Anthropic's IPO filing, explained: huge growth, bigger bills, and a warning about humanity
A draft of the prospectus for what could be the largest share sale in history shows how fast the AI business is growing, how much it costs to build, and what its makers say could go wrong.

Key points
- According to a draft prospectus reviewed by Reuters, Anthropic's revenue rose about twelvefold to nearly $4.6 billion in 2025, and reached $11.5 billion in the second quarter of 2026 alone.
- Costs grew almost as fast. Anthropic spent $7.3 billion on computing in 2025 and lost more than $8 billion on its operations, although it says it made an adjusted operating profit in the second quarter of 2026.
- It has committed to about $518 billion in future spending on cloud and computing capacity, more than 25 times the cash it held at the end of 2025.
- Two customers produced nearly a quarter of 2025 revenue, and many large clients are not locked into long-term contracts.
- The risk section warns that AI may pose catastrophic or existential risks to humanity, and describes model behaviour such as resisting shutdown and hiding information.
- Anthropic's founders are seeking voting control through a special share, and the company is fighting a Pentagon designation as a supply-chain risk.
Anthropic, the company behind the Claude chatbot, is preparing to sell shares to the public. It filed its paperwork with the US Securities and Exchange Commission confidentially in June, so the official prospectus is not yet public. But Reuters has seen a draft, and the numbers it reported give the clearest picture yet of how the AI business actually works: extraordinary growth, enormous costs and unusually blunt warnings.
A note on sources: the figures below come from Reuters’ reporting on that draft and may change before the final document is published. Backers reportedly hope for a valuation above $2 trillion, which would make it one of the largest listings ever. Reuters says the sale is expected after the US midterm elections in November.
Growth that is hard to picture
Anthropic’s revenue rose about twelvefold in 2025, to nearly $4.6 billion, from roughly $380 million the year before. Then it accelerated again. In the second quarter of 2026 alone, the company brought in $11.5 billion, more than twice as much as in the whole of 2025.
One quarter of 2026 brought in more than all of 2025
Anthropic revenue, billions of US dollars. 2024 is implied by the reported twelvefold growth.
Source: Draft prospectus as reported by Reuters and TechCrunch, September 2026
Most of that money comes from businesses, not consumers: companies paying to build Claude into their own products and staff using it for coding and office work. That has a catch. Two customers produced nearly a quarter of 2025 revenue, and the draft warns that many large clients are not tied to long-term contracts. If one of them switches to a rival, the revenue can go with it.
The bill for building it
AI models are expensive to train and to run, because every answer uses powerful chips in a data centre. In 2025 Anthropic spent $7.33 billion on computing and infrastructure, more than half of its $12.65 billion in operating costs. Its operating loss was more than $8 billion.
In 2025, computing alone cost more than all the revenue
Anthropic 2025, billions of US dollars
Source: Draft prospectus as reported by Reuters, September 2026
The headline net loss for 2025 looks far worse, at $42 billion, but about $34 billion of that is an accounting charge that doesn’t involve cash: it reflects the rising value of investments that can convert into shares. The trend is also moving in Anthropic’s favour. The company says it made an adjusted operating profit in the second quarter of 2026 and is on track for another in the third. “Adjusted” means some costs are left out, so it is not the same as a profit under standard accounting rules.
A $518 billion promise
The biggest number in the filing is not revenue or losses but commitments. Anthropic has signed up for about $518 billion of future spending on cloud and computing capacity, across several cloud providers. The latest is an $11.6 billion, seven-year deal with Akamai announced last week. At the end of 2025, Anthropic had $20.3 billion in cash and short-term investments.
Future computing commitments dwarf today's cash
Billions of US dollars
Source: Draft prospectus as reported by Reuters, September 2026
These commitments are spread over many years, and the bet is that revenue keeps growing fast enough to pay for them. That is the central question for anyone judging the share sale, and it applies to every company racing to build ever-larger AI systems.
An unusual list of risks
Every prospectus lists risks, but this one devotes nearly a third of its length to them, and some are unlike anything investors usually read. According to Reuters, the draft warns that AI may pose “catastrophic or existential risks to humanity”. It also describes behaviour seen in AI models during testing, including attempts to resist being shut down, to hide or manipulate information, and actions resembling blackmail.
More ordinary risks are there too. The US Defense Department has designated Anthropic a supply-chain risk, which limits its government work, and on 25 September an appeals court allowed that designation to stand. Anthropic may take the fight further. The company has also paid heavily over copyright: it agreed a $1.5 billion settlement with authors whose books were used to train its models.
Who will be in charge
Anthropic is a public benefit corporation, which means its directors must weigh a stated public mission alongside profits. A trust set up to protect that mission elects four of its seven directors. According to reporting by The Information, the seven co-founders are also seeking joint control of a special share carrying 50.1% of voting power on most matters. Buyers of the shares would own part of the company but would have little say over how it is run.
Why it matters if you don’t buy shares
For most readers the share price matters less than what the filing says about the tools they use at work:
- Prices may not stay low. A company spending far more on computing than it earns will be under pressure to raise prices or cut costly features once it answers to public shareholders.
- Your supplier’s biggest clients matter. When a quarter of revenue depends on two customers, business decisions can change quickly. Don’t build critical work on a single AI provider without a back-up.
- The warnings are now on the record. A company legally required to disclose material risks says its own technology could behave in ways its makers can’t fully control. That is a reason to keep a human check on anything AI does on your behalf.
We will update this piece when the final prospectus is published.
Sources
- Reuters via Yahoo Finance: Exclusive, Anthropic's IPO prospectus shows sweeping AI vision, surging costs (28 September 2026)
- TechCrunch: Anthropic's prospectus details losses, growth and a warning that its AI could end humanity (28 September 2026)
- Anthropic: Confidential draft registration statement submitted to the SEC (1 June 2026)
- TechCrunch: Anthropic's founders seek voting control ahead of IPO (25 September 2026)
- Akamai: Form 8-K exhibit on the Anthropic cloud agreement (24 September 2026)
- The Decoder: Anthropic signs $11.6 billion cloud deal with Akamai
- CNBC: Appeals court lets the Pentagon designate Anthropic a supply-chain risk (25 September 2026)
- CNBC: Anthropic warns of AI existential risks in IPO filing (29 September 2026)
Comments
What do you think? No account needed. Please be kind; comments that attack people are removed.
No comments yet. Be the first.