Anthropic has released the paperwork required before it lists on the US stock market — 335 pages of it. The company is widely expected to debut at a market valuation of $2 trillion or more, which would make it the largest stock listing in history.
Most of the coverage has led on one sentence: that “sufficiently advanced AI systems could, if not developed, deployed, and operated with appropriate safeguards, oversight, and security measures, pose catastrophic or existential risks to humanity”.
That is striking, but it is not new. Reuters obtained a version of this prospectus a month ago, and we covered it then: 80 pages on what could go wrong against 48 describing the business, including models that “resist shutdown”. A company that has spent years arguing AI is dangerous can hardly leave that out of a legal risk disclosure.
The genuinely new material is financial. And the number worth sitting with is not the loss.
Five customers, a quarter of the revenue
A large part of Anthropic’s business comes from roughly 250 major enterprise customers, each of which has paid at least $10m in the past year. Within that group, just five customers account for about a quarter of its current business.
Michael Monaghan, portfolio manager of Founder ETF, made the point that matters: “many top Anthropic customers have no long-term contract”. A quarter of the revenue of a two-trillion-dollar company can therefore walk out at renewal.
There is a second edge to this. Anthropic has enterprise relationships with Google and Amazon, both of which are also investors in the company and, as Monaghan noted, competitors. That is three roles at once — customer, shareholder, rival — and the filing says so.
Against that, the growth is real: the enterprise customer base went from roughly 1,500 last year to almost 12,000 this year, helped by a strategy of giving Claude away and converting users later. The company claims “the fastest enterprise adoption of any technology in history”.
The losses, and the cash that covers them
Anthropic lost $80.7bn in the first half of this year — double what it lost across the whole of 2025. Revenue for the same six months was $16.2bn.
That figure is worth lining up against what the company told investors privately. In October it briefed prospective backers that revenue had reached $16.8bn in the quarter that closed on 30 September, following $4.7bn in the first quarter and $11.6bn in the second. The filing’s half-year number is consistent with those quarters; the $16.8bn quarter sits outside the period the filing covers. Readers comparing headlines are not looking at a contradiction, they are looking at two different windows.
The company also has a great deal of money on hand: $131.6bn, almost all of it classed as working capital it can spend day to day. The filing puts its cash above Meta and Google combined.
It is spending at a matching scale. Leasing data centres alone has cost $321.4bn so far this year.
What the analysts actually said
Troy Hooper, co-head of equity capital markets at Mergermarket, called the revenue growth “exceptional” — and then set out the problem. Anthropic’s debt picture is unclear because of large accounting charges tied to certain forms of investment. As a result, he said, “investors do not yet have a clear view of Anthropic’s underlying profitability or whether its rate of cash generation is sufficient to support the business as it scales”.
His other observation is the one to keep. Investors are “less forgiving” of tech companies promising huge future returns than they were even months ago. Monaghan put the same thought more bluntly: anyone buying in “will be paying for future dominance, not present-day economics”.
The part that is a pitch
The filing also restates the upside, and it is written in the register of a mission statement rather than a balance sheet: bespoke treatments for illnesses, improved disease detection, tailored education, expert financial and legal advice, more efficient power grids and batteries, and help for the “developing world” in building technological capability. Decades of scientific progress, the company suggests, could be condensed into years.
None of that is measurable, and none of it is a commitment. It sits in the same document as the warning about catastrophic risk, which is the honest shape of the bet being offered: a company asking to be valued on what it might prevent and what it might cure, while disclosing that it cannot presently be valued on what it earns.
One structural promise from earlier this month is also now part of the picture: Anthropic said it would put 5% of its shares into a “public benefit fund”, with another 2% possible at the board’s discretion.
The filing notes, in passing, that less than a year ago a hacking group the company believed to be sponsored by China “manipulated” Claude to intrude on technology companies, financial institutions, chemical manufacturers and government agencies. For a business whose product is trust in a model, that is a risk factor in a more immediate sense than the existential one.
