Anthropic to run Claude on SpaceX’s Colossus 1 data‑centre capacity

AI models are starting to look a lot like steel mills. The bottleneck is not just algorithms anymore. It is power, cooling, land, and whoever can get tens of thousands of Nvidia chips online fast enough. That is why Anthropic’s new deal with SpaceX matters — on May 6, Anthropic said it signed for all of the compute capacity at Colossus 1 in Memphis, and then immediately raised Claude usage limits for paying customers. ### What actually changed for Claude users? The visible part is simple. Anthropic doubled Claude Code’s five-hour usage limits across paid tiers, removed peak-hour throttling for Claude Pro and Max, and raised API rate limits for Claude Opus models. Anthropic tied those changes directly to the new SpaceX compute agreement, so this was not a vague “future capacity” announcement — it was a capacity unlock that showed up in product limits right away. (anthropic.com) ### What is Colossus 1? Colossus 1 is a giant AI data-center footprint in Memphis tied to Elon Musk’s xAI orbit, but the agreement was announced through SpaceX. Anthropic said the deal gives it more than 300 megawatts of new capacity and over 220,000 Nvidia GPUs within the month. Those are huge numbers. At this scale, you are not renting a few racks to smooth demand. You are effectively adding another frontier-class compute estate. (anthropic.com) ### Why does 300 megawatts matter so much? Because AI infrastructure is now constrained by electricity almost as much as by chips. A modern frontier model needs massive parallel hardware for both training and inference, but inference is the part users feel every day — slower responses, tighter caps, more throttling. Three hundred megawatts is the kind of number utilities, governors, and transmission planners care about. Basically, Anthropic bought itself room to say yes more often when customers hit Claude hard. (anthropic.com) ### Why go to SpaceX at all? The short answer is speed. Building a new data center takes years. Leasing existing frontier-scale capacity can change your product this month. There is also a weird strategic twist here — SpaceX is controlled by Elon Musk, whose xAI competes in the same market. So this is not a cozy ecosystem deal. It is a reminder that in AI infrastructure, rivals will still transact if one side has spare power and the other side is desperate for it. (anthropic.com) ### Is this mostly about training or inference? Anthropic framed it around customer capacity, which points heavily to inference. The company said the new capacity will directly improve service for Claude Pro and Claude Max subscribers, and the immediate product changes were higher limits and fewer restrictions. Training may benefit too — giant GPU clusters are fungible to a point — but the near-term story is that Anthropic needed more headroom for people already using Claude, especially coding workloads that keep sessions running for hours. (cnbc.com) ### What is the stranger second part of the deal? Anthropic and SpaceX also signaled interest in orbital AI infrastructure. That does not mean Claude is moving to space next quarter. It means the companies are exploring whether future compute could be built off-planet at multi-gigawatt scale. For now, that is more roadmap than reality. But it tells you how constrained terrestrial power and permitting have become if serious AI companies are even willing to say this part out loud. (anthropic.com) ### Why should anyone outside AI care? Because compute concentration changes who can ship products reliably. If a handful of companies control the biggest clusters, they gain leverage over pricing, uptime, and who gets priority during demand spikes. That hits developers first, but it eventually reaches everyone building on these models — from coding tools to enterprise software to robots that cannot afford cloud outages. The catch is that frontier AI now depends on infrastructure that is very physical, very scarce, and increasingly political. (cnbc.com) ### Bottom line? This deal is really about bottlenecks. Anthropic found a fast way around one of them, and Claude users get the benefit now. But the bigger story is harder to miss — the AI race is becoming a race for power plants, substations, and GPU estates, and the companies that lock those up early will shape what everyone else can build. (anthropic.com)