US moves to close loophole letting Chinese firms rent AI compute abroad
Washington plans to require US licenses for Nvidia chips in third countries such as Singapore and Malaysia, targeting a route Chinese firms used to access cloud computing power. A Reuters investigation also describes how China is offsetting its hardware shortage by distilling US AI models.
The United States is tightening its export controls on advanced artificial intelligence chips in a move aimed at closing a route that allowed Chinese companies to access powerful computing capacity without owning the hardware. Data centers in third countries, including Singapore and Malaysia, will soon need a US license before they can buy Nvidia chips, according to the report on which this article is based.
The change targets a loophole that let Chinese firms rent cloud computing power from facilities located outside China. Because the chips sit in data centers abroad, they fell outside the direct reach of rules that restrict sales to Chinese buyers. Requiring licenses for purchases in third countries would bring that remote access under the same scrutiny that applies to direct exports, narrowing one of the main channels through which Chinese developers could train and run advanced AI systems.
The policy comes as Beijing faces a persistent shortage of high-end hardware. Export restrictions have limited China's access to the most capable Nvidia processors, which are central to training large AI models. The source material notes that this shortage has pushed Chinese developers to look for other ways to keep pace in the field.
A Reuters investigation cited in the report describes one such workaround. It reports that China is offsetting its hardware shortage through industrial-scale distillation of US AI models. Distillation is a technique in which a smaller or different model is trained to reproduce the outputs of a larger, more capable one. Applied at scale, it allows developers to capture much of the capability of American systems without access to the same chips used to build them.
Taken together, the two developments show the limits of relying on hardware controls alone. Restricting the chips and their physical location addresses one pathway, while the investigation suggests a parallel supply chain has formed around the outputs of US models themselves. The report leaves open how effective the new licensing requirement will be once it takes effect, and what additional measures might follow if distillation continues to grow.
For Washington, the challenge is to keep its technological lead without disrupting the broader market for chips and cloud services that many companies in Southeast Asia depend on. For Chinese firms, the combination of tighter access and alternative training methods will shape how quickly they can close the gap with US AI developers in the coming years.
Source: the-decoder.com
6
