Anthropic's IPO prospectus provides the most detailed public look yet at the company's business, finances and risks. The filing shows extraordinary growth and high revenue per employee, while also revealing hundreds of billions of dollars in compute commitments and concentrated revenue sources.
What the filing discloses
- The filing states Anthropic has nearly $650 billion in compute commitments (other parts of the filing cite about $637 billion), including non-cancellable minimum payments.
- Anthropic reports nearly 2,000 customers generating at least $1 million in annual revenue each, and 250 customers producing $10 million or more each year.
- In the first nine months of 2026, the company’s five largest customers accounted for roughly 25% of revenue.
- Nearly 50% of first-half revenue flowed through cloud marketplaces run by Amazon, Google and Microsoft — firms that are simultaneously investors, compute providers and competitors.
- The filing says 5% to 7% of revenue in the first nine months of 2026 came from customers using Anthropic models in ways that violated policies or terms of service; some outside experts attribute much of this to model distillation, where models are queried extensively for training purposes.
Financial position and efficiency
- The company reported substantial cash and short-term investments: earlier notes said more than $130 billion on hand as of June 30, while other parts of the filing list over $134 billion in cash and similar securities.
- Revenue per employee was $6.6 million at the end of the third quarter, a figure only slightly below that of Nvidia, a frequently cited efficiency benchmark in tech.
- Observers noted that relatively few costs appear in Anthropic’s reported cost of goods sold, which has the effect of inflating reported margins compared with the company's full operating costs.
Business-model and customer risks
- Many customers pay based on usage and do not make multi-year commitments, while Anthropic’s compute obligations extend for years and include minimum payments regardless of actual capacity usage — a mismatch that can create financial strain if customer spending does not keep pace.
- Heavy reliance on cloud marketplaces benefits hyperscalers in multiple ways: they collect compute fees and platform or marketplace fees, while also positioning themselves as competitors to Anthropic.
- Customer concentration — notably the outsized share from the top five customers — raises the risk of rapid revenue swings if one or more large customers reduce spending.
Notable deals, governance and executive pay
- The filing describes an expanding commercial relationship with Elon Musk: Anthropic’s payments to SpaceX are now presented as potentially reaching $3.1 billion per month and up to $120.5 billion in compute through 2029, up from $1.25 billion per month cited in SpaceX’s IPO filing. The agreement can reportedly be canceled by either party with 90 days’ notice.
- Anthropic’s filing shows large cash reserves while prompting some observers to question whether a substantial portion of bills has been covered via convertible notes, given that listed cash roughly matches the total capital raised.
- CEO Dario Amodei’s total compensation in 2025 was nearly $18 million, and the filing discloses millions spent on personal security for the company’s two top executives.
- The filing also details an unusual succession arrangement in which the founders can nominate successors for their voting power, subject to approval by the other co‑founders.
The bottom line for the IPO
The prospectus demonstrates that demand for powerful AI products can expand at a staggering pace and that Anthropic has converted that demand into strong revenue metrics and a sizeable paying customer base. At the same time, the IPO will be a test of whether investors believe that customer spending can remain high enough, and last long enough, to justify the enormous infrastructure and long-term compute commitments the company has entered into. Investing in Anthropic therefore represents a bet both on AI’s large promise and on sustained customer willingness to fund the costly compute that underpins these services.
