米国、中国企業向け Nvidia 最高級チップの海外販売ループホールを閉鎖へ

*New Commerce Department guidance ties export-licence rules to where a company is headquartered, not where it sits, snaring the overseas units of Chinese AI firms.*

For about a year, there was a way around America’s toughest chip controls, and it was a matter of geography. A Chinese AI company barred from buying Nvidia’s best processors at home could, in principle, have a subsidiary in a country like Malaysia buy them instead. On Sunday the US Commerce Department moved to shut that door.

The department issued guidance, posted to its website, extending export-licence requirements to advanced chips sold to any entity headquartered in China, regardless of where that entity is physically located.

The shift is subtle but consequential: the control now follows the parent company’s nationality rather than the address on the loading dock, which is precisely the seam that overseas subsidiaries had been operating in.

The chips at stake are the most capable on the market, including Nvidia’s Rubin and Blackwell processors and AMD’s MI350x. The scale of what may have slipped through is striking.

One industry source with deep supply-chain knowledge estimated to Reuters that hundreds of thousands of advanced chips may have reached Chinese-linked entities abroad during the window the loophole was open.

That window traces to a specific decision. In the last days of the Biden administration, the Commerce Department finalised the so-called AI Diffusion rule, a sweeping framework for governing where advanced chips could go.

In May 2025, the Trump administration said it would not enforce that rule, and the practical effect, on this reading, was to leave the overseas subsidiaries of Chinese firms in an ambiguous position for almost a year. The new guidance closes the ambiguity.

It stops short of the most disruptive option. The guidance does not require data centres already running the chips to stop using them, nor does it cut off servicing of advanced computing equipment such as servers.

The action is aimed at future flows, not at clawing back hardware that has already shipped, which limits the immediate operational shock while tightening the tap going forward.

The move fits a pattern of leakage and patching that has defined US chip policy. Washington has restricted China’s access to advanced chips since 2022 and widened the rules repeatedly, yet enforcement keeps running into workarounds, from third-country subsidiaries to outright smuggling.

US prosecutors have separately pursued a case alleging that a Thai company helped route Nvidia chips to Alibaba, a reminder that controls written in Washington are only as strong as their weakest border.

Nvidia and AMD did not immediately respond to requests for comment, and the companies are caught in a familiar bind: China remains a large potential market, and tighter rules narrow it further.

For Beijing, the closure removes one of the cleaner legal routes to frontier silicon, leaving it leaning harder on stockpiles, domestic chips, and the murkier channels Washington is still chasing. The harder part, as ever, is enforcement: a guidance document redraws the line, but it does not police it.