Huda Thomas@huda.thomas

Follow

Anthropic's $120.5B deal with Musk's SpaceXAI

Who actually keeps Claude running in 2027, Anthropic or Elon Musk? That question sits under the whole listing, and the filing makes it harder to wave off. The arrangement is with SpaceXAI, the SpaceX unit that used to be xAI, and it rents Anthropic the Nvidia chips Claude depends on. Ceiling on the spend: $120.5 billion. From early next year, either side can walk on 90 days' notice. May pricing ran $1.25 billion a month for a single Memphis data center. Reuters had the confidential draft at $84.5 billion, and the public version filed today shows $120.5 billion instead. Anthropic's own expectation for when all the capacity lands in December is around $3.1 billion a month. Compare that to the $16.8 billion reportedly booked last quarter, roughly $5.6 billion a month. More than half of what comes in would head to one supplier. The lock-in expires in January and March 2027, and after that a 90-day cancellation applies both ways. Fortune reported in July that this gives SpaceX a path to pull its chips back if Grok needs the room, and Grok is a direct Claude competitor. In February, Musk publicly called Anthropic's AI evil. So the lab marketing itself as the safe option would run a large chunk of its product on hardware belonging to its loudest critic, who also happens to own a rival and can start that clock in early 2027. The filing does flag that certain compute suppliers build their own models. Read that line with Musk as the supplier and it lands differently. He isn't doing Anthropic a favor either. SpaceX's CFO has pledged $100 billion in annual revenue by the end of this year, while Anthropic's rent at full speed reaches about $9.3 billion a quarter, ahead of the roughly $7.6 billion SpaceX's whole AI business brought in last quarter. Two of the biggest IPOs in history are holding each other up, and each one has a 90-day exit. There's a legitimate case for Anthropic. Compute is the scarcest resource in AI, Amazon's cloud capacity for 2027 is reportedly mostly spoken for, and the filing concedes the company badly underestimated its past compute needs. Taking every chip available is the rational play, and the same clause lets Anthropic bail if cheaper capacity turns up. Still, public buyers deserve two things before the listing. Anthropic should publish how much of Claude runs on SpaceX hardware and what happens to customers if that notice ever lands. And the 90-day exit should come out for the full contract term, matching how its other big compute bills are structured. Until then, anyone buying ANT at a $2 trillion valuation has to settle one thing for themselves: who do they trust to keep Claude running in 2027, Anthropic or Elon Musk?

1h

Huda Thomas@huda.thomas

Follow

Anthropic's $120.5B deal with Musk's SpaceXAI

Who actually keeps Claude running in 2027, Anthropic or Elon Musk? That question sits under the whole listing, and the filing makes it harder to wave off. The arrangement is with SpaceXAI, the SpaceX unit that used to be xAI, and it rents Anthropic the Nvidia chips Claude depends on. Ceiling on the spend: $120.5 billion. From early next year, either side can walk on 90 days' notice. May pricing ran $1.25 billion a month for a single Memphis data center. Reuters had the confidential draft at $84.5 billion, and the public version filed today shows $120.5 billion instead. Anthropic's own expectation for when all the capacity lands in December is around $3.1 billion a month. Compare that to the $16.8 billion reportedly booked last quarter, roughly $5.6 billion a month. More than half of what comes in would head to one supplier. The lock-in expires in January and March 2027, and after that a 90-day cancellation applies both ways. Fortune reported in July that this gives SpaceX a path to pull its chips back if Grok needs the room, and Grok is a direct Claude competitor. In February, Musk publicly called Anthropic's AI evil. So the lab marketing itself as the safe option would run a large chunk of its product on hardware belonging to its loudest critic, who also happens to own a rival and can start that clock in early 2027. The filing does flag that certain compute suppliers build their own models. Read that line with Musk as the supplier and it lands differently. He isn't doing Anthropic a favor either. SpaceX's CFO has pledged $100 billion in annual revenue by the end of this year, while Anthropic's rent at full speed reaches about $9.3 billion a quarter, ahead of the roughly $7.6 billion SpaceX's whole AI business brought in last quarter. Two of the biggest IPOs in history are holding each other up, and each one has a 90-day exit. There's a legitimate case for Anthropic. Compute is the scarcest resource in AI, Amazon's cloud capacity for 2027 is reportedly mostly spoken for, and the filing concedes the company badly underestimated its past compute needs. Taking every chip available is the rational play, and the same clause lets Anthropic bail if cheaper capacity turns up. Still, public buyers deserve two things before the listing. Anthropic should publish how much of Claude runs on SpaceX hardware and what happens to customers if that notice ever lands. And the 90-day exit should come out for the full contract term, matching how its other big compute bills are structured. Until then, anyone buying ANT at a $2 trillion valuation has to settle one thing for themselves: who do they trust to keep Claude running in 2027, Anthropic or Elon Musk?

1h