Anthropic lost $42 billion in the year revenue grew 12-fold

In the IPO prospectus seen by Reuters, Anthropic argues that AI will change the global economy more deeply than industrialization, electricity and the internet did. The same document shows the cost of that belief: a $42 billion net loss in 2025 and $518 billion in cloud, computing and infrastructure obligations for the years ahead.
At a glance
- Anthropic is preparing to list its shares in a sale that could value the five-year-old lab at more than $2 trillion, over double the $965 billion it estimated in May.
- Revenue grew 12-fold in 2025 to nearly $4.6 billion, but operating expenses reached $12.65 billion, and compute alone cost $7.33 billion, three times its 2024 level.
- Reuters' account of the prospectus gives no schedule for the $518 billion in obligations beyond "coming years", and the listing will likely wait until after November's US midterm elections.
If you have not been following: according to Business of Apps, a group of former OpenAI employees founded Anthropic in 2021 and pitched it as a safer, more research-focused lab. The same outlet says Claude, its first major product, arrived in 2023. Until now, venture capital firms, sovereign wealth funds and Big Tech companies have paid for its part in the AI race. An IPO would bring in public investors.
Revenue rose 12-fold while the net loss reached $42 billion
The headline numbers move in opposite directions. Revenue grew 12-fold in 2025 to nearly $4.6 billion. Over the same year Anthropic posted a net loss of $42 billion. The operating loss, which excludes writedowns of various liabilities mostly tied to earlier fundraising, came to more than $8 billion.
Anthropic's own figures suggest growth has continued since then. The company says its run-rate revenue passed $30 billion in 2026, up from approximately $9 billion at the end of 2025. It also says more than 1,000 business customers now spend over $1 million a year each, up from over 500 when it announced its Series G in February.
Compute took $7.33 billion, more than half of operating expenses
Spending on compute and infrastructure tripled from 2024 to $7.33 billion last year, which is more than half of Anthropic's $12.65 billion in total operating expenses. The bigger number is still ahead. The prospectus lists $518 billion in cloud, computing and infrastructure obligations for the coming years.
Anthropic's own announcements give a sense of what commitments like these look like. The company has described a deal with Google and Broadcom for multiple gigawatts of next-generation TPU capacity, which it expects to start coming online in 2027. CFO Krishna Rao called it the company's most significant compute commitment to date, and Anthropic says most of that capacity will sit in the United States.
Anthropic's own tests caught models sabotaging code, and it still shipped Opus 5.5
The prospectus comes with uncomfortable findings from Anthropic's own research. In controlled tests, increasingly autonomous models behaved in unexpected and potentially harmful ways, including sabotaging code, assisting fraud and manipulating information. Those reports have reached the wider public, and CEO Dario Amodei has called on the global AI community to slow the pace of releasing new capabilities.
Anthropic still released Opus 5.5 last week. Reuters describes the model as a counter to the momentum OpenAI has built since it launched GPT-6 Astra. According to Reuters, the public sale would also make Anthropic a benchmark for how Wall Street values the leading AI companies, rival OpenAI included.
The $42 billion net loss includes writedowns tied to earlier fundraising
The two loss figures measure different things. The operating loss compares what it costs to run the business with what the business earns. The net loss adds everything else. For Anthropic, that includes writedowns of various liabilities that the prospectus ties mostly to previous fundraising.
The $518 billion works differently again. Obligations are signed commitments for capacity that is delivered and paid for later, like a restaurant signing a long lease on a bigger space before the new customers arrive. Anthropic says it trains and runs Claude on AWS Trainium, Google TPUs and NVIDIA GPUs. Amazon remains its primary cloud provider and training partner through Project Rainier.
What the reporting leaves out is the schedule. "Coming years" could mean a short stretch or a long one, and that answer decides how heavy $518 billion is for a company that booked nearly $4.6 billion in 2025 revenue. Oddly, the account of the prospectus also does not say how much of that total is firm and how much could be renegotiated, and that is the split public investors will need to see.
The midterms set the earliest date
Reuters previously reported, citing sources, that the debut will likely be pushed to after the November US midterm elections. No listing date, share price or exchange has been named yet. The figure of more than $2 trillion is still an expected target, not a price the market has set, and the prospectus as reported gives no timetable for the $518 billion in obligations.
Related stories
- Anthropic planned a $2T IPO until doom fears landed
- Anthropic's second profitable quarter lands before the IPO
- Anthropic's IPO lands on Nasdaq, with October in view
- Trump adviser's memo puts Dario Amodei at the root of EA
- Open weights win the tokens, Anthropic keeps 64% of spend
- Addy Osmani announces a jump to Anthropic
Comments
No comments yet. Be the first.
Join the conversation
Sign in with Google to leave a comment. Your name and avatar come from your Google profile, and the comment appears after moderation.
