Anthropic on Wednesday said it has agreed to take all of the artificial intelligence compute capacity at SpaceX’s Colossus 1 data centre, a move that dramatically expands the startup’s training footprint and underscores an escalating race for hyperscale GPU power. Under the deal, Anthropic will access more than 300 megawatts of compute provided by over 220,000 Nvidia GPUs housed at Colossus 1, SpaceX and the AI company said. Financial terms were not disclosed.
Anthropic said the Colossus 1 capacity, combined with recent arrangements it has struck with Amazon, Google, Microsoft and Nvidia, will allow the company to scale services for Claude Pro and Claude Max subscribers and to extend some compute capacity to international markets that operate under stricter regulation, such as healthcare and financial services. The company framed the agreement as part of a broader push to meet demand from enterprise customers and regulated industries that require locality or specific compliance controls.
SpaceX confirmed Anthropic’s exclusive use of Colossus 1 and said the startup has also expressed interest in collaborating on orbiting AI data-centre concepts. Elon Musk, SpaceX’s CEO, wrote on X that he spent a week meeting with senior members of the Anthropic team and praised their competence and stated commitment to “doing the right thing,” adding in a now-public post: “No one set off my evil detector. So long as they engage in critical self-examination, Claude will probably be good.” He also warned that SpaceX reserves the right to reclaim compute if Anthropic’s systems “engage in actions that harm humanity.”
Musk said SpaceX had already shifted its internal AI training work to Colossus 2, which he cited as a reason the company felt comfortable leasing Colossus 1. SpaceX’s AI unit, xAI, runs a competing model called Grok and is viewed alongside Anthropic and OpenAI as one of several fast-growing AI players in the private market; Anthropic, SpaceX and OpenAI have each been the subject of media reports and market speculation about potential initial public offerings later this year.
Related Recipes
Claude Faces Elevated Error Rates on Opus 4.6/4.7 as Anthropic Investigates
Cerebras Opens U.S. Market Debut With $56 Billion Valuation After Oversubscribed AI Chip IPO
Cate Blanchett-Backed RSL Media Unveils Machine-Readable Human Consent Standard and Public AI Rights Register
Nokia debuts agentic AI for fixed networks to speed fibre rollouts and cut operational costs
The scale of the Colossus 1 commitment is notable. More than 300 megawatts and upwards of 220,000 GPUs represent one of the larger single-operator pools of AI compute capacity announced to date and highlights how large language-model developers are increasingly reliant on colocated, hyperscale GPU clusters to train and serve advanced models. Anthropic’s statement did not specify the deployment timeline for the Colossus 1 capacity or the precise allocations between training, fine-tuning and inference workloads.
Industry analysts say such large infrastructure agreements may accelerate competition for both raw compute and specialized chips, while also raising questions about concentration of AI training resources and contingency governance. Anthropic’s comment that some capacity will be dedicated to regulated international markets signals a commercial strategy to win customers that require data residency and compliance assurances — a growing commercial battleground as enterprises confront privacy and legal frameworks across jurisdictions.
Trending Now
Steve Kerr inks two-year extension to remain Warriors coach as franchise eyes changes around Steph Curry
Trump’s Beijing Visit Yields Few Breakthroughs as Boeing Deal Falls Short of Expectations
Beth and Rip Kick Off a New Yellowstone Chapter With Dutton Ranch’s Two-Episode May 15 Premiere Across Paramount+ and Paramount Network
Outlander Final Episode Runs 80 Minutes, Debuts on Starz May 15 With 8 PM ET Linear Air
For now, the deal’s commercial and technical details remain limited to the headline capacity figures and public statements from the two companies. Anthropic has rapidly expanded partnerships with major cloud and chip providers in recent months, and the SpaceX arrangement marks a high-profile addition to its infrastructure mix as the company prepares to broaden access to its Claude models.
Continue Reading
