Doctorow: AI Bubble's Roots Are Monopoly Power, Not Hype

In an Ars Technica interview, Cory Doctorow opens his case with blunt directness: "Of course AI is a bubble." But the argument in his book, "The Reverse Centaur's Guide to Life After AI," is less about forecasting a crash than about identifying who bears the risk and what structural changes would actually redirect it.

The core claim is a capital markets problem, not a capabilities debate. Monopoly-era tech firms with saturated core businesses need perpetual growth narratives to sustain stock valuations, and the "AI can do your job" pitch is the current installment. In practice, Doctorow argues, these deployments produce "reverse centaurs": workers forced to serve machine-paced workflows rather than the other way around. The financial math is stark. Sequoia has calculated that the AI sector needs to generate roughly $600 billion in annual revenues to justify current infrastructure spending, while OpenAI reportedly generated about $3.4 billion last year. Doctorow also notes that electrical power is a fundamental limiting factor, with chips being produced faster than they can be powered.

Who gets hurt when the math resolves is where the argument sharpens. Bubbles work, in Doctorow's framing, as a mechanism for insiders to pump and dump mania onto ordinary investors who rely on market-based pensions rather than defined-benefit ones. The insiders exit; the pension holders absorb the correction. His prescriptions aim at that structural dynamic rather than at hype: stronger antitrust enforcement, sectoral bargaining rights, unemployment cushions, and public procurement that favors open AI models over proprietary ones.

The piece is grounded by a specific counterexample. The Innocence Project of New Orleans used AI to identify arrest reports with linguistic correlates matching successful exonerations, helping lawyers prioritize cases and get innocent people out of prison. Doctorow frames this as evidence of a post-bubble residue that could prove durable: cheap hardware, open-source models running locally for tractable tasks rather than wholesale labor replacement.

What the interview does not settle is timing. The distance between "this is unsustainable" and "this ends soon" is where most bubble arguments run out of traction, and this one does not close that gap.