Anthropic Pays SpaceX $1.25B Per Month for Colossus Compute Access

The compute deal between Anthropic and xAI, which shocked the AI world earlier this month, now has a number, and it is amazing. Anthropic will pay xAI $1.25 billion a month until May 2029. This will guarantee that the Colossus 1 data center near Memphis, Tennessee, will produce nothing. With a lower rate for the first two months while xAI finishes ramping up, the deal could bring in more than $40 billion for the company over the course of the contract.

Money details were not given out when the deal was revealed. They came out this week in SpaceX's S-1 filing with the SEC, which is the company's official IPO registration document. This is one of the most important business news in a selling process that is already being closely watched.

With an annual payment of $15 billion, the amount is close to SpaceX's total reported yearly revenue of $18 billion from a single counterparty relationship. The scale changes the SpaceX-xAI-Anthropic triangle from a partnership in computing to a business relationship that all three companies depend on.

Why Anthropic Is Paying This Much — and Why xAI Needed the Deal

The deal shows the main imbalance in the current AI landscape: early-stage AI companies are limited by their ability to generate income before they are limited by their ability to generate compute. Anthropic is said to be making close to $45 billion a year and growing quickly, but it can't meet that demand because of the lack of real computing power. With a monthly payment of $1.25 billion, the Colossus deal is less about cost and more about getting more space. Anthropic is buying the right to keep growing at the rate it is now.

Tom Brown, co-founder and chief compute officer of Anthropic, said right before the S-1 went public that the company is already growing: "We're expanding our partnership with SpaceX and will be scaling up on Nvidia GB200 capacity in Colossus 2 throughout June." The deal has grown bigger than it was at first, and the paper hasn't even dried yet.

The way SpaceX describes the deal in its report is helpful. That the company says it lets xAI "monetise unused compute capacity in our infrastructure" is an important admission. Grok, xAI's most popular AI assistant, hasn't been used as much in recent months. This has freed up servers that the company is now renting to one of its main AI rivals. The Colossus facility was built to a size that xAI's needs have not yet met. The Anthropic deal turns that extra space into a source of income before the company's projected IPO valuation of several hundred billion dollars.

SpaceX is very clear about what it wants to do: "We expect to enter into additional similar services contracts." "Enough capacity to provide compute for our own AI models, including support of training and inference demands, and to satisfy the obligations under these agreements," the company says. This means that Colossus's capacity goes beyond what Anthropic has contracted.

The "Neocloud" Model Taking Shape

The deal is an example of a new way of doing business that most AI infrastructure players have not yet accepted. Most businesses build data centers either for their own use or to rent out to other businesses. One thing that xAI is doing at the same time is working as an AI developer for Grok and a cloud computing service for Anthropic.

This is what SpaceX calls a "dual monetisation strategy" that "provides multiple pathways to generate returns on invested capital." The way the question is framed makes sense from a strategic point of view: infrastructure that was built for peak demand in AI training sits empty during times of lower demand and can be rented during those times so that it doesn't lose value. The risk is what xAI's filing suggests: that Grok demand was higher than expected, which means that the Anthropic income was not only a good idea, but also necessary to run the business before the IPO.

Each side has 90 days to back out of the deal, which is the main business risk that comes with it. Both Anthropic and SpaceX could leave with 90 days' notice. Institutional investors will pay close attention to the fact that there isn't a locked, non-cancellable contract for a company that is about to go public and has a $1.25 billion monthly payment as a key income line.

What the S-1 Disclosure Means for the IPO

The Anthropic compute contract is one of the most important business details in SpaceX's S-1 filing, and it affects possible IPO investors in two ways. This is good news because it shows that SpaceX's infrastructure has a huge paying customer: one of the world's best-funded AI companies has agreed to pay $40 billion over the next 2029. Carefully, it shows that xAI built more compute than it needed, and the Anthropic deal is partly meeting a need rather than just making money off of extra margin.

The IPO wants to be worth at least $1.75 trillion, and a listing on Nasdaq on June 12 under the ticker SPCX is said to be a goal. At that price, the Anthropic deal adds about 2.3% to the company's implied market cap in annual compute income. This is a big deal, but it's not the main reason for the valuation, since Starlink is still the most important factor.

The $1.25 billion a month amount sets a new standard for how much AI work costs at large scale. For Anthropic, it's the cost of keeping up the current rate of revenue. This is a cost that can be covered by the company's $45 billion annualised revenue run rate, which is another reason why it's thinking about a $50 billion fundraising round at a $900 billion value. The Anthropic deal turns SpaceX's idle computing power into a source of short-term income for the IPO roadshow.

For the bigger AI infrastructure market, this deal's suggested price per compute unit will be used as a standard for all future large-scale compute talks between cutting edge AI labs and infrastructure providers.