Anthropic filed its S-1 publicly on Monday, 26 October, and will trade on Nasdaq under the symbol ANT. The share count and price range are still unspecified, but the document itself is the story: it is the first time the economics of a frontier lab have been laid out in full, and the shape is stranger than the headline numbers suggest.
The $80.7 Billion Loss Is Mostly a Number That Went Up
First-half 2026 revenue was $16.3 billion, up 1,300% from $1.16 billion a year earlier. The net loss was $80.7 billion — a figure that looks catastrophic and mostly is not. Roughly $72.4 billion comes from the change in accounting value of convertible notes and other financing instruments, which move with Anthropic’s own rising valuation, plus another $8.72 billion in non-cash charges.
Strip that out and the operating picture is ordinary: operating expenses of about $17 billion against $16.3 billion of revenue, an operating loss of $693 million (down from $2.14 billion), and positive operating cash flow of $3.93 billion. Excluding stock compensation and non-cash charitable contributions, adjusted operating income was $1.62 billion — real, but not profit under standard rules.
Two Clocks, Running at Different Speeds
The disclosure that should worry every enterprise buyer is not the loss. It is the mismatch of contract durations. Usage-based sales supplied 80% of first-half revenue — cancellable, per-token, effectively month to month. Infrastructure contracts run the other way: multi-year commitments that can require payment for years, including for capacity Anthropic does not use. One disclosed equipment lease alone entails approximately $125.2 billion in rent over five years, excluding interest, and Anthropic warns that a missed obligation could trigger defaults across other agreements and jeopardize access to essential resources.
Capacity is uncertain in the other direction too: the company says it has sought more compute than cloud providers were willing to supply, and its arrangements do not guarantee delivery on its preferred schedule. So the lab has five-year fixed costs, cancellable revenue, and no contractual right to the hardware it has already promised to pay for.
Enterprise scale is real — more than 1,900 customers generated at least $1 million each in the trailing twelve months, around 250 of them at least $10 million — and the same filing describes a long-term ambition of millions of autonomous collaborators working for weeks on end, a “country of geniuses in a data center.”
What Buyers Should Take From It
Two practical consequences follow. First, capacity guarantees, service continuity and the ability to move workloads between providers become procurement requirements, not operational nice-to-haves. Second, the liability section is unusually blunt: Anthropic says it remains unclear when an agent’s actions legally bind the deploying organization, warns that human reviewers may not keep pace with autonomous systems, and notes that a sufficiently severe AI incident could exceed the insurance industry’s capacity to cover the damage. The FTC issued an investigative demand in October covering consumer protection, privacy, data protection and product safety.
Related: Anthropic S-1 registration statement (SEC EDGAR) and VentureBeat’s reading of the filing