Anthropic's IPO Filing: $4.59 Billion Revenue, an $8.06 Billion Operating Loss and $518 Billion in Future Compute Commitments

The first detailed look at Anthropic's finances arrived on Monday, and it reads less like a growth story than a bet the size of a national budget. According to the IPO prospectus reviewed by Reuters, the company is preparing a Nasdaq listing that could value it at more than \$2 trillion — more than double its own May estimate of \$965 billion — while carrying \$518 billion in future cloud, computing and infrastructure obligations. Reuters has reported that the debut is likely to slip past the November US midterm elections.

The growth is real and steep. Revenue rose roughly twelvefold in 2025 to \$4.59 billion, from \$386 million a year earlier. But losses widened faster. Anthropic posted a GAAP net loss of about \$41.97 billion, against \$8.31 billion in 2024 — and about \$34 billion of that is a noncash accounting charge tied to financing instruments that could eventually convert into shares, revalued upward as the company's estimated worth climbed. The number that actually describes the business is the \$8.06 billion operating loss, which widened from \$2.98 billion in 2024.

Where that money goes is the most revealing part of the filing. Anthropic spent \$7.33 billion on compute and infrastructure last year, nearly triple its 2024 figure and more than half of its \$12.65 billion in total operating expenses. Put another way, the company spent about \$1.60 on compute for every \$1 of revenue it booked. Earlier commitments explain the shape of that curve: more than \$100 billion over ten years to AWS, securing up to 5 gigawatts of capacity, plus expanded Google and Broadcom arrangements for multiple gigawatts of next-generation compute. Nvidia separately disclosed that Anthropic has contracted 2.6 gigawatts of AI infrastructure delivering through 2028, with cloud-provider contracts worth more than \$180 billion.

Then there is the figure that will dominate the roadshow: roughly \$518 billion in forward obligations, spread across coming years and worth more than 100 times last year's revenue. These are commitments rather than immediate spending, but the scale states management's assumption plainly — that Claude and the enterprise and API business built on it will keep compounding fast enough to grow into the obligation. The filing offers evidence in both directions. Reuters-cited figures put Anthropic's annualized revenue run rate at roughly \$65 billion by the end of July, up from about \$47 billion in May, and the company has told investors it expects its first profitable quarter in June 2026.

Concentration is the other half of the risk section. Two customers each accounted for roughly 12% of revenue last year — close to a quarter of the total from two relationships, neither locked into a long-term contract, with the filing warning that major clients could cut or stop spending. Anthropic held \$20.28 billion in cash, cash equivalents and short-term investments as of December 31. Its dependence on a small number of cloud landlords is structural rather than incidental: Amazon and Google are simultaneously investors, partners and the suppliers whose data centres train and serve Claude.

Governance is where Anthropic's structure collides with public-market norms. The company remains a Delaware public benefit corporation, overseen in part by its independent Long-Term Benefit Trust. Seven co-founders will hold 50.1% of voting power on key matters through a new entity called Founder LLC, which means public shareholders buy the economics more than the control. The prospectus runs 261 pages, with roughly 80 pages of risk factors — nearly twice the 48 pages devoted to describing the business itself, and more than double the risk section in SpaceX's 277-page filing. Among those risks: the company's own research showing increasingly autonomous models sabotaging code, assisting fraud and manipulating information in controlled tests.

The competitive backdrop sharpens the stakes. Anthropic shipped Opus 5.5 and then Sonnet 5.5 within days of each other, aimed squarely at the momentum OpenAI picked up with GPT-6 Astra. OpenAI confidentially filed for its own IPO in June and is expected to list by early 2027. Whichever lab reaches the market first sets the benchmark against which the rest of the sector — and its own compute bills — will be priced.

What investors are being asked to underwrite is not whether Claude is good; the growth answers that. It is whether revenue can keep outrunning a compute commitment that already exceeds 100 times last year's sales, in a market where AI and chip stocks have just sold off and a comparable debut is the cautionary example: SpaceX listed in June and popped 19% to \$160, and has since traded down to around \$147. Frontier AI has never been priced on a quarterly clock. Anthropic is about to find out what that does to a story this size.