SpaceX supplies Colossus 1 to Anthropic

AI companies keep running into the same wall — demand is growing faster than the data centers behind the models. That wall just got a lot farther away for Anthropic. On May 6, Anthropic said it signed a deal with SpaceX to use all of the compute capacity at Colossus 1, a giant GPU cluster tied to xAI’s Colossus system. The point is simple: more room to train models, serve more users, and stop rate limits from feeling like a permanent feature. (anthropic.com) ### What actually got announced? Anthropic paired two announcements together. First, it raised usage limits for Claude Code and increased API rate limits for Claude Opus models. Second, it said the reason it can do that is a new compute agreement with SpaceX for Colossus 1. Anthropic described the deal as access to all of the facility’s capacity, with that capacity coming online within a month. (anthropic.com) ### How big is Colossus 1? Big enough that the numbers stop sounding like a normal cloud contract. Anthropic said the deal gives it more than 300 megawatts of new capacity and over 220,000 NVIDIA GPUs. xAI’s own writeup calls Colossus 1 one of the world’s largest AI supercomputers and says the system includes dense deployments of H100, H200, and GB200 chips. xAI has separ(anthropic.com)0 GPUs already, with a roadmap toward 1 million. (anthropic.com) ### Why does that matter for Claude users? Because compute is the hidden product. If you run out of it, users feel that as slower responses, tighter caps, and more peak-hour restrictions. Anthropic said this new capacity will directly improve service for Claude Pro and Claude Max subscribers. It also said Claude Code’s five-hour rate limits are doubling for several paid p(anthropic.com) API limits are being raised considerably. (anthropic.com) ### Why is SpaceX in this at all? The weird part is the branding. Colossus is xAI infrastructure, but xAI was acquired by SpaceX earlier this year, so the deal is now being framed as a SpaceX compute partnership. That makes this less like a standard hyperscaler contract and more like Anthropic tapping a newly integrated Musk-controlled compute stack — chips, power, facilities, and now a parent company with its own huge capital base. (x.ai) ### Is this just about one data center? No — and that is the bigger story. Anthropic is stacking capacity from everywhere it can get it. In the same announcement, it pointed to an up-to-5-gigawatt agreement with Amazon, a 5-gigawatt agreement with Google and Broadcom starting in 2027, a Microsoft-NVIDIA partnership that includes $30 billion of Azure capacity, and a $50 billion infrastructure plan with Fluidstack. (x.ai)ing like frontier AI is now a power procurement business with models attached. (anthropic.com) ### Why move this aggressively now? Because Claude’s growth appears to be forcing the issue. Anthropic said last month that its run-rate revenue had surpassed $30 billion, up from about $9 billion at the end of 2025. It also said the number of business customers spending more than $1 million annually had doubled in less than two months to over 1,000. If those numbers are (anthropic.com)oning success. (anthropic.com) ### What’s the strangest part of the announcement? Anthropic and SpaceX also floated interest in developing multiple gigawatts of orbital AI compute capacity. That is not a product roadmap yet — it is more like a statement of where the bottleneck is heading. The idea is that power, land, and cooling on Earth may become the limiting factor, so future compute(anthropic.com)hink about scale. (anthropic.com) ### Bottom line This is not just a flashy hardware lease. It is Anthropic buying itself room to keep Claude growing right now, while the whole frontier AI industry shifts from “who has the best model?” to “who can secure enough chips and electricity fast enough?” (anthropic.com)