Anthropic locks in SpaceX Colossus 1 for 300MW of new Claude compute

Anthropic has signed a deal for all of the AI compute capacity at SpaceX's Colossus 1 data centre in Memphis, Tennessee. The Claude maker raised rate limits on its paid plans the same day. Anthropic gets more than 300 megawatts of new capacity and over 220,000 NVIDIA GPUs within a month, the company said in a blog post on Wednesday.

Financial terms were not disclosed. The Colossus site was originally built for Elon Musk's xAI. It houses a mix of NVIDIA H100, H200 and GB200 accelerators, according to Invezz. Al Jazeera reported the 300MW load is roughly the electricity draw of 300,000 homes.

Anthropic doubled the five-hour rate limits on Claude Code for Pro, Max, Team and Enterprise subscribers, effective immediately. The company removed peak-hour throttling on Claude Code Pro and Max accounts. It also raised API rate limits on Claude Opus models. The Memphis capacity will feed those higher ceilings, Anthropic said.

A wider compute roster

The SpaceX deal sits alongside several multi-gigawatt commitments Anthropic has signed over the past year. Its Amazon Web Services agreement runs up to 5 gigawatts, with nearly 1 gigawatt of new capacity due online by the end of 2026. A separate 5 gigawatt arrangement with Google and chip designer Broadcom is scheduled for 2027.

Microsoft and NVIDIA are providing $US30 billion of Azure capacity under a separate strategic partnership. A $US50 billion build-out of US AI infrastructure with Fluidstack is also in train. Anthropic said future expansion would be selective. It would focus on democratic countries with strong legal and regulatory frameworks, and target enterprise customers in finance, healthcare and government that need data residency.

Musk approves

Musk has been a vocal critic of OpenAI chief executive Sam Altman, and is suing OpenAI on the basis it abandoned its non-profit mission. Asked about the Anthropic tie-up, Musk said: "No one set off my evil detector," Al Jazeera reported.

Anthropic also signalled longer-term ambition. The company said it would explore developing "multiple gigawatts of orbital AI compute capacity" with SpaceX. It acknowledged the project would be both capital-intensive and technically demanding. No timeline was attached.

AU enterprise read

For Australian Claude customers, the immediate effect is on rate-limit ceilings, not regional latency. Anthropic does not run dedicated infrastructure in Australia. It routes paid traffic through hyperscaler partners, including AWS Sydney. Larger ceilings on Claude Code and the Opus API mean fewer queue throttles for AU developers and enterprise teams running coding agents and long-context workloads.