Anthropic's IPO prospectus: a $2 trillion ask on $4.6 billion of revenue

Anthropic is asking public investors to finance one of the largest bets in the history of technology, and it wants a valuation of about $2 trillion to do it. The company's confidential IPO prospectus, seen by Reuters, argues that AI will transform the global economy "more profoundly than industrialization, electricity and the internet," and reports a $42 billion net loss for 2025 alongside $518 billion of future cloud and computing commitments. Revenue, meanwhile, reached nearly $4.6 billion. The filing has not been made public, so every number here carries a qualifier: this is what the prospectus says, as Reuters summarized it from a copy it reviewed.

Inside the numbers: a $42 billion loss that is mostly paper

The headline loss is real on paper and misleading in practice. Roughly $34 billion of the $42 billion shortfall was an accounting charge, reflecting a rise in the estimated value of financing that could eventually convert into Anthropic shares rather than cash the company burned operating the business. Strip that out and the operating loss still more than doubled, to above $8 billion, on revenue that grew twelvefold to nearly $4.6 billion.

The cost structure is where the story lives. Anthropic spent $7.33 billion on compute and infrastructure in 2025, roughly three times its 2024 figure and more than half of its $12.65 billion in total operating expenses. It closed the year with $20.28 billion in cash, cash equivalents and short-term investments. None of these figures had been published before; Reuters reported them for the first time from the confidential filing.

Why the valuation math only works forward

Reuters, from the prospectus:

Metric	Figure
2025 revenue	$4.6 billion
2025 net loss	$42 billion
2025 operating loss	over $8 billion
2025 compute and infrastructure spend	$7.33 billion
Forward cloud and computing commitments	$518 billion
Target IPO valuation	about $2 trillion

Trailing multiples are the first thing to break. About $2 trillion against $4.6 billion of 2025 revenue is a multiple in the hundreds, near 430 times. The company's own valuation history shows how fast the base has moved: $183 billion post-money after a $13 billion round closed in September 2025, then about $965 billion in May 2026. The company had already passed OpenAI on annualized revenue back in April.

The only way the $2 trillion works is on forecasts. Reuters reported in August, citing two people familiar with the company's finances, that Anthropic projects 2028 revenue of roughly $190 billion to $200 billion, which would put the same valuation nearer ten times forward sales. Set against that trajectory, the $518 billion of commitments is a claim on a revenue line that does not exist yet, more than a hundred times 2025 sales. Buyers are being asked to price a slope, not a business.

The partners who sell, supply and own the product

The most concrete finding in the filing concerns distribution. Anthropic routed 47% of its 2025 sales, about $2.16 billion, through the cloud marketplaces of Amazon and Alphabet's Google, and paid roughly $351 million back to those platforms in distribution fees, Reuters found. That works out to about 16 cents for every dollar of marketplace sales, booked inside Anthropic's "sales, marketing, and partnerships" expense line.

The ratio reframes the partnership story. The same two companies are among Anthropic's largest investors, its principal suppliers of computing power, the collectors of its customer bills and, through their own frontier models, its competitors. Their share of distribution has climbed quickly: 11% of revenue in 2023, 32% in 2024, nearly half in 2025. Two unnamed customers each accounted for 12% of revenue, and the platforms collected on 60% of the $909 million of customer bills outstanding at year-end.

Anthropic frames this as reach, not exposure. Selling through Amazon, Google and Microsoft, it argues, accelerates "market penetration at a scale we believe would be difficult for any single organization to directly replicate." Its risk factors concede the other reading: that reliance on a limited number of partners and suppliers "creates complex dynamics that could give rise to conflicts of interest and adversely affect our access to compute." That is a circular fact pattern rather than a scandal, and it is checkable later against the platforms' own disclosures.

A civilizational pitch with a mostly non-cancelable bill

Reuters' September 29 follow-up answers the question that matters most for those forward commitments: most of the $518 billion is money Anthropic must pay whether or not it uses the capacity. About 80% of the total is non-cancelable or payable regardless of usage, spread across six partners over roughly a decade. The named obligations include $111.1 billion with Google, $110 billion with Amazon and $31.4 billion with Microsoft, plus about $161.2 billion of Broadcom-related equipment leases that are largely non-cancelable. "If our actual spend falls short, we must pay Google the difference," the company said, and similar terms apply to its Amazon agreement. The xAI arrangement, up to $84.5 billion, is the notable exception, largely cancelable with 90 days' notice.

The risk section is proportionate to the ambition. Roughly 80 of the prospectus' 261 main-body pages are risk factors, against 48 pages describing the business, and they include warnings that Anthropic's own models could show "self-preserving behaviors," such as resisting shutdown or behavior "resembling blackmail," as The Guardian reported. The company has also spent this year managing a fight at home, after the Pentagon temporarily blacklisted it, a move a US judge blocked in August, with the appeal still live.

What would settle the open questions

Three things would move this from a wire-service summary of an inaccessible document to a verified picture. The first is the public S-1 itself: the full risk factors, a revenue breakdown by segment, and the customer-concentration percentages behind that "two customers, 12% each" line. The second is the mechanics of the offer, meaning pricing, allocation, and how much of the roughly $2 trillion is set at the offer rather than left to the market afterward. The third is whether the 2028 revenue slope holds, because the $518 billion bill is fixed while the revenue is not.

None of the standard listing details has surfaced. There is no share price, no share count, no listing date and no named group of underwriters in any of the reporting, and the filing remains confidential. Reuters has reported, citing sources, that the debut is likely to slip past November's US midterm elections.

Governance is the quietest open question. Per Reuters, seven founders would hold a single Class F share carrying 50.1% of voting power, giving public investors the economics without much of the control. Public shareholders would be buying into a company whose stated mission is to develop AI safely while selling it as fast as it can, with a risk factor conceding that structure could produce decisions conflicting with shareholders' financial interests. The prospectus, in other words, is an argument as much as a set of accounts.