Trump Administration Moves to Close AI Chip Loophole Before China Trip
A Commerce Department rule targets remote access to advanced Nvidia chips, putting overseas data center demand in play
Photo by Alex Konokh on Unsplash
Commerce moves to block Chinese firms from renting advanced Nvidia chips abroad, closing a gap ahead of Trump's China trip.
The rule targets access, not shipments
The Trump administration is moving to close the loophole that let Chinese companies use advanced Nvidia (NVDA) chips without buying them. A new Commerce Department rule aims at remote access, stopping Chinese AI firms from renting computing power in data centers outside China. The timing lands just ahead of the president's trip to China.
Until now, export controls focused on the physical movement of chips. Chinese developers worked around that by renting Blackwell-class compute in third countries, with Thailand and Malaysia the most frequently named. The new provisions cover both of those countries.
The administration has debated the rules internally, and the fight has been loud. The outcome could cut into Nvidia's sales to foreign data centers that serve Chinese customers.
How the gap opened
The loophole has a paper trail. On May 31, the Commerce Department issued guidance to close a related gap that let overseas subsidiaries of Chinese companies buy top-end Nvidia and AMD chips without a license. That guidance did not require existing data centers to stop using the chips or cut off service.
That left rental arrangements untouched. Data centers in Southeast Asia and Japan could keep provisioning GPU access to Chinese AI companies without verifying who was ultimately using the compute.
Pressure built over the summer. A White House official said Moonshot AI had reached Nvidia's chips despite the export ban, and reports named ByteDance, Alibaba and Tencent as users of overseas capacity. Moonshot's Kimi K3 model became the example Washington cited.
What the rule asks of operators
The Bureau of Industry and Security (BIS) may offer exemptions for American-operated data centers in the affected countries. Those exemptions would carry conditions. Operators would need strict limits on remote access from arms-embargoed nations such as China and Russia, plus detailed reporting on customers and computing capabilities.
That structure puts the compliance burden on the cloud operators, not Nvidia. It also gives U.S. operators a path to keep serving legitimate customers in the region.
What it means for Nvidia
The immediate question for Nvidia and the cloud providers renting its chips is demand. If Chinese customers lose remote access, part of the compute rental business in Southeast Asia could shrink. Nvidia's direct China data center share is already near zero by the company's own account, so the exposure sits in the indirect channel.
That channel is hard to size. No public figure separates compute rented by Chinese firms from the rest of Southeast Asian demand. Investors will look for any quantification in Nvidia's next earnings commentary.
The whale flow on NVDA options is worth tracking through the week on the [Whale Alerts dashboard](/whalealerts). Positioning into a policy headline often shows up in premium before it shows up in price.
Legal and diplomatic friction
The rule faces a legal question. Export-control lawyers argue Commerce lacks clear statutory authority to regulate rentals of computing time. They also note the administration widened the gap itself by rescinding the Biden-era know-your-customer approach.
The China trip adds a second variable. Chip policy has moved with the diplomatic calendar before. In April 2025 the administration backed off a planned crackdown on Nvidia's H20 chip, and the reversal followed Nvidia's promise of new U.S. data center investment.
The next read comes with the final rule text and any exemption criteria. Watch whether BIS publishes a compliance deadline before the president lands in China.