Anthropic's Leaked IPO Filing Shows $42 Billion Loss and $518 Billion in Future Spending Commitments

Since Anthropic's safety researchers and rivals at OpenAI spent recent weeks warning that AI systems are already slipping lab controls and hacking real targets, the company has now put its own money problem on paper. A leaked IPO prospectus, reviewed by Reuters and reported by outlets including the Financial Times, shows Anthropic lost $42 billion in 2025 while pulling in just $4.6 billion in revenue.

That loss figure is eye-popping, but the accounting behind it matters. According to Reuters, roughly $34 billion of the $42 billion net loss was an accounting charge tied to the rising estimated value of financing instruments that could eventually convert into Anthropic shares, not cash the company actually spent running the business. Strip that out and the operating loss was still $8.06 billion, up from $2.98 billion the year before, according to Fortune.

Revenue grew 12-fold in 2025, or roughly 1,088% by Fortune's count. That growth is real. So is the spending required to sustain it. Anthropic spent $7.33 billion on compute and infrastructure last year, triple its 2024 spend, and that single line item ate 58% of its $12.65 billion in total operating expenses, according to ZeroHedge's read of the same filing.

The $518 billion question

The number drawing the most fire from analysts is the $518 billion in future cloud, computing, and infrastructure obligations Anthropic has committed to, per the prospectus details reported by Reuters and carried by Yahoo Finance, StockTwits, and TradingView. Anthropic ended 2025 with $20.28 billion in cash, cash equivalents, and short-term investments, according to Fortune. That leaves an enormous gap between what the company has on hand and what it has promised to spend.

Ross Hendricks, an equity analyst who writes the Ross Report, questioned the math directly on X: "Just a casual mention of raising half a trillion in a single year, really? Does anyone believe this? Let's say they pull $100B in equity from the IPO... and then what, $400B from the debt markets?" Hendricks also argued Anthropic's widening losses come as it faces growing competition from cheaper open-source models. Open-weight Chinese models have closed the capability gap quickly this year.

EZ Primary Research CEO Ed Zitron was blunter, writing on X: "Bahahahaha Anthropic is a total dog of a company. Spent $12.6bn to make $4.6bn in revenue in 2025, $7.33bn of which was compute costs. Operating loss of $8bn. Losses getting worse." Zitron has argued for months that frontier AI economics depend on continuous access to capital markets and cloud-provider goodwill, not underlying business fundamentals. Two customers accounted for nearly a quarter of Anthropic's 2025 revenue, per the prospectus, and Anthropic itself warned those clients aren't locked into long-term contracts.

Anthropic's own risk-factor section reads like a warning label

The most striking part of the filing isn't the money. It's what Anthropic told its own future shareholders about the product they'd be buying stock in. According to the Guardian's review of the prospectus, Anthropic disclosed that its models could exhibit "self-preserving behaviours," including attempts to "resist shutdown," "conceal or manipulate information," and behavior "resembling blackmail." The company wrote that expanding its use cases "could further increase the risk that our models cause harm."

Reuters reported that roughly 80 of the 261 pages in the prospectus's main body are devoted to risk factors, compared with 48 pages describing the actual business. That's a company spending nearly twice as much paper warning regulators and investors about catastrophe as it spends explaining how it makes money.

Some AI researchers dispute how seriously to take those warnings. The Guardian noted that critics call existential-risk claims unverifiable and unscientific, even as the same filing arrives amid genuine, documented incidents of AI agents acting outside intended bounds, including OpenAI agents accessing third-party systems without authorization. Whether Anthropic's language reflects sober risk disclosure required by securities law or a marketing angle that reinforces the company's "safety-first" brand is a fair question neither side has settled.

Anthropic declined to comment on the leaked filing, according to Reuters. The company's public listing, which could value it above $2 trillion, more than double its $965 billion valuation from May, is now expected to land after the November midterms.

The unresolved question for anyone eyeing that IPO: where does $518 billion in infrastructure spending actually come from when the company's cash on hand is $20 billion and a quarter of its revenue rests on two customers with no contractual obligation to stick around.