Noah Smith just mapped the nightmare scenario where AI monopolies become as powerful as nations, and his math is uncomfortably plausible.
The Summary
- Noah Smith explores a future where a handful of AI companies control trillions in capital and reshape global power dynamics, calling it the "Robot Lords" scenario
- The mechanism: AI agents automate knowledge work at scale, profits concentrate in model owners, and capital compounds faster than governments can respond
- Key risk isn't just inequality, it's that these companies could become parallel power structures competing with nation-states for authority
The Signal
Smith isn't writing sci-fi speculation. He's war-gaming what happens when today's AI economics extend forward a decade. If AI agents replace enough high-skill labor, and a few companies own the models doing that work, you get revenue concentration that makes Big Tech's current dominance look quaint.
The mechanism he lays out: knowledge workers get automated, companies pay license fees instead of salaries, and those fees flow to maybe five companies controlling frontier models. Those companies reinvest in compute and talent, widening the moat. Capital compounds. Governments that moved slowly on antitrust now face entities with treasuries rivaling small nations.
"The real question isn't whether AI creates wealth, it's where that wealth accumulates and how fast."
Here's where it gets interesting for the Web4 thesis. Smith focuses on centralized AI labs becoming "Robot Lords," but he's describing the exact problem Web3 tried to solve with Web2 platforms. Concentrated ownership of infrastructure. Rent-seeking at scale. Users and builders creating value, owners capturing it.
The difference: AI agents are more powerful infrastructure than social networks ever were. A platform that mediates human connection is one thing. A platform that replaces human cognition is another order of magnitude. If OpenAI or Anthropic becomes the cognitive infrastructure layer for half the economy, "delete Facebook" isn't an option anymore.
What Smith doesn't explore deeply: the counterfactual where AI development stays fragmented. Open weights models, smaller specialized agents, crypto-based compute markets. The tools to build a decentralized AI economy exist today. Bittensor, Ritual, inference marketplaces on Solana. The question is whether they scale fast enough to matter before the concentration lock-in happens.
The Implication
This isn't a prediction, it's a blueprint for what to watch. If you're building in AI, the next two years determine whether you're building on rented land or owning infrastructure. If you're investing, the concentration thesis Smith outlines makes both the mega-cap AI plays and the credible decentralization projects interesting, for opposite reasons.
The real action is in governance. Governments are already behind. The question is whether they wake up when these companies hit $5 trillion market caps or $10 trillion. By then, the "Robot Lords" have enough capital to shape the rules being written about them.