Anthropic's Claude model series is now subject to BIS export controls -- the mechanism described in the Ars Technica reporting as one Anthropic actively shaped rather than resisted. The immediate consequence is not a revenue line. It is a map. Every inference contract Anthropic holds with a counterparty operating in Singapore, Hong Kong, or any jurisdiction where PRC-affiliated entities can route API calls is now subject to end-use review under the same framework that gates Nvidia H100 and H200 shipments to the same addresses. The gap between "globally available API" and "BIS-controlled model artifact" is now zero. Anthropic closed it, with the Commerce Department's signature.

What the Groq $650M raise and the Baseten $1.5B round price is the inference geography that remains after that closure. Groq runs LPU clusters in Santa Clara and a colocation node in Amsterdam; neither sits inside the entity-list perimeter. Baseten's inference infrastructure serves US and EU enterprise contracts with latency SLAs that PRC-routed traffic would violate anyway. The capital flowing into both firms in Q2 2026 is not a bet on model quality -- it is a bet that the BIS-controlled inference surface and the non-controlled inference surface are now two distinct markets, and that the non-controlled surface needs dedicated infrastructure. Demis Hassabis's prize researcher, now at Anthropic, sharpens the model stack on one side of that line. The capital markets are building the pipes for the other side, and the two sides will not reconverge before the BIS October 2023 control schedule's next revision window, which the Commerce Department has not yet scheduled.