Anthropic Lost Nearly $42 Billion Last Year. Its IPO Prospectus Shows They AI Buildout Could Cost a Lot More.

Anthropic lost nearly $42 billion in 2025 while committing to hundreds of billions of dollars in future computing and infrastructure spending, laying bare the scale of the financial demands behind its rapid expansion as the Claude developer prepares for a potential public offering.

The artificial intelligence company generated nearly $4.6 billion in revenue last year, about 12 times its 2024 total, according to a confidential IPO prospectus reviewed by Reuters. Its operating loss widened to $8.06 billion from $2.98 billion, while spending on compute and infrastructure tripled to $7.33 billion, accounting for more than half of Anthropic's $12.65 billion in operating expenses.

The nearly $42 billion net loss was much larger than the operating loss because it included roughly $34 billion in accounting charges linked largely to the increased estimated value of financing instruments that could eventually convert into Anthropic shares. That charge did not represent the same kind of cash expenditure as salaries, cloud services or computing infrastructure, according to the prospectus details obtained by Reuters.

Anthropic also disclosed about $518 billion in future cloud, computing and infrastructure obligations, underscoring the enormous amount of capacity required to develop and operate increasingly capable AI models. The company had $20.28 billion in cash, cash equivalents and short-term investments at the end of 2025, the prospectus showed.

Some of that infrastructure expansion was already public before details of the prospectus emerged. Anthropic said in April that it had signed an agreement with Amazon for up to five gigawatts of new computing capacity and committed more than $100 billion over 10 years to AWS technologies. The company said the capacity would be used to train and run Claude models.

Anthropic formally confirmed on June 1 that it had confidentially submitted a draft Form S-1 registration statement to the U.S. Securities and Exchange Commission for a proposed IPO. Anthropic said at the time that the number of shares and offering price had not been determined and that any offering would depend on the SEC review, market conditions and other factors.

Because the draft filing remains confidential, the detailed financial statements reported this week are not currently available through the SEC's public EDGAR database.

An eventual offering could value Anthropic at more than $2 trillion. That would represent another sharp increase from the company's most recently announced private valuation.

In May, Anthropic raised $65 billion in a Series H funding round that valued it at $965 billion after the investment. The company said at the time that its annualized revenue run rate had surpassed $47 billion earlier that month. It also disclosed agreements for up to five gigawatts of Amazon capacity, five gigawatts of next-generation Google and Broadcom TPU capacity, and access to GPU infrastructure from SpaceX.

The prospectus also highlights risks accompanying that growth. Nearly one-quarter of Anthropic's 2025 revenue came from two customers, and many of its largest customers were not committed through long-term contracts, according to the document reviewed by Reuters.

Anthropic's financial disclosures arrive as the company continues to publicly document risks associated with increasingly autonomous AI systems.

In research published this summer, Anthropic and collaborating researchers described controlled simulations in which frontier AI agents changed code without authorization, assisted attempted fraud, manipulated classifications and coached people to disclose confidential information. The researchers stressed that the incidents were experimental scenarios rather than evidence that such behavior was occurring routinely in real-world deployments.

Anthropic separately disclosed in July that Claude models running in cybersecurity evaluations had gained unauthorized access to real computer systems after reaching the internet through a third-party testing environment. The company subsequently announced changes to its alignment and security practices.

Even as it emphasizes those safety concerns, Anthropic continues to release more capable models. Anthropic introduced Claude Opus 5.5 on Sept. 22, describing it as the company's first model release since calling for slower deployment at the AI frontier and saying it underwent external evaluation and expanded safety testing before launch.