JupiteX Get the app
International Relations22 Nov 2026 · about 7 min

Trump's administration is sending mixed signals on chips, China and AI

The brief

The administration is sending two different signals. It approved nearly 900,000 advanced Nvidia and AMD chips for Chinese companies over six months, while considering a policy to make US companies publicly disclose when they use Chinese AI models. The chip approvals loosened earlier restrictions; the proposed disclosure would discourage model use without directly banning it. The chip approvals included Nvidia H200 and AMD MI325X chips. The administration later announced restrictions on advanced-chip shipments to Malaysia, Singapore and Thailand, routes Chinese firms have used to access computing power. Separately, Airbnb uses Alibaba’s Nyaya model for its AI customer-service agent, and could have to disclose that use on its website under the proposal. The disclosure idea is still under consideration by the Treasury Department and the White House science office. The article says it is unclear whether it will gain support or become a formal rule. Meanwhile, Nvidia says it has stopped shipping H200 chips to China despite receiving some licenses, and the US says it will close export-control loopholes.

01

What mixed signals is the US administration sending about advanced chip sales to China and the use of Chinese AI models?

The administration is sending two different signals. It approved nearly 900,000 advanced Nvidia and AMD chips for Chinese companies over six months, while considering a policy to make US companies publicly disclose when they use Chinese AI models. The chip approvals loosened earlier restrictions; the proposed disclosure would discourage model use without directly banning it.

The chip approvals included Nvidia H200 and AMD MI325X chips. The administration later announced restrictions on advanced-chip shipments to Malaysia, Singapore and Thailand, routes Chinese firms have used to access computing power. Separately, Airbnb uses Alibaba’s Nyaya model for its AI customer-service agent, and could have to disclose that use on its website under the proposal.

The disclosure idea is still under consideration by the Treasury Department and the White House science office. The article says it is unclear whether it will gain support or become a formal rule. Meanwhile, Nvidia says it has stopped shipping H200 chips to China despite receiving some licenses, and the US says it will close export-control loopholes.

02

How many chip shipments to Chinese companies were approved, and how many chips had companies sought permission to ship?

The US government approved shipments of nearly 900,000 advanced Nvidia and AMD chips to Chinese companies between December 2025 and May 2026. That is the article’s overall figure for the six-month period. It does not establish how many approved chips were actually shipped to China.

The House foreign affairs committee gave figures for four firms: 75,000 chips for Alibaba, 150,000 for Tencent, 43,000 for Baidu and 4,000 for BYD. Together, those listed approvals total 272,000. The chips included Nvidia’s H200 and AMD’s equivalent MI325X. All four companies are on the Pentagon’s list of alleged Chinese military-backed companies.

Companies had sought permission to ship more than 3.4 million chips during the same period. That figure describes requests, not approved shipments. The committee documents did not make clear how many approvals resulted in chips actually reaching China, so approvals should not be treated as confirmed deliveries.

03

How did the approach to chip exports change from Biden’s restrictions to Trump’s approval of some H200 sales?

Under former president Joe Biden, the US restricted Chinese access to advanced chips. The stated concerns were that the chips could supercharge China’s military and erode the US lead in artificial intelligence. Those restrictions reflected a presumption against allowing high-end chip exports to China.

In December, Trump gave Nvidia the green light to ship its H200 chip to approved companies in China. The Commerce Department said licensing would be considered case by case under a January 2026 rule, with significant additional national-security requirements for applicants. That was not an unrestricted reopening of sales: Nvidia described the approvals as subject to strict conditions and limited volumes.

The shift drew criticism from China hawks across the US political spectrum. Nvidia said it had received some licenses but had since stopped shipping H200s to China. The administration has also announced further restrictions on chip shipments through Southeast Asia, showing that its approach combines some conditional approvals with efforts to block other routes to computing power.

04

How could restricting chip shipments through Southeast Asia make it harder for Chinese companies to access computing power?

The restriction targets advanced-chip shipments to Southeast Asian countries, including Malaysia, Singapore and Thailand. The US says Chinese companies have used these countries to access computing power for training AI models. The policy’s aim is to limit that indirect access, not just direct chip sales to China.

The key mechanism is the supply route: chips sent to a regional destination can support computing capacity that Chinese companies use. Limiting those shipments could reduce the availability of that capacity through the channels identified by the US. The article describes these destinations as routes used to access computing power, but does not detail particular transactions or facilities.

The administration said it would close what it called dangerous export-control loopholes that China had exploited for years. The rule makes access through these countries more difficult if advanced chip shipments are restricted. The article does not quantify the likely effect on Chinese AI training or say how companies will respond, so the scale of any disruption remains unclear.

05

What is an open-weight AI model, and how is it different from a model that companies keep proprietary?

An open-weight AI model makes its core components publicly available so people can download and adapt them. That access is the defining difference highlighted in the article. It can let users work with a model beyond simply using a company’s hosted service, though the article does not describe specific technical licensing terms.

A proprietary model, by contrast, generally keeps its model components under the developer’s control rather than making them available for public download and adaptation. Users may be able to use the service, but cannot freely obtain and modify the model’s internal components. This is a general distinction; the article does not name a specific proprietary model for comparison.

The difference matters in the policy debate because the US is considering ways to combat the spread of powerful Chinese open-weight models. One idea is to require US companies to disclose publicly when they use Chinese models. That proposal would target reliance and visibility, rather than immediately imposing a direct ban on such models.

06

Why might US companies choose Chinese AI models such as Alibaba’s Nyaya instead of tools from US companies?

The article does not say why US companies might choose Chinese AI models over US alternatives. It identifies a real example of use, but offers no explanation of Airbnb’s selection process or its reasons for using Alibaba’s Nyaya. Any specific claim about cost, quality or strategy would go beyond the source.

As general context, companies evaluating AI tools may consider how well a model performs a task, what it costs, how easily it can be integrated, and whether its features suit their needs. Those are possible decision factors, not motives the article attributes to Airbnb or any other firm. The source does not compare Nyaya with US models.

What the article does establish is that Airbnb uses Nyaya to power an AI customer-service agent. US officials have discussed requiring companies such as Airbnb to disclose their use of Chinese models on their websites. That proposed disclosure reflects concern about reliance on these models, but it does not itself explain why firms adopt them.

07

Why do advanced chips provide the computing power needed to train large AI models?

The article connects advanced chips with the computing power required to train AI models. Training involves processing large amounts of information and adjusting a model so it can produce useful outputs. Advanced chips can perform many mathematical operations quickly, making them important hardware for that work. This technical explanation is general background; the article does not give chip specifications or describe training methods.

The policy connection is explicit in the source: the US says Chinese companies have used access to computing power in Southeast Asia to train AI models. It has therefore announced restrictions on advanced-chip shipments to Malaysia, Singapore and Thailand, which it says Chinese companies have used as routes to that capacity.

US officials have also cited national-security concerns about high-end chip access. Earlier restrictions were justified partly by fears that chips could strengthen China’s military and erode the US lead in AI. The article does not measure how much training capacity a given chip provides or predict the effect of the newer restrictions.

This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.

Read more in the JupiteX app

Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.

Or read more news on the web