The SpaceX S-1 IPO filing revealed something that didn't get nearly enough attention: Anthropic has contracted $1.25 billion per month — $15 billion per year — for access to Colossus 1 and 2 through May 2029.
Let that number sit for a second. $15 billion a year. For compute.
That contract is what's powering Claude Pro and Claude Max right now. Every time someone uses Claude for a complex task, there's a reasonable chance that inference is running on SpaceX hardware in Memphis or Mississippi.
What this tells you isn't just that Colossus is big — you already knew that. What it tells you is that Anthropic, one of the most well-funded and technically sophisticated AI labs in the world, looked at every available compute option and decided SpaceX's infrastructure was worth $15 billion a year. That's not a desperation move. That's a quality signal.
It also tells you something about the SpaceX IPO math. Guaranteed recurring revenue of $15 billion a year from a single customer, on a contract running through 2029, shows up in the S-1 as extremely predictable cash flow. That's the kind of revenue multiple that moves a valuation from $350 billion to $500 billion in analyst models.
The longer play here is orbital compute. SpaceX and Anthropic announced a partnership to develop multiple gigawatts of orbital AI compute capacity — essentially putting data centers in space, powered by solar arrays the size of city blocks, cooled for free by the vacuum of space. If that works at scale, the economics of AI compute change permanently. Ground-based data centers have land costs, power costs, cooling costs, and regulatory costs. Orbital compute has none of those.
SpaceX is the only company on Earth that can actually build this. They have the rockets, the satellite infrastructure, the compute expertise through Colossus, and the AI capability through Grok. Nobody else is even close to having all four.
The Anthropic contract is the proof of concept. The orbital compute partnership is the endgame.