AIO Apex
en

US Closes AI Chip Export Loophole, Requires Licenses for Data Centers in Five Countries

South China Morning Post
Share:
US Closes AI Chip Export Loophole, Requires Licenses for Data Centers in Five Countries

The U.S. Commerce Department's Bureau of Industry and Security (BIS) issued an interim final rule on November 16, 2026, closing a loophole that let Chinese companies tap advanced AI chips by renting computing capacity from data centers outside direct U.S. jurisdiction. The rule, titled “Addressing Remote Access to Advanced Semiconductors” (RIN 0694-AK50), adds new licensing requirements for data centers in five countries — Brazil, Indonesia, Malaysia, Singapore, and Thailand — that BIS says carry a higher-than-normal risk of diversion to restricted buyers.

For nearly three years, Washington has tried to keep its most advanced AI accelerators away from Chinese buyers through a shifting set of export bans on Nvidia and AMD chips. Those controls targeted direct sales. Chinese firms responded by sidestepping them entirely: instead of buying restricted chips, they rented GPU capacity from data centers in Southeast Asia and Latin America that had legally purchased the hardware, then accessed that compute remotely over the internet. The new rule is Commerce's attempt to shut that door, extending licensing obligations to cloud and colocation capacity rather than just chip shipments.

Under the rule, any data center that provides more than 12 million aggregate Total Processing Performance (TPP) units — a measure of chip compute capacity — to a customer headquartered in China, Macau, or another Country Group D:5 jurisdiction now requires an export license, regardless of where the data center itself is located. Operators already serving that level of compute to covered customers get a 90-day transition window from the rule's effective date to apply for a license if they want to keep the relationship running. The rule is scheduled for Federal Register publication on November 19, 2026.

The five named countries face a separate, broader licensing requirement for any advanced chip shipments headed their way, not just those tied to flagged customers — a recognition that GPUs shipped to data centers in these markets have proven easy to re-route. A narrow exemption applies to chips serving customers headquartered in the United States, but only if the exporter files monthly reports identifying that customer's ultimate parent company, closing off shell-company workarounds. The rule also formally rescinds the 2025 AI Diffusion Rule, the Biden-era tiered framework it replaces.

The timing is notable: the rule lands weeks after Nvidia came under renewed Congressional scrutiny tied to allegations that its chips had reached restricted markets through intermediaries, and amid broader turmoil atop the company following CEO Jensen Huang's departure. BIS has signaled it will watch for countries that tolerate evasion and may consider import restrictions against jurisdictions that do not enforce the new licensing regime.

For cloud and colocation operators in Brazil, Indonesia, Malaysia, Singapore, and Thailand, the rule means new compliance overhead on top of hardware they may have already purchased and deployed — tracking customer headquarters, aggregating TPP across linked facilities, and filing paperwork that didn't previously exist. For Nvidia and AMD, it narrows one of the few remaining channels through which their highest-end accelerators were reaching Chinese AI labs indirectly, though enforcement will depend on how rigorously the licensing regime is applied once it takes effect.

As reported by the South China Morning Post and legal analysis from Nixon Peabody, the rule is the most significant tightening of U.S. chip export policy since the original China GPU ban, and the first to treat remote data center access as functionally equivalent to a direct chip sale.

Originally reported by South China Morning Post. Read the original article for additional details.

View original source
Share: