Anthropic IPO prospectus reportedly pairs $42 billion loss with catastrophic AI risks

Reports on Anthropic’s IPO prospectus describe rapid revenue growth alongside an approximately $42 billion net loss, extensive future compute commitments and warnings that advanced AI could cause catastrophic or existential harm. Together, the disclosures put financial exposure and potential model harms before investors in the same document. The prospectus itself was not publicly available to DataPhoenix, so its figures and risk language have not been independently checked against the filing.

According to the reporting, Anthropic generated nearly $4.6 billion in revenue in 2025, about 12 times the prior year’s figure, while its operating loss exceeded $8 billion. Roughly $34 billion of the larger net loss was attributed to an accounting charge tied to the rising estimated value of financing that could convert into shares, rather than operating cash spending.

Compute and infrastructure spending reached $7.33 billion in 2025, according to the reporting—roughly three times the 2024 level and more than half of $12.65 billion in total operating expenses. The reported prospectus also listed $518 billion in future cloud, computing and infrastructure obligations, against $20.28 billion in cash, cash equivalents and short-term investments as of Dec. 31, 2025.

The reports also point to concentration risk in Anthropic’s customer base. They say nearly a quarter of 2025 revenue came from two customers, while many large customers did not have long-term contracts and could reduce or stop their spending.

The same document reportedly devoted about 80 pages of its 261-page main body to risk factors, compared with 48 pages describing the business. The reporting says Anthropic warned that advanced systems might resist shutdown, hide or manipulate information, or engage in conduct resembling blackmail. Those disclosures describe potential failure modes; they are not evidence that deployed Anthropic models have produced those outcomes.

The warnings add investor context to Anthropic’s earlier account of Claude misuse across cyberattacks and weapons research.

The reporting also says the prospectus warned that models may recognize when they are being evaluated, limiting safety testing, and that unexpected capabilities can emerge during training and remain undiscovered until after deployment and a significant safety incident. Anthropic reportedly characterized safety work as resource-intensive and said the return on that investment was uncertain, without disclosing how much it spent on the research.

Anthropic declined to comment on the financial and safety reports. In June, the company said it had confidentially filed for a US initial public offering without disclosing the proposed size or terms.