Inside Anthropic's Next AI Infrastructure Bet with SpaceXAI

Anthropic’s latest AI expansion is less about chatbot features and more about the thing quietly shaping the entire AI race: who can secure enough data centre capacity to keep the models running.

After striking a deal with SpaceXAI for access to Colossus 1, one of the world’s largest AI supercomputers, Anthropic is lifting Claude usage limits and scaling up compute access for enterprise users.

The agreement adds another heavyweight customer to the growing list of AI firms competing for hyperscale infrastructure, as demand for power, GPUs and cooling pushes data centre operators into a new era of buildouts.

Colossus 1 delivers more than 300MW of compute capacity through a cluster of over 220,000 NVIDIA GPUs, including H100, H200 and GB200 accelerators.

Built in record time, the system has been designed to support large-scale AI training, inference workloads, multimodal systems and scientific simulations.

“I spent a lot of time last week with senior members of the Anthropic team to understand what they do to ensure Claude is good for humanity and was impressed,” writes Elon Musk, CEO of SpaceXAI, in a post on X.

“Everyone I met was highly competent and cared a great deal about doing the right thing.

“After that, I was ok leasing Colossus 1 to Anthropic, as SpaceXAI had already moved training to Colossus 2.”

Claude usage limits increase

Anthropic says the new compute capacity will allow it to “substantially increase” availability across Claude services as enterprise demand accelerates.

The company has already announced that it is doubling Claude Code’s five-hour rate limits for Pro, Max, Team and seat-based Enterprise plans, while peak-hour reductions for Pro and Max accounts are also being removed.

Higher rate limits are also being introduced for Claude Opus models.

The Colossus agreement forms part of a broader infrastructure expansion strategy for Anthropic, which has signed multiple hyperscale compute partnerships over the past year.

Alongside the SpaceXAI deal, Anthropic cited a 5GW agreement with Amazon, additional capacity arrangements with Google and Broadcom and a partnership with Microsoft and NVIDIA providing US$30bn of Azure capacity as factors supporting the increased Claude limits.

Orbital compute ambitions emerge

The partnership also points towards longer-term infrastructure ambitions beyond terrestrial data centres.

Anthropic said it is interested in working with SpaceXAI on developing multiple gigawatts of orbital AI compute capacity, a concept aimed at addressing growing constraints around land, energy and cooling availability.

“SpaceX is the only organisation with the launch cadence, mass-to-orbit economics and constellation operations experience to make orbital compute a near-term engineering program rather than a research concept,” SpaceXAI claims.

“If engineering challenges can be overcome, space-based compute offers near-limitless sustainable power with less impact on Earth.”

The rapid expansion of AI clusters has already intensified competition for grid access, liquid cooling systems and suitable land for hyperscale campuses across major markets.

Compute bottlenecks reshape AI infrastructure

Anthropic is not alone in turning to large-scale infrastructure partnerships to secure compute access.

Earlier this year, Cursor announced its own agreement with SpaceXAI to train Composer coding models on Colossus infrastructure, saying compute availability had previously limited model development.

In its announcement, Cursor said each increase in compute translated into “meaningfully more capable models”, adding the company had been “bottlenecked by compute” before gaining access to the platform.

Anthropic CEO, Dario Amodei, said the company’s growth has significantly exceeded expectations, creating additional strain on compute resources.

“That is the reason we have had difficulties with compute,” Dario says, adding that the company is “working as quickly as possible to provide more”.