SpaceX's major AI compute deal with Anthropic could end a lot sooner than expected

By William Gavin

SpaceX CEO Elon Musk said his firm requested a 'short term' deal, despite touting plans that it could last for years

Elon Musk's SpaceX is gearing up for a massive initial public offering.

SpaceX's deal to rent out space in its data centers to Anthropic has been a bright spot, but it might be a fleeting one.

Anthropic is set to pay $1.25 billion per month through May 2029 to access artificial-intelligence computing capacity in SpaceX's Colossus I and Colossus II data centers in Tennessee. The deal could be terminated by either company as long as they gave 90 days' notice, according to SpaceX's initial public offering paperwork.

Assuming that neither company nixed the agreement, Anthropic could pay SpaceX up to $45 billion over three years. SpaceX isn't yet willing to commit to keeping the deal in place for that long, according to CEO Elon Musk.

"SpaceX has not committed to leasing Colossus for years, although it's possible that may be what happens," Musk said in a late-night post on X, which is owned by SpaceX.

"This is a 180-day lease with 90-day notice mutual cancellation thereafter," he added. "The short term was our request, not Anthropic's."

Read more: Dan Loeb says tech is the most attractive sector right now - unless you hold this 'draconian' view

While a financial win for SpaceX, the Anthropic deal was also viewed as a sign that the Grok model was lagging. Anthropic's use of SpaceX's capacity is set to ramp up in May and June at a "reduced fee," SpaceX said in its filing, adding that it has "sufficient capacity" for its own AI models.

The deal showed that Anthropic doesn't see SpaceX as much of a threat, PitchBook analyst Franco Granda wrote in a recent report. It revealed that SpaceX, which is also building a $20 billion data center in Mississippi, has "meaningfully underused" its capacity, he added.

In the IPO paperwork, SpaceX said that the Anthropic deal "highlights the increasing importance" of AI infrastructure and positions it to provide compute capacity. Last week, Musk said SpaceX was in discussions with companies other than Anthropic to offer access to AI compute.

For comparison, Meta Platforms (META) plans to spend as much as $145 billion on capital expenditures in 2026, mostly for AI-related plans. The company hasn't ruled out selling compute if it winds up with excess capacity, but it's not currently planned, according to CEO Mark Zuckerberg.

"Almost every week, there are different companies that come to us ...masking if we have compute that they could buy from us at some premium," Zuckerberg said on Wednesday at Meta's annual investor meeting. "We haven't done that yet because we think that we have a use for the compute."

SpaceX, which filed for an initial public offering last week, has had difficulty making its Grok popular. Enterprise adoption has been slow, according to Ramp data, and just 21% of SpaceX's monthly active users on the Grok and X.com platforms have used AI features as of March 31.

The company's AI division reported a $2.5 billion operational loss in the March quarter against revenue of $818 million, according to the company's IPO filing. SpaceX expects AI to eventually be its biggest business and account for the majority of its projected $28.5 trillion total addressable market.

See more: Space stocks are rocketing as SpaceX keeps making pre-IPO headlines with first Starship launch

On Thursday, Musk also said that SpaceX was nearly finished developing an in-house AI training stack that should help it train its models faster. He said earlier this week that Grok's next foundation model should be released sometime in June.

-William Gavin

This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

05-28-26 1157ET