing number of electric companies not in the Elonverse. Westmag will be the winning motor company in this ecosystem, and will, in turn, enable its customers to win.

In a time of experimentation and innovation, like the one drones and robots are in today, if you place component manufacturing near product manufacturing, the whole machine spins faster. This is particularly true when, to say nothing of geopolitics and everything of commerce, American companies are producing at volumes at which they are not Chinese component suppliers’ top priority.

But that is half the story, a geopolitical imperative, not a corporate strategic one.

The other half of the story is why a specific company like Westmag should exist, and how it can generate persistent differential returns in a market in which practically zero-margin Chinese alternatives exist. While we will tell the whole thing, that’s the half we’ll focus on today.

It’s a story about Westmag specifically, and motors and actuators specifically, but it’s also the story of how to ride the Red, White, and Blue premium for just as long as you need to without counting on it long-term in order to kickstart demand and ride that growing volume down the cost/performance curve until, without subsidies and without patriotic premia, you can use scale and proximity and flexibility and speed to compete on overall system cost and win.

This, to be sure, is a scale game, and that’s not the type of game we’ve played well recently. As David said, “Motors are capped at 100% efficiency, so most improvements are incremental. So you can go from 89% to 91%, but it turns out that doesn’t matter at all if you don’t make it and get it adopted at scale.” Previous attempts to compete in motors were focused on those efficiency gains or theoretical breakthroughs at the expense of the ability to quickly reach meaningful adoption and scale, he continued:

Actually building a lot of it is the only way to get good at building it. China’s strength, which we are replicating, is building a lot of things and then improving it along the way. Bespoke low volume doesn’t create a manufacturing powerhouse with compounding advantages.

Westmag plans to build a manufacturing powerhouse by focusing on scale. As Jordan put it:

We are first focused on scale: scaling what works now and what is in demand now, while in parallel innovating on, and through scaling that, we will drive this virtuous feedback loop of innovations in how we manufacture and how we design motors, both for manufacturability but also for performance. You only get good at the stuff if you build a lot of it. And then you only win the market if you can actually get it out to the market in large numbers.

So this is a story about why and how to build electric motors in America for economic reasons, how to build them at scale, and how to win. And it’s a story about what it will mean for the rest of America’s electric ecosystem, alongside which Westmag is growing up, if it does.

It begins with what electric motors are, why they’re important, and where they’re made today.

Electric Motor Primer

In The Electric Slide, we used the electric motor as the vehicle for understanding the whole Electric Stack, because the motor is where everything comes together. The batteries supply power to the electromagnets that create rotating fields that pull the Neodymium magnets around and around, coordinated by the embedded compute that takes in data from sensors and tells the power electronics how to flip the current thousands of times per second to create the smooth rotation that turns electrical power into mechanical power.

Specifically, in a brushless DC motor, which power drones and the actuators inside humanoid robots, the stator, the part that holds still, is a ring of copper coils wound around teeth of laminated electrical steel, and the rotor, the part that spins, is studded with permanent neodymium magnets.

The controller fires the stator coils in a precise rotating sequence. Three phases switch on and off thousands of times per second, so that the magnetic field appears to whirl smoothly around the stator. The permanent magnets in the rotor chase that rotating field and never quite catch it, and that perpetual chase is what spins the shaft. As we wrote:

The magnetic force is doing the spinning. Everything else is about getting the magnets in the right place with the right polarity; nature does the rest.

An electric motor simply directs electromagnetic forces that want to move towards equilibrium.

The “brushless” part is an upgrade from older motors, which used carbon brushes to physically scrape current onto a rotating commutator, a mechanical hack that worked well enough but wore out, sparked, and capped achievable speeds. If you swap the brushes for electronics, you get a motor that is quieter, more efficient, more controllable, capable of tens of thousands of RPM, and good for a billion rotations before anything wears out. Which is why it has become the motor of choice for almost everything that needs to move precisely under software control, which is an increasingly large number of things.

If you want a deeper understanding of how electric motors work, check out our explainer, How Electric Motors Work, or watch this video:

For our purposes, what you need to know is that any electric product that moves is basically a bunch of actuators, of which motors are a subset, converting some form of energy into physical motion, wrapped in bodies that allow them to understand the world around them and move certain ways in response. The bill-of-materials (BOM) for a humanoid robot, for example, is roughly 50% actuators with motors and magnets at their core, which act as their joints.

“Motors are simple,” Jordan told me, someone to whom motors are not simple, when I visited the proto-motor factory. “They’re just magnets and copper wire wrapped around some electrical steel.”

From a materials perspective, motors are simple. Making motors, however, turning those materials into a precision component at scale, is much more complex.

A brushless DC motor is a bundle of compounding tolerance requirements. The electrical steel arrives as thin sheets that have to be stamped into laminations, coated to prevent eddy currents, and stacked into a stator with the layers aligned to within a hair. Copper wire, sometimes thinner than a human hair itself, has to be wound around the stator teeth at a precise tension, in a precise pattern, with as much copper crammed into the available slot area as physically possible.

Then there are the magnets. Neodymium magnets get pressed into the rotor at orientations specified to a fraction of a degree, and magnetized in place by fixtures that fire enormous bursts of current through coils to align the magnetic domains. Increasingly, you can get blocks of neo magnets in the US, but what you get is a slab of metal that isn’t yet magnetized or cut to the right shape. To turn block into a usable motor magnet you have to cut it (small-motor magnets are curved, and the curve is not easy), shape it, coat it, and magnetize it. No one does this in America today, so if you want to do it right and quick (i.e. if you don’t want to send the American magnet block to China or Malaysia and back), you probably need to do it yourself. If you get the orientation slightly wrong, your field ends up lumpy. If your field is lumpy, your motor cogs, vibrates, and wastes energy.

Stack those tolerances on top of each other (lamination, winding, magnet orientation, rotor balance, bearing fit, etc…) and you start to see why motor manufacturing is mostly a process problem, not a materials problem. The materials are simple. The process is annoyingly precise.

That is why, as much as their total dominance of the rare earth magnet supply chain or (no longer) cheap labor, China is so good at this. Over the past thirty years, they have made a lot of motors, and in the process, they have written a library of institutional knowledge. The winding machines in Shenzhen have been refined by a thousand small revisions. Chinese line workers know what a properly-wound stator feels like in their hands. Their test rigs have been calibrated against millions of motors.

This knowledge compounds for you, just like the tolerances compound against you, and it’s only if the knowledge wins that you can produce a lot of motors, cheaply and reliably. For the past three decades, China has been doing all of the compounding.

China Spins Up Motor Production

Like every layer of the Electric Stack, electric motors were invented in the West and Japan. Michael Faraday of cage fame built the proto-motor in London in 1821. A decade later, in Princeton, Joseph Henry discovered that you could make incredibly powerful electromagnets by wrapping insulated wire around iron cores.

Forty years later, in 1871, Zénobe Gramme built the first commercially successful generator, using electromagnets for the field magnets, powered by some of the current it generated itself in a process called self-excitation. In the 1950s and 1960s, American researchers began replacing mechanical commutators with electronic switching, and in the 1960s and 1970s, Japanese firms like Yaskawa, Panasonic, Sony, and later Mabuchi aggressively productized compact permanent‑magnet motors as transistors and then MOSFETs got cheaper. These were the first truly mass‑manufactured, consumer‑scale BLDC motors.

In 1983, in a story that you need to read if you haven’t, Sumitomo’s Masato Sagawa (Japan) and GM’s John Croat (US) independently discover Nd₂Fe₁₄B, the modern neo magnet, and presented their findings at the same conference in Pittsburgh.

Sumitomo perfected sintered high‑performance blocks, while GM’s Magnequench division perfects bonded molded magnets, both of which were used in neodymium magnets’ alpha product: 3.5” hard-disk drives (HDD), which relied on two small electric motors, voice coil motors and spindle motors.

As 3.5” HDDs overtook 5.25” HDDs, neo magnets swept the market.

And as neo magnets scaled, they and the motors they powered got cheaper, unlocking new use cases, more scale, better price-for-performance, and therefore more use cases, and more scale, in a virtuous cycle that we are still riding today. You can read the full story in The Electric Slide.

Today, however, the products that run on electric motors don’t run on American-made electric motors or neo magnets.

In 1995, GM, under financial pressure, sold 80% of Magnequench for $70M to a “US‑led” consortium that was, in reality, two PRC‑controlled companies led by Deng Xiaoping’s sons‑in‑law. CFIUS approved the deal on the condition of a 5‑year pledge to keep production in the US. Long before the five-year pledge expired, Magnequench’s Chinese owners had already cloned the Indiana lines in Tianjin. By 2003, the US plant shut down.

In parallel, as part of Xiaoping’s long-term plan, China came to dominate rare earth mining and manufacturing, including the mining and manufacturing of neodymium. By the early 2000s, having undercut them on price and environmental standards, China forced the US’ only big rare earth mine, Mountain Pass, into bankruptcy, and came to control the full rare‑earth → NdFeB magnet chain that is essential for high-performance BLDC rotors.

But supply without demand does not an industrial superpower make. Enter Shenzhen.

Shenzhen was a small fishing village of about 30,000 people across the border from Hong Kong before Xiaoping, as part of his “reform and opening” policy, designated it the opening country’s first Special Economic Zone in May 1980.

Throughout the 1980s and into the 1990s, China’s quasi-capitalist city exploded. Cheap labor poured in from across China to fill the demand for hands: Shenzhen had quickly become a hub for assembly and low-end manufacturing. Soon, dozens of contract manufacturers were making toys, watches, and all manner of cheap electronic devices.

BYD started in Shenzhen, to execute an arbitrage: reverse engineer Japanese battery manufacturing processes and replace all of the expensive automation with Shenzhen’s cheap, abundant labor. Over time, Wang Chuanfu’s company built out Shenzhen’s battery supply chain while becoming the world leader in electric vehicles, one feeding the other. If you make the components, you can make better products.

Johnson Electric, founded in Hong Kong in 1959 by Wang Seng Liang and his wife, set out to manufacture miniature DC motors specifically for the booming Hong Kong toy industry. When Xiaoping opened the Pearl River Delta to ~capitalism, Johnson moved production across the border, like many of Hong Kong’s electronics companies. By the 1990s, more than 80% of Hong Kong’s factories had moved to the mainland, mostly into the Pearl River Delta. Per the (surprisingly shitty) Porter’s Five Forces website entry for Johnson Electric, “the 1980s show a shift toward application‑specific motion solutions as automation and automotive electrification rose.” By the 1990s and early 2000s, “Johnson Electric established mainland China production and verticalized stamping, molding, and magnetics to protect margins; it diversified from brushed motors into BLDC, stepper, linear actuators and subassemblies.”

Thanks in part to the competence that Johnson Electric had built in motor manufacturing, Shenzhen became the epicenter of the RC hobby industry, which refined brushless motors specifically. The next big thing will start out looking like a toy.

So by the time that Frank Wang used the proceeds from selling flight-control parts to universities and Chinese power companies to move to Shenzhen and start Da-Jiang Innovations, or DJI, the city was already the place where almost every component a flying camera needs was already being made within a couple hours’ drive of his apartment, by people who had been making them for years. RC hobbyists were refining the brushless motors, and the speed controllers to drive them. BYD had already been making lithium cells and packs for a decade. Camera modules were widely available hand-me-downs from the smartphone industry that had turned Shenzhen into the best place on Earth to turn a bare Sony sensor into a working, calibrated eye for a few dollars. Gimbals, plastics, PCBs, radios, and GPS units were available on a quick stroll through Huaqiangbei, the city’s electronics district, where Wang could pull from some 30 billion components crammed into a single square mile, get a custom PCB turned around in 90 minutes, and go from sketch to prototype in a few days.

Cheap labor had attracted BYD, Johnson Electric, and countless other manufacturers to Shenzhen, but it was the components and expertise they’d built with that labor that made the city an ideal place to start a drone company in 2006.

But precisely because components were so widely available, a hundred copycats could pull them off the same shelves. So DJI started vertically integrating. It built the flight controller first, because that was the part Wang understood best, and the cost fell from several thousand dollars in the mid-2000s to a few hundred by 2012. Then they developed the gimbal in-house and shrunk until it cost a tenth of the professional rig it replaced. Then the camera. Then, eventually, the propulsion system itself, the motor and ESC and propeller, designed as a single matched unit.

Today, DJI makes something like 70-80% of the drones in the world, and it is, by itself, supplying itself, the largest drone motor manufacturer in the world by a wide margin. It got so good at making motors that it’s even started selling its Avinox e-bike motors to other companies.

Thanks in large part to DJI, but also to the ecosystem it grew up in and Xiaoping’s foresight, to the fact that someone buying a motor can also buy a battery and power electronics and custom-made PCBs right next door, China absolutely dominates the world’s production of drone motors.

That same expertise allows them to dominate the world’s production of robot actuators, each of which has a drone motor at its core.

Which means, along with all of those batteries and power electronics and custom-made PCBs, that China absolutely dominates the world’s production of drones and robots.

Because if you want to build products on the Electric Stack, it is critical to have fast-turn components available nearby.

Why America Needs Its Own Motor Company

On its face, it is not bad that China dominates drone motor and actuator production. In a frictionless utopia, it would be great.

A company in one country (say, America) would design a drone or a robot or anything that moves, they would send the specs for the components they need to a bunch of other countries (like, for example, China) where they could be made best-for-the-cost, those countries would make the components and ship them near-instantly back to the buyer, the buyer would assemble those components into a finished drone or robot or whatever, and it would sell them to customers around the world, each of whom would benefit from a product made in the most efficient way possible.

Sure, America could design and manufacture here, and there was a time in the 1970s when we were better at both designing and manufacturing than China, but if we’re better at designing than manufacturing, and if design captures more of the value, we can design here and manufacture in China, which has a comparative advantage in manufacturing.

Comparative advantage is a concept coined by the economist David Ricardo in 1817 to explain why countries engage in international trade even when one country’s workers are more efficient at producing every single good than workers in other countries. When Nobel Laureate Paul Samuelson was challenged to “name me one proposition in all of the social sciences which is both true and non-trivial” by the mathematician Stanislaw Ulam, he thought for a few years and came back with comparative advantage: “That it is logically true need not be argued before a mathematician; that it is not trivial is attested by the thousands of important and intelligent men who have never been able to grasp the doctrine for themselves or to believe it after it was explained to them.”

This is the logic that policymakers and economists used to justify globalization and the World Trade Organization, and it fueled China’s rise.

From the reform era through WTO accession in 2001 and into the 2000s, China was labor-abundant and capital-and-skill-scarce, so it specialized in labor-intensive assembly while the US specialized in the capital-, IP-, and skill-intensive ends. For a while there, it worked as planned. Americans were smiling all the way to the bank.

The Smiling Curve depicts how “value added varies across the different stages of bringing a product on to the market in an IT-related manufacturing industry,” and therefore, where value is captured.

This was the dream. China would sit low in the middle while America captured value on both sides. The iPhone is a canonical example. Kraemer, Linden, and Dedrick’s 2010 teardown found that Apple captured 58.5% of the value of the iPhone 4/3G in profits, while China’s labor earned just 1.8% of the value.

But China, as we’ve seen, didn’t plan to stay on the bottom lip, and it didn’t, because doing the assembly taught it the adjacent capabilities. Assembling electronics pulled it into making components, then the tooling and machines that make components, then design itself. Serving as the world’s factory was a thirty-year education that pulled China up the smile curve and across the product space into denser, more complex nodes. Over time, the country’s edge became capability instead of labor cost.

Today, in a specific set of sectors, China holds an absolute advantage, with the lowest cost and highest capability and most complete stack, all at once, such that capital and capability rationally flow toward China rather than away. The advantage rests on scale economies (largest volume → furthest down the learning curve → lowest unit cost), agglomeration (the Shenzhen/Pearl River cluster where all of the inputs, tools, and skill sit within a short radius), accumulated process knowledge, and vertical integration.

This is the situation we described in The Electric Slide. China doesn’t hold an advantage in everything, but it certainly does hold the advantage in the Electric Stack.

“Today, China produces 75% of lithium-ion batteries globally and manufactures 90% of the neodymium magnets that make motors spin. In power electronics and embedded compute, it’s rapidly gaining ground.” As a result, the world’s leading drone company (DJI), electric vehicle company (BYD), and humanoid robotics (Unitree, although the market is still very small) are all Chinese. As the WSJ reported, even Tesla is turning to China for Optimus’ actuators.

Read statically, at a moment in time in the late 20th Century, comparative advantage correctly identified assembly as the low-margin thing for America to offload. Read dynamically, however, low-margin assembly was the tuition that China paid to climb up the value ladder into absolute advantage over America in a category that I believe will define the future.

Again, there is a way to read all of this as “China Bad,” which isn’t particularly interesting. Certainly, if China is currently America’s greatest adversary and largest geopolitical threat, it is not ideal that they produce the magnets, motors, and batteries on which our drones, future humanoid soldiers, and all manner of electric vehicles run. It is in America’s defense interest to incentivize the production and consumption of American components by American companies, and it is doing that, as we will discuss.

But America’s goal should not simply be to survive militarily, but to thrive economically through the Electric Era, when almost everything we combust fuel to power today, and many things that aren’t currently possible, will need to be rebuilt on the Electric Stack.

And to do that, we will need to manufacture key components here, right next to the companies that make the products that use them because maintaining design leadership without manufacturing leadership is not a coherent strategic position.

In 2010, former Intel CEO and Silicon Valley legend Andy Grove wrote an editorial for Bloomberg Businessweek titled How to Make an American Job.

In it, he argued that it was unsustainable for Silicon Valley to pay a small number of Americans increasingly high compensation while hollowing out manufacturing jobs, and correctly predicted that China’s lithium-ion battery dominance would lead to EV dominance.

There’s more at stake than exported jobs. With some technologies, both scaling and innovation take place overseas.

What microprocessors are to computing, batteries are to electric vehicles. Unlike with microprocessors, the U.S. share of lithium-ion battery production is tiny.

That’s a problem. A new industry needs an effective ecosystem in which technology knowhow accumulates, experience builds on experience, and close relationships develop between supplier and customer. The U.S. lost its lead in batteries 30 years ago when it stopped making consumer electronics devices. Whoever made batteries then gained the exposure and relationships needed to supply batteries for the more demanding PC laptop market, and then after that, for the even more demanding automobile market. U.S. companies did not participate in the first phase and consequently were not in the running for all that followed. I doubt they will ever catch up.

Grove disagrees with the then-(and maybe still-) popular idea “that as long as ‘knowledge work’ stays in the U.S., it doesn’t matter what happens to factory jobs… Not only did we lose an untold number of jobs, we broke the chain of experience that is so important in technological evolution. As happened with batteries, abandoning today’s ‘commodity’ manufacturing can lock you out of tomorrow’s emerging industry.”

Sixteen years on, it’s safe to say that Andy Grove was right.

“Without scaling, we don’t just lose jobs — we lose our hold on new technologies,” he wrote, long before LLMs — so put away your Pangrams. “Losing the ability to scale will ultimately damage our capacity to innovate.”

What I like about Grove’s analysis, apart from his predictions proving correct, is that it was written at a time when more Americans viewed China’s ascendency as a positive than a negative.

It rests on economic and industrial rather than geopolitical logic; it is offensive, not defensive. And he is arguing that to innovate in America, you need to manufacture the key components here, too.

To make great American drones and robots, for example, you need to manufacture motors and actuators in America, too.

The next question is: given the intense competition, is there a way to build a profitable drone motor and actuator company in America?

David and Jordan Decide to Start a Motor Company

It is de rigueur to care about drone motors, and to want to make them in America, but David Hansen has been obsessed with motors and the Chinese manufacture thereof for longer than some would-be motor mavens have been alive, and he has the tweets to prove it.

An archaeological dig through his account finds that he began reply-guying to tweets on IEEE electric motor-related articles in November 2018 with his own YouTube rabbit hole discoveries on BLDCs.

That’s the same year that David went to China for the first time, to walk Huaqiangbei for himself and begin sourcing for the self-balancing, AI-copiloted e-bike company he would cofound in Seattle in 2018, Weel.

Weel custom designed and built its motors and actuators by hand, due to the fact that there wasn’t yet an off-the-shelf Chinese actuator that worked for their needs. Most small component suppliers didn’t even have English-language websites; it just wasn’t worth it, they had all the demand they could get from customers who spoke Mandarin.

But something you learn about David is that he wants to get to the center of things, to speak to the suppliers’ suppliers’ suppliers directly, to see what he can buy from as close to the source as possible, and to learn how the source does what it does. Despite his best efforts, he was basically stonewalled, until COVID-19 happened and the world shut down.

All of a sudden, there wasn’t too much demand to waste time talking to the persistent American with the small orders anymore, and there was plenty of time to waste:

I learned the Chinese supply chain in 2020 and 2021. When China shut down for COVID, the suppliers were all stuck at home, downloaded WhatsApp, and started replying more to Western companies. Over 2020 you suddenly saw them building English websites — small suppliers with a dozen people who’d never have bothered before, since they already had the channels. A lot of factories went direct that year, just as a means of survival.

So 2020 and 2021 is when I started buying a lot of stuff direct from China: magnets, stators, other parts you couldn’t buy from the factory as easily before. Everybody’s stuck at home in both countries, talking over the internet and buying stuff.

A quick Twitter search verifies the timing, because around 2020 is when David, aka @boxcardavid, started becoming the motor guy on Twitter.

(Lore: David’s handle is @boxcardavid because he lived in this train car for most of his 20’s. If you’re trying to make motors or actuators, this is who you’re competing with.)

You could tell the man liked motors back in the COVID times, but he really let his motorhead flag fly starting in 2024, when he started figuring out what to do after Weel. Teardowns, comparisons, technical debates, knockoffs, factories. If you were one of the small handful of Americans who cared about brushless motors before they were cool, you probably followed David.

That same search for the next thing took David on a tour of the American companies whose products relied on motors and actuators. “David was the motor guy on Twitter, traveling around everywhere,” Jordan said. “Everyone said it was a clear problem. A lot of folks wanted to hire him as Head of Hardware or Head of Motors to solve it, but it was unclear what that role would mean. There was a desire, but not necessarily the urgency to reduce their reliance on Chinese suppliers.”

Chinese motors were just too cheap and too high quality, and besides, it didn’t make much sense for any given company, each of which had low volumes and other things to worry about, to spin up their own in-house motor and actuator assembly lines.

From all of these conversations, David was starting to realize that someone might have to build the American brushless motor company. His plan then, in November 2024, was to try to push someone else to do it.

Then our friend Sam D’Amico told him, “do it,” and David said, “k,” and two days later, Western Magnetics Company, a C Corporation named by Sam in a DM, was incorporated.

The company that may represent the West’s best hope to compete in drones and robotics began the way that most great companies begin: with an “lol”.

Incorporation docs are not a company, however, and David kept traveling, trying to figure out what everyone who needed motors and actuators was doing to get them. In early 2025, his odyssey took him to Georgia to see Jordan, who had been an early investor and advisor to Weel and was now Chief Commercial Officer at Slip Robotics, a company that builds robots that load and unload trucks.

“Everyone has motor supply and performance issues,” Jordan told me. “Having worked in robotics for a decade, I’d seen and been part of companies that thought about producing their own motors in-house. Then you look into it and realize, oh man, it’s actually pretty expensive and hard. It just didn’t make sense for one company to do it in-house just to fill their own demand.”

But, David and Jordan put their brains together and thought, it might make sense for one company to aggregate demand and build all of the motors and actuators for the rest of them as a vertically integrated horizontal play. To do all of the things that you’d have to do to get really good at making high-quality motors and actuators, like going as far as needed upstream into the supply chain and building automation and, most importantly, getting to scale to drive all sorts of efficiencies.

“Any competent engineer with focus can build a prototype actuator or a motor,” David told me. “You can build a bad motor really easily, and an okay motor without much practice and using open-sourced designs. Same with actuators. The design has literally been open-sourced since 2018, thanks to the great Ben Katz, who I really want to work with one day.” (Ben, if you’re reading this, go work with David and Jordan.)

“How to do it is known: you can order stuff from China, hand assemble it, and make it happen pretty easily,” he went on. “But doing it at scale is almost unrelated.” So scale is what they’d do.

David and Jordan teamed up to found Westmag, a company that was born to scale by growing up alongside the nascent American drone and robot industries, aggregating their demand, and providing them with the motors and actuators they’d need to make their products move, with the consistency, responsiveness, and respect that those companies were simply too small to get from their Chinese suppliers, on tight timelines, and eventually, with scale, at a similar price.

Their timing couldn’t have been better, although they didn’t know it at the time.

Soon after they decided to start Westmag, before they’d made a single hat, let alone a motor, the US Government would sanction T-Motor, China’s largest drone motor supplier, sending drone companies scrambling for a Western second source, and a little later, robotics would become the topic du jour in San Francisco and in the White House, which had no interest in letting China get another 10+ year head start in one of the future’s most important industries.

None of that was part of the plan. That part was just pure luck.

“What’s that phrase?” David searched his brain. “You can’t count on luck, but luck counts, and I definitely count on it. It’s turned out great. Good to be lucky.”

It’s not that Westmag wouldn’t have worked without the new government-mandated necessity of its products, it’s just David and Jordan expected it to take a lot longer. “Our bet was that this stuff would happen in five to ten years,” David said, “but the wind shifted last year, in the middle of us doing all this, with the government, regulators, and customers.”

And truly, I don’t want to leave you hanging here, and the stories about this stuff are some of the best inside baseball in the piece, including T-Motor’s hasty launch of a sub-brand “Lig Power” (as in, we strongly suspect, Ligma, which, if so, nice work, China…) to get around sanctions, and I promise that I will tell them to you, and that they make Westmag’s position in the market stronger than it would have been otherwise at this point, but the whole point of this essay is that Westmag is not a geopolitical bet, that it stands on its own industrial and economic logic, and that, sanctions or no, Great Power Competition or no, Westmag’s strategy is a smart and well-supported one that will require a Herculean grind to pull off but is certainly possible to pull off, and if it does, answers the question “is there a way to build a profitable drone motor and actuator company in America?” in the affirmative, so we have to pause the storytelling for a moment here, in early 2025, before Westmag knew that it would become more necessary, sooner than it expected, to unpack the strategy.

Westmag’s Strategy: Going Vertical to Go Horizontal

When I wrote the Vertical Integrator series, I had a whole section that I ended up cutting on Porter’s Five Forces and why I don’t normally like the position of doing something really hard in order to earn the right to sell components into a handful of large, powerful, and slow-moving incumbents. Maybe, if there are only a couple of key buyers, they’ll hammer you on price and force you to build to spec. Almost certainly, they’ll move on a procurement timeline so slow it will bleed you dry.

Plus, given where we are, at the dawn of a new techno-economic paradigm, in what Carlota Perez calls an Installation Period, book says vertically integrate: “the Installation Period favors Vertical Integrators and the Deployment Period favors modularized suppliers.” If you’ve created a magical new technology, use it as the core of a new system that competes directly with them. In Power in the Age of Intelligence, I summarized my thinking in a question:

If your technology is so good, why aren’t you using it to compete?

Which is to say, in a very simple reading of the situation, Westmag is not the kind of company I’ve been looking to back.

But the fun thing about all of this, and why I find business strategy so endlessly fascinating, is that there are exceptions to every rule, and understanding where to apply the rule or where to grant an exception requires a deep dive into the specific details of the case at hand.

Take Westmag’s key thesis at the beginning, which Jordan described as: “at this moment, maybe these industries are taking off enough that you can actually aggregate this demand and jump to scale.”

There are no large, powerful American incumbents in drones or humanoid robotics. There is a fragmented market of younger companies that are too small to have much buyer power. Each, as Jordan realized, is too small or stretched too thin to spin up meaningful motor or actuator manufacturing themselves, which means there’s an opportunity. It also means that the normal pitfalls of trying to sell to large, powerful incumbents don’t apply: startups can move fast, and for now, no individual company has enough power (or alternatives) to exert pricing power.

The “for now” part is important, because the plan requires that some of the companies Westmag serves get very big. The key is, Westmag will ride to scale alongside them, and aggregate demand among many of them, so that by the time any drone or robot company gets really big, Westmag will have achieved a scale that none of them can match alone.

Scale, they realized early, is everything in this game, and getting to scale first means that Westmag will be able to do all of the hard things required to actually do this well.

Specifically, it will vertically integrate its own supply chain where needed in ways that are deeply impractical for any single drone or robotics company to do.

Two examples, both of which Jordan and David said were the things that kept them up at night early on but no longer do (as much).

First, as discussed, while there are American companies like MP Materials and Vulcan Elements that can provide neo magnet block today, there is a gap in western suppliers that can cut, coat, and magnetize the magnets to the requirements needed for drone motors. Today, the block gets sent to Malaysia or China for cutting and coating, and back to the customer in the US. Leaving aside the added time or export restrictions, this is an issue: “magnets are fragile and like to stick to each other, so they’re difficult to buy the further away you are,” David explained. “That worried us for a long time. We’re less worried now because we know that world, we know how magnets flow around the world and how to get them.”

Westmag is exploring, by itself or with partners, opening up capacity to cut, coat, and magnetize its own magnets in America. This will be annoying, and come with high upfront costs, but it will give Westmag greater control over its supply chain, lower costs, simpler logistics, and the ability to get to a wider product mix.

Second, Westmag will buy electrical steel from Japan (they are clear that the material supply chain will be US and allied countries, not just US), but it will stamp and powder-coat them here to make their own stators.

Doing those things only makes sense if you’re planning to get to scale, but if you can do them, they allow you to build out a broader catalog more quickly, which helps get more scale, because you can serve a larger number of customers.

In motors and actuators, broadening the catalog to serve a high mix means making “smaller circles and bigger circles,” which sounds simple, but it really helps if you control your design and inputs.

“Planning out the product catalog,” David explained:

We realized that if we make our own stators, it’s not too hard to build a different stator size the next day. But if we have to design everything, send it out, and have someone else stamp it, you’ve got built-in cost and friction. That’s another huge reason to move upstream: it’s what enables higher mix.

On the actuator side, if you want to build a different actuator, controlling the motor matters, because the core of an actuator is the drone motor. If you don’t control that, your constraint is whatever motors you have available. I feel like I should say something smart after that, but it’s just true.

A couple minutes later though, he found something smart to say and came back. “Something else on the actuator side, since you control the motors…”

The design-cycle time for actuators right now, for almost anyone in the US, is in the months, because you rely on overseas suppliers to order pretty much everything. We saw the same thing with drone motors: you don’t just start making American motors that are better than China’s. To get good, you have to make a bunch. Same with actuators. So cutting the iteration time on actuators is super important: if you can only redesign four times a year, how far can you get? We think it’s key to making them reliable.

Moving fast and breaking things, along with scale economies, will bring other advantages, namely, process power.

“You can’t easily pull process knowledge out without literally pulling the people out of the neighborhood,” David acknowledged, “but a way to get it is to build and break much, much faster.”

You can start to see how all of this fits together into a strategy.

First things first, you need to get to scale.

That means, Jordan said, “Going to high-volume customers and asking, ‘What do you have? What are you using now?’ and benchmarking our initial SKUs against what they need in order to get larger offtake agreements against an aggressive manufacturing ramp, which justify making larger buys and investments of input materials.” Quite literally, the plan is to not reinvent the wheel, but to make the wheel that people need right now so that they can get enough wheel orders that it makes sense to make bigger input material orders and even to integrate upstream.

Then, if you can integrate upstream, you can spin your whole process faster.

If you can stamp your own stators, you can offer a larger motor on short-notice, which means cost efficiencies plus more demand and larger orders. Plus, if you control motors of a bunch of different sizes, you can iterate on actuator designs more quickly, and Tasmanian Devil your way into process power in months that would otherwise take years. You can then automate the parts of the process that are amenable to it, and start to gain small advantages over Chinese suppliers.

Eventually, riding drone motor volume demand that exists right now, and using it to fuel its parallel work on actuators, each of which has a drone motor at its core, you get to “high volume, high mix” offering on par with the Chinese, but on the same continent as its customers and more responsive to them.

To get there, they’ll need to operate like a fab in the near-term, a bit like a TSMC for motors and actuators.

Being a fab - serving the demand that customers have today - is how you get volume, and you can use that order to standardize, so that your processes get easier, your margins get fatter, and you become the platform on top of which anyone, from a scaling drone manufacturer to a pre-seed robotics team to a hobbyist high schooler, can build all manner of electric things that fly and roll and grab.

The North Star is essentially the T-Motor catalog with a “Buy Now” button and fast shipping.

All of this is hard but not impossible, and most of it requires getting to scale first and dominating the space. This doesn’t work as well if the drone motor and actuator supplier landscape becomes really fragmented, because it could mean no one gets to the scale required to offer great quality at a great price.

To that end, Westmag has some advantages.

While there are a number of motor startups popping up now that the space is hot, it really helps that Westmag has been building out its production capacity, relationships, know-how, and supply chain since before it was cool. It has a head start.

It also has the right investors. In August 2025, Westmag raised its $11 million Seed Round from a16z, Founders Fund, Lux Capital, NFDG, Menlo Ventures and a group of angels including SendCutSend’s Jim Belosic, Sam, and me. These firms are among the most likely to back something like this over the many years and hundreds of millions, or billions, of dollars it will take to win. Their portfolios are also full of the companies that will be Westmag’s first customers if they can deliver.

As I wrote in Vertical Integrators Part IV, “Among startups, I expect we’ll see much less competition. The companies that show an early ability to execute against a big and credible enough vision will attract the top talent and the limited pool of investors willing to back such hard-to-underwrite companies, sucking the air out of the room for would-be challengers.” In an industry that requires scale more than raw innovation, this is a feature.

Still, the longer it takes to get to scale, the more opportunity there is for new entrants to come in and fragment the market.

Which is why it helps that Westmag got a little lucky.

Westmag Becomes Urgently Needed

Ok we’re back.

When David was running around talking to drone companies in 2024, most of them weren’t urgently trying to get out of China. They knew cerebrally they needed to diversify their supplier base as they scaled, as any company does, and probably would have preferred to diversify into a country other than China, but they were hooked on China’s cheap prices and availability. It just didn’t make sense to spend the reps figuring it out only to get more expensive and possibly lower quality motors, so at most, they’d diversify to two or three companies in China just in case T-Motor, the gold standard, had an issue.

All of that changed on January 15, 2025, five days before President Trump’s second inauguration and a couple of months after David had incorporated Westmag, when the Treasury’s Office of Foreign Assets Control (OFAC) sanctioned Jiangxi Xintuo / T-Motor by name for sending more than $9 million worth of items to Russian companies, including entities involved in Russian UAV production.

“There was a bit of a delayed reaction,” David remembered. They followed the rules and stopped buying, but they figured that some sort of alternative would pop up, and they had some inventory stored up in the meantime.

And there were workarounds. T-Motor unsubtly spun up “Lig Power” - like, if you go to T-Motor’s website and click “North America,” it just takes you to Lig Power, which is not sanctioned but also probably not your best long-term bet.

It wasn’t until April or May, when their motor shelves started to go bare, that they began to freak out. So while there had been companies taking China seriously - Skydio had already been sanctioned by China, and Neros was already serious about the risk - it was May 2025 when the American drone industry as a whole started thinking urgently about dual-sourcing.

Then Jordan left Slip and the duo really got to work, realizing they’d need to scale up faster than previously anticipated, which was both a blessing and a challenge. And the blessings kept coming.

In December 2025, the Federal Communications Commission (FCC) gave Westmag an early Christmas present when it added Uncrewed Aerial Systems (UAS) and UAS Critical Components Produced Abroad to its Covered List. Already-approved systems were grandfathered in, and it didn’t make using drones or motors you’d already bought illegal, but the Covered List meant that the FCC wouldn’t authorize new foreign-made drone components, and the list of components was broad: data transmission devices, communications systems, flight controllers, ground-control stations/controllers, navigation systems, sensors/cameras, batteries/BMS, motors, and associated software.

The FCC’s move was divisive in the industry. On the one hand, it makes sense: we probably don’t want to rely on Chinese motors for drones that may be used in a war with China. On the other hand, it included foreign countries beyond China (although there are exemptions for some allied suppliers) in an effort to support a Buy American agenda, while kneecapping American drone companies’ ability to produce.

In either case, it meant that American drone companies need to dual-source and find reliable domestic suppliers pronto.

That was the drone side. From the beginning, the plan was: there is a drone industry today, with fast-growing order volume today, so start with the drone motors, and use them to ramp production volume. Then, eventually, since every robot actuator has a drone motor inside of it, use the motor to expand into the potentially much larger robot actuator business over time, as that industry ramps.

Each drone has 4 motors, and Ukraine will need 28 million drone motors for 7 million drones this year; each humanoid robot has 20-40 actuators, and by some estimates, we will have 10 million of them in a decade, and three billion by 2060. That puts actuator demand in the tens to low hundreds of billions.

Westmag would have been golden riding drone motors down the learning curve, but the gifts of urgent demand just kept on coming.

In late 2024 and early 2025, when David was talking to robotics companies, he kept hearing that they just couldn’t get the changes they wanted made, that orders were inconsistent, and that Americans were simply at the bottom of the priority list.

You’d order 100 actuators, it’d come in two shipments, and on the second shipment they’d move the connector location or type, change the firmware, or completely change the mechanicals inside.

Jordan and David like to tell the story of a friend of theirs who’s a famous robotics guy over here who found a bug in the firmware for one of these Chinese actuators and sent them a bug fix, and the Chinese manufacturer just didn’t care.

Still, while they were casually looking for other paths, their #1 question back then was, “What is this going to cost?” China wins that question 100 times out of 100 today, especially at lower volume. They wanted to spec their own actuator designs to work with their specific robots, but you could find a contract manufacturer in China who was willing to do that pretty easily. As long as it was roughly to spec and cheap for the quality, they didn’t really care where it came from, so they stayed.

That started to change late last year. Something was in the air.

People began to realize that with AI taking off, robots would be the next big thing, and that, like AI, robots would be hardware constrained. Specifically, they’d need a lot more actuators than we can currently produce.

Quickly, the discussion went from “How much do my actuators cost?” to “Where are we going to get all of the actuators we need in one, two, five years? Can or should all the actuators come from China?”

David is friends with the head of hardware at this one robotics company, and every time they talked last year, the conversation was always about cost. Now, “every text conversation is about it not being in China, having more than one supplier, and planning for if China cuts us off.”

This is the definition of sovereignty in the future as David and Jordan see it: controlling your own robot supply chain.

The US Government seems to be coming around to this definition. Across government and industry, there is a growing belief that while we are about a decade behind in drones and their supply chains, and are taking actions to play catch-up, robotics is so nascent that we can try to be competitive from the jump. People are starting to think about what we can do to be more proactive in the robotics supply chain, which means eating more of the Electric Stack alongside our allies.

Because as we wrote of China’s bet in The Electric Slide, “for intelligence to truly matter, it needs energy and action.” Or as Aaron Slodov put it more pictorially:

Having the smartest computers doesn’t really matter if they can’t act on the physical world. To act on the physical world, they need bodies (whether humanoid, arms, or vehicles). And to build bodies, you need to control enough of the components to build them well.

God Bless the Red, White, and Blue Premium

Westmag’s initial bet was that if you aggregated demand from the nascent American drone and robotics industries, and made it easier for them to iterate quickly and grow, you could grow with them. It was a true bet, because both drone and robotics companies cared about cost above all else.

And Westmag could get to cost parity, with enough volume and time, but how do you solve that initial chicken and egg? How do you get the early orders that give you the volume to bring your price down to competitive levels?

For motors and actuators, as with many industries, from chips to solar panels, the initial push down the learning curve has come from Defense demand.

With the government’s actions, and the industry’s waking up to the fragility of relying on China for its components, price has become a secondary concern for a certain buyer, particularly one selling into Defense. The DoW is willing to pay more for American drones with American components, which means that the companies making American drones can afford to pay more for American motors. This is industrial policy through Defense demand like the type we discussed in Thank God for Data Centers.

This is referred to as the Red, White, and Blue Premium. It is not a long-term strategy. But it is a hell of a bootstrap and jumpstart.

“DC and Defense are near-term feral markets,” Sam D’Amico told me on a call last week, “but long-term, that’s actually a small part of the TAM. Robot actuators and consumer and commercial demand are. All of the electric cars, appliances, computers… brushless motors are in everything. And if you’re long robotics, you’re long actuators.”

Strategically, the RWB Premium is less about the actual price, and more about the fact that today, it’s where the volume is. Everything else flows from that.

Westmag has begun signing contracts with customers focused on Defense-related applications. In just the last few weeks, it has signed offtake agreements for hundreds of thousands of motors, and is in late-stage discussions with many other drone and robotics companies for similar deals. Commercial demand will be important to getting to scale, but it’s not where the highest volume demand is today. As Sam has pointed out many times, American demand is our greatest advantage.

The government could also potentially get more directly involved in lowering Westmag’s prices and increasing its volume, through subsidies, loans, and offtake agreements. There is precedent: it did deals with American neo magnet manufacturers MP Materials and Vulcan Elements last year, giving both access to cheap capital to fund CapEx, and offtake agreements to support scaled manufacturing. Even if the private market won’t buy the magnets (which is unlikely), the government set a price floor.

In either case, the trick is not to rest on Defense-related contracts, but to use them to fund the things that will make Westmag commercially competitive, potentially as a second-source and then as a primary source for growing drone and robotics companies.

That means doing the things that scale allows you to do, quickly.

As discussed, it will mean moving further upstream into magnet cutting and stator stamping. It will mean deeper partnerships with suppliers in the ecosystem, from electrical steel producers in Japan to machine parts manufacturers in Tennessee.

It will also mean expanding into a larger manufacturing facility in the Bay Area this year. While there are good reasons to spread its upstream facilities throughout the country, it makes sense to manufacture right next to the drone and robotics companies that will be its biggest customers, as Shenzhen has proven.

Speaking of things that Shenzhen has proven, in the beginning, it will mean making the motors the exact same way customers expect. That means designs optimized for Asian supply chains, for now: CNC-ing parts with complex shapes out of aluminum, very little design-for-manufacturing, manual assembly.

Over time, however, Westmag plans to adapt its manufacturing to modern American processes. As volume grows, designing a new way of making motors begins to make sense.

Westmag will design for manufacturing and, given that much of its initial costs come from labor, automation. It will redesign the same motor over time so it uses simpler machined parts and more processes that scale with capital instead of labor, like stamping, die-casting, molding, and near-net-shape parts, and so the geometry is something a machine, not a person, can make and assemble.

This is the other reason to build in San Francisco: it’s where the automation and robotics engineers are.

Working with these engineers, it will design the motor and the factory together, because as David put it, “The design of the factory floor and the design of the motor go hand-in-hand.”

One of China’s gifts is that all of its suppliers are in the same neighborhood, but that also means that it doesn’t make sense for any one of them to put everything under one roof. Westmag doesn’t have that luxury, so it will have to.

Surprisingly, Chinese suppliers still do a lot of this stuff manually, like running wires by hand. There aren’t many automated lines. There isn’t really even one “line,” they’ve discovered:

The “factory” in China isn’t usually in one place. There’s a dense neighborhood of specialized subcomponent shops connected by couriers. There are automated machines, but they’re at one address; a part gets built there, a courier takes it an hour later to another location a few blocks away where it is joined with another part, then a couple hours later that goes via courier somewhere else.

So part of what we’re doing, and will continue to do, is to vertically integrate, bring more processes and machines in house, and connect them. Essentially: connect everything by conveyors, not couriers.

The more it can do in-house, in this Installation Period for American motors and actuators, the faster it can spin, and therefore the faster its customers can, too.

The one place it won’t integrate is downstream, into end products like a full drone or robot, because the point of Westmag is to make affordable, reliable, fast, high-quality motors and actuators on top of which every other American electric product can build.

Without knowing exactly where the peculiarities of the supply chain will take it, it’s hard to know when Westmag will get to cost parity with China, but “it’s not crazy five-to-ten-year math at insane scale to be competitive. It’s classic industrial logic. The Idiot Index denominator on these things (the cost of the raw materials) is very low,” David said, “so even when they’re selling them for $20, there’s room for us to grow, scale, and lower the Index.”

That is a high bar, the unit-to-unit cost comparison, a hard-but-achievable-but-probably-not-necessary one to clear, because the real value if Westmag succeeds, and its raison d’être even if the US and China become best friends, is that if you can make critical components near the customers that use them and supply them reliably, the whole innovation machine spins faster.

Spinning America Faster

“A new industry needs an effective ecosystem in which technology knowhow accumulates, experience builds on experience, and close relationships develop between supplier and customer,” Andy Grove wrote, and time has proven him right.

China has built this way and is now winning the Electric Stack; America has not, and is not.

Westmag’s key insight is that while China has an early lead, the race is just beginning. Both drones and robotics are very new industries.

As much as drones dominate the conversation today, thanks to the war in Ukraine, and as important as people think they will continue to be, I expect that they’ll be even bigger. For large industries, Defense is usually a very small initial market in retrospect. Packages and people will fly through the air faster, cheaper, and more efficiently than they crawl through terrestrial traffic today.

Robotics is even more nascent, with unit volumes in the thousands, and live debates as to which form factors and model architectures will win out. What everyone seems to agree on is that robotics is going to get much bigger.

OpenAI is getting back to its roots and back into robots…

Jensen Huang is getting excited about robots…

And just this week, The Wall Street Journal shared PitchBook data showing that venture investment in physical AI and robotics is on pace to pass an already-record 2025 in the first half of 2026 alone.

Wall Street estimates vary, but they all project unit volumes in the millions over the next decade, and an installed base in the billions in the decades beyond. We are currently operating in bars that will be barely visible above the x-axis looking backward.

The question is: who’s going to make all of the drones and robots, and therefore, whose economy will most benefit from their growth?

If you had had to bet, back when Nvidia was founded as a gaming graphics chip company in 1993, whether it, Intel, or AMD would be the biggest thirty years hence, you would have gotten ludicrously good odds on Nvidia. The chip market seemed set, even though, as we now know, it was barely in its infancy.

“Real men have fabs!”, former AMD CEO Jerry Sanders declared about chip manufacturing in the 1980s, and were he operating today, he might say the same thing about motors and actuators. DJI and Unitree make the most drones and robots, respectively, today, and both leveraged the know-how from their ecosystems to vertically integrate down to the motors and actuators, respectively.

But in 1987, a former Texas Instruments engineer named Morris Chang launched TSMC, which offered to fab everyone else’s chips for them. Nvidia, founded six years later, was born fabless and free to focus on architecture, software, developer ecosystem, cadence, and market selection. These are the things that have compounded into Nvidia’s moat over time.

Nvidia is now the largest company in the world by market cap, and it still fabs with TSMC. The second and third largest companies, Google and Apple, do too.

TSMC, by vertically integrating the hard, CapEx-intensive, process knowledge-driven work of fabbing chips and serving horizontally, made it possible for an entire ecosystem to flourish. It’s grown up alongside that ecosystem, and even now that the companies it’s enabled are cash-rich enough to build their own fabs, the compounding that their collective business has paid for would make it almost impossible to catch up. So even at scale, TSMC and its customers keep winning together.

A bet on Westmag is that we are at the same place in the Electric Age today that we were in the Information Age then.

Westmag’s hope is that by vertically integrating the hard, CapEx-intensive, process knowledge-driven work of making motors and actuators and serving horizontally, it will make it possible for the American drone and robotics ecosystem to flourish.

Because the company is still small, like the industries it serves, Westmag can afford to give American drone and robot companies the time of day. Because it’s vertically integrated, it can be responsive to their needs and spin up prototypes in days instead of months. Because of that, and because it’s in their backyard, it can shorten the iteration loop, so that its customers can iterate their way to better drones and robots than are made anywhere else in the world.

The real big bet on Westmag is that it’s not too late. That if you build a machine that marries American-style innovation with scaled manufacturing, the way we used to, and spin it really fast, slope will outrace intercept and the future might be built in America again after all.

Thanks to Jordan, David, Sam, and many others for helping me get smarter on motors.

That’s all for today. We’ll be back in your inbox Friday with another Weekly Dose.

Thanks for reading,

Packy

Hi friends 👋,

Happy Friday, and welcome to our 195th Weekly Dose of Optimism.

Man, just when I thought it was going to be hard to top last week’s Dose, we have an LDL cholesterol-fighting gene therapy, evidence that GLP-1s slow cancer, supersonic flight, and nanotech, and a Moon Base. We also have the coolest Science Breakthroughs roundup yet. Even a massive Blue Origin explosion can’t slow us down.

If you get through this and want even more optimism injected into your veins, check out this week’s essay:

Let’s get to it.

Today’s Not Boring is brought to you by… Deel

Hiring globally doesn’t have to be complicated

Hiring globally can unlock growth, but local laws, payroll, and compliance often slow startups down.

Deel’s free guide breaks down what an Employer of Record (EOR) is, how startups use EORs to hire internationally without opening entities, and when it makes sense to use one so you can scale with confidence.

Download Now

(1) Eli Lily Does it Again, Now with LDL Cholesterol Gene Therapy

For those keeping score at home, this is the second week in a row that we’ve led off with Eli Lily. Last week, their Reta Phase 3 trials showed astonishingly good results, and this week, they published the results of a phase 1, open-label, single-ascending-dose study on the VERVE-102 gene therapy targeting PCSK9, which is responsible for LDL cholesterol, which is in turn responsible for a lot of heart disease and deaths.

Now this is just a phase 1, and there were only 35 people involved, but a single dose gene therapy reduced PCSK9 levels “from 51% at the 0.3-mg-per-kilogram dose to 88% at the 1.0-mg-per-kilogram dose,” and showed corresponding reductions in the LDL cholesterol level “from 9% at the 0.3-mg-per-kilogram dose to 62% at the 1.0-mg-per-kilogram dose.”

This is a big deal! Globally, about 4.4 million deaths a year are attributable to high LDL cholesterol, roughly 7.8% of all deaths. Cardiovascular disease causes roughly 18.6 million deaths a year worldwide, so high LDL accounts for somewhere around a quarter of those.

Now, we might be able to knock out 4.4 million deaths a year with a shot.

PCSK9 inhibitors are not new. On a recent Invest Like the Best episode, Braidwell Managing Partner Alex Karnal called PCSK9 medicines “pretty much a free lunch”:

The headline here is PCSK9 medicines are amazing, because what they do is they today can lower our bad cholesterol, that LDL cholesterol, by 50%. And we now have outcome studies in patients at different degrees of having high cholesterol, having significant protection from ever developing and having a heart attack or a stroke. And so in patients that have previously had a heart attack or stroke, we can reduce that risk by over 20% in the future. And for people that are at high risk of having a heart attack or a stroke, the medicines that are approved and on the market today can lower that risk by about 25%.

With previous PCSK9 medicines, people would have to stay on them forever to continue to eat that free lunch. Now, if Lily’s early results hold, they will get all of the benefits in a one-and-done shot. Essentially, the drug edits your gene to be like the “population of people in the world that have a genetic mutation that conveys a massive advantage. They have a mutation in their PCSK9 gene, which means they don’t produce the PCSK9 protein.”

We are still a few years away from this miracle drug hitting the market, and there are still trials to be done, but this is a really big deal, not only because it’s an assault on LDL cholesterol and therefore heart disease, but also because it’s another example of our ability to see an advantageous mutation in a certain population, make a drug to mimic it, and knock more and more diseases off the list.

Plus, as the WSJ reported, a number of pharma companies are developing drugs to attack LDL’s cousin, Lp(a), or lipoprotein(a). Lp(a) is almost entirely genetic and currently untreatable. About 90% of a person’s Lp(a) level is fixed at birth, diet and exercise don’t lower it, and statins can actually raise it. So unlike LDL, which we have many tools for, there’s essentially no approved therapy for high Lp(a) today. Roughly one in five people globally have dangerously high levels, and that’s bad. Lp(a) hardens arteries and promotes clotting, which is why it’s so noxious in driving cardiovascular events.

So far, science has been powerless against Lp(a), both in treating it and in proving that lowering a person's Lp(a) will actually reduce heart attacks and strokes. Earlier drugs lowered the particle moderately but failed to reduce events in studies. The current bet from Novartis, Amgen, and, you guessed it, Eli Lilly is that new technologies can cut Lp(a) far more dramatically, with late-stage trials underway and Novartis's first Phase 3 results (pelacarsen) expected this year. Human genetics suggests it should work, but they don't know for sure.

Heart disease is the leading cause of death in the US, and now we have even more data points suggesting that you should really figure out how to stay alive for the next few years, because thereafter, you may be able to stay alive for a very long time. “Medicine keeps getting better and better.”

(2) Weight-Loss Drugs May Have Surprising Side Effect: Stalling Cancer

Xavier Martinez for The Wall Street Journal

You thought we were done with Eli Lily? We’re not done with Eli Lily.

Just as the ink was drying on last week’s Dose, The Wall Street Journal published an article about GLP-1s’ cancer-fighting abilities. “A suite of four new studies suggest that people taking so-called GLP-1 drugs like Novo Nordisk’s Ozempic and Eli Lilly’s Mounjaro saw reductions in tumor progression, lower overall chance of death and less risk of developing breast cancer.”

In lung cancer patients, the rate of progression to advanced disease was cut roughly in half—10% in GLP-1 users versus 22% in the comparison group. Breast cancer patients showed a similar pattern, with progression rates of 10% versus 20%. Colorectal and liver cancers also showed statistically significant reductions.

These studies are from places like UT MD Anderson, University of Pennsylvania, and Cleveland Clinic, and while researchers don’t yet understand the mechanism, the research points to yet another miracle for this miracle drug.

I’m not even sure what to say at this point, so say it with me… get fucked, cancer.

(3) Hermeus Quarterhorse Mk 2.1 Completes First Supersonic Flight

In Riding the Leopard, I wrote, “I would like to live in the future in which we have spaceships and abundant energy and supersonic planes, the future in which car crashes and cancer are a thing of the past.”

I wasn’t expecting that future to come quite so soon. While GLP-1s go to work on cancer, Hermeus became “the world’s first privately developed, unmanned supersonic jet and the fastest unmanned aircraft flying today” when its Quarterhorse Mk 2.1 went supersonic at Mach 1.21 in an unmanned test flight. Huge congrats to AJ and the Hermeus team. I got chills watching the video.

Now, they’ll have more fuel to go even faster, more often. Yesterday, they announced that the Defense Innovation Unit expanded its contract by $159M to $219M to tackle high-Mach flight and payload release.

I may have dreamed too small. Hermeus is now gunning for hypersonic flight.

(4) Atomically precise mechanosynthesis of carbon structures on hydrogenated Si(100) by inverted-mode STM

One thing that I forgot to mention in my vision for the future, but that I would like to add now and would very much like to see, is nanotechnology.

At the end of Where Is My Flying Car?, J. Storrs Hall paints a picture of the future we could have with abundant energy: flying cars, utility fog, the Weather Machine, a “space pier,” and maybe most tantalizingly, atomically precise manufacturing (nanotech), nanofactories and self-replicating machines that build objects atom by atom, collapsing the cost of physical goods the way semiconductors collapsed the cost of computation.

Nanotech is the vision Richard Feynman laid out in Plenty of Room at the Bottom, the one Eric Drexler theorized in Nanosystems, the one Hall imagines in Where Is My Flying Car?, and the one Neil Stephenson painted in The Diamond Age. If you can manipulate atoms one-by-one, you can build anything you can dream up.

Until now, though, it’s remained a dream. Drexler’s theory of positional mechanosynthesis, alongside Ralph Merkle and Robert Freitas, was attacked, most famously in the Drexler–Smalley debate, where Smalley's "fat fingers / sticky fingers" objection held that you couldn't do controlled positional chemistry at that scale.

Well, we can tell Smalley exactly where to stick his fat fingers now.

On Tuesday, Merkle, Freitas, and a number of researchers published a paper demonstrating simultaneous spatial and chemical control over the mechanosynthetic fabrication of carbon structures. Concretely, using a technique they call inverted-mode STM, carbon dimer (C₂) units are donated from surface-deposited molecules onto pre-patterned reactive sites on a hydrogen-passivated Si(100) surface. They show three escalating things: single-site C₂ donation, spatially patterned multi-site donation, and the stepwise assembly of polyyne structures through successive C–C bond formation. Their framing is that this establishes controlled mechanosynthetic donation as a foundational capability for programmable atomically precise fabrication.

For roughly forty years, Mechanosynthesis— using mechanical positional control to drive site-specific chemistry, building structures atom by atom — lived almost entirely in theory and computational chemistry. The canonical proposed primitive was a "dimer placement tool" that deposits C₂ units onto a workpiece to grow diamondoid structures. That is almost exactly what this paper demonstrates experimentally!

There is still a massive gap between this work and Hall’s nanofactory. They've built short carbon chains, one dimer at a time, with an STM tip, which is a primitive. The chasm to the vision is throughput and dimensionality: a single tip placing dimers serially is astronomically slow versus the massively parallel, self-replicating systems APM actually requires, and going from 1D polyyne chains to 3D diamondoid objects is its own huge challenge.

But come on! This thing that people said was impossible was just demonstrated to be possible! I am going to spend the weekend dreaming up all of the things I want to order from the nanofactory, like my own APM island a la The Diamond Age.

In the meantime, if you want to get smart on nanotech, I recommend Hall’s primer for the Abundance Institute and Jacob Rintamaki’s A Technical Review of Nanosytems.

(5) MOON BASE

On Tuesday, NASA Administrator Jared Isaacman and the NASA team held a press conference at its HQ in Washington to provide updates on the Moon Base program, a long-term lunar exploration and infrastructure initiative under the Artemis program aimed at enabling sustained human presence and expanded scientific and commercial activity at the lunar South Pole.

It also launched a Moon Base Website, complete with the hype video above and a timeline for the Moon Base. Because we are establishing a Base, on the Moon.

Isaacman & Co laid out the plan for three initial Moon Base missions. Moon Base I, targeted for no earlier than fall 2026, will use Blue Origin’s privately funded Blue Moon Mark 1 Endurance lander to deliver NASA science payloads, including the Stereo Cameras for Lunar Plume-Surface Studies instrument and a Laser Retroreflective Array, to the Shackleton Connecting Ridge. Moon Base II, planned for later in 2026, will use Astrobotic’s Griffin lander to deliver more than 500 kilograms (over 1,100 pounds) of cargo, including Astrolab’s FLIP rover, to mature lunar terrain vehicle mobility, autonomous operations, and logistics. Moon Base III will prioritize scientific payloads to expand understanding of the lunar surface.

Then, eventually, we’ll have people living on the Moon, which is a harsh mistress but a potentially perfect launchpad for humanity’s mission to Mars. What a universe.

EXTRA DOSES: Science Breakthroughs, Telescope Ranchers, Encyclical

Quick note: I think this might be the coolest Science Breakthroughs yet, with entries on the genetic architecture of complex traits, hallmarks of aging and mortality, homing pigeons relying on superparamagnetic macrophages for navigation, and armadillo-inspired morphing skeletons for robots. Science Breakthroughs is a roundup of the bleeding edge, the stuff that’s even earlier in its development than what we cover in the Dose.

I think it’s worth the subscription alone, and in general, I’m trying to share a lot more value with not boring world subscribers, including recent pieces like Riding the Leopard, Cowboy Space Corporation Case Study, and Thank God for Data Centers. I hope you’ll subscribe and join us.

Read more

Subscribe now

Hi friends 👋,

Happy Wednesday!

A couple weeks ago, I asked if you wanted me to start sharing more off-the-cuff notes with not boring world subscribers, and the response was great, so we’re back. It’s a Wednesday afternoon, not my normal send time, but these are meant to be less formal and more, “I noticed something interesting, here are my quick thoughts.” This one happens to be a little longer than it is quick, but it’s one I wanted to get out for two reasons:

1. People hate AI Data Centers, and I think they’re wrong, even if they don’t like AI.

2. Because I keep hearing, reading, and seeing that AI Data Centers are funding new technologies before they’ve come down the learning curve, which might be a providentially big boon to Reindustrialization and all of the hard, physical things we want to see in the world.

It’s pretty beautiful that gaming chips that evolved from Apollo-funded integrated circuits are creating a product with so much demand that their houses can pay for all sorts of novel technologies, like the Apollo Program did.

Let’s get to it.

Today’s Not Boring is brought to you by… Deel

Hiring globally doesn’t have to be complicated

Hiring globally can unlock growth, but local laws, payroll, and compliance often slow startups down.

Deel’s free guide breaks down what an Employer of Record (EOR) is, how startups use EORs to hire internationally without opening entities, and when it makes sense to use one so you can scale with confidence.

Download Now

Thank God for Data Centers

There exists a vast pool of technologies that are potentially superior to those we employ today, but which require scale and learning curves to reach their potential.

Advanced nuclear reactors are one such technology - they are more expensive than alternatives today, but manufactured at scale, and benefiting from the learning curves required to get there, may become cheaper than other generation technologies. The cost physics are on their side, and nuclear is reliable, safe, firm, and clean.

The challenge with these technologies, in normal times, is that there is little economic incentive for the buyers who would enable the scale to stick their necks out. Natural gas is cheap and abundant, and it’s not that bad for the environment, compared to coal and oil at least, and the environment is someone else’s problem, anyway. And so, in normal times, we remain stuck in local maxima without the demand to push towards global ones.

Historically, these stalemates have cracked in a couple of main ways: Alpha Products and extraeconomic Buyers of Capabilities. These are essentially the same mechanism at different scales and with different motivations.

In The Electric Slide, we discussed the role that Alpha Products played in providing the initial demand that eventually brought each layer of the Electric Stack down their respective cost-performance curves. For lithium-ion batteries, the alpha product was the Sony Handycam. For neodymium magnets and motors, it was the 3.5” hard‑disk drive. For power electronics (IGBTs / inverters), it was the variable‑frequency drive (VFD) for industrial motors. For microcontrollers (MCUs), it was the calculator. Etc.

For each of these, the new technology was advantageous enough in a specific way to the end product that it was worth paying higher costs or sacrificing on other capabilities to capture those benefits.

Alpha Products, however, typically support components that are not multi-hundred million or multi-billion dollar projects in their own right.

The role of the extraeconomic Buyer of Capabilities in the development of new technologies is even better-understood. This is the DoD or NASA, mainly, buying technologies to confer a specific advantage, almost irrespective of their cost. This category of customer cares less about price than about capability.

Theirs is an important role because it gives new technologies the opportunity to get to scale, come down the learning curve, and ultimately compete in the much larger commercial market.

For a while now, but particularly over the past couple of weeks, I’ve heard some version of the same story over and over again:

“We are still going after our long-term mission, but to fund it, we’re planning to sell to data centers.”

Today, Data Centers are increasingly serving as Buyers of Capabilities, acting as something between a government and a commercial buyer. The Data Center is the meta-Alpha Product. If you can sell them something they need, fast, they have an almost bottomless bid.

This is true for obvious things like GPUs, inference chips, and DRAM, but it’s also true for companies that you wouldn’t typically associate with AI data centers, like supersonic turbines, enhanced geothermal, modular construction, high-voltage direct current grids, solid-state transformers, silicon photonics, optical fiber, lasers, batteries, and nuclear.

Many of these technologies have the potential to be better and cheaper than the incumbent technologies they aim to replace, but they have been too expensive and unproven to compete. With backlogs in all of the traditional inputs to Data Centers, however, developers are willing to pay up for new technologies that can deliver fast, which gives them the opportunity to scale up and cost down.

For these technologies, Data Centers act as a third type of Buyer of Capabilities, a commercial analog operating on DoD-style procurement logic but commercial timescales.

Given the size of the budgets, the relative smallness of any one input’s cost relative to the overall project cost and revenue opportunity, and the speed with which Data Centers are making decisions and putting down deposits, Data Centers may meaningfully increase the odds of success of hard tech companies and Vertical Integrators more than the market realizes.

Far from being the villains they are painted as (often using misinformation and largely due to their association with deeply unpopular AI), Data Centers may be the greatest accelerant of American Reindustrialization and a built-world future that benefits all people that we’ve ever seen.

They offer dilution-free capital (real revenue on a negative working capital cycle) to fund the big vision, and more importantly, the opportunity to get to scale and down the learning curve years earlier than would otherwise have been possible. This both accelerates timelines of things that might have worked, but more slowly, and makes companies that might otherwise have died in the Valley of Death viable.

Whatever your feelings are on AI, the furor towards Data Centers is misplaced. Hell, whether or not you think we’re in an AI Bubble barely matters here. In five years, this could all fall apart, and the world will be much better off. Data Centers are funding the future where no one else will.

This isn’t the first time that people have gotten mad that something that seems frivolous is sucking up so many resources. The immediate stuff - like how much money is being spent or power is being consumed - is an easy target, while the long-term benefits are hard to see.

We Choose to Go to the Moon Not Because it is Popular

With the benefit of hindsight and distance, the Apollo mission has become one of America’s proudest accomplishments. At the time, though, not everyone loved JFK taking us to the Moon. There were too many problems on Earth to be solved to waste all that time, money, and smarts on a lunar boondoggle.

In May 1961, Gallup asked Americans, “It has been estimated that it would cost the United States 40 billion dollars-or an average of about $225 per person-to send a man to the moon. Would you like to see this amount spent for this purpose, or not?” 58% of respondents said that they would not, another 9% had no opinion, and only 33% supported the mission.

President John F. Kennedy gave his canonical “We choose to go to the Moon” speech not because it was popular, but because it was unpopular and he needed to rally support.

It didn’t work that well. In a 1964 Gallup poll that asked “Do you think the United States should go all out to beat the Russians in a manned-flight to the moon–or don’t you think this is too important?”, 66% of respondents said that they did not think it was too important, and another 8% said “Don’t know,” which amounts to the same thing. In the summer of 1965, “one-third of the nation favored cutting the space budget, while only 16% wanted to increase it.”

Even a decade after NASA pulled off the near-impossible, in 1979, only 41% of Americans told an NBC/AP poll that the benefits of the space program outweighed the costs. By 1995, that had risen to 47%, by 1999 it hit 55%, and by the 50th anniversary of the Moon Landing, in 2019, support had reached 64%.

One explanation is that, not having to pay the costs ourselves but getting to bask in the memory of victory, today’s Americans can of course look back and say it was worth it.

Another, though, is that over time, the real benefits, not at all obvious at the time, have become more clear.

Apollo’s critics were ultimately proven wrong, both because beating the Soviets to the Moon was awesome…

… and more directly because the absurdity of the task’s ambition coupled with the bottomlessness of its budget that they complained about were exactly the conditions needed to create terrestrially-useful innovations that would otherwise have taken much longer, or never been invented at all.

While the role that the Apollo Program played in inventing new technologies is probably overblown, it accelerated, scaled, and de-risked a lot of things that otherwise may not have gotten to sufficient scale or cost to impact our lives. Some of the technologies and products for which they served this role include:

• Fireproof fabrics and flame-retardant materials (developed after the Apollo 1 fire), which made their way into firefighter suits and racing gear

• Improved freeze-drying processes for food preservation

• Mylar-based reflective insulation, which became the emergency space blanket

• Advances in composite materials and ablative coatings

• CAT and MRI imaging benefited from digital image processing techniques developed to enhance lunar photographs at JPL

• Implantable cardiac pacemakers improved from bidirectional telemetry developed for astronaut biosensors

• Cool suits (liquid-cooled garments) used for MS patients, burn victims, and racing drivers came directly from the spacesuit undergarment

• Kidney dialysis machines used a chemical process developed to remove toxins from astronauts’ water

• Black & Decker developed the technology to make cordless power tools for collecting lunar samples

• NASA developed Memory foam (Temper Foam) for crash protection in seats

• Water filtration using silver ions, based on the system that purified the Apollo crew’s water

• Scratch-resistant lens coatings, derived from coatings developed for astronaut visors

• Improved smoke detectors (the modern ionization-type was refined for Skylab, which leveraged Apollo hardware)

Even cleanroom protocols for handling lunar samples spread into pharmaceuticals and, relevant to today’s discussion, semiconductor manufacturing.

That’s just one program, albeit a very large one. If you zoom out to include the technologies that received early support from the DoD, the list includes pretty much everything that defines modern life.

The internet and TCP/IP directly, and Ethernet indirectly. Satellite communications, GPS, the inertial navigation systems that run in our phones and cars, and radio navigation. The mouse, windowing interfaces, and hypertext; the work that Doug Engelbart showed off in his “Mother of All Demos” was ARPA-funded. Public-key cryptography was invented for UK SigInt, Grace Hopper developed COBOL on the Navy’s dime, and DARPA funded the first AI labs at MIT, Stanford, CMU, and Stanford Research Institute for decades. The chips AI runs on, GPUs, and the CUDA software behind them, owe something to DARPA-funded parallel computing research. Siri’s lineage, for better or worse, through DARPA’s CALO speech recognition program at SRI. There are the more obviously military products, like the jet engine and the planes jet engines power, stealth coatings, night vision, radar and synthetic aperture radar, Lidar, ultrasound, drones, and infrared and thermal imaging. There are materials, like composites, including carbon fiber, titanium alloys, and advanced ceramics, all of which scaled through defense procurement. LEDs were funded through early signaling work, and digital photography found a buyer in early spy planes. In medicine, we have the DoD to thank for EpiPens, tourniquets, hemostatic agents like QuikClot, prosthetics, blood banking, and plasma storage. Penicillin was first mass-produced in a wartime crash program. And energy? Civilian nuclear power descended directly from Admiral Rickover’s Naval reactor program, lithium-ion battery research had defense funding, and solar photovoltaic cells were pulled forward by satellites’ unique power needs.

Military procurement is the closest thing that the United States has to a national industrial policy, and its worked. The military has funded the development of practically every general purpose technology of the past century, before the commercial market took those de-risked, cost-downed technologies and figured out how to bring them to the masses.

Of all of these, the cleanest case study and the one that maps most directly onto what’s happening in 2026, is the integrated circuit.

Minuteman, Apollo, and Moore’s Law

Fairchild Semiconductor was founded in 1957 as a transistor company, and its first big customer was the military.

Specifically, facing a threat from larger Soviet boosters that could launch intercontinental ballistic missiles (ICBMs) that could fit vacuum tubes, the US military, with its smaller boosters, had to invest in miniaturization, which meant transistors. Between 1958 and 1960, Fairchild’s revenue grew from $500k to $21 million, largely on the back of the Minuteman I program, for which it produced custom designs. Challenge was, as the number of transistors grew, the electronics industry ran into the “tyranny of numbers”: now that they could, engineers wanted to design circuits with thousands of components, all of which had to be wired together by hand.

So in Dallas, Texas Instruments’ Jack Kilby came up with his “monolithic idea.” “He realized that,” Carl Leonard writes, “instead of connecting separate components, an entire electronic assembly could be made as one unit from one semiconducting material by overlaying it with various impurities to replicate individual electronic components, such as resistors, capacitors, and transistors.” He had invented the integrated circuit (IC).

Three months later, in Mountain View, Fairchild co-founder Bob Noyce arrived at a similar idea from a different angle. Starting from the planar process, invented by Jean Hoerni in early 1959, Noyce realized that you could build transistors flat on a silicon wafer, with all the connections on the top surface, protected by a layer of silicon oxide. He wrote in his notebook, “In many applications now it would be desirable to make multiple devices on a single piece of silicon in order to be able to make interconnections between devices as part of the manufacturing process, and thus reduce size, weight, etc., as well as cost per active element,” and filed a patent that year.

While there are differences between the ICs they developed (Kilby’s used Geranium and Noyce’s silicon, for example), the two are credited as the co-inventor of the IC. What matters for our story is 1) the IC may not have been developed without demand from the Air Force, Army, and Navy, which were each spending real money on parallel attempts to solve it, and 2) Fairchild was now in the IC business.

In 1961, after starting at $1,000 per chip on tiny pilot runs the year before, Fairchild brought the integrated circuit to market at $120 per chip. The challenge was, no one really needed an integrated circuit, not enough to pay that price. Any electronics firm could wire together discrete transistors to do the same thing for a fraction of the price. As Britannica puts it, “a buyer had to have a serious space constraint to justify purchasing ICs.” Fortunately for Fairchild, NASA had a serious space constraint.

In early 1962, MIT’s Instrumentation Lab, which was responsible for the Apollo Guidance Computer (AGC), placed a test order for 100 ICs at $43.50 per unit. Meanwhile, getting ahead of volume and hoping to use scale to rev aerospace demand, Noyce cut the price of the IC from $120 to $15, an 87.5% drop, while still charging NASA and MIT premium prices to fund scale. “Noyce slashed prices, too, gambling that this would drastically expand the civilian market for chips, Chris Miller wrote in Chip War. “In the mid-1960s, Fairchild chips that previously sold for $20 were cut to $2. At times Fairchild even sold products below manufacturing cost, hoping to convince more customers to try them.”

Meanwhile, the government funded the gap. That November, MIT decided to go with Fairchild’s IC, or more specifically, its Micrologic computer made up of ICs. By 1963, MIT was consuming 60% of US IC production for the AGC; other military and aerospace buyers made up the rest. In a 1964 article for IEEE, Noyce wrote, “Military and space applications accounted for essentially the entire integrated circuits market last year, and will use over 95 per cent of the integrated circuits produced this year.”

In 1965, the year that Gordon Moore wrote the Cramming More Components onto Integrated Circuits paper that birthed Moore’s Law, ICs had reached cost parity with discrete components, at around $10, and were beginning to beat them. It was around this time that the Minuteman II program, which now used Texas Instruments ICs, became the technology’s largest buyer. That meant two things – 1) there were multiple large buyers of ICs and 2) there was competition – which combined caused Noyce to cut prices again, down to $2, and again, to $1. The cash coming in from Apollo allowed him to attack the commercial market with lower prices.

And it worked. By the end of the decade, America beat the Russians to the Moon, and Burroughs released the B2500 computer, the first to use ICs.

From that point forward, Moore’s Law has largely been driven by demand from the much larger and faster-moving commercial market. But the IC would not have come down the cost curve so quickly – and Moore’s Law, that self-fulfilling prophecy, may never have been coined or executed against – without the early space and military demand. It is to these Buyers of Capabilities that we owe the computer-powered world we inhabit today.

Something similar might be happening with Data Centers today.

Data Center Demand

There is a sentiment floating around that we can’t build hard things in America anymore, but by any measure, modern AI data centers are hard assets to build and operate, and we are building a lot of very big ones. Western hyperscalers, labs, and neoclouds will spend something like$750 billion this year and more than $1 trillion next year building them. Goldman estimates that AI CapEx will take $7.6 trillion of capital between 2026 and 2031 across Compute, “Data Centers,” and Power.

This is an enormous amount of CapEx, absolutely, historically, any-way-you-slice-itly. With US GDP around $32 trillion, this year’s spend represents 2.4% of GDP. Assuming GDP grows by an aggressive 3% next year, AI CapEx will account for 3.1% of GDP. For context, the Manhattan Project reached 0.4% of GDP, the late-90s telecom bubble reached 1.2%, and even the Apollo Program only hit 0.4%. To find a more GDP-dominating project, you’d need to look to one of the two World Wars, the New Deal, or the Railroad Boom.

For a little more context, the $15 billion per year that Anthropic will pay SpaceX for the use of two of its Colossus data centers is roughly 60% of NASA’s entire annual budget.

Read more

Hi friends 👋,

Happy Friday and welcome to our 194th Weekly Dose of Optimism. I’m emailing you from beautiful Utah, where I attended the Abundance Institute Gala last night and got to talk nuclear with Oklo’s Jake DeWitte and General Matter’s Scott Nolan, and where I’ll be spending today at the state’s Operation Gigawatt Conference.

It’s hard not to feel the optimism in the mountain air. When we started writing the Dose a few years back, nuclear was a controversial dream. Today, we’re heading into the Memorial Day Weekend that kicks off the summer during which multiple new reactors will go critical for the first time.

Plus, you’re about to read one of the most action-packed Doses in Dose history.

Things are moving fast, and mostly for the better. Happy Memorial Day Weekend, y’all.

Today’s Weekly Dose if brought to you by… Lightwork

It turns out, my house has been making me worse at writing. Sorry about that. I have it under control now.

Lightwork came by my place recently and tested our air and water quality, light sources, likelihood of mold, electromagnetic exposure (turns out electromagnetism does run the world), and household products. It was an awesome experience. I learned a lot, including that my home’s air quality and lighting were undermining my energy levels, our shower water was full of VOCs whose vapor I was inhaling, and a whole lot more.

Lightwork’s assessments are concierge evaluations where they test every aspect of your home that could be impacting your health, produce a science-backed, in-depth report on all their findings, and then help you fix all the issues they uncover. I recommend them. Check them out!

Book a Free Consultation Today

(1) Retatrutide Stuns In Phase 3 Obesity Trial

Eli Lily

Yesterday morning, Eli Lilly reported topline results from TRIUMPH-1, the Phase 3 trial for retatrutide in adults with obesity and at least one weight-related comorbidity. Here’s a summary of the results from Superpower founder and peptide enthusiast Max Marchione:

• 28.3% bodyweight lost on 12mg over 80 weeks

• 70.3 pounds on avg. or 31.9 kg

• 45.3% of patients hit 30%+ weight loss (this is bariatric surgery territory)

• 30.3% weight loss (85 lbs) at 104 weeks in higher-BMI patients

• 65.3% of 12mg patients dropped below the obesity BMI threshold

• 19% loss on 4mg over 80 weeks (47.2 lbs) with fewer dropouts than placebo (4.1% vs 4.9%)

• significant drops in blood pressure, triglycerides, non-HDL cholesterol, waist circumference, and hsCRP

• no cardiac or liver signals

These are bariatric surgery results in a shot, with a bunch of freebies thrown in.

People have been excited about reta on social media for a while, but it’s still gray market, so this is a key step towards getting it to the general public, and just an astonishing set of results. People are losing 28.3% of their bodyweight, or 70 pounds, in less than two years!

For context, Zepbound (tirzepatide) tops out around 20-22% weight loss in its big trials, and the previous-generation Wegovy (semaglutide) at around 15%.

Retatrutide is the first triple hormone receptor agonist (GIP + GLP-1 + glucagon). We previewed this back in Dose #175, in our piece on grey-market peptides and Gen3 GLP-1/GIP/glucagon agonists. The author of the essay we covered, Not for Human Consumption, called these drugs “the holy grail of weight loss medications” before any of them had a Phase 3 readout. Now one of them does.

This is Lebron-level. Unbelievably hyped out of the gate, and then exceeds the hype.

As with all of these miracle drugs, there are positive side effects beyond weight loss. Retatrutide reduced osteoarthritis knee pain by an average of 75.8%, with more than 1 in 8 patients reporting they were completely free of knee pain at the end of the trial. LDL cholesterol dropped 20%. 72% of prediabetic participants returned to normal blood sugar levels. Seven more Phase 3 readouts are expected this year, including diabetes, sleep apnea, chronic back pain, and cardiorenal outcomes.

Get ready to get hot and healthy. It’s Reta Summer (in 2027).

(2) Colossal Biosciences Hatches Chick From Artificial Eggs

“Before we named the stars, before we mapped the atom, we asked ourselves an impossible question: which came first, the chicken or the egg?”

On Monday, Ben Lamm's Colossal Biosciences announced that it has hatched 26 healthy chicks from a fully artificial egg, an oval printed shell coated in an oxygen-permeable membrane that lets an embryo develop from poured-in yolk to pipping bird, without a biological eggshell. Researchers crack a freshly laid fertilized egg, pour the contents into the artificial shell (Colossal adds ground-up calcium back in, because the embryo eats the shell as it grows), and watch through the top window as eyes form, vasculature spreads, and, 18 days later, a chick taps its way out.

Real scientists will tell you (and MIT Tech Review's Antonio Regalado dutifully did) that growing birds outside the shell isn't new. A Japanese group hatched quail in 1998. Katsuya Obara hatched chickens under transparent plastic film in 2024. Colossal's marketing is, per Regalado, "pure Hollywood." Which is kind of fair but too clucking cynical.

Colossal’s real engineering advance is the membrane, which lets the embryo pull enough oxygen from ambient air that the system doesn't need supplemental gas. Previous shell-less methods did, and chicks tended to die. I like the version in which the chicks live, Regalado.

While this experiment was on a chicken, we have plenty of those. Colossal likes to de-extinct. Its stated goal is the 12-foot, 500-pound South Island giant moa, which laid four-liter eggs no living bird is large enough to surrogate. Colossal staff are reportedly already calling the prototype scaled-up version the "salad spinner." More immediately, the artificial egg is a tool for genetic rescue of endangered birds and, as per our discussion in Dose #186 with not boring capital portfolio company Neion Bio, a substrate for transgenic chickens producing therapeutic proteins in egg whites at a fraction of mammalian cell culture costs.

This is the first member of what Colossal calls its "exogenous development" family. Up next: artificial wombs for marsupials.

With this chicken breakthrough, it feels like humans have crossed a road.

(3) SpaceX Launches S-1

For the entirety of its existence, human civilization has lived on a single celestial body: Earth. The current paradigm, in which human civilization is confined to one planet, exposes humanity to existential threats that are unpredictable and uncontrollable on a planetary scale. By moving beyond the only home we have ever known, we ensure species-level redundancy and that the light of consciousness will not be tied to a single planet subject to the inevitable hazards of a harsh and vast universe. We do not want humans to have the same fate as dinosaurs.

On Wednesday, SpaeX dropped what may have been the most highly-anticipated S-1 in the history of the planet (until Anthropic files later this year). You can read it here.

Financial highlights include:

• Revenue: $4.69B (run-rate ~$19B)

• Operating loss: $(1.94B), but Adjusted EBITDA of +$1.13B

• Connectivity (Starlink) prints cash: $3.26B revenue, $1.19B operating income in just the first quarter. Its revenue is growing ~50% per year, and profitability is growing faster.

• AI is losing money for now…: $818M revenue, $(2.47B) operating loss in Q1. Capex on AI alone was $7.7B in Q1, more than Space ($1.05B) and Connectivity ($1.33B) combined, by far.

• But has the monster TAM: AI infra, consumer subscriptions, digital advertising, and enterprise applications - the AI segment - has a company-estimated TAM of $26.5 trillion, which seems small when you consider the size of the universe and the potential to spread Enterprise Applications throughout it.

  

SpaceX also disclosed that it’s doing a second compute deal with Anthropic, in which Dario & Co. will pay SpaceX $1.25 billion per month for Colossus 2, which is nearly as much as entire SpaceX makes now. For now, it’s selling compute on earth, but per the S-1, it plans to begin deploying orbital compute in 2028 and scale it to 100 GW of compute launched annually, which would make Colossus seem like a pale blue dot.

Starlink, which had been the bull case until AI, is at 9,600 satellites, 10.3 million subscribers across 164 countries, and now accounts for ~75% of all active maneuverable satellites in orbit.

There is no one operating at a higher level than Elon and the SpaceX team, across everything from terrestrial data centers to rockets to satellites to, soon, Terafab. After more than 20 years, it’s awesome to see the company go public (and generate some much-needed DPI for the venture and seven-layer-SPV ecosystems). Is the price a little high? Maybe, yeah, sure. But what would you pay to own a Scarce Asset that catches Statue of Liberty-sized rockets with chopsticks…

We’ll find out in June.

In the meanting, after scrapping the launch of Starship v3 last night, SpaceX is giving it another go today. What a way to launch summer.

(4) GPT 5.5 Pro Makes Breakthrough on Erdős Planar Distance Problem

Take n dots on a piece of paper. How many pairs of dots can sit exactly one unit apart from each other?

Paul Erdős posed this in 1946, conjectured the answer grows just barely faster than n, and offered cash for a resolution. For 80 years, the best constructions anyone could find were boring square grids, and most of combinatorial geometry treated Erdős's upper bound as essentially right.

This week, an internal general-purpose OpenAI reasoning model produced a proof that disproves it. The model isn’t math-specific, it just thought for a long time and burned an estimated ~$1,000 max worth of tokens, which is pretty affordable for an 80 year old unsolved problem! It found that there's an infinite family of point arrangements that beat the grid, polynomially. Erdős was wrong, and his “favorite problem” is settled.

How’d it pull it off? Instead of more clever geometry, it pulled in algebraic number theory (specifically, infinite class field towers and Golod–Shafarevich theory, of course, why didn’t I think of that) to construct number systems with the right symmetries, then read the geometric configurations off of those. No human had tried this connection. Thomas Bloom thinks it will unlock other long-stuck problems in discrete geometry.

And, importantly, it’s verified. Last October, OpenAI claimed GPT-5 had solved ten Erdős problems; on inspection, it had just surfaced answers from the literature. This time, it got external review by Fields Medalist Tim Gowers and Will Sawin at Princeton, with a companion paper. Gowers's verdict: "if a human had written the paper and submitted it to the Annals of Mathematics, I would have recommended acceptance without any hesitation." It also got endorsements from Noga Alon, Melanie Wood, Arul Shankar. Shankar said the models “are capable of having original ingenious ideas, and then carrying them out to fruition.” The full proof is available as a public PDF.

Ten months ago, frontier models were at IMO-gold level. Really smart high schoolers, basically. Now, they’re doing Annals-worthy novel work.

I’ve been more skeptical about AI than I normally am about new technologies, and less excited than the average person in tech. Per the evidence that keeps coming out, I’ve been wrong. On the business side and the capabilities side, the pace has been truly remarkable, and I can’t wait to see what problems they’ll solve for us in the years ahead.

(5) SendCutSend Raises $110M at $1B Valuation

Jim Belosic (SendCutSend)

Jim Belosic started SendCutSend in 2018 because he couldn't get custom metal parts fast enough for the cars and motorcycles he was restoring. Eight years later, bootstrapped to this point on credit cards, savings, a PayPal loan, and bank-financed machines, the Reno laser-cutting and bending shop is doing roughly $200 million in revenue, growing 100% year over year, and turning into one of the most important on-demand manufacturers in America.

Austin Vernon's February essay Speed Can Reindustrialize America is the best piece on why the company is so impressive and potentially important. Vernon's argument is that the US is great at high-volume manufacturing and terrible at low-volume, short-lead-time custom work because soft costs (quoting, programming, billing, all the white-collar overhead) swamp the actual fabrication cost on small orders. SendCutSend's whole company is an attack on soft costs: instant quotes, one-click buying, software-routed production. It’s generating ~$275K revenue per employee, and the market would support a hell of a lot of employees.

Interestingly, Patrick Collison was the one person Vernon thanked for feedback, right above the “Funding Opportunities Apendix.”

And I guess Collison took it literally, because he reached out to Belosic about investing, and although SCS had never taken outside capital, he decided to take money from the Collisons, Sequoia (Andrew Reed), and Paradigm (Matt Huang) to go bigger, faster.

Jim and SCS are easy to root for and have become a cult classic over the past few years. Now, the cat’s out of the bag, the money is in the bank, and IT’S TIME TO BUILD.

In other great news for building stuff here, Amca raised $300M at $1B+ from Caffeinated and crew. Amca builds and modernize factories to design and produce critical aerospace and defense components that are supply-constrained and often sole-sourced.

Jai Malik’s eighteen-month-old El Segundo startup already runs factories in California, Iowa, and New York, supplies components for the F-35, and uses its RAPID software platform to deliver parts to BAE, Airbus, Textron, Honeywell, and GE Aerospace 67% faster than the legacy defense supply chain. Now, it can make more of the critical components on which the country runs.

Extra Doses: Science Breakthroughs, Anthropic, Valar Atomics, Quantum Investment, Grindslop

Read more

I saw a great case study on the Great Differentiation this week, so put together some quick thoughts off the dome, as I continue to study how companies tell their stories in new and interesting ways. Let me know what you think.

Cowboy Space Corp Differentiation Case Study

Earlier this week, I saw a headline that a company called Cowboy Space Corporation had raised over $200 million to build rockets whose upper stages are just foldable data centers. My first thought was who would be dumb enough to compete with Elon in launch at this point and my second was that this had to be the peak.

Then, I saw that it was Robinhood co-founder Baiju Bhatt’s company, Aetherflux, rebranded and expanded, and I got a little more interested, because at least they’d been planning to do energy in space since before data centers were cool, and because turning Robinhood money into sci-fi energy and compute moonshots is exactly how you should billionaire. This is Choose Good Quests.

I wrote about it in the Dose yesterday, because if nothing else, it’s bold.

But in that process, I watched the videos. There’s the one I included in the Dose, which was pretty cool because it had a tumbleweed, and some cowboy-adjacent music, and because the product itself - the upper stage that unfurls into a data center powered by a bunch of solar panels - is arresting, and because calling space “The High Frontier” is both cool and relevant for a company now called Cowboy Space Corporation.

If it were just for that video, though, I wouldn’t be writing this note. This is the one that made my fingers do the keyboard two-step:

It’s so fucking weird when you watch it, right? Baiju, who I’ve never met and I’m not sure watching this if he’s really like this or if this is a bit, is just talking into the camera while he slaps cowboy hats on his team’s head.

Read more

Hi friends 👋 ,

Happy Friday and welcome back to our 193rd Weekly Dose of Optimism.

The sun is shining, the birds are chirping, Demis is going to solve all disease, Varda is making space drugs, pancreatic cancer keeps getting punched in the mouth, we’re turning whole rocket upper stages into space data centers, and the IPO window is OPEN. Y’all mind if the optimists have another good week?

Let’s get to it.

Today’s Weekly Dose is Brought to you by… SimpleClosure

We celebrate ambitious companies every week in the Dose, but ambitious also means long odds, and not every ambitious company ends the same way. Some shut down. Some sell assets. Some get acquired. Some simply reach the point where winding down is the right move.

But however a company ends, there are still final steps to handle: state filings, investor communications, distributions, compliance requirements, asset decisions, and remaining obligations.

SimpleClosure helps founders bring structure to closure, so the process is handled clearly, responsibly, and without becoming another months-long burden.

Whether you’re actively winding down or trying to understand what comes next, SimpleClosure can help.1

Learn More

(1) Isomorphic Labs Raises $2.1 Billion to Solve All Disease

I recently read Sebastian Mallaby’s book on Demis Hassabis and DeepMind, The Infinity Machine, after having already watched the documentary on DeepMind’s development of AlphaFold, The Thinking Game, and a big takeaway for me is that you don’t want to bet against Sir Demis.

In addition to DeepMind, now part of Google, Demis spun out a bio-focused company called Isomorphic Labs whose mission, right there on the homepage, is to “Solve All Disease.” As of today, they have $2.1 billion more in the anti-disease arsenal thanks to a Series B led by Thrive Capital with participation from Google vehicles Alphabet, GV, and CapitalG, and sovereign vehicles, MGX, Temasek, and the UK’s Sovereign AI Fund. It is the second largest biotech round ever, after Jeff Bezos and Yuri Milner’s AltosLabs came out with $3 billion in 2022.

Back in Dose #180, we covered IsoDDE, Isomorphic Labs’ unified AI drug design engine that crushed AlphaFold 3 on the hardest protein-ligand benchmarks and, in one of the more striking results, rediscovered cereblon’s second binding pocket from sequence alone, an achievement it took human researchers 15 years to confirm experimentally. Our take then was that the engine was real but the proof would come when AI-designed molecules actually entered humans. The money will go towards that, with the first of their wholly-owned drug candidates expected to enter human trials at the end of this year, and to expand the pipeline, continue to improve IsoDDE, and keep hiring the very best talent.

While there are a number of companies promising to solve drug discovery with AI, Isomorphic Labs seems to have the track record, talent, and war chest to do it.

Nobel Laureate Sir Demis Hassabis has a history of planning to do things that sounded crazy at the time and then doing them, and the craziest part is that if he does it again this time, the thing that sounds crazy that he will actually pull off is solving all disease. God Save the Queen, and all of us as well.

(2) Varda and United Therapeutics Testing Space Drugs

Sana Pashankar for Bloomberg

Isomorphic seems promising for Earth Drugs. But what about Space Drugs?

Way back in June 2023, almost three years ago, I wrote a Deep Dive on not boring capital portfolio company Varda in which we discussed its plans to manufacture drugs in space, where the lack of gravity lets you do things you can’t on earth, and send them back down to earth, where the people who need them are.

After three years of launches and re-entries with defense customers, the company is igniting stage 2 of the plan: start testing space drug manufacturing.

This week, it announced a partnership with $25B publicly-traded pharmaceutical company, United Therapuetics, to “explore the use of microgravity in creating drugs for chronic diseases in a deal that could prove out the value of in-space development for the pharmaceutical industry.”

United makes drugs to treat cancer and high blood pressure in the lungs, and it’s expected that this partnership will start by testing molecular samples of drugs intended to treat lung diseases specifically.

That’s one big step for Space Drugs, and one potentially giant leap for mankind.

(3) Pancreatic Cancer Just Met Its Match

for

Back on earth… it’s been a bad few weeks for pancreatic cancer, which means it’s been a great few weeks for humans. We led off Dose #190 with a pancreatic cancer mRNA vaccine, Dose #191 with pancreatic cancer-detecting AI, and now, Ruxandra Teslo drops a banger on a different pancreatic cancer fighting drug, Revolution Medicines’ daraxonrasib.

Last month, Revolution shared daraxonrasib’s Phase 3 results. The oral, once-daily pill roughly doubled median survival for metastatic pancreatic cancer, pushing it to 13.2 months. More importantly, it represents a new class of drugs that use a “molecular glue” strategy to hit targets that don’t offer the usual drug-binding pocket. Daraxonrasib targets RAS, a cancer-driving protein found in roughly 25% of human cancers and more than 90% of pancreatic cancers, long considered “undruggable.” This isn’t a cure, but it is a crack, an early sign of the undruggable becoming druggable.

Go read Ruxandra’s excellent essay. Get fucked, cancer.

(4) AI Chipmaker Cerebras Climbs 68% After Year’s Biggest IPO

Bloomberg

Do you feel that breeze? That’s the IPO window, baby. It’s wide open.

Yesterday, fast inference chipmaker Cerebras had a very successful IPO, pricing well above initial range at $185, and then closing 68% above that. Bill Gurley has got to be conflicted… on the one hand, the man hates when a company leaves money on the table in an IPO. On the other, his old firm, Benchmark, is going to make an extraordinary ~$5 billion+ at the $311 opening day closing price.

Fervo, the geothermal company whose S-1 we covered recently, also IPO’d this week. Its shares popped 33% on Day 1 from its $27 price, and then jumped another 11% today.

Look, things are a little frothy. I’m not buying Cerebras at $311. But we love to see tech companies win, but I’m also just psyched to see some liquidity flow back to LPs to flow back into venture to fund the next batch of companies whose IPOs we’ll cover right here in the Dose in ~8-10 years.

Plus…

In older recent tech IPO news, Figma defied the SaaSpocalypse / Anthropic design plugin narrative by crushing its first quarter financial results. The stock popped 8%. Not dead.

(5) Cowboy Space Corp. Gets in the Saddle

Speaking of AI infra exuberance, Robinhood co-founder Baiju Bhatt’s space-based energy company, Aetherflux, has rebranded to Cowboy Space Corp, a company that will, of course, be doing orbital data centers.

On the one hand, it’s a little 2026 Mad Libs. On the other, though, it kind of rocks.

Choose Good Quests ✅ - Baiju is Robinhooding himself, stealing his own gains from the fintech app and giving them to the space company. It’s what we want out of our billionaires.

Also, the idea is pretty interesting. Instead of an upper stage that holds and releases a data center payload, the upper stage is the data center payload. That means that Cowboy has to build its own rockets, in competition with SpaceX, but is focusing fully on orbital data centers by making them half of the rocket.

Maybe this is a sign that we’re nearing a peak. Maybe it’s exactly what we need to provide AI with all of the compute it will need. I don’t know.

I think we’re at the point in the cycle where we just sit back, enjoy, crack open a High Life, and root for these crazy Cowboys to wrangle their space GPUs for the benefit of all mankind. Giddy up baby. Yeehaw.

Extra Dose for not boring world subscribers down below. We got…

Science Breakthroughs, Vacuum Energy, Starship v3, Magic Words

Read more