Anthropic picked the most loaded co-dependency in the industry

On the day of their first developer conference, Anthropic announced a compute deal with SpaceX. Colossus 1. Memphis. 300 megawatts. 220,000 Nvidia chips. The same data center xAI built to train Grok.

Musk's take, for the record: the Anthropic team didn't set off his "evil detector." He'll keep them in the facility so long as Claude doesn't "engage in actions that harm humanity." That's the SLA. That's the covenant. Anthropic's compute access is contingent on one man's ongoing opinion of their alignment. The org that has spent more time than any other explaining, in careful, footnoted detail, why it is not like the others, just handed a killswitch to the person most likely to pull it for reasons unrelated to safety.

The math on the deal is real. Rate limits up, Opus caps loosened, orbital compute in the roadmap. None of that is vaporware. But the counterparty you choose when your whole brand is responsible AI is a statement whether you mean it to be or not. They also use Google for chips and cloud. Fine. Google doesn't tweet about your evil detector. This is the most safety-branded org in the industry signing its most operationally critical dependency with the most charged figure in it, and announcing it on the day they're asking developers to trust them. The deck says safety. The cap table says something else.