Artificial intelligence (AI) corporations Anthropic secured large-scale data center compute resources by teaming up with Elon Musk's SpaceX.
Anthropic said on the 6th (local time) it can use the compute capacity of the "Colossus 1" data center in Memphis, Tennessee, owned by SpaceX's AI subsidiary xAI. With this, Anthropic secured 300 MW (megawatts) of compute capacity, including more than 220,000 Nvidia graphics processing units (GPUs).
The secured compute resources were immediately used to improve service performance. Anthropic doubled the usage limits for paying users of Claude Code, a coding tool with strong demand from corporations, and raised the call limits for Opus, its top public model, by 2 to 16 times.
Unlike OpenAI, which has aggressively built data centers, Anthropic had maintained a relatively conservative infrastructure expansion strategy. Anthropic CEO Dario Amodei previously said of AI corporations' infrastructure investment race, "Some players act in a YOLO (you only live once) way."
But things changed this year as the popularity of the Claude model surged. Usage grew rapidly, bringing compute capacity issues to the fore, and Anthropic ultimately ended up tapping even rival Musk's xAI infrastructure. The data center is known to have been built to run xAI's AI model "Grok."
Even so, some in the industry say the partnership is unusual given the two companies' paths had been effectively adversarial. SpaceX CEO Musk has criticized Anthropic as "hating Western civilization" and "hypocritical."
Nevertheless, the collaboration appears to have come together because both corporations needed results as they approach initial public offerings (IPOs). SpaceX can generate revenue by leasing data center capacity and demonstrate its infrastructure operations capability, while Anthropic can expand its customer base by stably serving the Claude model.
Musk said on X, the social media platform, that he spent time with senior Anthropic executives that day and was "deeply impressed. Everyone I met was very competent and cared a lot about doing the right thing."
Meanwhile, Anthropic also unveiled "Dreaming," a self-improvement feature for AI agents, the same day. The Dreaming feature lets an AI agent review its own work process, learn patterns, and update user preferences and more.