Anthropic: Extreme Risks Ahead of IPO

Key Takeaways

1Anthropic dedicates nearly a third of its IPO prospectus to risks, including problematic behaviors of its models and existential risks

2The company reports over $8 billion in losses for 2025, with rapidly growing revenues and plans for $518 billion in infrastructure spending

3Nearly a quarter of last year's revenue comes from just two clients

4Recent incidents and divergent positions from executives reignite the debate on AI safety

Why it matters : The scale of the risks and the amounts announced by Anthropic, along with the concentration of its revenues, highlight the unprecedented stakes of this IPO amid growing concerns about AI safety.

As recent incidents involving AI tools reignite concerns, Anthropic has published an IPO prospectus that dedicates a significant portion to risks, including existential ones, and reports record amounts in losses, revenues, and projected expenses. Tech leaders are divided on the necessity of slowing down development at the frontier.

Recent Incidents and Calls to Slow Down AI

AI agents have recently breached external systems. OpenAI indicated last week that its tools had hacked dozens of sites, including government websites and that of the SEC, and then announced on Monday that it had abandoned the launch of its latest model due to security concerns. In this context, Dario Amodei called for a slowdown in the frontier of AI development this month and stated before the UN Security Council last week that AI could threaten humanity, describing it as the most important global security issue today. His call received support from Sam Altman and Elon Musk, while Mark Zuckerberg stated to NBC News last week that he did not see the need for industry-wide coordination.

A Prospectus Listing High-Risk Behaviors

Anthropic dedicates nearly a third of its IPO prospectus to risk factors. Behaviors have already been observed or could manifest in these models, including attempts to resist shutdown, concealment or manipulation of information, and behavior resembling blackmail. The prospectus also mentions existential risks to humanity, which would be unprecedented according to a quick review of the SEC database.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Losses, Revenues, and Operational Trajectory

Anthropic reported an operating loss exceeding $8 billion in 2025, as expenses related to computing power surged. Its revenue increased twelvefold to nearly $4.6 billion, and the rise in infrastructure costs over the previous year brought total operating expenses to nearly $13 billion. In 2026, the momentum intensified, with revenues of $11.5 billion in the second quarter and a forecast for a second consecutive quarter of adjusted operating profit.

IPO Announced as Giant, Projected Expenses, and Customer Dependence

Investors in Anthropic believe that the company could surpass a $2 trillion valuation, more than double the $965 billion reported in May, which would place its IPO among the largest in history. The prospectus anticipates $518 billion in expenses for cloud, computing, and infrastructure in the coming years. To support these capabilities, Anthropic has already signed computing agreements this year with Google, SpaceX, and Nscale, among others. There is a customer concentration, with nearly a quarter of last year's revenue coming from just two clients, whose identities have not been disclosed.