Markets
US Closes the Cloud Loophole on AI Chips for China
A US rule issued on November 16, 2026 requires licences to ship AI chips to any data centre that serves Chinese-headquartered firms above a set computing threshold.
For three years, US export controls stopped Nvidia's best chips from being shipped to China but not from being used by Chinese companies. A Chinese AI lab could rent time on a cluster in Malaysia or Singapore and train its models there. On Monday, November 16, the Commerce Department's Bureau of Industry and Security issued an interim final rule aimed squarely at that gap.
The rule works through the hardware. Exporters now need a licence to export, re-export or transfer advanced AI chips and servers to a data centre anywhere in the world if it provides, or will provide, large-scale computing to companies headquartered in Macau or in the countries of Country Group D:5, an arms-embargo list that includes China, Russia, Iran and North Korea. The trigger is capacity of more than 12 million in total processing performance, or TPP, at a single site or across linked sites. TechCrunch put that at roughly the computing power of 750 Nvidia H100 or H200 chips.
Several details give it reach. Access through brokers or intermediaries counts, so exporters and operators have to check who ultimately controls their customers. New licence requirements also apply to chip shipments to Malaysia, Singapore, Indonesia, Thailand and Brazil, which the department said present "a higher-than-normal risk of diversion," although chips going to US-headquartered companies in those countries are exempt. Operators already serving covered customers get 90 days from the effective date to obtain a licence, and recommendation systems used for e-commerce or social media are excluded. The rule also formally rescinds the Biden administration's AI Diffusion Rule of January 2025, and comments are due within 60 days.
The politics are as notable as the text. Commerce said the rule was "written by and at the direction of the Department of Treasury and the National Security Council," an unusual line in a regulation, and Yahoo News reported, citing three people familiar with the matter, that Treasury Secretary Scott Bessent pushed it through over Commerce's objections. It comes days after Jensen Huang resigned as Nvidia's chief executive amid an investigation into whether Singapore-based Megaspeed diverted Nvidia chips to China, and after Reuters reported that Chinese AI companies were rushing to lock in overseas computing capacity before the rule landed.
Investors read it as one more cap on Nvidia's addressable market, a week before the company reports. Nvidia fell 4.12% on Monday, and on Tuesday the selling moved to Asia: Korea's Kospi was down about 2% in the morning and an Investing.com tally put the technology sub-index of the MSCI Asia Pacific gauge 3.6% lower. Data-centre operators in Johor and Singapore, which have grown quickly on Chinese demand, now face a compliance check on every large customer.
The open question is enforcement. A licence requirement only works if Washington can see who is renting the machines, and cloud capacity is easy to resell through layers of companies. The 60-day comment period, the first licence decisions and any names added to the entity list, including the long-debated Megaspeed listing, will show whether the rule closes the loophole or just makes it more expensive to use.