Anthropic's SpaceX announcement was already large enough on Earth: more than 300 megawatts of Colossus 1 capacity, over 220,000 Nvidia GPUs, and higher Claude limits attached to a supplier whose owner also runs a rival lab. [1]

Thursday's paper said Anthropic had added SpaceX capacity before the filing caught up. Friday's smaller but stranger receipt is one sentence in the same file: Anthropic said it is also interested in developing multiple gigawatts of AI compute in orbit. [1]

That line should not be inflated into a procurement plan. It names interest, not a contract, launch schedule, power architecture, cooling design, jurisdictional answer or orbital debris policy. CNBC's account kept the emphasis on the SpaceX data-center capacity and the Claude rate-limit relief, while noting the orbital-compute ambition as part of the partnership language. [2]

The divergence is useful because mainstream coverage reads the deal as capacity relief. X will read the orbital clause as destiny or dystopia. The paper's frame is colder: when frontier labs start using orbit as a capacity horizon, AI infrastructure has escaped the ordinary data-center map.

The filing gap still governs the story. Compute in orbit would make permitting, power, launch economics and sovereign control harder, not easier. The fact that Anthropic can say the sentence publicly is itself the receipt.

-- THEO KAPLAN, San Francisco