Back

Alex Imas And Philip Trammell On Owning The AI Upside

"It's not enough for them to just own the S&P 500."

Watch the recap video here

Dwarkesh Patel and Dwarkesh PodcastDwarkesh Patel and Dwarkesh PodcastsourceAlex ImasAlex ImasspeakerDwarkesh PatelDwarkesh PatelspeakerPhilip TrammellPhilip TrammellspeakerWatch full clip

Source Moment

Dwarkesh Patel asks whether normal people and developing countries can share in AI wealth by owning broad market exposure if the gains concentrate in private labs, AI-memory suppliers, fabs, and infrastructure. His concrete version is Nigeria: owning the S&P 500 may not be enough if the important assets are things like SK Hynix, Anthropic, and other pieces of the AI supply chain.

Context

Dwarkesh Patel published this YouTube interview on June 4, 2026. The guests are Alex Imas and Philip Trammell.

Imas is a professor at the University of Chicago Booth working in behavioral science, economics, and applied AI. Trammell is an economist connected to Stanford's Digital Economy Lab and Epoch, with work on economic growth and AI.

The episode is framed as an economics-of-AGI discussion. AGI means artificial general intelligence: AI systems able to perform a broad range of economically useful work rather than one narrow task. The conversation covers labor share, capital share, scarcity, redistribution, demand, machine workflows, and developing-country strategy.

Full Recap

Dwarkesh Patel interviews Alex Imas and Philip Trammell about the economics of advanced AI. The source asks what remains scarce if AI can automate much more work, and who owns the assets that collect the gains.

Better AI does not automatically make everyone richer in the same way. It can make some goods cheap, shift spending into new categories, raise demand for compute and infrastructure, or concentrate returns in private companies and hard-to-own bottlenecks. The hard part is predicting what stays scarce and who owns the scarce assets when AI changes prices, demand, supply chains, and what people want.

The weird word that unlocks the middle of the episode is "satiation." It means reaching the point where more of a category stops adding much value. Trammell's Mongolian economist example makes it vivid: if you only know horses, yogurt, yurts, and singers, you might predict automation makes the first three cheap and everyone spends the future on singers. History did something else. It created many new non-singer things to want.

  • 0:00-19:36:Capital Share, Labor Share, And What Remains Scarce: The opening defines the core economics. Labor share is income paid to workers; capital share is income paid to owners of assets. AI can automate tasks, but economy-wide shares depend on prices, demand, supply chains, human-valued services, and whether AI creates new capital-made goods faster than people satiate on old ones.
  • 30:02-39:26:Why Demand Collapse Is Hard To Get: Imas pushes against a simple story where automation makes things abundant and the economy shrinks. If AI makes something cheaper, people may buy more of it, firms may invest the savings, and new product categories may appear. Negative growth from abundance requires hard demand limits, weak investment demand, and little new variety.
  • 1:01:28-1:16:08:Developing Countries, Indexing AI Gains, And Concentrated Returns: The final section turns the episode into an ownership problem. India, Nigeria, and other countries outside the AI supply chain may not benefit enough from global AI growth if the biggest returns sit in private model labs, chip suppliers, fabs, data centers, and infrastructure companies. The electricity-versus-social-media comparison asks whether AI rents spread broadly through downstream users or stay concentrated in a few platforms and bottlenecks.
  • 19:36-25:57:The Messy Middle: Slow Displacement And Political Pressure: The transition risk is a period where AI pressures wages and jobs before society sees a clear abundance dividend. That can create underemployment and political stress without an obvious compensation target.
  • 25:57-30:02:Taxing And Redistributing AI Wealth: The policy section separates taxes, transfers, and ownership. Universal basic capital sounds cleaner than a fragile benefit program, but it depends on owning the right assets before the winners are obvious.
  • 39:26-43:08:Why Humans May Be Hard To Integrate Into Machine Workflows: Humans may remain valuable in trust-heavy or relational settings, but they can become slow, noisy parts inside fast machine-native production. That distinction matters for whether "humans still matter" preserves labor income.
  • 43:08-1:01:28:Intrinsic Wealth Accumulation, Status, And Capital Demand: The most speculative section asks whether some people, institutions, or future AIs keep accumulating capital because they value status, influence, control, or expansion itself. That could keep demand for capital high even when ordinary consumer demand saturates.

Technical Need To Knows

  • Labor share: The share of economic income paid to workers as wages. The episode asks whether this shrinks when AI can do more work.
  • Capital share: The share of economic income paid to owners of assets such as companies, machines, chips, data centers, land, and infrastructure.
  • Scarcity: Something remains scarce when people still compete for it and pay for it. The source asks what remains scarce after AI changes production costs.
  • Satiation: The point where more of a good or category stops adding much value. If people satiate on AI-made goods quickly, spending can move elsewhere.
  • Demand elasticity: How much buying changes when prices change. If AI makes a service cheaper and people buy much more of it, total activity can rise.
  • Jevons paradox: Efficiency can increase total use when cheaper access expands demand. The episode uses this as a guardrail against lazy demand-collapse stories.
  • Universal basic capital: A redistribution idea where people receive ownership stakes in capital assets rather than only cash transfers.
  • Indexing: Owning a broad basket of assets so you share in market gains without picking a single winner. The episode asks whether broad indexes are enough if AI returns concentrate in private or specialized assets.
Back to allocation feed