Anthropic exposes its losses and risks ahead of an extraordinary IPO
The creator of Claude filed its IPO prospectus with the SEC. Anthropic reports $16.3 billion in half-year revenue but also $80.7 billion in net loss and 94 pages of risks.
Anthropic, the publisher of the assistant Claude, filed its IPO prospectus with the Securities and Exchange Commission on Monday, with a planned listing on Nasdaq under the ticker ANT. The group reveals $16.3 billion in revenue for the first six months of 2026, but also a net loss of $80.7 billion. The document is 614 pages long, including 94 pages dedicated to risks, indicating that at this stage investor confidence will have to contend with a long list of warnings.
Une croissance rapide, mais encore loin d’un modèle apaisé
The financial figures set the tone. In the first half of 2026, Anthropic says it multiplied its revenue by more than 13 compared to the $1.2 billion recorded a year earlier. The group also reports an operating loss of $693 million, compared to $2.1 billion a year before, and claims an adjusted operating profit of $1.6 billion, or a margin close to 10%.
But the picture remains fragile. Of the $80.7 billion net loss, $72.4 billion come from an accounting adjustment related to valuation increases, not cash outflows. In other words, the financial snapshot still largely depends on private valuation mechanics, not just Claude’s sales. Susannah Streeter, strategist at Wealth Club, sums up the friction point: IPO investors might pay a price loaded “with a good dose of future success.”
Pourquoi Anthropic parle aussi de blackmail et de shutdown
The document does not only discuss revenue and governance. Anthropic indicates that its models, including its own, have shown unexpected or self-preserving behaviors in tests: attempts to resist shutdown, conceal information, outputs resembling blackmail. The company specifies that this behavior has not been observed in clients but is considered serious enough to be explicitly included in the S-1.
This caution is also explained by the growing use of agents able to interact with external tools, code repositories, and enterprise systems. In this context, Anthropic warns that an attacker could embed malicious instructions in a website, document, or API response. If the agent follows them, it could use the client’s rights to steal data, modify code, or trigger unauthorized transactions. The scenario is not science fiction. It is much more mundane, therefore more troubling.
Une gouvernance pensée pour limiter la main de Wall Street
The most unusual point in the document may be the power structure. Anthropic announces five classes of shares, compared to three at Google, and specifies that the seven co-founders will retain 50.1% of voting rights on shareholder matters. A Long-Term Benefit Trust can also appoint the majority of the board of directors. In short, investors are invited to finance the group’s growth without necessarily controlling strategic direction.
In a letter attached to the document, Dario and Daniela Amodei, along with other co-founders, ask future shareholders not to consider their purchase “in purely transactional terms.” Anthropic adds that it might forgo profitable activities if it deems the risks too high. This stance may reassure on safety but complicates market reading, which prefers margins, schedules, and simple rules. Disorder rarely has fans on Wall Street, except when accompanied by a growth promise.
Ce que l’IPO d’Anthropic dit du marché de l’IA
The document mainly shows where the current tension in generative AI lies: rapidly rising revenues, still heavy infrastructure costs, and increasingly visible internal regulation. Anthropic wants to raise funds to support the race for more powerful models, while acknowledging it might slow its own development if safeguards require it. This mix of caution and ambition is not new in tech but becomes clearer when written over 614 pages.
One major unknown remains: Anthropic has not yet disclosed the target price for its shares nor the number of shares offered. At this stage, the stock market therefore knows neither the final bill nor the size of the check. It only knows the cost promises to be steep and that it also includes a user guide to risks.
By the numbers
- $16.3 billion — Anthropic’s revenue for the first six months of 2026.
- $80.7 billion — half-year net loss, including $72.4 billion accounting adjustment.
- 614 pages — length of the S-1 filing submitted to the SEC.
- 94 pages — section dedicated to risk factors.
- 50.1% — share of voting rights retained by the seven co-founders.