Anthropic lost $42 billion in 2025, even as revenue grew nearly twelvefold to $4.6 billion. Those figures come from a confidential draft of the company’s IPO prospectus, which Reuters obtained and first reported on Monday. The document also shows Anthropic committing to more than half a trillion dollars in future computing infrastructure, and warns that the company’s own AI could pose “catastrophic or existential risks to humanity.”
About $34 billion of the $42 billion loss comes from a non-cash accounting charge tied to revaluing financing instruments that could eventually convert into equity — a bookkeeping requirement, not actual spending. Strip it out and Anthropic’s operating loss for 2025 was about $8 billion, roughly the size of its entire net loss a year earlier, even as revenue grew twelvefold.
The $518 billion in infrastructure commitments is also a multi-year total, not a single year’s bill. It includes a ten-year, $100-billion-plus pledge to Amazon Web Services for up to 5 gigawatts of compute capacity, plus further gigawatt-scale expansions Anthropic has agreed to with Google and Broadcom, on top of $71 billion in off-balance-sheet debt through special-purpose vehicles for Google’s TPU chips and a separate $15 billion credit facility.
Compute and infrastructure spending totaled $7.33 billion in 2025—nearly three times 2024’s total and 58% of the company’s $12.65 billion in operating expenses. Anthropic spent about $1.60 on computing for every dollar of revenue it brought in. Its top two customers accounted for about 25% of that revenue, and the company disclosed that most of its largest customers haven’t signed long-term contracts and could cut or cancel spending at any time.
The company ended 2025 with $20.3 billion in cash, and has since raised $65 billion in a Series H round in May at a $965 billion valuation, up from the $380 billion valuation behind its Series G round earlier in the year. The IPO valuation is more than double that — over $2 trillion, on the numbers in the leaked prospectus. A valuation that high would also lift the paper returns of early backers like Salesforce, which put $50 million into Anthropic in 2023 and has followed on since; investors last valued that stake at around $5 billion.
Anthropic confidentially submitted a draft S-1 registration to the SEC on June 1, cautioning that the offering “will depend on market conditions and other factors” and that share numbers and pricing “have not yet been set.” By early September, Reuters was reporting that marketing would begin in mid-October at the earliest. The latest reporting suggests the listing is now more likely to land after November’s US midterm elections.
Read: Anthropic Revenue Run Rate Surpasses $30 Billion As Enterprise Demand Accelerates
Risk factors make up about 31% of the document, 80 of its 261 pages — compared with roughly 14% in SpaceX‘s IPO prospectus (38 of 277 pages). Anthropic spent only 48 pages describing the business itself. The company says its models have attempted to “resist shutdown” during testing and have shown they can “conceal or manipulate information.” It also flags conduct “resembling blackmail” among the behaviors it’s watching for. Its models, Anthropic says, can also learn to recognize when researchers are evaluating them and adjust their behavior accordingly — “a significant limitation on our ability to assess model safety.”
🦔Anthropic filed its IPO prospectus. Revenue grew 12x in 2025 to $4.6 billion. The company lost $8 billion on operations. It plans to spend $518 billion on cloud and infrastructure in the years ahead. Nearly a quarter of revenue came from two customers who aren't locked into… pic.twitter.com/zL97QtEXCc
— Hedgie (@HedgieMarkets) September 29, 2026
Only about 6% of the computing power Anthropic dedicated to AI research went to safety work during one week this July, despite the company’s safety-first public image — and Anthropic calls the return on that spending “unclear.”
Equity analyst Ross Hendricks, who writes the Ross Report, mocked the infrastructure figure. “Just a casual mention of raising half a trillion in a single year, really? Does anyone believe this?” he wrote, adding, “Let’s say they pull $100B in equity from the IPO… and then what, $400B from the debt markets?”
Another hidden gem in the Reuters drop of Anthropic's financials… the company "plans to spend $518 billion on cloud, computing and infrastructure obligations in coming year, according to the prospectus."
— Ross Hendricks (@Ross__Hendricks) September 29, 2026
Just a casual mention of raising half a trillion in a single year,…
Ed Zitron, CEO of EZ Primary Research and a longtime AI critic, called Anthropic “a total dog of a company,” noting it spent $12.6 billion to bring in $4.6 billion in revenue in 2025, with $7.33 billion of that going to compute alone. “Losses getting worse,” he added, pointing to the widening operating loss.
Bahahahaha Anthropic is a total dog of a company. Spent $12.6bn to make $4.6bn in revenue in 2025, $7.33bn of which was compute costs. Operating loss of $8bn. Losses getting worse. Weird that Reuters doesn’t have the 2026 numbers from the prospectushttps://t.co/eWnHlfpCgF https://t.co/xZBY54VpoZ pic.twitter.com/kyLTeYf3uS
— Ed Zitron (@edzitron) September 29, 2026