Elon Musk has decided Anthropic isn’t evil after all. The SpaceX chief, who spent much of February calling the AI lab misanthropic and accusing it of hating Western civilization, now says that he sat down with senior Anthropic executives last week and walked away impressed. “No one set off my evil detector,” he wrote on X. The reversal came alongside news that Anthropic has signed a deal to use the entire Colossus 1 supercomputer in Memphis—the data center built by Musk’s xAI before its merger with SpaceX earlier this year. The agreement gives Anthropic more than 300 megawatts of capacity and access to over 220,000 Nvidia GPUs within the month, spanning H100, H200, and GB200 chips. Inference workloads are expected to begin shifting onto the cluster in the coming days.
How Elon Musk went from calling Anthropic ‘misanthropic’ to powering Claude with 220,000 GPUs, in just three months
Musk has not been subtle about his views on Anthropic. In February he greeted the company’s $30 billion funding round by calling it doomed by its own name, accused its models of demographic bias, and predicted it would never win. He later piled on over Anthropic’s clash with the Pentagon, which blacklisted the lab in March as a supply chain risk.His tone on Wednesday was different. He told followers that the Anthropic team he met was highly competent and serious about doing the right thing, and that Claude would probably turn out fine so long as the lab kept questioning itself. He also said he was comfortable leasing out Colossus 1 because xAI had already shifted training to its newer Colossus 2 facility, and reserved the right to pull the compute back if Claude ever turned harmful.Anthropic chief product officer Ami Vora unveiled the partnership on stage at the company’s Code with Claude developer conference in San Francisco. Co-founder Tom Brown, posting on X, said keeping up with AI demand will require moving a lot of atoms—and that nobody moves atoms faster than Musk’s operation, on or off the planet.
Anthropic’s compute crunch made Musk’s spare cluster hard to refuse
The lab has spent months apologising to paying users over rate limits and outages on Claude Code, especially during peak hours. Its other recent compute deals—a 5 GW agreement with Amazon, a 5 GW partnership with Google and Broadcom, $30 billion of Azure capacity from Microsoft, and a $50 billion Fluidstack build-out—won’t start coming online until late 2026 at the earliest. Colossus 1, by contrast, can begin running Claude inference within days.Alongside the SpaceX announcement, Anthropic doubled its five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans, removed peak-hour throttling for Claude Code, and raised API limits for its Opus models.The two companies also said they have begun talking about something stranger: building multiple gigawatts of AI compute in orbit. SpaceX, which absorbed xAI in a merger earlier this year, made that consolidation official in Wednesday’s announcement—referring to itself publicly as “SpaceXAI” for the first time. Musk has since confirmed that xAI will be dissolved as a separate company, with its AI products folded into SpaceX under the SpaceXAI name. The combined entity is preparing for an IPO expected as soon as next month, slated to be the largest in corporate history.
