Anthropic filed for a trillion-dollar IPO. The S-1 shows what Claude costs.

Anthropic publicly filed its S-1 with the U.S. Securities and Exchange Commission on October 26, Beijing time. The company plans to list on Nasdaq under the ticker ANT, with Goldman Sachs and Morgan Stanley leading the underwriting. The 335-page document opens up Anthropic’s revenue, losses, customers, compute contracts, and governance for the first time.

Market discussion puts the potential IPO valuation at $1.5 trillion to $2 trillion, with a possible raise of about $100 billion. If completed, the listing could rank among the largest technology IPOs in history and set a public benchmark for the generative AI business model.

The numbers in the filing are large enough to justify the attention: $16.29 billion in half-year revenue, up about 13-fold year over year; an $80.75 billion net loss; more than 11,900 customers contributing at least $100,000 in annual revenue; and long-term compute commitments above $500 billion.

The S-1 gives the first public answer to two questions: how much money Claude can make, and how much cash the business consumes.

Revenue growth

Anthropic’s revenue was $386 million in 2024, $4.6 billion in 2025, and $16.3 billion in the first half of 2026, about 14 times the year-earlier period. Second-quarter revenue reached $11.6 billion. Third-quarter revenue rose to $16.8 billion, up about 45% from the second quarter.

Enterprise customers have moved from trials to large-scale procurement. In the second quarter, Anthropic generated about $1 billion in operating profit, an 8% margin, compared with a $1.2 billion loss a year earlier. Compute costs explain much of the shift. In the first quarter of 2025, compute spending was 2.5 times revenue. By the second quarter of 2026, it had fallen to 48% of revenue. In the first half of 2025, Anthropic spent about $2.01 on compute and infrastructure for every $1 of revenue. In the first half of 2026, that figure fell to about $0.55.

Anthropic books compute 12 to 18 months in advance. Revenue is growing faster than those fixed bills are released. The company still carries losses, but the operating numbers show that a large model company can turn profitable at scale.

Who pays

As of the end of September 2026, Anthropic had more than 11,900 customers that had spent at least $100,000 over the previous 12 months. More than 1,900 had spent at least $1 million. About 250 had spent more than $10 million.

That customer mix looks like an enterprise AI platform, not a consumer chatbot.

Net revenue retention reached 900% as of the end of September 2026, excluding Claude Pro and Max subscriptions. For the same eligible customers that were already paying a year earlier, revenue in the third quarter of 2026 was about nine times the year-earlier level.

In the first half of 2026, $13.1 billion of Anthropic’s $16.3 billion in revenue came from usage-based pricing, or 80%. Subscriptions contributed about $3.2 billion, or 20%. Enterprises call Claude through APIs and pay by token consumption and model usage. Claude Code and AI agents are pushing that usage into software development and office workflows. Model calls are becoming a production cost rather than a one-time purchase.

Customer retention still needs time to prove itself. For now, Anthropic has grown from a frontier model lab into an enterprise AI platform with hundreds of customers spending at the $10 million level.

The $80.7 billion loss

Anthropic’s first-half net loss was $80.748 billion. About $72.388 billion came from fair value changes in convertible notes, derivative liabilities, and SAFEs. The company’s rising valuation pushed up the fair value of some outstanding convertible promissory notes. That produced a large accounting loss.

The loss is mostly non-cash. As of the end of June, Anthropic had raised $141.1 billion in cumulative capital and held $112.4 billion in cash and cash equivalents plus $19.3 billion in short-term investments, about $131.7 billion in total. Operating activities generated about $3.93 billion in positive cash flow in the first half of 2026, though customer prepayments and settlement timing affected that figure.

The net loss reflects accounting treatment. The compute bill is the cash issue.

The compute bill

Anthropic has locked in compute through long-term contracts. The filing lists about $161.2 billion for TPU equipment leased from Broadcom; up to $120.5 billion for Nvidia GPU compute from SpaceX’s SpaceXAI; at least $114.5 billion from Google; at least $110 billion from Amazon; at least $44.9 billion from Microsoft; up to $44.6 billion from Nscale; and about $41.3 billion from Lambda. AMD’s contract has no fixed amount, but the first batch could exceed $20 billion. The longest contracts run to 2045.

The listed contracts add up to more than $600 billion. Some media tallies put the total closer to $800 billion. Contracts signed or modified after June 30 carry minimum payments of about $245 billion. Anthropic also expects to buy another batch of TPUs from 2028 to 2031 for $125 billion to $150 billion.

The bet is straightforward: enterprise AI usage keeps growing, Claude holds share in model competition, and falling inference costs do not turn these contracts into a heavy burden.

Cloud providers also control a large share of distribution. In 2025, 47% of revenue came through cloud marketplaces. In the first half of 2026, the figure was 48%. Anthropic depends on cloud providers for both revenue and compute. That brings reliable infrastructure and customer access, along with bargaining power and channel risk.

Control after the IPO

Seven co-founders, including CEO Dario Amodei and President Daniela Amodei, will hold special Class F shares that give them 50.1% of shareholder voting power after the IPO. Amazon’s conversion of some convertible notes is capped at 12% ownership. Google’s share issuance under its 2026 financing agreement is capped at 14.99%. Both companies mainly hold Class N shares without general voting rights. The Long-Term Benefit Trust will elect a majority of the board through a special governance mechanism.

IPO investors will bear stock price risk without proportional control over major decisions. Anthropic is a Public Benefit Corporation, and its mission includes developing safe and controllable advanced AI. The risk disclosures discuss autonomous agent behavior, information security, regulation, copyright litigation, and accidents that could exceed insurance coverage.

A company that warns about extreme risks from its own technology is now seeking a trillion-dollar valuation. Public markets have not priced that combination before.

The valuation

A $1.5 trillion to $2 trillion valuation implies 46 to 61 times annualized first-half revenue of about $32.6 billion. That calculation uses disclosed revenue, not forecasts. Annualizing third-quarter revenue of $16.8 billion gives about $67.2 billion, which lowers the multiple but still leaves it high for a growth company.

Wall Street is pricing more than the Claude product. The bet is that AI models become infrastructure for the next digital economy. If agents spread through office work, software development, financial services, and research, token usage could keep growing. Anthropic could become the model platform behind enterprise AI applications.

Competition could weaken that outcome. Open-source models are improving. Inference costs are falling. Model quality may matter less than enterprise integration, data, and workflow control. The moat remains unclear.

What the IPO means

Anthropic’s IPO is the first large public pricing test for a frontier foundation model company. The S-1 shows that enterprises will pay for advanced AI. It also shows how much capital the business requires.

OpenAI, Google, and open-source model teams are still pushing model capability and price competition. Future AI winners will need to convert model capability into revenue and revenue into profit. Anthropic has delivered its first public scorecard. Wall Street now decides what that scorecard is worth.

Under SEC rules, a company using the confidential filing process must release its prospectus at least 15 days before starting its roadshow. Anthropic filed publicly on October 26, so a roadshow could begin around November 10.