Anthropic said it has signed an agreement with SpaceX to connect to the Colossus 1 AI supercomputer, a system described as running on more than 220,000 NVIDIA GPUs. Under the deal, Anthropic expects to receive more than 300 megawatts of new compute capacity within one month, adding a major block of infrastructure for its large language models and Claude services.
New capacity will be used to expand Claude services and API access
The company said the added compute will go directly toward improving service quality and system capacity for Claude Pro and Claude Max subscribers. At the same time, Anthropic announced a set of product changes tied to the larger compute pool. The five-hour rate limits for Claude Code across Pro, Max, Team, and Enterprise plans will double. Peak-hour throttling for Pro and Max accounts is being removed, and API rate limits for the Claude Opus model are also being raised.
According to the source material, Colossus 1 uses NVIDIA H100, H200, and next-generation GB200 accelerators. The cluster was built for large-scale language model training, multimodal systems, scientific simulation, and generative AI workloads. That makes the partnership more than a short-term capacity boost; it gives Anthropic access to infrastructure built for heavy training and inference demand.
Anthropic and SpaceX also outlined orbital AI compute ambitions
The headline part of the announcement goes beyond terrestrial expansion. Anthropic said it wants to work with SpaceX on developing multi-gigawatt “orbital AI compute capacity.” The company argued that next-generation AI systems are increasing in complexity fast enough that their compute and power requirements are pushing against physical limits on Earth, including electricity supply, land availability, and cooling systems.
Its proposed answer is to move compute infrastructure into space, where solar energy is abundant and low-temperature conditions could ease heat dissipation. In that framing, the deal is not limited to access to an existing supercomputer. It also sets out a longer-range infrastructure target tied to off-Earth deployment.
Deal adds to Anthropic’s growing list of compute agreements
The SpaceX arrangement extends a broader infrastructure buildout by Anthropic. The source says the company has a 5-gigawatt agreement with Amazon, with nearly 1 gigawatt of new capacity expected by the end of 2026. It also has a 5-gigawatt compute agreement with Google and Broadcom beginning in 2027, along with access to as much as $30 billion in Azure compute resources through Microsoft and NVIDIA.
Separately, Anthropic has partnered with Fluidstack on up to $50 billion in strategic investment for AI infrastructure in the United States. With Colossus 1 now added to that list, Anthropic is increasing its position in training capacity, inference scale, and future high-density AI deployment.