
The US Commerce Department is moving to limit remote access to advanced artificial-intelligence chips, targeting a route by which Chinese companies can use restricted Nvidia hardware without owning it. The step comes as scrutiny of semiconductor exports intensifies after an alleged diversion scandal involving Nvidia’s customer Megaspeed, a Singapore-based data center operator.
The draft would require licences for shipments of top-end processors to additional Southeast Asian countries, among them Malaysia, Singapore, Thailand and Indonesia, and would bar Chinese entities from renting the compute once it is installed there. Final text had not been published in the Federal Register as of Monday, and the scope of the licensing requirement, including any performance thresholds, has not been confirmed.
The move targets a gap in the existing framework. US export controls have focused on the physical chip, while remote use of hardware already sitting in a foreign data center sits in a legal gray area. A January 2026 rule from the Bureau of Industry and Security already reaches part of the problem: cloud providers and colocation operators offering access to controlled compute face conditions that include collecting know-your-customer information on ultimate users.
Legislators are pushing from a second direction. The House passed the Remote Access Security Act 369-22 in January. It would widen the legal definition of an export to cover cloud-based access to restricted hardware, and it still needs to clear the Senate. Several Republican lawmakers have separately demanded that Commerce Secretary Howard Lutnick explain why Megaspeed was able to obtain frontier chips.
Records cited in connection with the Megaspeed case indicate the company arranged roughly $12 billion of Nvidia chips for China between March 2024 and June 2025, though how much was actually delivered is unclear. Megaspeed says it operates lawfully, complies with export controls and has opened an internal investigation. Multiple federal agencies have been said to favor adding it to the Entity List, and no such listing has been made.
The policy shift lands in a week when Nvidia faces its first earnings report since Jensen Huang’s resignation as chief executive, and with US stock futures lower ahead of the results and Anthropic’s initial public offering.
Some trade lawyers argue that Commerce lacks clear statutory authority to police remote access until Congress acts, which would leave a rule relying on licence conditions and customer checks open to legal challenge. A narrower rule aimed only at listed entities would also differ materially from one that obliges every cloud operator to vet customers, and the final design has not been settled.
One carve-out has been described for US companies operating data centers abroad: they would not need individual export licences if they put technical barriers in place against access from China, Russia and other embargoed jurisdictions. US cloud operators would face revised reporting duties, including disclosing customers from restricted jurisdictions and the total compute they lease.
Washington has been tightening in stages. The Biden-era AI diffusion rule was set aside in May 2025 on the administration’s announcement that it would replace it, and the gap that followed let entities with overseas subsidiaries in places such as Malaysia obtain Blackwell-class chips without licences. Guidance issued earlier this year applies only to future exports.
Sources: Bloomberg, mayerbrown.com, aiweekly.co.









