Anthropic signs deal to run models on SpaceX’s Colossus 1 data centre

AI this week is really a story about electricity, chips, and who can get them first. Anthropic’s news with SpaceX matters because it turns “compute” from a vague buzzword into something brutally concrete — 300-plus megawatts, 220,000 Nvidia GPUs, and a full data center in Memphis. The gap was simple: Claude demand was rising faster than Anthropic’s available capacity. On May 6, Anthropic said it solved part of that by signing a deal to use all of SpaceX’s Colossus 1 compute capacity, with the new power coming online within a month. ### What is Colossus 1? Colossus 1 is a giant AI data center in Memphis, Tennessee — basically a warehouse-scale cluster built to train and run frontier models. Anthropic said the agreement gives it more than 300 megawatts of new capacity and over 220,000 Nvidia GPUs. That is not a side contract for overflow traffic. It is the kind of capacity reservation that changes what products a lab can actually ship. ### Why does Anthropic need this much? Because Claude has been hitting limits. Anthropic tied the SpaceX deal directly to higher usage caps for customers. It doubled Claude Code’s five-hour rate limits for paid plans, removed peak-hour reductions for Pro and Max accounts, and raised API rate limits for Claude Opus models. In plain English — Anthropic ### Why is SpaceX in this story? Because SpaceX is no longer just rockets. Earlier this year, Elon Musk merged SpaceX with xAI, and Colossus sits inside that broader AI infrastructure push. Leasing the full site to Anthropic gives SpaceX a very visible customer just as it tries to sell investors on its AI ambitions. It also shows Musk is willing to be public. ### Wait — weren’t Musk and Anthropic fighting? Yes — that is part of why this stands out. Musk had been publicly hostile to Anthropic as recently as February. Then, on May 6, he flipped and said time with Anthropic’s leadership left him impressed. The business logic won. Anthropic needed compute. SpaceX had it. In the AI race, yesterday’s rival can become today’s landlord fast. ### Why not just buy more cloud? Anthropic is already doing that. The company said this SpaceX agreement joins other huge infrastructure commitments — up to 5 gigawatts with Amazon, 5 gigawatts with Google and Broadcom starting in 2027, a Microsoft-Nvidia partnership that includes $30 billion of Azure capacity, and a $50 billion bet on every supply channel they can find. ### What is the weird “in space” part? Anthropic also said it has expressed interest in working with SpaceX on multiple gigawatts of orbital AI compute capacity. That is still an ambition, not a deployed system. But the fact that it is even being discussed tells you where this market is headed. When terrestrial power, land, cooling, and grid are sounding like pure sci-fi and starts sounding like a roadmap sketch. ### So what actually changed for users? The immediate change is not some abstract strategic win. It is capacity relief. Anthropic says the new compute lets it raise Claude Code and API limits now, especially for paid users and developers pushing Opus hard. That makes this deal unusually tangible — customers should feel it in fewer caps, fewer slowdowns, and more room to build. ### Bottom line? This is what the AI race looks like once the easy phase is over. The scarce thing is no longer just talent or model ideas. It is access to industrial-scale power, chips, and facilities — and Anthropic just locked down a lot more of all three.