Anthropic's IPO filing shows $895 billion in compute commitments as revenue growth cools

Anthropic's revised S-1 filing puts total compute commitments near $895 billion, pushing past $1 trillion with additional needed capacity, while Q3 revenue growth slowed to 46 percent. The filing warns the 1,206 percent year-over-year growth rate it posted through September won't repeat.

Status

  1. Announced
  2. Demonstrated
  3. Available
  4. Independently confirmed

Sources

Anthropic's revised S-1 filing, submitted Monday ahead of its planned IPO, puts the company's compute and infrastructure commitments at close to $895 billion, up from the $800 billion figure reported last week. Add the $150 billion to $175 billion in additional compute the company says it still needs to secure, and the total tab climbs past $1 trillion.

The filing lays out a payment schedule: Anthropic expects to pay nearly $70 billion against these obligations by September 2027, with the cumulative total reaching about $447.5 billion by the end of 2030. That schedule assumes revenue keeps climbing. The company's own prospectus says that's not a safe bet.

"In the nine months ended September 30, 2026, our revenue was $33.2 billion, representing a year-over-year growth rate of 1,206%," the filing states. "We do not expect our revenue will continue to grow at comparable rates, and we cannot assure you that it will continue to grow at all."

The numbers inside the filing back up that caution. Revenue grew nearly 2.5x between the first and second quarters of 2026. In the third quarter, Anthropic reported close to $17 billion in revenue, a 46 percent increase over Q2. Compute and infrastructure costs grew 77 percent over the same stretch, and the company kept signing contracts that lock in more spending ahead.

That math matters because of how the contracts are structured. Of the $895 billion in commitments, Anthropic says only about $213 billion can be avoided through termination rights or similar conditions. The remaining $682 billion binds the company regardless of whether revenue keeps pace.

Roughly $320 billion of the total reflects existing contractual obligations already in force. Another $451 billion covers leases Anthropic has signed but that haven't started yet, with the company expecting them to begin sometime between 2026 and 2028, some still subject to termination rights or other conditions.

The filing names the counterparties behind the biggest piece of that exposure: Microsoft, Google, Amazon, and Broadcom, whose combined deals with Anthropic now total $437 billion. Those agreements are non-cancellable. Under some of them, Anthropic owes payment whether or not it ever activates a single GPU.

The same three cloud providers also sit on the revenue side of Anthropic's business. Cloud marketplaces run by Amazon, Google, and Microsoft accounted for about 44 percent of Anthropic's revenue in the nine months ended September 30, according to the filing. Anthropic states that no single customer accounts for more than 10 percent of revenue, so that cloud-marketplace concentration sits across multiple accounts rather than one buyer.

That dependence creates a structural tension the filing doesn't fully resolve. Amazon, Google, and Microsoft are Anthropic's largest compute suppliers and its largest distribution channels, and each is building or hosting competing AI products at the same time. Anthropic needs their infrastructure to run its models and needs their marketplaces to sell access to those same models, while those companies pursue their own AI lines.

Anthropic frames the spending as the price of a much larger bet. The prospectus argues that AI will reshape the economy "in ways greater than the impact of industrialization, electricity, semiconductors, and the internet." Whether that bet pays off now runs through a set of non-cancellable contracts worth $437 billion, payable on a schedule the company's own revenue growth may not keep up with.

Daniel Koo

Products & Policy Desk

Splits time between what a product actually does in a working session and what a policy or corporate announcement actually obligates. A feature exists when it runs outside a demo; a rule takes effect on the date the text says it does, not the date of the press release.

Keep reading

  1. Amazon's AI memo tells engineers non-use is fine. Its performance process says otherwise.
  2. Behind Apple's new CEO disavowing "AI for AI's sake": a Google licensing deal, not a slogan
  3. Want to build a UK datacenter? Please hold while Whitehall transfers you
  4. Pennsylvania's new AI ad law reaches from Super Bowl spots to wing-special flyers
stagirus