The Commerce Department moved on Sunday, May 31, to close a year-old loophole that allowed the world's most advanced AI silicon to flow to Chinese-owned subsidiaries based outside China. CNBC's reporting names Nvidia Rubin and Blackwell and AMD's MI350x as the products in scope, and notes the gap opened in May 2025 when Commerce said it would not enforce the Biden-era AI Diffusion rule. One industry source with deep supply-chain knowledge estimated to CNBC that shipments through the gap landed in the hundreds of thousands of chips. Malaysia is named as a primary transit market.
Hundreds of thousands of leading-edge accelerators do not sit on a shelf. They go into racks, and racks need heat rejection. If even a portion of that volume landed in Southeast Asia for Chinese-affiliated buildouts, the cooling capacity to support it was already procured, installed, and is now operating. That demand was absorbed by regional liquid cooling, cold plate, and CDU vendors during a year when the official US position on the policy was ambiguous. The announcement does not unbuild any of that. It freezes the next tranche.
For cooling vendors that forecast against Nvidia and AMD volume, the immediate effect is uncertainty in the regional split. A portion of Blackwell and Rubin shipments that had been quietly routed to facilities outside China now has to find a different home. Either it stays in the US and European hyperscale pipeline, tightening supply on the West Side, or it gets rejected and the order falls through entirely. The liquid cooling supply chain now has a near-term swing variable it cannot model without inside-track export-control intelligence.
Chinese firms that lose access to the rerouted Blackwell and Rubin supply will accelerate the domestic alternative, which means more volume for Chinese accelerator programs and the cooling vendors that serve them. The earlier round of export controls already produced a domestic supply response. This announcement tightens the rule and reinforces that response. For Western cooling vendors trying to sell into Chinese end users through indirect channels, the policy environment just got harder to navigate, not easier. For Chinese cooling vendors operating domestically, the address book just got longer.
The practical takeaway for anyone planning a 2026 or 2027 deployment is that accelerator availability is now subject to a policy signal that can change on a weekend. The conservative move is to lock supply on the cooling side against firm GPU allocations, not against vendor roadmap projections. The roadmap projections built in the rerouted Asian volume. The firm allocations do not.