The mechanism: BIS's January 2026 rule moved advanced GPU exports to China (Nvidia H200, AMD MI325X) from presumption-of-denial to case-by-case review — but wrapped it in a 25% tariff, a 50% volume cap, mandatory third-party testing, and KYC/remote-monitoring conditions so onerous most hyperscalers won't bother routing China-bound capacity through them. Then in May, BIS closed the offshore-subsidiary loophole entirely: any buyer whose ultimate parent sits in China or Macau now needs a license, wherever the entity is registered. Every turn of this ratchet raises the compliance tax, timeline risk, and political liability of building AI infrastructure on merchant GPUs — anywhere, not just in China, because export-control uncertainty pushes hyperscalers to diversify their supply chain away from a single controlled chokepoint. The policy doesn't ban Nvidia. It makes owning your own silicon roadmap look like the safer capital allocation.
Who cashes in: