Anthropic Will Pay SpaceX $1.25 Billion a Month for Computing Power
The cost of running frontier AI is almost hard to believe. Anthropic has agreed to pay SpaceX 1.25 billion dollars every month through 2029 for computing power, about 15 billion dollars a year to a single vendor. It is a staggering number that shows how the AI race has become a contest of who can afford the most compute.
The cost of running frontier AI is almost hard to believe. Anthropic has agreed to pay SpaceX 1.25 billion dollars every month through May 2029 for computing power, about 15 billion dollars a year to a single vendor. It is a staggering number that shows how the AI race has become a contest of who can afford the most compute.
The scale is the story. Fifteen billion dollars a year, to just one supplier, for the raw computing power to train and run Claude, is more than the total revenue of many large companies. Anthropic's overall compute spending for 2026 is estimated around 50 billion dollars, with a similar figure at OpenAI. The bill for building modern AI is enormous and growing.
The choice of SpaceX is notable. Elon Musk's rocket company, tied to his AI venture xAI, is now selling computing capacity to a direct AI rival, turning its data-center infrastructure into a revenue stream. That a company like SpaceX can be a major compute provider shows how the buildout of AI power has become its own giant business, separate from the models themselves. The infrastructure is where a lot of the money flows.
It underlines the real barrier to entry in AI. The best models require access to vast, expensive computing power, which only a handful of extraordinarily well-funded companies can afford, and locking in years of supply is how they stay in the race. This is why the leading labs are raising tens of billions and pursuing IPOs, to pay for exactly these commitments. Money buys compute, and compute buys a seat at the table.
The honest question is whether the spending pays off. Committing 15 billion dollars a year assumes Claude will generate enough value to justify it, and Anthropic, like its rivals, is not yet clearly profitable while spending at this pace. If AI demand grows as expected, the investment looks smart, but if it slows, these fixed multi-year bills become a heavy weight. Enormous spending is a bet, not a guarantee.
So Anthropic is committing to one of the largest compute deals ever, a reminder that modern AI runs on almost unimaginable amounts of money and hardware. A 1.25 billion dollar monthly bill, 15 billion a year, all for computing power. The AI race is an infrastructure race first. Watch whether the revenue ever catches up to the spending.