Three months ago, Elon Musk wrote in X that humans are “evil” and “misanthropic” and that the AI Institute hates Western civilization. On Wednesday, he leased one of his most valuable assets, the world’s largest supercomputer, to Anthropic.
But humanity lovers shouldn’t soak in Mr. Musk’s newfound praise for too long (even if Mr. Musk decides that “nobody set off my evil detector”). Analysts said the deal had little to do with them as a company. luck, and everything related to the upcoming prospectus.
SpaceX plans to open to the public next month, and a confidential S-1 filed on April 1 targets a valuation of $1.75 trillion to $2 trillion. Wednesday’s announcement, along with the dissolution of Musk’s AI company xAI into SpaceX (which became SpaceXAi), gave the IPO something it didn’t have a week ago. It was a leading AI customer in the trusted cloud infrastructure business.
New Street Research analyst Antoine Chikaiban estimates that the deal with Anthropic will generate annual revenue of $3 billion to $4 billion for SpaceX and more than $2.5 billion in cash. The profit margins seem extremely high, but that’s because the data centers have already been built. Fixed capital expenditures are buried, and the only meaningful operating costs are electricity, plus relatively minimal labor costs.
“He doesn’t want billions of dollars of GPUs sitting idle,” Chkaiban said. luck. “This is a very good business decision.”
And it appears to be the beginning of Elon Musk’s transition from aiming to be at the forefront of the modeling race to becoming the mastermind of AI.
“Whoever controls the data center is now actually controlling the application of artificial intelligence,” said Andrew Moore, former head of Google Cloud AI and current CEO of defense AI startup Lovelace AI. luck. “Well, I think it’s going to be a little stressful for both parties in this marriage of convenience.”
Hyperscaler pivot
Colossus 1 has around 220,000 Nvidia GPUs and was built in 2024 to train Grok, Musk’s AI assistant. But Grok doesn’t meet that. Chkaiban estimates Grok’s annual revenue to be less than $1 billion. Anthropic is on track to earn more than $40 billion. Disparity is a deal. Musk has so much computing power that Grok can’t fill it, despite endless “Ask Grok” about X. Anthropic has too many users and not enough compute. We will fund that shortfall by leasing Colossus 1 to Anthropic.
But it would also be a step up for Mr. Musk. Frontier AI Labs’ biggest cost line is the 30%+ margin paid to AWS, Microsoft Azure, or Google Cloud for computing. SpaceX is reaping the benefits of hyperscalers instead of paying for them in stressful debt deals like AI Labs.
The framework for SpaceX to be the fourth hyperscaler will need to be accepted by investors before Musk can set a price, analysts said. luck. SpaceX, which can compete with AWS, is worth a hyperscaler multiple, not a rocket company multiple. These days, Alphabet, Microsoft and Amazon have estimated earnings multiples that are about twice that of Boeing and Lockheed Martin.
But Moore was skeptical that the pivot would be easy. Large enterprise customers such as government agencies, or luck 500 companies choose data center storage primarily based on location. Whether the energy costs are low and whether there are fail-safe features in case something goes wrong. You can’t recreate AWS’ global and legal footprint by building one large data center in Memphis. “The battle is not just about who has the most compute servers,” he said.
“I would never bet on Elon doing anything great,” Moore added. “But he’s really doing his best to compete against AWS.”
Kill switch clause
Whether Musk wins that battle or not, he already has something that other computing providers don’t have. In a reply to X, he wrote that SpaceX “reserves the right to take back the computing” if Anthropic’s AI engages in “acts that are harmful to humanity.” This clause was not mentioned in the official press release, and it is unclear whether it is included in the contract. However, if enforceable, it would strongly restrict Musk from one of the three major AI labs in existence, and he would be able to challenge OpenAI’s leadership in federal court.
That wasn’t the case two weeks ago, but what Mr. Musk has now is great power. And even if Musk doesn’t change his mind much about AI, it probably won’t matter much.
Moore, who served as dean of Carnegie Mellon University’s computer science department at a time when Musk was most vocal about his existential crisis, remembers Musk as “one of the people who vociferously argued that artificial intelligence was an existential threat to humanity.” He now says that AI will bring about a richer world.
Anthropic almost certainly has a backup plan. Frontier AI Labs doesn’t have a habit of single-sourcing data centers on which its entire product depends, Moore said, and actively works on computational efficiencies in the background. “They will have contingency plans three, six, 12 months out,” he said.
Still, neither side can come away with a clean result. Gene Munster, managing partner at Deepwater Asset Management, said there was an 80% chance the deal would still be viable after two years. The remaining 20% is a bet on Mr. Musk himself. “What makes it unique is Elon’s history,” Munster said. “He can change his mind. It’s not about the actual clause, it’s about who is enforcing the clause.”
There’s no question about the odds of the deal. Even if Munster is correct and the contract is valid for two years, one in three frontier AI labs in the world would now be running on infrastructure controlled by a competitor’s CEO.
“The stakes are very high,” Moore said. “Everyone is trying to get through the next six months. We’ll do whatever it takes.”
