--- title: "Is AI a bubble? What the record says · The AI Compute Race" source: https://dryburgh.com/vol1/bubble description: "Is AI a bubble?, from The AI Compute Race. Is AI a bubble? What the record says, and what would settle it." --- # Is AI a bubble? Updated 31 October 2026 **Is AI a bubble? What the record says, and what would settle it.** Every sentence below is the book’s own, from chapters 11 and 12 of THE AI COMPUTE RACE. The book does not answer the question with an opinion. It defines what a destructive outcome would be, names the readings that would show it, and dates each one. ### What the book claims, and has not found Destructive finance has not been found. This section defines it first, so that a finding, if one comes, cannot move the definition. The claim, put plainly, is that this is a mania in shape and not yet in outcome. ### What would count The other direction is read on 31 December 2028. Destructive, here and wherever this claim uses the word, means a default, a restructuring or a write-down of debt raised for AI infrastructure or of the assets behind it, of $1 billion or more in a single instance, in an issuer’s or a lender’s own filing by that day. The thresholds are three readings that can each fire alone. None of this is the destructive definition. It is the shape arriving before the outcome, which is what the claim says a mania looks like from inside. ### The bulls, in their own words The seller of the chips, Jensen Huang, said the same at greater length. He dated his own bubble: “The bubble will come someday. It’s just not today.” Asked “Five years, ten years?” he answered “It’s very unlikely in the next five years”; asked “5 to 10?” he said it again and went on: “Yeah, it’s very unlikely in the next 5 years, and we’ll see about the 5 to 10 years. It just depends on how fast we can build.” The bulls’ own 2024 defense has expired on its own terms: “Prior bubbles also didn’t have the most profitable companies that humanity has ever created investing and they were debt financed. This is not debt financed yet,” Dylan Patel said in October 2024. It is now. Brett Winton, ARK Invest’s chief futurist, called the bubble narrative “flat wrong” on The Information’s program in November 2025, in a segment presented by AWS, which sells the infrastructure he was forecasting. ### The buyers, in theirs OpenAI’s Sam Altman, the largest buyer, had conceded in August 2025 that investors were overexcited, and a year later said a version of it again. “Are we in a phase where investors as a whole are overexcited about AI?” he asked, and answered, in an interview with *The Verge* and other reporters on 14 August 2025, a week after GPT-5 shipped: “My opinion is yes.” “Someone is going to lose a phenomenal amount of money. We don’t know who, and a lot of people are going to make a phenomenal amount of money,” he said, and in the same interview: “You should expect OpenAI to spend trillions of dollars on data center construction in the not very distant future.” Mark Zuckerberg had taken the same position with the same interviewer a year earlier, on the first episode of Heath’s podcast, published on 18 September 2025. Asked about the bubble talk, he said, “I mean, I think it’s quite possible.” Reminded that he had just told Trump that Meta would spend “like 600 billion,” he answered, “I did, yeah, through 2028,” and then: “And if we end up misspending a couple of hundred billion dollars, I think that that is going to be very unfortunate, obviously. But what I’d say is I actually think the risk is higher on the other side.” The risk on the other side is the wager of chapter 1, and the couple of hundred billion is the price he puts on being wrong about it. ### The economists Mohamed El-Erian, the economist who ran the bond manager PIMCO and now presides over Queens’ College, Cambridge, gave the same shape a sequence, at a Stanford policy forum on 20 November 2025, and he gave it as a qualified bull. It is “more likely than not that we are in an AI bubble,” he said, and the characteristics arrive in order, four of them, numbered in the strip below. His conclusion is not the bear’s. All of it, he said, “will lead to a much better outcome,” because it lets the capital markets put into the field “much more than government financing could ever do”: “this is not the Tulip Bubble,” and “the outcome is going to be that we’re better off.” Nick Srnicek, the political economist whose *Silicon Empires* gives an account of AI’s value-capture strategies, said in a talk released on 3 May 2026 that the bubble section was “the section that I would revise the most” of his own book. The bust he had written about “wasn’t going to have these sort of systemic effects,” and now “the impact of a bubble bursting would be much more significant today.” Two economists, Jared Bernstein and Ryan Cummings, who had called AI a financial bubble the year before, ran the test on six named spenders in a post of 25 September 2026, by a discounted-cash-flow model, which values future cash at today’s worth, whose method they set out. ### What the official bodies have measured The official bodies read the circle the same way through 2026, in eleven statements over seven months, each in its own words and each with its qualification. The shape has been measured; the outcome has not arrived. *Chapters 11, What the Machine Has to Earn, and 12, Cross-Examination. The dated tests are in The Dated Tests; the monthly reading is the Scorecard.* - [Campuses](https://dryburgh.com/vol1/ledger) - [Tests](https://dryburgh.com/vol1/tests) - [Scorecard](https://dryburgh.com/vol1/scorecard) - [Data](https://dryburgh.com/vol1/data/releases) - [Chapter by chapter](https://dryburgh.com/vol1/companion) - [The record at a glance](https://dryburgh.com/vol1/observatory) - [The machine](https://dryburgh.com/vol1/machine) - [One campus, four futures](https://dryburgh.com/vol1/hyperion) - [Glossary](https://dryburgh.com/vol1/glossary) - [People](https://dryburgh.com/vol1/people) - [Sources](https://dryburgh.com/vol1/sources) - [Notes](https://dryburgh.com/vol1/notes) - [Where the book was wrong](https://dryburgh.com/vol1/corrections) - [Where its sources were wrong](https://dryburgh.com/corrections?book=vol1) - [The Paper](https://dryburgh.com/vol1/paper) - [Consent](https://dryburgh.com/vol1/consent) - [Who builds and pays](https://dryburgh.com/vol1/parties) *The AI Compute Race: Five Races, One Machine* is a book by Lee S. Dryburgh, the first volume of THE COMPUTE ECONOMY, published by Dryburgh, an imprint of Contestra OÜ, Tallinn, Estonia; its record closes on 8 November 2026. [About the book](https://dryburgh.com/vol1)[Tell me when it is out](https://dryburgh.com/vol1#notify)