US to require licenses for Nvidia chips in third countries; China distills US models
The US is moving to require a license for Nvidia chip purchases by data centers in third countries such as Singapore and Malaysia, closing a route that let Chinese firms rent AI compute abroad rather than buy restricted hardware directly, The Decoder reports. The measure is described only as coming “soon”: no rule text, effective date, licensing authority or scope beyond those example countries is given, so it is a stated plan rather than an active requirement. Separately, the outlet cites a Reuters investigation finding that China is offsetting its hardware shortfall through industrial-scale distillation — training domestic models on the outputs of US AI models — to build a parallel supply chain out of American AI output.
Verified and unconfirmed
- Verified: The Decoder reports that Washington is tightening chip export controls so that data centers in third countries such as Singapore and Malaysia will soon need a US license to buy Nvidia chips, closing a route Chinese firms used to rent cloud compute abroad, and that a Reuters investigation described China offsetting its hardware shortage by distilling US AI models. US officials say they distinguish legitimate distillation from large-scale extraction they characterize as intellectual property theft (tool-3-1).
- Not yet confirmed: The article gives no rule text, agency, license criteria or effective date, so the timing and scope of the licensing requirement rest on secondary reporting. The “industrial-scale” distillation claim is attributed to the Reuters investigation and US officials rather than independent measurement, with US intelligence reportedly naming DeepSeek, Alibaba, Moonshot AI and three other firms (tool-3-2) — an accusation China’s Commerce Ministry rejects and has threatened to counter (tool-3-3). No figures quantifying how much Chinese model capability derives from US outputs are provided.
Background
US export controls have until now restricted physical shipments of advanced Nvidia chips to China, leaving a gap for chips installed in third-country data centers whose compute could be rented back to Chinese firms. AI Daily Radar’s November 15 digest cited a Wall Street Journal report that Nvidia chips were reaching Chinese companies through large Southeast Asian data centers that exploited that gap. The following day’s digest reported that the Commerce Department had issued an interim rule requiring export licenses when AI chips may provide compute to China or other restricted countries, with reporting naming Singapore, Malaysia, Thailand, Indonesia and Brazil.
Why it matters
Data-center and cloud operators in Singapore, Malaysia and similar jurisdictions would have to clear a US licensing step before buying Nvidia chips, which adds compliance work for capacity that serves Chinese customers; those customers would then face delivery delays or have to shift workloads to licensed or domestic supply. A licensing regime aimed at hardware does not touch the distillation route, which consumes model outputs rather than chips.
What to watch
Watch for the formal rule text and its effective date, and for whether the license requirement is extended beyond Singapore and Malaysia to other transshipment and hosting hubs.
Further reading
- 2026-11-16 — US Commerce tightens AI chip export rules to cover remote compute access
- 2026-11-15 — WSJ: Nvidia chips reached Chinese firms via SE Asian data centers
- China accuses US of ‘ AI hegemonism’, threatens countermeasures…
- Chinese AI Firms Accused of Stealing US Frontier Models
- China Rejects US AI Distillation Claims: 6 Firms Named
Comments
Comments use a GitHub account and are moderated. Comment rules