Dwarkesh Podcast
Dwarkesh Podcast

Dylan Patel – Anthropic & OpenAI will have most of the world’s compute by 2028

August 25, 2026

AI Summary

5 min read

Dylan Patel – Anthropic & OpenAI will have most of the world’s compute by 2028

"By the time you're towards the end of 2028, if this trend continues, you've got them just controlling most of the usable flops in the world." That is the central forecast from Dylan Patel, founder of Seminalysis, who returns to the Dwarkesh Podcast to walk through the numbers behind AI compute concentration — and why the path to that outcome is anything but straightforward.

The arithmetic of concentration

The headline numbers are stark. At the beginning of 2024, Anthropic and OpenAI each operated roughly 2 gigawatts of compute. By the end of the year, both are above 5 gigawatts — a 3-4x increase. Of all new compute being added globally this year, about 30% is going to these two labs. Next year, based on contracts already signed, that share rises to 40-50%. And the trend is accelerating, not slowing.

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

What you'll learn

  • 1 (00:29) **Where AI Compute Stands Today** - Dylan Patel frames the current state of lab compute and revenue, noting that AI infrastructure is a major driver of US GDP growth.
  • 2 (03:34) **Centralization of Compute: The Labs' Growing Share** - Patel explains how the labs are taking an increasing percentage of the world's incremental compute, projecting when they will dominate.
  • 3 (05:24) **The Tripling Trajectory and World Compute Constraints** - Patel projects the labs' compute growth and addresses the physical limits of global compute supply.
  • 4 (07:10) **The Fab Economics: A 100x Discrepancy** - Patel explains why capitalism should drive massive expansion: fab capex yields enormous AI revenue.
  • 5 (10:29) **Cash Flows vs. Capex: The Labs' Funding Gap** - The labs' revenue is growing, but it still can't fund the needed expansion without massive debt.
  • 6 (13:16) **How High Can Compute Prices Go?** - Patel discusses the pricing dynamics as labs outbid everyone for compute.
  • 7 (16:15) **The Safety Slowdown: Regulation as a Brake** - Patel argues that safety regulation is slowing the labs more than it slows open-source or Chinese models.

+ Full timestamped outline available in the app

Guests on this episode

Dylan Patel

Show Notes

Had a lot of fun chatting again with my twin brother Dylan Patel.

We went through lab economics over the next few years - the shift from inference to training as RSI draws near; and how Anthropic and OpenAI are on track to control most of the world’s usable FLOPs within the next few years (because they can monetize compute better and thus outbid everyone).

And then we discuss whether the >$10T of total AI capex we’ll see by the end of the decade will cause a sovereign debt crisis, where hyperscaler debt raises interest rates, drives non-AI exposed countries into bankruptcy, and crashes non-AI equities.

One question we weren’t able to resolve is whether there’s anything that can counter all the forces barrelling towards centralization in this industry - the economies of scale in training, the scarcity of compute, and eventually continual learning and RSI.

Watch on YouTube; read the transcript.

Sponsors

* Grok Bot has been quite helpful with my search for a new editor. I created a recruiter bot and described the type of editor I was looking for. That bot then spun up a handful of subagents that combed through my emails and X DMs, read the end credits of various documentaries I like, and figured out who edits for some of my favorite YouTubers. It took all of those results, and then delivered me a shortlist of candidates that matched my criteria. Try Grok Bot for yourself at x.ai/bot

* Antithesis lets you add time travel to your software testing toolkit. Since the Antithesis platform is fully deterministic, everything that happens inside of it is perfectly reproducible. So if your software crashes, you can rewind to the exact right moment, freeze time, and investigate. Or you can test different hypotheses by perturbing the system: kill a node or disable a feature, see what happens, then reset the trajectory and try something else. Learn more at antithesis.com/dwarkesh

* Jane Street is hiring for two separate ML internships right now, one focused primarily on research and one focused on engineering. In both cases, interns are expected to contribute to real work, not contrived exercises: one common project is adapting a frontier LLM paper to financial markets, which tend to come with a ton of different gnarly challenges. Importantly, you don’t need any finance background to apply. 2027 applications are open now at janestreet.com/dwarkesh<

Dwarkesh Podcast

More from this podcast

 Dwarkesh Podcast →