Anthropic's registration statement went public on Monday morning, and within the hour the number moving across timelines was $80.7 billion — a first-half net loss roughly five times the revenue the company booked in all of 2025.

As a description of the business, that number is close to meaningless.

About $72.4 billion of it comes from the accounting value of convertible notes and other financing instruments shifting as Anthropic's own valuation rose. The cash-flow statement attributes a further $8.72 billion to non-cash charges on other financial instruments. None of that is money spent running Claude. Some of it exists precisely because investors decided the company was worth more.

Strip the accounting out and a different company appears. Revenue of $16.3 billion in the six months to June, against operating expenses of roughly $17 billion, produced an operating loss of $693 million — down from $2.14 billion a year earlier. Computing and infrastructure alone reached $8.95 billion, up from $2.33 billion; personnel cost $3.15 billion, sales, marketing and partnerships another $3.17 billion. Excluding stock-based compensation and non-cash charitable contributions, the company reports adjusted operating income of $1.62 billion, a measure that is not profit under standard rules. Operating cash flow was positive at $3.93 billion, which Anthropic partly credits to the timing of customer receipts and vendor payments.

Growth is genuinely extraordinary: $16.3 billion in the first half against $1.16 billion a year earlier, or about 1,300%. Enterprise concentration is real. More than 11,900 customers spent at least $100,000 in the trailing twelve months, more than 1,900 spent at least $1 million, and roughly 250 spent at least $10 million.

Then comes the paragraph that should outlive the headline. Usage-based sales supplied 80% of first-half revenue. Much of that revenue carries no long-term customer commitment. The infrastructure contracts on the other side of the ledger can require payment for years, including for capacity Anthropic does not use. One disclosed equipment lease alone entails approximately $125.2 billion in rent over five years, excluding interest — a figure that dwarfs the $16.3 billion booked in the half year. Anthropic warns that failing to meet certain obligations could trigger defaults across other agreements and jeopardise access to essential resources. It also concedes that it has previously sought more computing capacity than cloud providers were willing to supply, and that contractual arrangements do not guarantee delivery on its preferred schedule.

That is the shape of the bet: multi-year, largely fixed obligations on one side, cancellable demand on the other. It is the same structure that has made investors nervous about every layer of the AI build-out, except that here it sits in a document signed by the company itself.

The filing's risk section is unusually frank about a second seam. Anthropic wants Claude to take on open-ended assignments lasting weeks and to run millions of instances at once — "a country of geniuses in a data center," in its own phrasing. It then admits the law has not caught up: it is unclear when an agent's actions legally bind the organisation deploying it, or whether contractual liability limits will hold. Some agreements with outside model evaluators expose Anthropic to uncapped liability for specified third-party claims, and the company cautions that a severe enough AI incident could exceed the insurance industry's capacity to cover the damage.

The regulatory backdrop is already concrete. Anthropic discloses a Federal Trade Commission investigative demand issued in October, seeking information on consumer protection, privacy, data protection, and product safety and security; the filing establishes no wrongdoing. It also flags the Trump administration's decision to designate the company a supply-chain risk, and warns it cannot be certain that resolving related litigation will not affect its ability to sell to government agencies.

The listing would be on Nasdaq under the ticker ANT, and The Wall Street Journal reported the IPO could come later in November. The prospectus specifies neither share count nor price range; reporting in August suggested a raise above $100 billion at a valuation near $2 trillion. As of 30 June, the company had raised $141.1 billion in aggregate capital.

The comparison it invites is unavoidable. OpenAI filed confidentially in the summer, then announced it would not go public this year, citing the risks AI poses to humanity. Anthropic, founded by seven former OpenAI employees who left over exactly that concern, is now asking public markets to underwrite it — and telling them, in the same document, that the obligations it has signed may outlast the customers who are meant to pay for them.

Sources