One thing I’m seeing across multiple sectors is that AI isn’t just accelerating innovation, it’s exposing how little spare power capacity our grids actually have. The next wave of breakthroughs won’t come from chips alone, but from the ability to deliver reliable, high-density electricity at scale.
That’s why firm power, especially nuclear and modular reactors, is starting to show up in the planning discussions for AI infrastructure. Optimism accelerates when the physical layer can keep up.
But totally agreed- I'm 2/3 through writing a series on 1) why energy (among other things) is an AI bottleneck 2) what that means for hyperscalers 3) what happens when they get through that bottleneck and inference costs crater
Also agree that nuclear's generation style is a great fit for the smooth/consistent energy needs of data centers, so super excited to see progress there both in established methods and startups like Valar and Radiant
100%, energy is becoming the primary rate limiter for hyperscale AI. Once inference cost curves flatten, the next competitive edge is who can secure stable, high-density power with predictable costs.
What’s interesting with Valar and Radiant is that they’re designing around that exact requirement: high availability, low operational complexity, and fast siting. That’s where modular nuclear starts looking less like “energy” and more like infrastructure strategy.
The Physical Intelligence demo really shows where we're at with embodied AI. What's interesting is how diffrent the training approaches are between π*0.6 and Sunday's Act-1. The RL approach makes sense for repetitive tasks where you can iterate quickly, but I wonder if the human demonstration method will scale beter for more complex, nuanced movements. Either way, having robots that can actually do useful household stuff feels like it's finally happening.
Curious about the Bezos image generation prompt, since it was Icarus who flew too close to the sun. Prometheus stole the fire from the gods. It was maybe the expected outcome but with a mythological mistake ignored?
One thing I’m seeing across multiple sectors is that AI isn’t just accelerating innovation, it’s exposing how little spare power capacity our grids actually have. The next wave of breakthroughs won’t come from chips alone, but from the ability to deliver reliable, high-density electricity at scale.
That’s why firm power, especially nuclear and modular reactors, is starting to show up in the planning discussions for AI infrastructure. Optimism accelerates when the physical layer can keep up.
More energy = more vibes
But totally agreed- I'm 2/3 through writing a series on 1) why energy (among other things) is an AI bottleneck 2) what that means for hyperscalers 3) what happens when they get through that bottleneck and inference costs crater
Also agree that nuclear's generation style is a great fit for the smooth/consistent energy needs of data centers, so super excited to see progress there both in established methods and startups like Valar and Radiant
100%, energy is becoming the primary rate limiter for hyperscale AI. Once inference cost curves flatten, the next competitive edge is who can secure stable, high-density power with predictable costs.
What’s interesting with Valar and Radiant is that they’re designing around that exact requirement: high availability, low operational complexity, and fast siting. That’s where modular nuclear starts looking less like “energy” and more like infrastructure strategy.
The Physical Intelligence demo really shows where we're at with embodied AI. What's interesting is how diffrent the training approaches are between π*0.6 and Sunday's Act-1. The RL approach makes sense for repetitive tasks where you can iterate quickly, but I wonder if the human demonstration method will scale beter for more complex, nuanced movements. Either way, having robots that can actually do useful household stuff feels like it's finally happening.
Curious about the Bezos image generation prompt, since it was Icarus who flew too close to the sun. Prometheus stole the fire from the gods. It was maybe the expected outcome but with a mythological mistake ignored?
Oh very odd! Good catch
Thank you. In case anyone is new to all the new slides/infographics generation from Gemini/Nano Banana/NotebookLM, here's a writeup https://acaiberry.substack.com/p/google-gemini-creates-informative
Irelands gigantic offshore wind potential: https://open.substack.com/pub/yenurmasem/p/the-great-atlantic-engine-a-600gw?r=6w0ag&utm_medium=ios
Another super uplifting post... want 3 of the Sunday robots for the holidays... spare my back.. :-)
If power is the bottleneck, it feels like China has a huge advantage and perhaps factors into our geopolitical decisions beyond rare earths