Anthropic publicly filed for an initial public offering Monday, revealing a business growing at extraordinary speed while warning that the autonomous AI systems underpinning its ambitions could expose customers—and the company—to legal obligations, security failures and liabilities that insurance may not cover.
The Claude developer’s S-1 registration statement reports $16.3 billion in revenue for the first six months of 2026, up 1,300% from $1.16 billion a year earlier. Anthropic has applied to trade on Nasdaq under the symbol ANT, although the filing leaves the offering’s share count and price range unspecified.
For enterprise technology leaders, the document offers a detailed look at the economics and dependencies behind a vendor increasingly embedded in business operations. It also spells out where Anthropic wants to take Claude: beyond answering questions and writing code, toward millions of autonomous collaborators handling assignments that stretch across weeks.
The $80.7 billion loss needs unpacking
Anthropic reports a first-half net loss of $80.7 billion. But that figure does not represent cash burned running Claude.
About $72.4 billion comes from changes in the accounting value of convertible notes and other financing instruments, primarily reflecting Anthropic’s rising valuation. Its cash-flow statement identifies another $8.72 billion in noncash charges involving other financial instruments.
The operating picture is substantially different. Revenue of $16.3 billion fell short of operating expenses of approximately $17 billion, producing a $693 million operating loss—down from $2.14 billion a year earlier.
Computing and infrastructure expenses reached $8.95 billion, up from $2.33 billion. Personnel expenses totaled $3.15 billion, while sales, marketing and partnerships accounted for another $3.17 billion.
Excluding stock-based compensation and noncash charitable contributions, Anthropic reports adjusted operating income of $1.62 billion. That measure is not a profit under standard accounting rules. Operating cash flow was positive at $3.93 billion, with the company attributing the improvement partly to the timing of customer receipts and vendor payments.
Enterprise adoption helps explain the growth. As of September 30, Anthropic says more than 1,900 customers generated at least $1 million each in revenue over the preceding 12 months, including approximately 250 that generated at least $10 million.
Computing commitments outlast customer commitments
One of the filing’s most consequential warnings concerns a mismatch between Anthropic’s spending obligations and its revenue.
Usage-based sales supplied 80% of first-half revenue. Much of its revenue lacks long-term customer commitments, while infrastructure contracts can require payments for years, including for capacity it does not use.
One disclosed equipment lease alone entails approximately $125.2 billion in rent over five years, excluding interest. Anthropic warns that failure to meet certain obligations could trigger defaults across other agreements and jeopardize access to essential resources.
Capacity is also uncertain. The company says it has previously sought more computing resources than cloud providers were willing to supply, and contractual arrangements do not guarantee delivery on its preferred schedule.
For enterprises, these disclosures make capacity guarantees, service continuity and the ability to shift workloads between providers important procurement considerations alongside model performance.
A workforce of agents—and unresolved responsibility
Anthropic’s longer-term ambition is to give Claude increasingly open-ended assignments, have it work independently for extended periods, and eventually deploy millions of instances simultaneously. The filing describes the destination as a “country of geniuses in a data center.”
That vision also creates a practical question for customers: What happens when an agent makes a consequential mistake?
Anthropic says it remains unclear when an agent’s actions could legally bind the organization deploying it, or whether contractual liability limits will hold up. It also warns that human reviewers may be unable to keep pace with autonomous systems’ speed and complexity.
Some agreements with outside model evaluators expose Anthropic to uncapped liability for specified third-party claims. The company cautions that a sufficiently severe AI incident could exceed even the insurance industry’s ability to cover the damage.
Regulatory scrutiny is already concrete. Anthropic discloses that the Federal Trade Commission issued an investigative demand in October seeking information about consumer protection, privacy, data protection, and product safety and security. The filing does not establish wrongdoing or predict the inquiry’s outcome.
For enterprise buyers, the prospectus makes the trade-off unusually explicit. Anthropic is pursuing AI that can assume more responsibility across an organization. Its own disclosures show why buyers still need to define that responsibility carefully: which systems agents can access, which decisions require approval, who bears losses, and how operations continue when either the model or its infrastructure fails.