Anthropic's Compute Commitments Hit $900.9 Billion

The amended SEC filing shows most of the spending goes to compute and infrastructure, with the SpaceX portion cancellable from early 2027.

Updated on Nov. 19, 2026 in Artificial Intelligence

Bold vector editorial illustration of a stack of steel cable spools beside a looping fiber-optic line against a brick wall.
Anthropic disclosed $900.9 billion in compute and infrastructure spending commitments in an amended SEC filing, with up to $120.5 billion allocated to SpaceX. AI Illustration. Upload story photo >

Anthropic's amended S-1 filing with the Securities and Exchange Commission discloses $900.9 billion in spending commitments, most of it for compute and infrastructure. Up to $120.5 billion of that is allocated to SpaceX's SpaceXAI subsidiary. The development is a disclosed contractual commitment, not a shipped product.

Why it matters

The figures show how much of frontier AI development now depends on long-term compute contracts with cloud and aerospace providers rather than owned infrastructure. The SpaceX deal's 90-day cancellation clause, which only activates after noncancellable periods end, makes it more flexible than the Microsoft, Google and Amazon agreements, which are noncancelable absent material breach.

The disclosed commitments are contractual obligations rather than measured performance results, so no benchmark or capability figures are attached to them. The filed numbers are the total spending commitments by provider, with the SpaceX agreement's $3.1 billion monthly payment conditional on capacity delivery.

The players

Anthropic

An AI developer building the Claude family of large language models, whose compute spending now totals $900.9 billion in commitments.

SpaceX

A rocket and satellite company whose SpaceXAI subsidiary is set to receive up to $120.5 billion of Anthropic's compute commitments.

The details

Anthropic's commitments are largely for compute, meaning the processing hardware and data-center capacity used to train and run large AI models. The SpaceX agreement runs through 2029 but has noncancellable periods ending in January and March 2027. After those periods, either party can exit with 90 days' written notice, so Anthropic could send notice as early as February 1 or April 1 of 2027. The Microsoft, Google and Amazon agreements, by contrast, cannot be cancelled absent a material breach.

Timeline

  1. December 2026: Anthropic expects capacity delivery and payments to SpaceX to begin.

  2. January 2027: The first noncancellable period of the SpaceX agreement ends.

  3. February 1, 2027: Earliest date Anthropic could send a cancellation notice to SpaceX.

  4. March 2027: The second noncancellable period of the SpaceX agreement ends.

  5. April 1, 2027: Second date Anthropic could send a cancellation notice to SpaceX.

The Tech Race

Anthropic's commitments with Google ($121.3 billion), SpaceX (up to $120.5 billion), Amazon (at least $110 billion) and Microsoft (up to $49.2 billion) place it among the largest long-term compute buyers disclosed in a public filing. The SpaceX agreement stands out because it is the only one of the four with a stated no-fault cancellation right after a noncancellable window.

The commitments affect the capacity available to Anthropic's models, which shapes the compute behind the products people use. The filing does not indicate any change in pricing or availability for end users.

The takeaway

The scale of these commitments signals that frontier AI capability is now tied to multi-year infrastructure contracts measured in the hundreds of billions. Watch whether Anthropic gives 90-day notice to SpaceX on or after February 1, 2027, the first date the SpaceX agreement can be cancelled.

What happens next

Anthropic expects SpaceX capacity delivery and payments to begin in December 2026, and the SpaceX agreement's noncancellable periods end in January and March 2027, after which either party may give 90 days' notice.

Further reading

For background on the infrastructure these deals fund, see the Artificial Intelligence section.