Episode summary
Dylan Patel – Two labs will soon control most of the world's workforce
Dwarkesh Podcast
· · 1 hr 17 min · Hosts: Dwarkesh Patel · Guests: Dylan Patel
Summary
The conversation argues that AI labs are rapidly becoming the central buyers of the world’s compute. The host and Dylan Patel discuss a shift from inference-led economics toward training-heavy spending as models improve, with Anthropic and OpenAI already taking a growing share of new capacity and possibly controlling most usable FLOPs by the end of 2028 if current trends continue. Patel frames this as a result of improving monetization: if a lab can turn $10–15 million per megawatt into $50 million or more, it can keep bidding up compute and reinvest profits into even larger training runs.
A major theme is that this buildout is not just a software story but a supply-chain and capital-markets story. Patel says the industry could require trillions of dollars of annual capex by the end of the decade, with hyperscalers, debt markets, and infrastructure suppliers all pulled into the race. He thinks the pressure could raise interest rates, crowd out other borrowing, and hurt non-AI equities and debt-heavy countries. He is especially bearish on countries like Pakistan or Nigeria in that regime.
Patel also argues that regulation and safety constraints may be one of the few forces slowing concentration. In his view, if frontier labs are prevented from releasing their best models, their revenue per megawatt rises more slowly. But absent that, he expects the largest labs to keep shifting more compute toward training, not inference, to pursue AGI and future earnings growth.
Key takeaways
- Dylan Patel expects Anthropic and OpenAI to keep taking a larger share of new compute and says they could control most usable FLOPs by the end of 2028 if current trends persist.
- Patel says frontier labs can already monetize compute at roughly $50 million per megawatt in some cases, versus a much lower build cost, which lets them reinvest operating profits into training.
- The host and Patel argue that AI buildout could require trillions of dollars of annual capex by 2028-2030, forcing hyperscalers and the credit markets to fund more of the expansion.
- Patel thinks higher AI-driven capital demand could push up interest rates, crush long-duration equity valuations, and leave debt-heavy countries such as Pakistan or Nigeria especially exposed.
- Patel’s main counterforce to centralization is regulation: if frontier labs cannot release their best models, their revenue-per-megawatt growth and bidding power could slow.
Notable quotes
“The labs are also taking an increasing percentage of this”
“By the time you’re towards the end of 2028 — if this trend continues, and I see nothing that’s stopping it — you’ve got them just controlling most of the usable flops in the world on their own”
“The way the supply chain is expanding currently, 100 is roughly still the right number”
“I think the US will be fine at the end of the day”
How the episode opens
Okay, I’m back with Dylan Patel, founder of SemiAnalysis. Our version of a family Thanksgiving dinner is a regular yearly podcast. But we're not actually related.
Don’t tell the people this. It will destroy the myth. Basically where the world economy is headed is more and more becoming a function of where lab economics are headed, where the compute market is headed, et cetera.
I want to understand where the crazy future ends up within a few years. But let’s start with where we are today. Walk me through lab compute and lab revenue right now, and maybe project out a year or two.
When we go back to last year, even at the end of the year, most of GDP growth in America was just AI infrastructure. As we look towards this year, about a third of the compute coming online is for the labs, for OpenAI and Anthropic. It may be built by others and then rented to them, but at the end customer, it’s them.
As we go forward into the future, the numbers for compute are ballooning. We’re at a little bit over a trillion dollars of CapEx this year. As we go out into ’28, it’s going to be more than $2 trillion.
The labs are also taking an increasing percentage of this. So ultimately, you’ve got a very interesting situation where the labs are going from companies that spend tens of billions of dollars a year to hundreds of billions of dollars a year, to forecasting to spend trillions of dollars a year even towards the end of the decade. This is at least some of the contracts they’ve begun signing with their partners.
This requires a big reshaping of what happens with their economics. Up until now, they have been companies that mostly lost money.
From the episode’s captions: roughly the first two minutes, lightly cleaned. Keep watching on YouTube →
Topics
- ai compute centralization
- lab economics
- inference to training shift
- ai capex and sovereign debt risk
- compute pricing and surplus capture
Companies and tickers mentioned
- Anthropic
- OpenAI
Watch the episode
Watch “Dylan Patel – Two labs will soon control most of the world's workforce” on YouTube
Summary, takeaways and quote selection by Octagon. Quotes are short excerpts; the full conversation is in the episode.
More from Dwarkesh Podcast
-
What Robert Caro learned about power after studying it for 50 years -
What will computer chips look like in 2045? -
How did a few hundred Spanish soldiers topple two empires? – Si Sheppard -
OpenAI researcher on agent swarms & recursive self-improvement -
AI researchers debate how close we are to recursive self-improvement -
General relativity from first principles – Adam Brown
Dwarkesh Podcast, this episode and its artwork belong to Dwarkesh Patel. Octagon is not affiliated with the show; this page is an independent summary.
