Anthropic IPO filing shows $16.3 billion in first-half sales

Anthropic, the maker of the Claude AI models, reported $16.3 billion in revenue for the first six months of 2026, up 1,300% from a year earlier, in a public registration statement filed Monday with the Securities and Exchange Commission for an initial public offering.

The company said it has applied to list its Class A shares on the Nasdaq Global Select Market under the symbol “ANT.” The filing leaves the price range and the number of shares blank, so the size of the offering and the company’s valuation are not yet set. Goldman Sachs and Morgan Stanley head the list of underwriters on the cover.

The same filing shows a net loss of $80.7 billion for the half, after a $42.0 billion loss for all of 2025. Most of the first-half loss, $72.4 billion, came from changes in the fair value of convertible notes, derivative liabilities and simple agreements for future equity, which the filing says were “primarily driven by” an increase in the company’s enterprise valuation. The company warned that it does “not expect our revenue will continue to grow at comparable rates.”

What’s driving it

The filing points to rapid adoption by large customers. As of Sept. 30, Anthropic said it had about 250 customers contributing $10 million or more in revenue over the trailing 12 months, more than 1,900 at $1 million or more and more than 11,900 at $100,000 or more, the level it uses to define enterprise customers. It reported net revenue retention of about 900%.

Second-quarter revenue was $11.6 billion, up from $787 million, a 1,368% increase, while compute and infrastructure expenses rose 300% over the same period, the filing says. Anthropic reported operating income of $1.0 billion for the quarter, an 8% margin, and $2.2 billion of Adjusted Operating Income, a non-GAAP measure the company defines. For the half, its operating loss narrowed to $693 million from $2.1 billion a year earlier.

Why it matters

The filing gives investors a detailed look at a major AI model developer’s books, though the first-half figures are unaudited, and the headline numbers pull in opposite directions. The revenue and operating figures suggest the core business moved toward profitability in the second quarter, at least by the company’s own measures. The $80.7 billion net loss, by the filing’s account, mostly reflects accounting for financing instruments whose value rose as Anthropic’s valuation rose, not cash spent running the business. Both matter: the charge is non-cash, but the notes behind it are real obligations, and the filing names repaying them as a risk. The bigger long-term question is compute. Anthropic disclosed commitments that dwarf its current revenue, including up to $120.5 billion with SpaceX’s AI unit and at least about $114.5 billion with Google. The Google commitment is non-cancelable except for breach, the filing says, while the SpaceXAI deals can be ended on 90 days’ notice after initial periods that run into early 2027. Either way, the bet depends on demand keeping pace for years. The company itself cautions that its growth rate will not last.

Who has a stake

Compute suppliers: The filing says Anthropic agreed to spend up to $120.5 billion over terms extending into 2029 with SpaceXAI, a SpaceX subsidiary formerly known as X.AI, for Nvidia-powered capacity, about $3.1 billion a month once fully delivered, which it expects in December 2026. It also committed to spend at least about $114.5 billion with Google between April 2026 and July 2033. It further committed to buy more TPU compute from providers including Google for delivery from 2028 to 2031, which it estimates at about $125 billion to $150 billion.

Founders and the trust: Anthropic is a public benefit corporation. A Long-Term Benefit Trust generally can elect and remove a majority of the board, and a special share class will be held by an entity owned by the seven co-founders, including CEO Dario Amodei, President Daniela Amodei and Jared Kaplan, the filing says. Krishna Rao has been chief financial officer since April 2024.

Future shareholders: Among its risk factors, the company says models, “including our own,” have exhibited behaviors that include attempts to “generate outputs that could be interpreted as coercive, deceptive, or manipulative, including behavior resembling blackmail.”

What’s next

The S-1 must still be updated with a price range and share count before the offering can be marketed. Timing and valuation will depend on market conditions. AI Early Edition reported details of a draft version on Oct. 22, citing The Wall Street Journal.