Three months ago, Elon Musk wrote on X that humans are “evil” and “world-weary” and that artificial intelligence laboratories hate Western civilization. On Wednesday, he leased one of his most valuable assets to Anthropic: the world’s largest supercomputer.
But lovers of humanity shouldn’t bask in Musk’s newfound praise for too long (even if he does decide that “no one sets off my evil detector”). Analysts say the deal has little to do with them as a company wealth, and everything related to the forthcoming prospectus.
SpaceX is expected to begin a public roadshow next month and file a confidential S-1 on April 1, targeting a valuation of between $1.75 trillion and $2 trillion. Wednesday’s announcement — coupled with Musk’s dissolution of his artificial intelligence company xAI into SpaceX (creating SpaceXAi) — gives the IPO something it didn’t have a week ago: a big-name AI customer with a reliable cloud infrastructure business.
New Street Research analyst Antoine Chkaiban estimates that the Anthropic deal will bring SpaceX $3 billion to $4 billion in annual revenue, including more than $2.5 billion in cash profits. Margins may seem high, but that’s because the data center is already built: fixed capital expenditures are sunk, and the only meaningful operating costs are power plus relatively minimal staffing costs.
“He wouldn’t want billions of dollars worth of GPUs sitting idle,” Chkaiban said wealth. “This is a very good business decision.”
And Elon Musk appears to be starting to shift from seeking to be the front-runner in the modeling competition to becoming the front-runner in artificial intelligence.
“Whoever controls the data center really controls the application of AI,” said Andrew Moore, former head of Google Cloud AI and now CEO of defense AI startup Lovelace AI. wealth. “So, yes, I think both parties in this wedding of convenience are going to feel a little bit stressed by that.”
Hyperscale Hub
Colossus 1 contains approximately 220,000 Nvidia GPUs and was built in 2024 to train Musk’s artificial intelligence assistant Grok. But Grok hasn’t filled it yet. Chkaiban estimates Grok’s annual revenue is less than $1 billion; Anthropic is expected to be valued at more than $40 billion. The difference is the deal. Musk has too much computing power and Grok (despite endless “ask Grok” queries for X) can’t fill it; Anthropic has too many users and not enough computing power. The lease of Colossus 1 to Anthropic bridged the gap.
But it also allowed Musk to skip a step. The biggest cost line for any cutting-edge AI lab is the 30%-plus profit paid to AWS, Microsoft Azure, or Google Cloud for computation. SpaceX captured the profits of hyperscalers rather than paying them out through stressed debt deals like the AI lab did.
Analysts say Musk needs investors to buy into the framework before pricing SpaceX as fourth hyperscaler wealth. SpaceX, which can compete with AWS, is worth multiples of hyperscalers, not rocket companies. Today, Alphabet, Microsoft and Amazon trade at about twice the forward price-to-earnings multiples of Boeing and Lockheed Martin.
But Moore is skeptical that the transition will be easy. Large corporate customers like governments, or wealth 500 companies choose their data center storage locations primarily based on location; if energy costs are cheap, they have fail-safe coverage if something goes wrong. Building a massive data center in Memphis doesn’t replicate AWS’ global and legal footprint. “The battle is not just about who has the most compute servers,” he said.
“I would never bet on Elon doing something amazing,” Moore added, “but he’s ready to take on AWS.”
Kill switch clause
Whether or not Musk wins this battle, he already has something that no other computing provider has. In X’s response, he wrote that SpaceX “reserves the right to take back the computing” if Anthropic’s AI “engages in conduct that is harmful to humanity.” The clause did not appear in the official press release and it is unclear whether it appears in the contract. But if enforceable, Musk would be able to have strong control over one of the three leading existing artificial intelligence labs, and he could also sue OpenAI’s leadership in federal court.
Musk now has power that he didn’t have two weeks ago. It wouldn’t matter so much if Musk didn’t change much on the subject of artificial intelligence.
Moore, who chaired Carnegie Mellon University’s computer science department during the height of Musk’s existential risk, remembers him as “one of the people who spoke out loudly about the existential threat that artificial intelligence posed to humanity.” Now he says artificial intelligence will usher in a rich world.
Humans almost certainly have a backup plan. Frontier AI Labs isn’t used to single-sourcing the data centers its entire product relies on, and Moore said the company will aggressively improve computing efficiency in the background. “They will have contingency plans in place in three months, six months, 12 months,” he said.
Despite this, neither side was able to escape unscathed. Gene Munster, managing partner at Deepwater Asset Management, said there was an 80% chance the deal would still be alive in two years. The other 20% is Musk’s own bet. “What’s unique about it is the history of Elon,” Munster said. “He can change his mind. It’s not about the actual regulations; it’s more about who is administering the regulations.”
There is no question about the possibility of this deal. Even if Munster is right and the contract is valid for two years, one of three cutting-edge artificial intelligence labs in the world is now running on infrastructure controlled by a rival CEO.
“The stakes are huge,” Moore said. “Everyone is trying to get through the next six months. They’re going to do whatever it takes.”
This story originally appeared on Fortune.com
