Commerce Dept Closes GPU Loophole; China-Linked SE Asia Data Centers Face License Requirement
This closes the most accessible arbitrage in US chip export controls. Chinese firms had been using Singapore and Malaysia as legal jurisdictions to purchase H100s and H200s—the same chips banned for direct export to China.
BIS's move forces them back to either (a) domestic alternatives (Huawei's Ascend chips, which lag Nvidia by 2-3 generations), (b) acquiring older-generation GPUs from the gray market at 40-50% premiums, or (c) building their own (a 3-5 year R&D cycle). For Nvidia, this eliminates a reported $1.5-2B annual revenue stream from Southeast Asia data center operators serving Chinese customers.
For US competitors (Intel, AMD), it removes an indirect sales channel—they were also selling through the same loophole.
Nvidia's China revenue stream just contracted overnight.
The BIS guidance forces every Singapore and Malaysia data center operator to either apply for an export license (which will be denied for Chinese-linked entities) or liquidate their Nvidia inventory—a $2-4B stranded asset problem across Southeast Asian cloud platforms.
The retroactive application means existing contracts are now in violation, exposing Chinese tech companies like ByteDance, Alibaba Cloud, and Baidu to secondary sanctions if they continue operating these facilities. Watch whether China retaliates with restrictions on US semiconductor IP or rare earth exports by end of Q2.
Did BIS grandfather any existing licenses, or are all current SE Asia GPU deployments now in technical violation? The difference between a 90-day compliance window and immediate enforcement is the difference between an orderly shutdown and a fire sale.
Fault Lines
Strategic intelligence, synthesized daily — with a public track record. Every call graded against what actually happened.