The example of Claude chipping away at the CUDA lock-in in just a few days highlights the central paradox of this entire cycle: frontier intelligence naturally eroding its own defensibility.
And how most technological surplus eventually leaks to consumers through competition and commoditization.
What makes AI unique is that the technology is actively accelerating that process.
The moment an AI model becomes capable enough to port complex software across different hardware or reverse-engineer specialized techniques in an afternoon, switching costs evaporate.
The smarter the model, the faster it eliminates cognitive moats.
That’s why the Vanderbilt parallel hits the mark. Vanderbilt didn’t win long-term because he had superior steam engines; he won because he owned the physical right-of-way into Manhattan.
If software architectures and model weights can be rapidly imitated by the very tools they produce, the only durable moats left won't be intellectual or digital. They will be the stubborn, physical bottlenecks that code cannot optimize away...
grid capacity, interconnect queues, physical upkeep, and real-world distribution.
The defining question for this boom isn't just whether capital is overbuilding supply, but whether frontier models are inherently engineered to destroy the pricing power required to pay for them.
Curious to know where you're concentrating your effort with this insight in mind?
There's a story that I remember Charlie Munger used to tell about Warren Buffett. There was a new weaving technology that was rolling out across the weaving industry, and the management came to Warren and said: "Check this out! We're going to invest in this and it's going to lead to higher quality cloth at lower cost, and it's going to be great!"
And Warren's immediate response was sad: "This is really too bad. We're going to have to shut down the mill."
Charlie used the story to illustrate the intuitive grasp of the incidence of costs due to price elasticity. Everybody in the industry had been sold on the benefits of the new technology, and everybody was making the capital investments necessary to improve the quality of their cloth simultaneously. As a result, none of them were going to be able to pass those costs onto the customers in the form of increased prices.
I watched the same thing happen to the display advertising industry in the 2010s. The price elasticity of demand and the concomitant effects it has on the incidence of capital investment (or taxation) is a surprising subtle concept that is difficult for people to grasp, much less translate into strategy.
Now let's say that we have a technology that truly is the last of its kind — a technology that is so wondrous and important in value, that nobody who does not own this technology is going to be capable of meeting the needs of customers in the same way. Well then, naturally, every seller of goods and services is going to invest in that technology. It's either that or unwind.
So then what moat is left to decide which of them is going to be able to pass their capital expenditures on to customers in the form of increased prices? What is the last symmetry broken among these firms?
There are, of course, different answers for different types of goods and services. But if I could pick just one word, which I think sums up what it is that ties all the winners together: trust.
the Anthropic IPO feels like a live test of this, the valuation came in at up to $1.2tn which is well under the $2tn some bankers had been floating and that's still a huge number for a business where the next lab can undercut you on price, and I'll be watching how it trades once it's on Nasdaq because that's where the moat vs consumer surplus question actually gets answered
Vertically integrate to Drive down costs is the way and I think the Models are just so good to bring about so many Vertical Integrations that suddenly it is about whether you keep investing to make your models better or you invest to build out your Vertical Integrations. But it is simply not a matter of Money here. I was at CONFIG, the Figma conference, I realised that they are the ones who are poised to integrate the entire stack of building software. It takes a different kind of company altogether. And I wonder if they should try to acquire Figma like xAI acquired Cursor.
A miracle tool is not a business. I see agents grab a pile of new AI apps and think they're ahead, then the next person has the same stack a week later.
The part that still holds is the follow-up, the relationship, and whether people trust you when it's time to list. Cheap tools just make that part more obvious.
Fantastic Article.
The example of Claude chipping away at the CUDA lock-in in just a few days highlights the central paradox of this entire cycle: frontier intelligence naturally eroding its own defensibility.
And how most technological surplus eventually leaks to consumers through competition and commoditization.
What makes AI unique is that the technology is actively accelerating that process.
The moment an AI model becomes capable enough to port complex software across different hardware or reverse-engineer specialized techniques in an afternoon, switching costs evaporate.
The smarter the model, the faster it eliminates cognitive moats.
That’s why the Vanderbilt parallel hits the mark. Vanderbilt didn’t win long-term because he had superior steam engines; he won because he owned the physical right-of-way into Manhattan.
If software architectures and model weights can be rapidly imitated by the very tools they produce, the only durable moats left won't be intellectual or digital. They will be the stubborn, physical bottlenecks that code cannot optimize away...
grid capacity, interconnect queues, physical upkeep, and real-world distribution.
The defining question for this boom isn't just whether capital is overbuilding supply, but whether frontier models are inherently engineered to destroy the pricing power required to pay for them.
Curious to know where you're concentrating your effort with this insight in mind?
good read! i read it in intervals when my agents were puttering, pondering, dily-dalling (and whatever weird verbs claude uses for its magic).
i can't wait for when i can capture 98% of the consumer surplus, i want opus in my fucking microwave. hate these buttons with everything in me.
Like clockwork, Packy puts out another banger here!
There's a story that I remember Charlie Munger used to tell about Warren Buffett. There was a new weaving technology that was rolling out across the weaving industry, and the management came to Warren and said: "Check this out! We're going to invest in this and it's going to lead to higher quality cloth at lower cost, and it's going to be great!"
And Warren's immediate response was sad: "This is really too bad. We're going to have to shut down the mill."
Charlie used the story to illustrate the intuitive grasp of the incidence of costs due to price elasticity. Everybody in the industry had been sold on the benefits of the new technology, and everybody was making the capital investments necessary to improve the quality of their cloth simultaneously. As a result, none of them were going to be able to pass those costs onto the customers in the form of increased prices.
I watched the same thing happen to the display advertising industry in the 2010s. The price elasticity of demand and the concomitant effects it has on the incidence of capital investment (or taxation) is a surprising subtle concept that is difficult for people to grasp, much less translate into strategy.
Now let's say that we have a technology that truly is the last of its kind — a technology that is so wondrous and important in value, that nobody who does not own this technology is going to be capable of meeting the needs of customers in the same way. Well then, naturally, every seller of goods and services is going to invest in that technology. It's either that or unwind.
So then what moat is left to decide which of them is going to be able to pass their capital expenditures on to customers in the form of increased prices? What is the last symmetry broken among these firms?
There are, of course, different answers for different types of goods and services. But if I could pick just one word, which I think sums up what it is that ties all the winners together: trust.
the Anthropic IPO feels like a live test of this, the valuation came in at up to $1.2tn which is well under the $2tn some bankers had been floating and that's still a huge number for a business where the next lab can undercut you on price, and I'll be watching how it trades once it's on Nasdaq because that's where the moat vs consumer surplus question actually gets answered
miracles without moats. wine knows that feeling.
Vertically integrate to Drive down costs is the way and I think the Models are just so good to bring about so many Vertical Integrations that suddenly it is about whether you keep investing to make your models better or you invest to build out your Vertical Integrations. But it is simply not a matter of Money here. I was at CONFIG, the Figma conference, I realised that they are the ones who are poised to integrate the entire stack of building software. It takes a different kind of company altogether. And I wonder if they should try to acquire Figma like xAI acquired Cursor.
A miracle tool is not a business. I see agents grab a pile of new AI apps and think they're ahead, then the next person has the same stack a week later.
The part that still holds is the follow-up, the relationship, and whether people trust you when it's time to list. Cheap tools just make that part more obvious.