Anthropic's S-1 filing shows compute commitments of up to $120.5 billion with SpaceXAI, which already accounts for a third of SpaceX revenue
The company best known for landing rockets on barges now has a new top client, and it isn’t a government space agency. Anthropic, the AI lab behind the Claude models, is SpaceX’s largest single customer.
The relationship came into focus on October 26, 2026, when Anthropic disclosed its compute deals with SpaceXAI in its S-1 filing. The numbers are large enough to make a satellite constellation look like a modest line item.
What the filing shows
An S-1 is the registration document a company files with US securities regulators before going public. It is also where companies lay out their biggest financial commitments, whether they enjoy the attention or not.
Anthropic’s filing describes multiple agreements with SpaceXAI, a SpaceX subsidiary. The deals give Anthropic access to computing power running on NVIDIA GPUs.
The contracts were finalized between May and October 2026. Combined, they commit Anthropic to spend up to $120.5 billion with SpaceXAI through 2029.
At full capacity, the spending works out to approximately $3.1 billion per month. Anthropic expects to reach that full run rate by December 2026.
A third of SpaceX’s quarter
The impact is already visible on SpaceX’s books. The company’s Q3 results showed total revenue of $14.58 billion.
Approximately $5 billion of that came from Anthropic. That works out to 34.5% of the quarter’s revenue from one customer.
The partnership started in May 2026. Anthropic first tapped SpaceX’s Colossus 1 supercomputer, then expanded onto Colossus II as its needs grew.
The fine print on cancellation
The headline figure comes with an important qualifier. The $120.5 billion is a maximum, not a guaranteed sum.
Both Anthropic and SpaceXAI hold cancellation rights starting after early 2027. Either side can walk away by giving 90 days of written notice.
That structure stands apart from Anthropic’s arrangements with Google and Amazon, which the filing characterizes as more binding. The SpaceXAI deals, by comparison, look more like a flexible lease than a long-term mortgage.
Background: compute as the new oil field
Anthropic has historically leaned on hyperscale cloud partners for that capacity, notably Google and Amazon. The SpaceXAI deals show the company diversifying its suppliers rather than relying on a small set of providers.
What this means
For SpaceX, the Anthropic relationship is both a windfall and a concentration risk. Having 34.5% of quarterly revenue tied to one customer can supercharge growth, but it also makes results sensitive to that customer’s decisions.
The cancellation terms sharpen that point. Once early 2027 passes, Anthropic could shift workloads elsewhere with 90 days’ notice.
For Anthropic, the filing shows how capital-intensive frontier AI has become. A commitment of up to $120.5 billion through 2029 places compute among the company’s largest obligations, and prospective public-market investors will weigh it against revenue growth.
The contrast with the Google and Amazon deals is also telling. Anthropic appears to be keeping its tighter commitments with established cloud partners while using SpaceXAI as a large but more adjustable source of capacity.
The key things to watch are whether Anthropic hits its full run rate by December 2026, how SpaceX’s revenue mix shifts in coming quarters, and whether either party signals intent around the cancellation window in early 2027.
