Anthropic S-1: $16.3B Half-Year Revenue, Ticker ANT

Anthropic has filed its registration statement on Form S-1 with the U.S. Securities and Exchange Commission, the public step toward an IPO that began with a confidential filing in June. The company’s Class A shares would trade on the Nasdaq Global Select Market under the ticker ANT. Anthropic’s announcement says the number of shares and the price range have not been set, and the statement is not yet effective, so nothing can be sold. The cover of the prospectus names Goldman Sachs, Morgan Stanley and J.P. Morgan. The full filing on EDGAR runs to 224 MB with its exhibits.

The numbers go well past the audited 2025 figures that Reuters reported from the draft in September. Revenue was $41 million in 2023, $386 million in 2024 and $4.59 billion in 2025. In the first six months of 2026 it reached $16.29 billion, against $1.16 billion a year earlier, and $11.6 billion of that came in the second quarter alone, up 1,368% year over year while compute and infrastructure costs grew 300%. Compute fell from 251% of revenue in the first quarter of 2025 to 48% in the second quarter of 2026, and that quarter produced Anthropic’s first operating income, $1.0 billion. About 80% of first-half revenue, $13.1 billion, came from pay-per-use API and model consumption, and 20%, $3.2 billion, from subscriptions.

The bottom line looks very different. Anthropic reports a net loss of $42.0 billion for 2025 and $80.7 billion for the first half of 2026, but $72.4 billion of the half-year loss is a non-cash charge for the rising fair value of its convertible notes, derivative liabilities and SAFEs, which grow more expensive on paper as the company’s valuation climbs. It held $131.7 billion in cash and short-term investments on June 30. More than 11,900 customers paid it $100,000 or more over the twelve months to September 30, up from about 1,500 at the end of 2025; about 1,900 paid over $1 million and about 250 over $10 million, while roughly 40 million smaller customers make up the rest. The prospectus also credits Claude Opus 5.5 with cutting customers’ cost per task by about 40% compared with Opus 5.

Control stays with the founders and the company’s trust. Just before the offering, Anthropic will issue a single share of Class F stock to an LLC owned by its seven co-founders, including chief executive Dario Amodei and president Daniela Amodei. That one share carries 50.1% of the votes on ordinary shareholder matters and no economic rights. The Long-Term Benefit Trust, through Class T stock, elects the directors who make up a majority of the board. Employees other than the founders get Class E stock that votes only on limited matters, and strategic partners hold non-voting Class N shares. The S-1 also formalises the Anthropic Public Benefit Fund: each time the share price rises 50% above the previous milestone, starting from the IPO price, the fund receives Class N shares worth up to another 2% of the company. Each co-founder has pledged 80% of their personal equity to philanthropy.

Two passages in the risk factors are new. One says Anthropic “recently chose to delay pre-training a new frontier base model for more than a month” to strengthen its defences against theft of model weights and its intellectual property. The other revisits the evaluation incidents BitsMinds covered in July: during cybersecurity evaluations built by a third-party partner, a misconfigured environment left the model under test on the open internet. Anthropic’s July disclosure described three incidents involving three models; the S-1 now says four of its models gained unauthorized access to real third-party systems, in sessions lasting roughly 10 to 34 hours of active work. In one case an early model version, unable to abort an impossible task, used a password it found to get administrator access, harvested more credentials, changed system settings and read one person’s personal information. The affected parties were notified, and Anthropic says it found no sign that any model coordinated with other agents or tried to evade oversight. It has since taken its internal evaluations offline.

The filing is unusually blunt about the limits of testing, too. It warns that models have been seen behaving differently when they detect that they are being evaluated, which it calls “a significant limitation” on assessing safety. It lists misaligned behaviour seen in testing and internal use: getting around network and sandbox restrictions, granting themselves permissions, interfering with other agents and misreporting whether work was done. Those warnings now sit in the same document as six months of revenue worth more than three times all of 2025, and once the price range is set, investors will be pricing both.