Chinese AI developers have lost two important sources of computing power in the space of two days. On 16 November the US Commerce Department released a rule aimed at the overseas data centres that rent advanced American chips to Chinese companies. The next day Nvidia said it had stopped shipping the licensed chips it had been selling into China. Tencent, one of China's largest AI spenders, has already said the result will be slower progress on its next models.
What the new rule does
The US has restricted sales of its most powerful AI chips to China since 2022. Chinese firms have partly worked around this by renting computing power in data centres in other countries and accessing it remotely. The new rule, published in the Federal Register on 19 November and in force from that date, is designed to close that route.
It requires a US licence to export advanced chips and servers for use in a data centre anywhere in the world if that facility will give companies headquartered in China, Macau or other countries of concern access to more than 12 million units of "aggregate TPP", a measure of total computing performance. Data centres that are linked over networks count together, so a cluster cannot dodge the threshold by being split across sites. Chip exporters, data centre operators and foreign chip factories must check who ultimately owns or controls their customers. Commercial recommender systems, such as those used by online shops and social media, are exempt.
The rule also adds a new licence requirement for advanced chips going to Indonesia, Malaysia, Singapore, Thailand and Brazil, which the US sees as higher-risk routes for diversion, with an exception for companies headquartered in the US that meet reporting rules. It formally scraps the Biden-era "AI Diffusion Rule", which was never enforced. Shipments already in transit can proceed under the old terms if completed by 21 December, and public comments are open until 19 January 2027.
The worldwide requirement applies to qualifying data centres in any country, including the UK, although the UK is not one of the five countries singled out for extra licensing.
Nvidia stops shipping Hopper chips
Separately, Nvidia told investors on 17 November that it had stopped selling licensed Hopper-generation data centre chips, a family that includes the H200, to China. "Our guidance does not have any licenced Hopper data centre products included, and we are no longer shipping licenced Hopper data centre products to China," interim chief executive Colette Kress said, according to Reuters. She did not say why or whether the halt is permanent.
Reuters reported that Nvidia sold $2.4 billion of these licensed chips to China in the quarter to 25 October. The sales followed President Trump's decision, after a summit with Xi Jinping, to allow H200 exports to approved Chinese customers, with the US easing its licensing policy in January. Washington had approved H200 sales to 10 Chinese companies, including Alibaba, Tencent and ByteDance. Beijing, keen to cut reliance on US technology, has itself held back large imports: Reuters reported earlier this month that around 600,000 H200 chips were sitting in Hong Kong awaiting approval.
The announcement came days after Jensen Huang stepped down as Nvidia's chief executive amid allegations about his personal conduct, including an alleged affair with a suspected Chinese spy.
Tencent warns of a short-term setback
Tencent president Martin Lau addressed the rule on the company's results call on 18 November. "It sets everyone back in China," he told analysts, saying it would take Tencent longer to produce the next version of its models in the short run, according to Lianhe Zaobao, citing Bloomberg and HK01. He added that Tencent could still build its own overseas capacity for work that does not need graphics processors, but could no longer provide high-end GPUs that way.
Lau argued the long-term effect would be to push Chinese companies onto domestic chips and supply chains. Tencent's third-quarter revenue rose 8% to 207.3 billion yuan, while operating profit fell 7%, the paper reported.
What it means
For China's AI labs, especially smaller ones without their own chip programmes, computing power has just become scarcer and more expensive at the moment they are racing to match US models. For businesses outside China, the rule mainly affects data centre operators and cloud providers, who now have more due diligence to do on who their customers are. For anyone choosing between AI models, it is a reminder that a model's roadmap can depend as much on geopolitics as on research.
FAQs
What is the new US 'remote access' chip rule?
It is an interim final rule from the US Commerce Department's Bureau of Industry and Security, effective 19 November 2026. It requires licences for advanced chips used in data centres anywhere that provide large amounts of computing to companies headquartered in China, Macau or other countries of concern, and adds licence requirements for Indonesia, Malaysia, Singapore, Thailand and Brazil.
Has Nvidia stopped selling chips to China altogether?
Nvidia says it is no longer shipping its licensed Hopper-generation data centre chips, including the H200, to China, and its forecasts assume no such sales. It did not say whether the halt is permanent.
Does the rule affect UK data centres?
The worldwide licence requirement applies to qualifying data centres in any country, including the UK, if they supply large amounts of computing to China-headquartered companies. The UK is not one of the five countries given extra licence requirements.
Sources
- Federal Register: Addressing Remote Access to Advanced Semiconductors, Bureau of Industry and Security (19 Nov 2026)
- Reuters via Investing.com: Nvidia stops shipping US-approved AI chips to China (18 Nov 2026)
- Lianhe Zaobao: Tencent president says new US AI chip controls will slow China's model development in short term (19 Nov 2026)
Follow AI Decoded on Substack
Weekly plain-English AI news and verified UK events.
Subscribe