In fact, we’ve gotten so good at scaling AI with GPUs that power not compute is the primary limitation.
Words found on the source page on Nov 6, 2026.
More from Martin Casado
EssayBitter EconomicsNov 19, 2025
AI projects, on the other hand, are taking far more capital, by far smaller teams, and turning that into historic growth rates for both new users and revenue.EssayBitter EconomicsNov 19, 2025
I’d argue that whether or not AGI emerges from a single model, the impact of this meta economic machinery is functionally similar. That is, it reduces a massively broad class of technical problems to simply a matter of economics.EssayBitter EconomicsNov 19, 2025
There’s a reason the fastest-growing AI companies are hiring software engineers as fast as they can.EssayBase AI Policy on Evidence, Not Existential AngstDec 16, 2024
But despite all the hubbub and competing interests, there actually is a reasonable policy position the United States can take: to focus on marginal risk and apply our regulatory energy there.EssayBase AI Policy on Evidence, Not Existential AngstDec 16, 2024
In the end, we might conclude the best policy for human welfare is to invest aggressively in AI rather than to encumber it.EssayBase AI Policy on Evidence, Not Existential AngstDec 16, 2024
We should only depart from the existing regulatory regime, and carve new ground, once we understand the marginal risks of AI relative to existing computer systems.
On the same topics
BlogBanning self-recursive improvement in AI models?Sep 22, 2026
If the top U.S. labs have to give up on what is supposed to be really important, they will look very “catchable.” The policy creates a big incentive for some other parts of the world to invest a lot more in compute.BlogFrontier OverhangsSep 21, 2026
A world with sufficient compute is one where intelligence is a commodity, and in commodity markets margin comes from a superior cost structure, which I would expect the leading model providers to have, in part because they have a meaningful lead in scaling and can apply superior AI to their infrastructure.BlogWhy I still haven't bought into true RSISep 19, 2026
I have my doubts that the labs can afford to spend a constant portion of this compute on internal R&D as the total volume goes up, especially with plans to IPO, as they face increased scrutiny on basic economics.BlogWhy I still haven't bought into true RSISep 19, 2026
It is important to not confuse massive steps in inference-time scaling, a dynamic which should be fairly predictable, with being the outputs of RSI, which is highly uncertain.PodcastDwarkesh Podcast: Agent swarms, alignment, and recursive self-improvementSep 17, 2026
If you have five hours, you’re probably going to do a lot better. The AI models are pretty similar.EssayAI Needs Evidenced Standards, Not A Cartel (Who Gets to Define the Rules for AI?)September 2026
Within a generation, the way we discover medicine, manage power grids, and secure our national infrastructure will be completely transformed.
About this quote
Quotesby Glyph