Anthropic is writing SpaceX a very large check: $1.25 billion a month through May 2029 for compute, according to SpaceX’s IPO filing. The deal is a reminder that frontier AI is now as much about warehouse-scale hardware as model quality, and that the companies with spare capacity can turn themselves into very expensive landlords.
The numbers are awkwardly beautiful. Anthropic says demand is running ahead of supply, while SpaceX gets a fresh revenue stream that is hard to ignore next to its roughly $18 billion in annual revenue. In other words: one company desperately needs more GPUs, and the other can rent them out while it waits for its rockets to do the heavy lifting.
What Anthropic is buying from SpaceX
The agreement covers massive compute capacity and appears to be tied to SpaceX’s Colossus 1 site, with Anthropic also expanding to Colossus 2. Tom Brown, Anthropic’s co-founder and chief compute officer, said the company would scale up on Nvidia GB200 capacity there through June.
SpaceX also signaled it may not stop with Anthropic. In its filing, the company said it expects to sign ”additional similar services contracts,” which is a polite way of saying it sees demand for selling spare AI infrastructure. The pitch is straightforward: if you have the power, cooling, and chips, someone will rent them.
- Payment: $1.25 billion per month
- Term: through May 2029
- Ramp-up: payments reduced for May and June
- Exit clause: either side can walk away with 90 days’ notice
Why SpaceX wants to rent out AI capacity
This kind of deal is becoming the new side hustle for infrastructure-heavy tech companies. Nvidia gets to sell the chips, data-center operators get to monetize the floor space, and AI labs get to keep training models without waiting in line for scarce hardware. The scramble has been public for a while; what changes here is the scale, which is now large enough to matter on a corporate filing.
There is, of course, a catch. A 90-day exit clause is not the language of a marriage, even by Silicon Valley standards. It gives both companies flexibility if demand changes, chip supply improves, or the economics of renting compute suddenly look less heroic and more expensive.
The AI compute race keeps getting pricier
Anthropic’s appetite for compute fits the broader pattern across AI: more users, larger models, and a constant shortage of the hardware needed to serve both training and inference. The companies with the deepest pockets and the best access to GPUs are the ones setting the pace, which is why a lease agreement can now read like a strategic weapon.
Expect more deals like this. The next question is whether this is a one-off workaround for Anthropic’s capacity crunch or the start of a much more formal market for premium AI compute, where spare clusters are treated less like infrastructure and more like a tradable commodity.