Anthropic on Wednesday announced a commercial arrangement with SpaceX to access Colossus 1, a large-scale AI supercomputer that the companies say is composed of over 220,000 NVIDIA GPUs. The agreement is designed to add substantial short-term compute capacity and to explore longer-term options for scaling AI infrastructure beyond terrestrial constraints.
Under the deal, Anthropic will receive more than 300 megawatts of fresh capacity within the coming month. The company said it intends to use that compute to expand service capacity for subscribers to its Claude Pro and Claude Max offerings.
Colossus 1 is described as a cluster built for demanding AI tasks. It comprises H100, H200, and GB200 accelerators and is configured to support large language model training, multimodal systems, scientific simulations, and generative AI workloads.
In conjunction with the SpaceX announcement, Anthropic indicated interest in a potential partnership to build orbital AI compute capacity on the scale of multiple gigawatts. The company noted that compute requirements for next-generation systems are outpacing what can be delivered by available terrestrial power, land, and cooling, and framed orbital compute as a prospective avenue for meeting those needs.
Anthropic also disclosed product-level changes intended to expand developer and customer throughput. The company said it is doubling Claude Code’s five-hour rate limits across Pro, Max, Team, and Enterprise plans. It is removing the peak hours limit reduction on Claude Code for Pro and Max accounts, and it is increasing API rate limits for Claude Opus models.
The SpaceX arrangement supplements a slate of existing and announced compute agreements the company has in place. Those include an up-to-5-gigawatt agreement with Amazon that provides nearly 1 gigawatt of new capacity by the end of 2026; a separate 5-gigawatt agreement with Alphabet’s Google and Broadcom beginning in 2027; a partnership with Microsoft and NVIDIA that includes $30 billion of Azure capacity; and a $50 billion investment in American AI infrastructure with Fluidstack.
Taken together, the deals reflect an effort to aggregate large volumes of compute across multiple providers and delivery models, including terrestrial hyperscale data centers and exploratory plans for orbital compute. Anthropic characterized the SpaceX relationship as both an immediate capacity source and a potential stepping stone toward new compute architectures.
Summary
Anthropic has signed an agreement with SpaceX to access Colossus 1, a cluster with more than 220,000 NVIDIA GPUs, delivering over 300 megawatts of capacity within the month to scale Claude Pro and Claude Max services. The company also expressed interest in partnering with SpaceX on orbital compute measured in multiple gigawatts and announced several product rate-limit increases.
Key points
- Anthropic will receive more than 300 megawatts of new compute capacity within the month via SpaceX's Colossus 1 cluster.
- Colossus 1 uses H100, H200, and GB200 accelerators and is aimed at LLM training, multimodal systems, scientific simulations, and generative AI workloads.
- The company has additional multi-gigawatt agreements and commitments with Amazon, Google and Broadcom, Microsoft and NVIDIA, and a $50 billion investment with Fluidstack.
Risks and uncertainties
- Anthropic's stated interest in orbital compute is exploratory - the announcement reflects interest rather than a binding commitment, introducing uncertainty about whether and when such capacity would materialize.
- The company noted that compute needs for next-generation systems are exceeding what terrestrial power, land, and cooling can provide, signaling a capacity constraint risk that affects data-center operators, cloud providers, and infrastructure investors.
- Timelines for the various multi-gigawatt agreements differ by provider and start dates, which may create uncertainty in when specific capacity increments will become available to support production workloads.