Anthropic, the company behind the Claude chatbot, has disclosed that it could spend up to $84.5 billion through 2029 renting computers from SpaceX, the rocket maker that now also runs artificial intelligence. Reuters reported the number on September 29 from Anthropic's confidential paperwork for a stock listing, and it is close to double the roughly $45 billion SpaceX had disclosed in May. The higher figure is a ceiling on a deal that already existed, not a new payment, and most of the agreements behind it can be ended with 90 days' notice. The disclosure puts a number on an unusual business for a rocket company: renting out the machines that run AI.
A rocket company in the computer rental business
SpaceX got into this business by buying xAI, the developer of the Grok chatbot, in February 2026. The purchase gave SpaceX two things at once: its own AI models to run, and computing capacity to lease to outside customers when it was not using all of it.
The site at the centre of the Anthropic deal is Colossus 1, a data center in Memphis, Tennessee. A data center is a building filled with the specialized chips, called GPUs, that train and run AI models, along with the power and cooling they need. Anthropic said in May that it had signed an agreement to use all of the computing capacity at Colossus 1, which it described as more than 300 megawatts of capacity and over 220,000 Nvidia GPUs.
Those two numbers describe the facility differently. The GPU count is the number of chips. The megawatt figure is the electric power the site can draw, a very large load. Renting this capacity by the month is what the SpaceX agreements do. It works like renting servers from a cloud provider, except the machinery here is built for AI.
Why $84.5 billion is a maximum, not a bill
The figure needs care because it is written as "up to." SpaceX's own listing paperwork in May 2026 set out the earlier version of the arrangement: Anthropic agreed to pay $1.25 billion a month through May 2029, about $15 billion a year, or roughly $45 billion over the term. Either side could end the agreement on 90 days' notice.
The prospectus Reuters reported on September 29 raises the possible total to $84.5 billion, about 88% above the earlier estimate, but it keeps the same shape. Most of the agreements can still be cancelled with 90 days' notice. So the number is the most SpaceX could earn, and the most Anthropic could spend, if the relationship ran its full course at its highest level. It is not revenue either company can count on. A contract that can be ended in a quarter is a different kind of asset from one locked in for years.
That distinction matters more here than it usually would, because Anthropic's other computing commitments are mostly rigid. In the same prospectus, Anthropic said it expects to spend at least $518 billion over the next decade across six partners, and about 80% of that is non-cancelable or must be paid regardless of how much computing it actually uses. By contrast, most of the SpaceX agreements can be cancelled.
What SpaceX has actually booked
Booked revenue is a smaller and firmer number. In its second-quarter 2026 results, SpaceX reported $2.6 billion of revenue in its AI segment, up 247% from a year earlier. SpaceX said it ended the quarter with 1.4 gigawatts of computing capacity, up from one gigawatt in the first quarter and 400 megawatts a year earlier. The two numbers measure different things: $2.6 billion is one quarter of money already recorded across SpaceX's whole AI segment, while $84.5 billion is a multi-year ceiling on contracts that can be cancelled.
The pattern beyond one deal
Anthropic is not SpaceX's only tenant. Google has reportedly agreed to pay $920 million a month to rent about 110,000 Nvidia GPUs from October 2026 to June 2029. That figure comes from secondary reporting rather than the contract itself, so it is best treated as reported rather than confirmed. What it shows, alongside the Anthropic agreements, is the shape of a business: AI companies that need far more computing than they can build quickly are renting it from a company that built it for another purpose.
The rivalry is worth noting too. SpaceX owns xAI, the maker of Grok, a direct competitor to Claude, and it supplies Anthropic with the capacity to run Claude.
What changes for people who use Claude
For Claude's users, the effect is concrete. When Anthropic announced the SpaceX agreement in May, it said the deal, together with its other recent computing deals, let it raise usage limits for Claude Code, its tool for writing software, and for its developer service, the Claude API. The gain came from all the combined deals, not from SpaceX alone.
The estimate, and what comes next
Bank analysts are modelling the growth. TD Cowen began covering SpaceX with a Buy rating and a $200 price target, and projected that its computing capacity would reach 6 gigawatts by the end of 2027. That is an outside estimate, not a company figure.
The disclosure is also a step in Anthropic's path to a public listing. Anthropic filed a draft registration statement with the US Securities and Exchange Commission in June. Whether Anthropic proves to be a lasting customer for SpaceX will depend on those 90-day cancellation terms.