Anthropic SpaceX deal hands Claude all of Colossus 1

UPDATED · News · 6 May 2026 · MTW News Desk

Anthropic SpaceX deal is the biggest AI infrastructure story of the week: xAI confirmed on 6 May 2026 that Anthropic will take all of the compute capacity at SpaceX’s Colossus 1 data centre in Memphis, gaining access to more than 300 megawatts and roughly 220,000 Nvidia GPUs. The Anthropic SpaceX deal also flagged a longer-term plan to develop AI compute capacity in orbit.

Why the Anthropic SpaceX deal matters

The Anthropic SpaceX deal is the answer to a problem Claude users have been complaining about for months. Anthropic’s user base grew roughly 80x in the first quarter of 2026, hitting reliability ceilings that triggered two Claude outages in 24 hours. The Colossus 1 transfer is the most direct fix possible: Anthropic gets dedicated access to a working hyperscale facility built around Nvidia GB300 systems, and SpaceX gets a long-term anchor tenant for a site Elon Musk no longer needs for xAI training. xAI has already shifted its work to the larger Colossus 2.

For developers, the immediate visible effect is the Claude Code limit change. Anthropic doubled the five-hour rate limit window for paid plans on 6 May and removed peak-hour throttling for Claude Code Pro and Max accounts. Anyone who follows the model-by-model assistant story can compare this with the long arc of ChatGPT vs Claude vs Gemini in 2026 and see that capacity, not model quality, has been the gating factor for the last six months.

Inside the Anthropic SpaceX deal terms

The Colossus 1 site is Memphis-based, built originally as an xAI training facility and now reconfigured as an Anthropic-only Claude inference and training site. Anthropic gets immediate access to more than 300 megawatts of compute, which xAI confirmed scales up over the coming month. The 220,000-plus Nvidia GPU count is one of the largest single-site GPU deployments anywhere; for context, that is the rough order of magnitude of an entire mid-sized hyperscaler. Financial terms were not disclosed.

The unusual part is the orbital footnote. xAI’s announcement and Anthropic’s keynote both flagged that the two companies have agreed to study multiple gigawatts of compute capacity in space. That is exactly the kind of sentence a normal data centre story would not include, and it is also exactly the kind of sentence SpaceX would like to put in a press release ahead of a future Starship IPO conversation. For now, the orbital piece is research interest only, not a project plan, but it is a tell that AI infrastructure is now being designed for the next decade rather than next quarter, in much the same way the Anthropic Amazon £79 (about $100)bn deal already reshaped Claude’s pricing curve.

What the Anthropic SpaceX deal changes for UK Claude users

UK developers should feel the Anthropic SpaceX deal land within days, not months. The Claude Code rate limit doubling is global, the peak-hour throttling removal applies to Pro and Max accounts regardless of geography, and Anthropic explicitly mentioned extending capacity for “regulated international markets” including UK healthcare and financial services. That is a direct response to the FCA’s AI live-testing programme, and it suggests Anthropic is preparing to compete with OpenAI and Google on Enterprise contracts in Britain rather than leaving the UK market to its rivals by default. Anyone tracking the FCA Cohort 2 with Barclays, Lloyds and UBS can see where this is heading.

There is a governance angle, too. Musk’s own xAI announcement included a line about reserving the right to claw back compute if Anthropic’s models “engage in actions that harm humanity”. That is the first time a compute lease has been publicly framed around safety contingencies, and it is a precedent both Microsoft and Amazon will likely be asked about in their next data centre deals. Anthropic, for its part, has put more public weight on its Responsible Scaling Policy in the last six months than at any time since the Claude 3 launch.

What to watch next on the Anthropic SpaceX deal

Three things will decide whether the Anthropic SpaceX deal is the inflection point Anthropic claims or just another headline. First, capacity. 300 megawatts is a strong number on paper, but the real test is whether Anthropic uses it for inference (helping current Claude users), training (the next Claude model) or both. Second, latency. Memphis is a long way from London, so UK enterprise users will want to know whether Anthropic is pairing this with European edge capacity. Third, the orbital research piece. If anything concrete emerges by the end of 2026, the AI infrastructure conversation has moved into space.

For now, the Anthropic SpaceX deal does the most important thing a compute deal can do: it ends the headline problem of “Claude is down” and gives developers a reason to trust Pro and Max plans again. That matters more than any feature drop, because the conversation about whether Claude is the best general-purpose model has been gated for months by whether Claude was up at all. With the Anthropic SpaceX deal closed, Anthropic finally gets to argue on model quality instead of capacity.

The wider AI compute landscape is now visibly shifting. Anthropic has SpaceX, OpenAI has its Stargate footprint and Cerebras commitment, Google has TPU v6e, and Microsoft is leaning on Anthropic via Azure as well as on OpenAI. The Anthropic SpaceX deal is the loudest sign yet that the 2026 AI race is about kilowatts as much as it is about parameters.