Anthropic pays – Anthropic agreed to pay SpaceX $1.25 billion per month for compute capacity through May 2029, tied to SpaceX’s Colossus data centers in Tennessee. The deal, detailed in SpaceX’s S-1 for an initial public offering, underscores how high demand for AI inference i
For SpaceX. the question is no longer whether it can sell compute for artificial intelligence—it’s how quickly the money will start flowing. In an agreement laid out in its initial public offering filing. SpaceX is set to receive $1.25 billion each month from Anthropic. with payments running through May 2029.
The deal isn’t just a headline number; it’s rooted in a specific operational arrangement. Earlier this month. Anthropic announced it would buy compute capacity from SpaceX’s Colossus data center in Tennessee to run tools for its growing customer base. SpaceX’s S-1, published Wednesday, says Anthropic will also get compute from its newer Colossus 2 data center.
Under the terms described in the filing, Anthropic pays a discounted rate for May and June. Either side can terminate the agreement with 90 days’ notice. If the contract remains in place, SpaceX would earn more than $40 billion from Anthropic through the life of the deal.
Anthropic’s spokesperson confirmed the $1.25 billion per month tally. SpaceX’s compute, the company says, will be used for inference—how AI models draw conclusions or create outputs—as Tom Brown, Anthropic’s compute chief, wrote earlier in May.
SpaceX’s filing frames the structure as a way to turn “unused compute capacity” into cash. The S-1 says the arrangement “allows us to monetize unused compute capacity in our infrastructure, while still permitting reallocation of the capacity for our own internal initiatives if needed in the future.”
That flexibility matters because the math behind selling AI compute has been brutal. SpaceX’s S-1 makes clear that it wasn’t cheap to build the chip supply required to sell spare capacity to Anthropic. The document describes how SpaceX is burning large sums on GPUs and cloud services.
AI losses from operations ballooned fourfold last year to more than $6 billion, driven by higher cloud costs and GPU depreciation. In the first quarter of this year, losses more than doubled to almost $2.5 billion.
Even hardware is part of the squeeze. The cost of producing chips is so large that the filing lists “manufacturing our own GPUs” among the “substantial capital expenditures” SpaceX plans. If SpaceX moves forward with chip production. it would place the company in direct competition with Nvidia. which dominates advanced GPUs.
The S-1 also depicts a second. more familiar tension: SpaceX is both dependent on and competing with major cloud and infrastructure players. SpaceX uses Google Cloud through its subsidiary, Starlink. The two divisions announced a partnership in 2021, when SpaceX installed Starlink ground stations at Google data centers.
But SpaceX has also become a competitor to Google Cloud by selling hundreds of megawatts of compute to Anthropic. The filing adds that this is not the only overlap. SpaceX’s ambitious plan to build data centers in space is something Google is in talks to assist with. the Wall Street Journal reported.
SpaceX’s broader ambitions extend well beyond a single customer contract. The S-1 lists “AI Infrastructure” among its markets and estimates that, based on demand for AI compute and current GPU rental rates, the market could be worth $2.4 trillion.
The filing also suggests that Anthropic is not meant to be a one-off. SpaceX expects to ink similar deals in the future—signaling that the compute-for-cash model, rather than remaining an experiment, is meant to become a core part of its financial runway.
In the meantime, the Anthropic agreement offers a rare kind of certainty in a business where costs have been spiraling and losses have been enormous—$1.25 billion a month, through May 2029, tied to the compute horsepower housed in SpaceX’s Tennessee data centers.
Anthropic SpaceX Colossus Colossus 2 AI compute inference GPUs data centers initial public offering S-1 Starlink Google Cloud Tom Brown Nvidia AI infrastructure Tennessee