Anthropic expands compute capacity with SpaceX partnership Proactive uses images sourced from Shutterstock
Anthropic announced that it has reached an agreement with SpaceX (Unlisted (US):SPACEX) to access computing resources from the company’s Colossus 1 artificial intelligence facility in Memphis, Tennessee.
The deal gives Anthropic access to the facility’s full capacity, which SpaceX said includes more than 220,000 Nvidia GPUs spanning H100, H200 and GB200 systems.
The cluster is designed for large-scale AI training and inference workloads and provides roughly 300 megawatts of additional compute power. Anthropic expects to begin integrating the capacity within the next month.
The company behind Claude said the added infrastructure will help support rising demand, particularly for products such as Claude Code, which has seen increased usage among developers.
Anthropic also said the expanded supply will allow it to raise rate limits for paid users and ease usage constraints during periods of heavy traffic.
For SpaceX, the agreement brings a major external customer to its growing AI computing business as it broadens its ambitions beyond aerospace and communications.
The company has positioned Colossus 1 as a rapidly deployed, large-scale system built for frontier AI development and high-performance workloads.
SpaceX additionally said Anthropic has indicated interest in exploring longer-term collaboration on orbital computing infrastructure at multi-gigawatt scale. It suggested that future AI systems may outgrow terrestrial limits around power, cooling, and land availability, although such space-based compute remains conceptual and would require substantial engineering advances.
The announcement coincided with an Anthropic developer event in San Francisco, where the company introduced a new Claude feature called “dreaming,” which enables models to review prior interactions between sessions and update internal context based on learned patterns.