Michael Burry, the investor who made his name shorting the 2008 housing bubble, has turned his attention to Anthropic's IPO filing and concluded he wants no part of it. His core objection is structural, not sentimental: Anthropic has committed to $895 billion in compute spending — chips, infrastructure, AI capacity — while even the most optimistic analysts project next year's revenue at roughly $100 billion. That's a 12:1 ratio of committed costs to hopeful income. The numbers surfaced in Anthropic's amended S-1 prospectus filed November 2. Q3 2026 revenue surged to $17 billion, a figure that sounds impressive until you read the fine print about computing costs that are already high and projected to keep climbing indefinitely. Burry's read: the market sees these numbers but doesn't fully process what they mean for the company's margin structure and survival. Burry's specific concern centers on Anthropic's "take-or-pay" contracts with major hyperscalers. These agreements obligate Anthropic to pay for computing capacity whether or not it needs it. In a downturn or a period of lower demand, the company still pays full freight. This inflates costs during slack periods and erodes competitive position against leaner language-model providers who can scale spending with demand. This wasn't always Burry's position. In April 2026, he praised Anthropic as "completely winning against Palantir" and noted the market was shifting toward plug-and-play AI model integration that would benefit Claude. Something changed between April and November, and it was the S-1's disclosure of the sheer scale of fixed commitments. The shift from bullish-on-product to bearish-on-structure is telling — Burry isn't questioning whether Claude is good, he's questioning whether the business model can survive its own cost base. "The company is not merely priced at a valuation that bakes in optimism — it has bet its very existence on that optimism," Burry wrote on Substack. He has no plans to acquire shares or become a shareholder in the near term. The implied argument: if AI demand growth slows even modestly, the fixed-cost structure turns Anthropic from a growth story into a cash incinerator. The broader signal here extends past Anthropic. Every frontier AI lab is making massive compute bets, but Anthropic's S-1 is the first to put exact dollar figures on paper for public scrutiny. The $895 billion figure is a stress test for the entire AI investment thesis — not just whether these models work, but whether the economics of building them can ever pencil out at the prices the market demands. Burry's track record is mixed — he's famous for being early and right on housing, but his post-2008 calls have been uneven. Still, the structural concern here isn't about timing or sentiment. It's about arithmetic: can any revenue trajectory grow fast enough to service $895 billion in obligations before those obligations crush the balance sheet?
