← Back to issue
Dwarkesh Podcast

The better AI gets, the smaller its share of the economy might get – Alex Imas and Phil Trammell

1h 16m · Transcribed via assemblyai · Watch on YouTube

Alex Imas (Director of AGI Economics at Google DeepMind, Professor of Economics at U Chicago) and Phil Trammell (head of Economics at Epoch, research scholar at Stanford) with Dwarkesh Patel. The throughline is the title's counter-intuitive thesis, and it cuts directly against this week's **$4T AI-IPO-wave euphoria**: the better AI gets, the *smaller* the AI/capital sector's share of the economy might become — a Baumol-effect argument that value accrues to what stays scarce, and the only durably scarce thing is **humans in the loop ('the relational sector')**. The mechanism: as automation drives the price of machine-made goods toward zero, demand for them satiates, and a constant or rising fraction of spending flows to the human-intrinsic services AI can't replace — which is exactly how, after two centuries of automation, **labour share has stayed 'over 60%'** and US prime-age employment sits at its highest since 2000. The load-bearing caveat for AI-company value capture is the compute counter-case: *'an H100 costs more to rent now than it did three years ago, even though we have much superior technology'* — *'because as models get smarter, the opportunity cost of compute gets higher'* — the one good whose demand may never satiate, which is the whole bull case for the labs. Imas's discipline note is that **the debate is data-free** (*'we need a Manhattan Project for data'* on demand elasticities), so the right move is scenario-mapping, not point forecasts. The most investable conclusion lands squarely on the IPO thread: the value-capture question reduces to whether AGI is **'like electricity or social media'** — a commoditised utility whose gains diffuse to users (you can't capture it, so *'just buy the index'*), or a platform that keeps the rents. Both economists *want* the labs commoditised — Imas because concentration makes the labs *'a very tangible, clear political target'* (he cites the Defense Production Act threat against Anthropic), Trammell flagging the one cost: commoditisation removes the safety buffer of a leader–laggard gap. The bear read on the entire **OpenAI/Anthropic IPO cohort**: the more transformative and broadly-adopted AI becomes, the harder it is for any single lab to keep the surplus — the same logic Mercor's CEO ran this week ('application-layer companies have no defensibility').

Key points

Notable quotes

Something like the relational sector, which is what I defined as basically services and goods, where the fact that the human was in the loop was actually part of the value of that product. And so because humans are naturally scarce, if we have automation, where a lot of other things stop being scarce, we will still have scarcity in things that humans are kind of involved in and in the loop for.

Alex Imas · 0:44

If you don't take anything out of this conversation for me, we don't have any data. I've been kind of saying we need a Manhattan Project for data. We don't have data on basically consumer demand elasticities. We don't know what they are.

Alex Imas · 2:24

It's incredibly surprising that it's over 60% after the industrial revolution, after all of the automation we've ever seen... Andy Atkinson has this paper showing that actually if you keep the accounting concep distant over the years, labor share hasn't even fallen ever.

Alex Imas · 7:14

The famous fact here is that an H100 costs more to rent now than it did three years ago, even though we have much superior technology and we have much more compute in the world. Because as models get smarter, the opportunity cost of compute gets higher.

Dwarkesh Patel · 15:33

If that increasing variety is fast enough, and there is no such increasing variety in the human sector, then you can get all of the relational that you want. But it doesn't matter for labor share, it goes to zero.

Alex Imas · 11:51

A lot of the downstream benefits actually came to like the users of the electricity rather than the, rather than the actual entity producing the electricity. On the other hand, with social media it was the opposite case. Right? Social media, you know, it was everywhere. Everybody uses social media, but the rents went to the platform.

Alex Imas · 1:06:47

So I think there is a world where it is concentrated in which case it's going to be really hard to index AGI. There is another world where it is not. It's electricity. Then like basically every company has access to AGI. So you just buy, you just buy the index.

Alex Imas · 1:09:34

One big cost of having commoditized frontier AI models, which is the tech race dynamic, that for safety purposes, you might want fewer frontier companies so that each one has a buffer in case they want to slow things down to make things safer.

Phil Trammell · 1:14:10

Themes

Mentioned

People

Companies

Ideas