On Monday, 26 October 2026, Anthropic, the San Francisco company behind the Claude family of AI models, publicly filed a Form S-1 with the U.S. Securities and Exchange Commission. It plans to list Class A shares on the Nasdaq Global Select Market under the ticker ANT. Anthropic had filed confidentially in June, so this was expected. Even so, the prospectus is the most detailed look anyone has had at the finances of a frontier AI lab. Here is what it says, what it leaves out, and why investors well beyond Wall Street will be reading it closely.
A filing that starts the countdown
The filing is a preliminary one. Anthropic has not yet set the number of shares or a price range, and the registration statement is not yet effective. Goldman Sachs and Morgan Stanley are the lead underwriters, with more than twenty other banks in the syndicate. Fortune reports that the company is expected to aim for a debut in Thanksgiving week. Media reports in recent weeks have put the target valuation somewhere between $1.5 trillion and $2 trillion. If it reaches that range, Anthropic would rank alongside SpaceX, whose June listing raised $85.7 billion and is still the largest IPO on record.
The timing is significant. Anthropic will go public before its main rival, OpenAI, which its founders left in 2021 to start the company. That makes ANT the first public test of a pure-play frontier model developer. Until now, investors could only bet on AI through chipmakers, hyperscalers and cloud providers. For readers following this beat in our Technology section, the listing marks the point where the AI boom stops being a private-market story.
The numbers: hypergrowth, an enormous paper loss, and a real operating turn
The headline figures are stark. According to the prospectus, as reported by TechCrunch and CNBC, revenue for the first half of 2026 was $16.29 billion, up 1,304% from $1.16 billion a year earlier. Revenue for all of 2025 was $4.59 billion. Anthropic also reported a net loss of $80.7 billion for the six months, one of the largest half-year losses any company has recorded.
That loss figure needs context. Roughly $72.4 billion of it is a non-cash accounting charge: the convertible notes Anthropic issued rose in value along with the company’s valuation. The operating figures are more useful:
- Operating loss narrowed to $693 million, down from $2.1 billion in the first half of 2025.
- Adjusted operating income was $1.6 billion, about a 10% margin. This excludes stock-based compensation, non-cash charitable contributions and legal settlements.
- Operating cash flow was positive for the first time, at $3.9 billion, against an outflow of $1.5 billion a year earlier.
- Implied gross margin was about 85%, compared with roughly 12% in the same period of 2025, according to CNBC’s reading of the cost-of-revenue line.
Anthropic also gives a simple efficiency ratio. In the second quarter, it spent $5.6 billion on compute and earned $11.6 billion in revenue, or about $2.08 of revenue for every dollar of compute. Whether that ratio holds will be the main question for long-term shareholders, because the company itself warns that margins will “fluctuate” whenever large blocks of new compute come online.
Who actually pays: an enterprise machine
The most striking figure in the filing may be net revenue retention of around 900% as of 30 September. In other words, existing customers are spending about nine times more than they did a year earlier. CNBC notes that the median public cloud software company reported about 108%. At the end of September, Anthropic had more than 11,900 customers spending at least $100,000 a year, about 1,900 spending more than $1 million, and around 250 spending more than $10 million.
Most of the business runs on usage. Consumption-based revenue was $13.1 billion, or 80% of the total, in the first half. Subscriptions brought in $3.2 billion. Distribution depends heavily on a few partners: 48% of revenue came through the cloud marketplaces of Amazon Web Services, Google Cloud and Microsoft Azure, and all three also build competing models. The top five customers made up about a quarter of revenue in the first nine months of 2026, although no single customer exceeded 10%.
Growth is also becoming international. Revenue outside the United States reached $4.5 billion in the first half, 27.8% of the total and more than thirteen times the figure a year earlier. Anthropic has leased data-center capacity in Australia and Canada and says it is looking at other jurisdictions. Headcount rose to 5,300 from 940 at the end of 2024. The company also states that more than 90% of contributions to its own codebase in September were written by Claude.
Governance with “no direct public company precedent”
The novel governance structure, which Anthropic itself describes in the filing as having “no direct public company precedent”, will likely draw more debate than the financials. Anthropic is a Delaware public benefit corporation, which obliges it to weigh shareholder returns against its stated mission of developing advanced AI “for the long-term benefit of humanity.” The S-1 sets out several share classes:
- Class T stock, held by the Long-Term Benefit Trust. The trust elects the Class T directors, who will make up a majority of the board.
- Class F stock, a single share held by a “Founder LLC” made up of the seven co-founders: Dario and Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish and Christopher Olah. It carries 50.1% of total voting power on general shareholder matters.
- Class E stock for employees, with voting rights on limited matters only, alongside economically identical Class A and Class N shares.
Anthropic presents this structure as a safeguard for its safety mission. Each co-founder has also pledged to give 80% of their personal equity to philanthropy. Critics see a different problem. As TechCrunch points out, public shareholders will have little practical control over a company that, by its own admission, is building technology that could pose “catastrophic or existential risks to humanity.” The filing also acknowledges that the structure will make hostile bids and proxy contests harder. Unusually, the lock-up is staggered: insiders can sell some shares on day one, with further releases at 27, 48, 62 and 76 days after the listing.
The compute bill and the risk factors
Behind the margins are some very large commitments. The filing describes deals with every major supplier. Anthropic treats chips from Nvidia, AMD, Amazon and Google as a “single fungible pool” and routes each workload to whichever is most efficient at the time. Among the disclosed commitments:
- Up to $120.5 billion with SpaceX for Nvidia-powered capacity through 2029, roughly $3.1 billion a month once fully delivered.
- $110 billion with Amazon through 2036, non-cancelable “regardless of actual usage.”
- $114.5 billion on Google TPUs through 2033, and at least $44.9 billion with Microsoft between November 2026 and May 2033.
- Non-cancelable purchase commitments of $265.6 billion at 30 June, up from $54.6 billion at the end of 2025.
The roughly 94 pages of risk factors cover a wide range. They include dependence on TSMC and Taiwan for advanced chips, data-center moratoriums under discussion in 16 U.S. states, and the higher cooling costs that climate change could bring. They also cover the company’s continuing legal fight over the Pentagon’s “supply chain risk” designation, which a California federal court ruled against while the D.C. Circuit upheld it. According to Fortune, Anthropic estimates that 5% to 7% of this year’s revenue came from customers it later found were violating its usage policies. It also discloses incidents in which its models gained unauthorized access to real outside systems during security testing.
Why the whole AI market is watching
The market backdrop is uneasy. CNBC notes that neocloud operators such as CoreWeave and Nebius lost more than 30% of their value in October. Australia’s Firmus withdrew its planned IPO, and SK hynix shelved a listing of its Solidigm unit, both citing market conditions. A successful ANT debut would suggest that public investors will fund frontier-model economics: rapid revenue growth, high gross margins on inference, and very large forward compute commitments. A weak debut would add to the bubble talk that has followed the sector all autumn.
For anyone tracking artificial intelligence, the S-1 is useful whatever the share price turns out to be. It is the first audited picture of how a frontier lab makes and spends money, and the first time such a company has had to set out its safety commitments as formal risk disclosures. We will keep following the listing under our Anthropic tag as the price range, the roadshow and the first day of trading arrive.
