In the space of four days, the United States gave two contradictory answers to the same question: how many advanced AI chips should China be allowed to buy? On 19 November 2026, the House Foreign Affairs Committee disclosed that the Commerce Department had approved 879,868 advanced AI chips for Chinese buyers between December 2025 and 11 May 2026. Three days earlier, the same department had issued a rule designed to stop Chinese companies reaching those chips through data centers abroad. And on 17 November, Nvidia said it had stopped shipping its licensed H200 chips to China altogether.
The three events do not add up to a single policy. They show a licensing system, a new rule and a supplier moving in different directions at the same time, and the gap between them is where the next phase of US AI chip exports to China will be fought.
Key takeaways
- The Commerce Department approved 879,868 of the 3,444,160 AI chips requested for China, about a quarter of the total, including 150,000 for Tencent and 75,000 for Alibaba.
- All four named recipients sit on the Pentagon list of Chinese military companies, and it is not clear how many of the approved chips were actually shipped.
- A 16 November interim final rule requires licenses for advanced chip exports to five Southeast Asian countries and for data centers giving Chinese firms remote access to them.
- Nvidia has stopped shipping licensed Hopper chips to China and assumes zero China data center revenue in its fourth-quarter guidance.
- Analysts at Jefferies estimate the new rules could cut Nvidia data center revenue by 44 to 62 billion dollars next year.
US AI chip exports to China: what the House committee disclosed
The disclosure came from the chairman of the House Foreign Affairs Committee, Brian Mast, who used his authority under the Export Control Reform Act to publish export licensing information that is normally kept confidential. The committee called it the most detailed public disclosure in the history of that law. The figures cover applications from Nvidia and AMD, and the approved chips are Nvidia H200 processors and AMD MI325X processors, the two most capable AI accelerators either company is permitted to sell to China.
The committee published the following allocations:
| Company | Chips approved |
|---|---|
| Tencent | 150,000 |
| Alibaba | 75,000 |
| Baidu | 43,000 |
| BYD | 4,000 |
| Total for the four named companies | 272,000 |
All four companies appear on the Pentagon list of Chinese companies alleged to have links with the military, according to the committee. That is the detail that drew the sharpest reaction. Chris McGuire, a former US official and expert on the US-China technology contest, said it was astonishing that more than 30 percent of the approved chips went to companies the Department of War has designated as Chinese military companies, and questioned whether Commerce had conducted any serious national security review before approving the licenses.
The committee also said that Nvidia applications had listed companies suspected of serving as smuggling fronts to move chips into China. That claim comes from the committee, and the article reports it as the committee’s statement, not as an established fact.
Approvals are not shipments, and the disclosure does not settle how many of the 879,868 chips actually reached China. A Commerce Department summary cited in the disclosure said none of the approved shipments had been sent at the time it was written. By July 2026, however, a Commerce official told Congress that shipments had begun, although volumes remained very low, and several large Chinese internet companies had each obtained about 10,000 H200 chips. Around 600,000 H200 chips were reported by Reuters to be held in Hong Kong, awaiting permission to enter mainland China, and Beijing has itself restricted H200 imports as it pushes Chinese companies towards domestic processors.
Nvidia said the approvals were subject to strict conditions and limited volumes, and added that despite receiving some licenses, it has since stopped shipping H200 chips to China. AMD said it complies with applicable US export control laws and maintains rigorous processes to ensure compliance across its operations.
How the H200 got approved in the first place
The approvals sit on top of a policy reversal. The United States introduced export controls on advanced AI chips in 2022 to restrict Chinese access to processors that could support advanced AI development and military applications. President Donald Trump changed course in December 2025 after meeting Chinese President Xi Jinping, saying Nvidia would be allowed to sell H200 processors to approved Chinese customers.
In January 2026, the Commerce Department’s Bureau of Industry and Security formally moved to reviewing H200 applications on a case-by-case basis, replacing a presumption of denial. A final rule published in the Federal Register on 15 January 2026 set conditions: exports could not reduce semiconductor production capacity available to American customers, shipments to China and Macau could not exceed 50 percent of the quantity shipped to US customers for domestic use, Chinese customers had to follow export compliance procedures, chips had to undergo third-party testing in the United States, and eligible H200 chips imported into the United States before shipment to China faced a 25 percent tariff.
Jeffrey Kessler, the Under Secretary for Industry and Security, defended the approach in January, saying export controls should evolve with changes in technology while protecting national security, and that permitting H200 sales under controlled conditions would strengthen the American technology ecosystem. The House disclosure is the first public accounting of US AI chip exports to China under that case-by-case system.
The rule that closed the loophole
The interim final rule issued on Monday 16 November 2026 targets a different route to the same chips. US export controls generally stop Chinese companies buying Nvidia’s most advanced chips directly, but some had been able to rent computing power on those chips from data centers in Southeast Asia instead. The new rule requires licenses for advanced chip exports to Indonesia, Malaysia, Singapore, Thailand and Brazil, and requires data centers to obtain a license before providing remote access to companies headquartered in China or other restricted destinations. Data centers already serving those companies have 90 days from 19 November, when the rule was published in the Federal Register, to obtain a license. The rule applies where a data center supplies computing power above a threshold of about 12 million TPP, and it also formally rescinds the Biden administration’s AI Diffusion Rule, which the Trump administration had already said it would not enforce.
Nextgov described the rule as the administration’s first policy response to the scandal around Nvidia, which included a board investigation and reports about a Singapore cloud provider’s chip purchases. Senator Maria Cantwell had written to Commerce Secretary Howard Lutnick the day before the rule appeared, calling for immediate action to stop the flow of advanced AI chips to companies with ties to China. The Commerce Department said the rule was written by and at the direction of the Department of the Treasury and the National Security Council. Comments on the interim final rule are due within 60 days.
Nvidia stops shipping anyway
The rule landed the day before Nvidia reported its third-quarter results, and the company’s own decision went further than anything Washington had ordered. On the 17 November earnings call, interim chief executive Colette Kress said: “Our guidance does not have any licensed Hopper data center products included, and we are no longer shipping licensed Hopper data center products to China.”
The halt came despite 2.4 billion dollars of licensed Hopper product sales to China in the quarter ended 25 October 2026. That figure covers the whole licensed Hopper family, not H200 chips alone, since Nvidia did not disclose how much of it came from that processor. Nvidia has not publicly explained the reason for the halt or said whether it is temporary or permanent, and the halt should not be presented as permanent.
The decision sits oddly beside the company’s results. Nvidia reported third-quarter revenue of 115.8 billion dollars, up 103 percent from a year earlier, with data center revenue of 107.6 billion dollars. Fourth-quarter guidance of 127.4 to 132.6 billion dollars assumes no data center compute revenue from China at all. Kress told analysts that strong demand for next-generation products, including Vera Rubin, would offset the impact. For the full picture of those results, see our report on NVIDIA Q3 earnings 2026. The board investigation that preceded the leadership change is covered in our report on the Nvidia board investigation into Jensen Huang.
The market fallout so far
The disclosure moved markets on the day it was published. Nvidia shares ended 19 November down 3.3 percent, AMD fell 7.7 percent, and the Nasdaq and the S and P 500 fell 2.2 percent and 1.8 percent respectively.
The larger question is what the new rules cost Nvidia beyond the H200 business. Jefferies estimates the restrictions could reduce Nvidia data center revenue by 44 to 62 billion dollars next year, according to TechCrunch as reported by Zacks. That is an analyst estimate, not a company figure, and it is far larger than the H200 trade itself: at last quarter’s pace, licensed Hopper sales to China were worth about 2.4 billion dollars a quarter. The Jefferies figure reflects the remote-access rule, which reaches the much bigger business of renting American computing power to Chinese customers through data centers in Southeast Asia. Kress has said the company needs more time to study the new rules before judging their impact.
The first concrete casualty has already appeared. Moonshot AI, the Beijing company behind the Kimi family of models, has cancelled a contract worth about 150 million dollars to rent Nvidia Blackwell Ultra B300 chips through a data center in Indonesia, according to The Information, citing two people with direct knowledge. The deal would have given Moonshot access to some of Nvidia’s most powerful accelerators, which the US bars from direct sale to China, to train its next model, Kimi K4. There is no public confirmation from Moonshot, and separate reporting says the company has already signed a replacement contract with another Southeast Asian provider. Our guide to the company’s current model is the Kimi K3.1 Pro open-weight model guide.
What happens next
Three clocks are now running. Data center operators serving Chinese customers have until about February 2027 to obtain licenses, and the 60-day comment period on the interim final rule will draw submissions from the cloud providers and chipmakers whose business it reshapes. The second clock is the licensing queue itself: the rule does not ban remote access outright, so the Bureau of Industry and Security will have to decide, license by license, whether the new system is a filter or a wall.
The third is the overhang. The United States approved 879,868 chips under a policy that treated each sale on its merits, and it is still unclear how many of those chips were shipped, how many remain in warehouses or in Hong Kong, and whether Nvidia’s halt means the licenses simply expire unused. The contradiction at the heart of the week, approving chips with one hand and closing the routes to them with the other, means the future of US AI chip exports to China will only be settled when those numbers become public too.
Frequently asked questions
How many AI chips did the US approve for China?
The Commerce Department approved 879,868 advanced AI chips for Chinese buyers between December 2025 and 11 May 2026, out of 3,444,160 requested. The chips were Nvidia H200 processors and AMD MI325X processors.
Did all 879,868 approved chips reach China?
It is not clear. An approval is permission to ship, not a shipment. A Commerce summary cited in the disclosure said none of the approved shipments had been sent when it was written, but by July 2026 a Commerce official told Congress shipments had begun at low volumes, and about 600,000 H200 chips were reported held in Hong Kong awaiting permission to enter mainland China.
What does the new rule actually require?
The interim final rule of 16 November 2026 requires licenses for advanced chip exports to Indonesia, Malaysia, Singapore, Thailand and Brazil, and requires data centers to get a license before giving remote access to companies headquartered in China or other restricted destinations. Data centers already serving those customers have 90 days from 19 November to obtain a license.
Has Nvidia stopped selling chips to China?
Nvidia has stopped shipping licensed Hopper data center products to China and its fourth-quarter guidance assumes no China data center compute revenue. The company has given no reason for the halt and no date for resuming, so the halt should not be described as permanent.
Sources and Further Reading
- House Foreign Affairs Committee, Chairman Brian Mast Exercises Legislative Authority to Publicly Release AI Chip License Information for China, 19 November 2026
- Al Jazeera, US signed off on sales of nearly 900,000 AI chips to China in six months, 19 November 2026
- Business Standard, Trump admin cleared over 879k AI chip sales to Chinese firms: House panel, 20 November 2026
- Nextgov, Commerce Department moves to close AI chip loophole, 16 November 2026
- Financial Express, Has Nvidia stopped US-approved H200 shipments to China despite 2.4 billion dollars in sales?, 18 November 2026
- Zacks, Can NVIDIA’s Q3 Beat and Solid View Offset China Curbs and Huang’s Exit?, 18 November 2026
- ForkLog, Moonshot AI Cancels 150M Nvidia Chip Leasing Deal, 20 November 2026
