Michael Burry has ruled out buying Anthropic stock. What risks is he warning about?

Investor Michael Burry, who predicted the 2008 mortgage crisis, said he has no plans to buy shares in the AI startup Anthropic, which is preparing for an IPO. On his blog, Cassandra Unchained, he analyzed the company’s financial metrics from its updated Form S-1 and explained why he has doubts about the sustainability of the business. What risks does Burry see for the company ahead of its potential IPO?

Anthropic is so dependent on extremely high revenue growth rates that, without them, it will be unable to cover its computing and infrastructure costs, Burry writes. In its updated IPO filings, the company disclosed liabilities totaling $895.4 billion. But in the future, these liabilities could turn out to be significantly higher: the figure does not account for expenses related to contracts with AMD and purchases of Google TPU computing power for 2028–2031, the investor noted. Anthropic itself estimates these costs at $150–175 billion—taking this into account, Burry calculates the total liabilities to be $1.05–1.07 trillion.

But even this figure may not be final: Burry believes that it is unclear from the company’s prospectus whether contracts with SpaceXAI totaling up to $120.5 billion are included in the disclosed $895.4 billion in liabilities. The investor speculates that if these contracts are included only partially or not at all, Anthropic’s total liabilities could rise to $1.17–1.19 trillion.

Burry believes that the uncertainty surrounding the SpaceXAI contracts also makes it difficult to estimate expenses for the coming year: Anthropic’s payments through the end of September 2027 could reach $100.8 billion, compared with the $69.8 billion disclosed in the filings.

Burry emphasizes that Anthropic needs extremely high revenue growth to cover all its expenses. According to his calculations, revenue must grow by 11.6% each month to cover expenses in 2027. To achieve a 20% GAAP profit margin, revenue would need to grow by 16.1% per month.

However, as Burry notes, Anthropic’s revenue growth has slowed significantly: in the first quarter, it grew by 132% compared to the previous quarter; in the second quarter, by 144%; and in the third quarter, by only 46.7%. He suggests that this slowdown is due to a reduction in excessive token usage: according to media reports, companies have begun to use computing resources more sparingly.

At the same time, Anthropic’s expenses for computing power and infrastructure are growing faster than its revenue: in the third quarter, they increased by 77.5%. The company is forced to purchase additional capacity under more expensive short-term contracts and warns that using this capacity will continue to erode profitability.

To reduce these costs, Anthropic is developing its own computing infrastructure, Anthropic Direct. However, Burry sees risks in this approach as well: to do this, the company enters into five-year leases for Broadcom equipment and data center leases for up to 20 years, thereby assuming the risk of technological obsolescence. Essentially, Burry argues, Anthropic is entering into an irrevocable “take-or-pay” contract with itself: the company becomes the sole consumer of its own computing capacity and assumes all costs associated with maintaining it, regardless of whether it actually needs it.

If Anthropic’s financial performance deteriorates, it could lose access to some of the funding from its strategic partners, Burry warns. This refers to agreements with Google, Amazon, and Broadcom, under which the company can raise up to $77 billion. Google accounts for $20 billion of this funding, Amazon for $15 billion, and Broadcom for $42 billion. In its IPO prospectus, Anthropic lists this support alongside its own liquid assets and an unused bank credit line: these are agreements under which the partners are willing to accept Anthropic securities as payment for equipment rentals and other services.

Another risk for the company stems from its obligations to Amazon. The tech giant has invested billions of dollars in Anthropic and also holds convertible bonds issued as part of financing agreements in 2023 and 2024. Converting the bonds into shares could dilute the stakes of Anthropic’s other shareholders following the IPO, Burry warns. According to his calculations, if Amazon converts all of its bonds into shares, its stake in the company could reach 20.6%. However, under the terms of the agreements, Amazon’s stake in Anthropic is capped at 12%, even after conversion. Burry believes that Amazon could circumvent this limit by retaining a portion of the bonds and converting them into shares at a later date, once their total number increases following the IPO.

Another challenge for Anthropic stems from the rising value of the convertible bonds held by Amazon: the AI startup records this as a loss in its financial statements, even though no actual payments are being made. From late June through late September, the book value of the bonds increased from $108 billion to $130.5 billion. Burry expects that even after the IPO, their revaluation will continue to have a significant impact on Anthropic’s reported earnings.

Optimistic expectations for revenue growth are already factored into Anthropic’s valuation, but for the company itself, maintaining that pace has become a matter of “survival,” writes Burry. The investor believes the market has already taken note of the scale of its obligations, but fears that investors will only realize the full gravity of the situation in hindsight.

This article was AI-translated and verified by a human editor