A leaked IPO prospectus shows Anthropic lost $42 billion in 2025 while revenue grew twelvefold to nearly $4.6 billion, and the company plans $518 billion in cloud, computing and infrastructure spending in the years ahead. The filing, first reported by Reuters, puts hard numbers on the frontier lab capital debate for the first time.
What the numbers say
Revenue grew roughly 12-fold in 2025 to nearly $4.6 billion. Operating losses ran above $8 billion excluding writedowns, and the headline $42 billion net loss includes a roughly $34 billion accounting charge for financing that could convert to Anthropic shares, not cash the company burned running servers.
| Metric | Figure |
|---|---|
| 2025 net loss | About $42 billion |
| Accounting charge included | Roughly $34 billion |
| Operating loss (ex-writedowns) | Above $8 billion |
| 2025 revenue | Nearly $4.6 billion, up 12-fold |
| Compute and infrastructure spend | $7.33 billion, up about 3x |
| Total operating expenses | $12.65 billion |
| Planned infrastructure spend | $518 billion |
| Cash and short-term investments | $20.28 billion at Dec. 31 |
Compute and infrastructure spending hit $7.33 billion last year, a threefold surge from 2024 and more than half of the $12.65 billion in total operating expenses. Anthropic held $20.28 billion in cash and short-term investments at the end of December, which buys a runway that most private companies could only dream about, though not one that covers $518 billion in planned obligations.For scale, the planned $518 billion figure is larger than the annual infrastructure budgets of most developed countries’ entire technology sectors. It is spread across what the prospectus outlines as cloud, computing and infrastructure obligations in coming years, the largest single number any private AI lab has put in a public document.
The valuation question
The filing sets up a public market debut that could value Anthropic at more than $2 trillion, more than double the roughly $965 billion the company reportedly fetched in May. Reuters reported the sale will likely slip to after the November US midterm elections. It would follow SpaceX’s recent IPO, which valued Elon Musk’s rocket company at $1.77 trillion.Not everyone is buying the math. New Constructs called it, per CNBC, “the most ridiculous IPO of the year” and wrote that “we don’t think Anthropic has a viable business” given the operating losses and competition from open-source models. The bull case rests on Claude’s positioning against OpenAI, the quality of the enterprise book, and the expectation that the winner of a winner-take-most market justifies nearly any entry price today.Concentration risk sits inside the document itself. Nearly a quarter of last year’s revenue came from two customers, and the risk factors warn that many of the largest clients are not locked into long-term contracts and could cut or stop spending. That kind of sentence rarely makes it into glossy investor decks, but it did make it into the prospectus.
“Since the arrival of open-source models, it’s been clear that the closed models would struggle to generate profits,” New Constructs wrote.
The short-model competition critique has teeth. Chinese open models and open-weight Western releases have been closing the gap on benchmarks, and enterprise buyers increasingly treat model choice as swappable. The counterargument is that enterprise Claude deployments are sticky in workflow, integrations and compliance, not in model downloads.
Context against OpenAI
The filing lands the same month OpenAI told investors its annualized revenue was approaching $50 billion as of end-September, a figure Fortune noted comes with an accounting gap: Anthropic counts sales made through cloud partners, and OpenAI historically has not. The two labs’ revenue figures are not directly comparable, which is half the reason chip stocks crashed on October 8 when the discrepancy became public, erasing roughly $170 billion of Nvidia’s market value in a single session before clawing back.Financing has become the real battleground. Broadcom has reportedly held talks to raise around $30 billion in debt to help OpenAI buy chips, after a similar $35 billion Anthropic deal and another $60 billion package in the works. Nvidia has mobilized over $500 billion with six finance firms to fund customer buildouts. AMD has offered Meta and OpenAI warrants for up to 320 million shares at a penny each, effectively paying customers to buy its hardware.Amazon and Google remain Anthropic’s strategic partners, both investing in the company while supplying the cloud infrastructure that trains and hosts Claude. That arrangement gives both cloud firms a stake in the IPO going well, and gives Anthropic dependencies on two companies who also fund its chief rival.
What it changes
The prospectus is the first public-scale look at real frontier lab economics, and it confirms the thing analysts have been arguing about all year: the business is growing at an extraordinary rate and losing money at one too. Public investors are being asked to fund a race that has so far been underwritten by venture capital, sovereign wealth funds and Big Tech balance sheets.Whether public money stabilizes spending or accelerates it is one of the open questions for 2027. Listed companies face quarterly scrutiny that private ones do not, but they also get access to a deeper pool of capital.Dario Amodei, Anthropic’s CEO, has publicly called for the AI community to slow the pace of releasing new capabilities, a position that sits awkwardly beside a $518 billion growth budget, and market observers have flagged the tension between the safety voice and the growth engine. The company declined to comment on the filing.If the debut lands near $2 trillion, it sets the benchmark valuation every subsequent AI listing will be measured against, from xAI to Mistral to whatever OpenAI eventually does. That is a heavy weight for one filing to carry, and precisely why the roadshow will be watched as closely as the numbers themselves.