Cory Doctorow's new book pegs global AI capital expenditure at $1.4 trillion against roughly $50 billion in annual sector revenue, a gap he argues makes the current AI mania structurally larger than any prior tech bubble. The Reverse Centaur's Guide to Life After AI, out from Macmillan, lands as seven AI companies account for more than a third of the US stock market and, in Doctorow's framing, pass around the same $100 billion IOU among themselves. Doctorow followed last year's Enshittification with a book he says he didn't want to write but felt compelled to: the numbers, in his telling, no longer leave room to ignore the question.
The CapEx figure has doubled in the time it took to write the book, climbing from $700 billion to $1.4 trillion. Meta alone spent $150 billion on AI over the last three years and has guided to another $150 billion this year, after writing off roughly $60 billion on the metaverse. The asset base behind those numbers depreciates fast: Doctorow notes AI infrastructure typically needs full replacement every 24 to 30 months, meaning the capital cycle has to keep spinning just to stand still.
Doctorow's thesis ties the AI buildout to a structural problem in public-company finance. Firms with 90% market share in their core business cannot grow organically and need a growth-stock narrative to maintain equity liquidity. Prior cycles cycled through the metaverse, crypto, and Web3 as imaginary markets to absorb that need. AI, he argues, is different in size and durability because the underlying computer science is real — a point he dates to roughly 10 years ago, when researchers applied existing techniques in a new way and got linear returns on investment that rarely show up in research.
Key facts
- 01Global AI CapEx has doubled from $700 billion to $1.4 trillion since Doctorow began writing the book.
- 02Seven AI companies now account for more than a third of the total stock market, per Doctorow.
- 03The AI sector turns over roughly $50 billion in annual revenue against that $1.4 trillion in CapEx.
- 04Meta spent $150 billion on AI in the last three years and plans another $150 billion this year, after wasting $60 billion on the metaverse.
- 05AI infrastructure assets need full replacement every 24 to 30 months, compounding the capital problem.
But the returns are tapering. Doctorow quotes the finance maxim that anything that can't go on forever has to stop, and argues the low-hanging fruit in scaling is largely gone. The economic structure underneath the bubble — billions in annual losses at the foundation-model layer, hyperscaler CapEx that lapses every two and a half years, and a customer base nowhere near large enough to amortize either — does not, on his read, support the current valuations.
The ideological pull, Doctorow argues, is the appeal to executives and political leaders of a world without workers. He cites DOGE's firings of government workers as an expression of the fantasy that government can run without government employees, and frames corporate AI adoption similarly: a way for leaders to bypass the people who actually know how the work gets done. In his automation framework, a centaur is a human augmented by a machine; a reverse centaur is a human reduced to a peripheral for one. The Amazon delivery driver surrounded by AI cameras is his canonical example.
The medical analogy in the book is sharper. Doctorow contrasts using AI to help radiologists catch tumors they would otherwise miss with firing nine out of ten radiologists and making the remaining one liable for checking the machine's work. The first is augmentation. The second, he argues, is the actual deployment pattern the economics of the bubble require — because cheap useful tools do not justify $1.4 trillion in spend.
“The capital markets have the object permanence of a toddler, and they would lose a game of peekaboo if they were drafted to play in the league.”— Cory Doctorow, Author and tech journalist
Doctorow is not anti-AI in the categorical sense. He uses AI tools and acknowledges genuine utility in many of the products shipping today. His objection is to the scale of capital being raised against products that, on his numbers, cannot generate the revenue required to service it. When the investment mania halts, he argues, most foundation models will disappear because running the data centers will stop being economical.
There are reasonable counterarguments he does not fully engage. Revenue at the application layer is growing fast — OpenAI, Anthropic, and a handful of vertical players have moved from near-zero to multi-billion-dollar annualized run rates inside three years. Inference costs are falling by orders of magnitude per model generation, which changes the unit economics of the depreciation cycle he cites. And the comparison to the metaverse is uneven: metaverse spend produced almost no consumer traction, while ChatGPT alone reports hundreds of millions of weekly users. Whether those data points add up to a defense of $1.4 trillion in CapEx is the open question.
The case Doctorow makes lands hardest on the financing structure rather than the technology. A sector turning over $50 billion against $1.4 trillion in CapEx, with a 24-to-30-month replacement cycle, needs either revenue to grow roughly an order of magnitude or capital costs to fall sharply — and probably both — for the math to work without a sharp correction. That is a falsifiable claim, and the next four quarters of hyperscaler guidance will go a long way toward testing it. For AI companies, the strategic implication is straightforward: the firms that survive a correction will be the ones that can show real revenue per dollar of compute, not the ones with the largest cluster commitments. The bubble debate is now an underwriting debate, and underwriters eventually ask for numbers.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
