Three months ago, Elon Musk wrote on X that Anthropic was “evil,” “misanthropic,” and that the AI lab hated Western civilization. On Wednesday, he leased Anthropic one of his most valuable assets: the world’s biggest supercomputer.
But Anthropic-lovers shouldn’t bask too long in Musk’s newfound praise (even if he did decide that “nobody set off my evil detector” ). The deal has little to do with them as a company, analysts told Fortune, and everything to do with an upcoming prospectus.
SpaceX is expected to begin its public roadshow next month, with a confidential S-1 filed April 1 targeting a valuation between $1.75 trillion and $2 trillion. Wednesday’s announcement—paired with Musk’s dissolution of his AI company xAI into SpaceX (to make SpaceXAi)—gives the IPO something it didn’t have a week ago: a marquee AI customer for a credible cloud-infrastructure business.
According to estimates from Antoine Chkaiban, an analyst at New Street Research, the Anthropic deal will generate $3 billion to $4 billion in annual revenue for SpaceX, with more than $2.5 billion in cash profit. The margins seem extreme, but that’s because the data center is already built: the fixed capital expense is sunk, and the only meaningful operating cost is electricity plus the relatively minimal costs of staffing the place.
“He’s not going to want multiple billions of dollars of GPUs sitting idle,” Chkaiban told Fortune. “It’s a very good business decision.”
And, it seems, the start of Elon Musk’s transition from seeking to be a frontrunner in the model race, to being the landlord of AI.
“He who controls the data center, really does control the application of artificial intelligence right now,” Andrew Moore, the former head of Google Cloud AI and now CEO of defense AI startup Lovelace AI, told Fortune. “So, yeah, I think both sides of this wedding of convenience will be a little stressed out by it.”
The hyperscaler pivot
Colossus 1 contains roughly 220,000 Nvidia GPUs and was built in 2024 to train Grok, Musk’s AI assistant. But Grok hasn’t filled it. Chkaiban estimates Grok generates less than $1 billion in annualized revenue; Anthropic is on track for more than $40 billion. The disparity is the deal. Musk has too much compute and Grok–despite endless “ask Grok” inquiries on X–can’t fill it; Anthropic has too many users and not enough compute. Leasing Colossus 1 to Anthropic funds the gap.
But it also lets Musk skip a step. The biggest cost line for any of the frontier AI labs is the 30%-plus margin paid to AWS, Microsoft Azure, or Google Cloud for compute. SpaceX captures those margins of the hyperscalers instead of paying it in stressful debt deals, like the AI labs.
That framing—SpaceX as the fourth hyperscaler—is what Musk needs investors to accept before pricing, analysts told Fortune. A SpaceX that can compete with AWS is worth a hyperscaler multiple, not a rocket company multiple. These days, Alphabet, Microsoft, and Amazon trade at roughly twice the forward earnings multiple of Boeing and Lockheed Martin.
But Moore was skeptical that the pivot is easy. Big enterprise customers like governments, or Fortune 500 companies pick where to store data centers mostly based on location; if the cost of energy is cheap, if they have failsafes if something goes wrong. Building one massive data center in Memphis doesn’t replicate AWS’s global and legal footprint. “The battle is not just who’s got the most compute servers,” he said.
“I would never bet against Elon doing something amazing,” Moore added, “but he’s got his work cut out to really take on AWS.”
The kill-switch clause
Whether or not Musk wins that battle, he already has something other compute providers don’t. In a reply on X, he wrote that SpaceX “reserves the right to reclaim the compute” if Anthropic’s AI “engages in actions that harm humanity.” The clause was not in the formal press release, and it’s unclear whether it appears in the contract. But if enforceable, it gives Musk a powerful leash on one of the three leading AI labs in existence, while he sues OpenAI’s leadership in federal court.
That’s a lot of power Musk has now that he didn’t two weeks ago. And it would matter less if Musk didn’t change his mind so much on AI.
Moore, who was dean of computer science at Carnegie Mellon during Musk’s loudest existential-risk phase, remembers him as “one of the loud voices saying that artificial intelligence is an existential threat for the human race.” Now he says AI will usher in a world of abundance.
Anthropic almost certainly has fallback plans. Frontier AI labs are not in the habit of single-sourcing the data center their entire product depends on, and Moore said the company will be working aggressively on compute efficiency in the background. “They will have contingency plans in three months, six months, twelve months,” he said.
Still, neither side gets to walk away clean. Gene Munster, managing partner at Deepwater Asset Management, put the odds the deal still exists in two years at 80%. The other 20% is a bet on Musk himself. “What makes it unique is Elon’s history,” Munster said. “He can change his mind. It’s less about the actual provision; it’s more just about who’s running the provision.”
The deal’s odds aren’t in question. Even if Munster is right and the contract holds for two years, one of three frontier AI labs in the world now operates on infrastructure controlled by the CEO of a competitor.
“The stakes are enormous,” Moore said. “Everyone is trying to get through the next six months. They’ll do whatever it takes.”
This story was originally featured on Fortune.com
Leave a Reply