“Big Short” Investor: One Company Could Break The AI Boom

A fifty-minute quarter-end interview with Steve Eisman, the investor known from The Big Short. He makes two separate arguments about AI, and the second does not depend on the first.

The first is about motive. Eisman does not believe the extinction warnings coming from the labs. He thinks the last six months changed their business: companies that had told staff to use AI without limit ran through their budgets and stopped, and open-weight models began taking share. With “no moats around their business whatsoever”, he says, the labs are trying “to manufacture a crisis” that brings regulation they can shape against open models. He calls this a conspiracy theory himself, and adds a second: that Anthropic is going public now because the first half of 2026 looks better than the second will.

The second is about structure. Around 70% of the AI revenue at the big cloud companies comes from OpenAI and Anthropic, and 70% of Nvidia’s receivables sit with five customers. If either lab stumbles in the next year, the whole chain does. He adds debt to the picture: data-centre borrowing kept off balance sheets to protect credit ratings, which he compares to Enron and the vehicles of 2008, and about $500 billion of AI debt this year competing with the Treasury for buyers while the ten-year yield sits above 5%.

He does not call a bubble. He is mostly long, has hedged a little, and says the failure of a lab would mean a recession, “not the end of the world”. He has no interest in buying Anthropic at $2 trillion and thinks its 2025 results are beside the point. The one short he describes is FICO, whose regulator has just favoured a rival credit score.

The macro facts are right in outline and loose in detail. The Fed did raise rates in September for the first time since 2023, but the ten-year closed the month at 5.29%, its highest since 2007. The deficit figure in the transcript, $6 trillion, is about three times the real one. The Bear Stearns funds failed in June and July 2007, not May. The 70% concentration figure is an aggregation of analyst notes published by Ed Zitron, not something the companies report; the Nvidia figure is in its filing. His claims about spending caps and open-weight share, which the first argument needs, could not be checked.