AI accelerator boards in protective trays with shipping containers and US and Chinese flags.

US Approved Nearly 900,000 AI Chips for China in Six Months, House Committee Reveals

Key Takeaways

  • The Trump administration approved shipments of nearly 900,000 advanced AI chips made by Nvidia and AMD for Chinese companies from December 2025 through May 2026, the House Foreign Affairs Committee said on Thursday 19 November.
  • The Commerce Department's approvals covered 879,000 chips, according to the committee's figures — the two numbers are both reported and are flagged here, not reconciled, because the reporting does not explain the difference between them.
  • The approvals included 75,000 chips for Alibaba, 150,000 for Tencent, 43,000 for Baidu and 4,000 for BYD — all four on the Pentagon's list of alleged Chinese military-backed companies, according to the committee.
  • Companies sought permission to ship 3.4 million AI chips during the same period, which means roughly a quarter of the chips requested were approved.
  • California has opened a probe into Nvidia over the China chip sales, according to the BBC, and Forbes has warned that the disclosures expose a wider AI threat.
  • The disclosure lands days after Jensen Huang's resignation as Nvidia's chief executive, and it reframes the export-control debate: the question is no longer whether advanced chips reach China, but how many, for whom, and under what rules.

The numbers: nearly 900,000, or 879,000?

The headline figure is nearly 900,000. The Trump administration approved shipments of nearly 900,000 advanced AI chips made by Nvidia and AMD for Chinese companies from December 2025 through May 2026, the House Foreign Affairs Committee said on Thursday, in what the committee's chair described as the most detailed public disclosure in the history of the Export Control Reform Act.

The detailed figure is 879,000. The Department of Commerce's approvals covered exports of 879,000 chips, the committee said, and the reporting that carries the detailed figure — Reuters, Al Jazeera and others — lists the same breakdown: 75,000 semiconductors for Alibaba, 150,000 for Tencent, 43,000 for Baidu, and 4,000 for BYD.

The gap between the two figures is about 21,000 chips, or roughly 2.4 percent. That is small in percentage terms, but it matters because the two figures are presented in the same reports without an explanation of what accounts for the difference. One possibility is that the "nearly 900,000" figure rounds up from a total that includes approvals outside the Commerce Department's count, or that it counts a different category of chip. Another is that the two figures come from different stages of the licensing process — approvals granted versus shipments authorised, for example. The reporting we reviewed does not say, and we are not going to guess.

What is clear is the scale. Companies sought permission to ship 3.4 million AI chips during December 2025 to May 2026, according to reporting on the committee's release. Against that, 879,000 approvals means roughly a quarter of the chips requested were approved. That is a substantial flow of advanced hardware to a country the United States has spent years trying to deny it, and it is the reason the disclosure has landed as a scandal rather than as a routine data release.

The chips in question are advanced AI chips — the kind used to train and run large AI models — made by Nvidia and AMD. Reuters has reported separately that the United States cleared about ten companies to buy Nvidia's H200 chip in May, but that no deliveries had taken place at that point, and that several major Chinese internet companies had each since obtained about 10,000 H200 chips. In testimony before the committee in July, Commerce Department official Jeffrey Kessler said shipments had begun but there had been "very few". The committee's release is the first detailed public account of the scale of the approvals behind those shipments.

For the background on Nvidia's leadership change, which frames this story, see our pieces on Jensen Huang's resignation as Nvidia chief executive and on Nvidia's earnings after Huang's resignation.

Who got the chips — and why the recipients matter

The breakdown of the approvals is as important as the total. The 879,000 chips included 75,000 for Alibaba, 150,000 for Tencent, 43,000 for Baidu and 4,000 for BYD. All four companies are on the Pentagon's list of alleged Chinese military-backed companies, according to the committee, and that fact is the heart of the political argument the disclosure has triggered.

Alibaba, Tencent and Baidu are China's largest internet companies, and they are also among the largest buyers of AI computing in the world. Their cloud businesses sell AI services to thousands of customers, and their own AI models compete directly with American systems. BYD is the world's largest maker of electric vehicles, and its interest in AI chips is tied to autonomous driving and robotics as much as to the data centre. The approvals mean that these companies — all of them designated by the Pentagon as alleged military-backed entities — were licensed to receive advanced American AI hardware in substantial quantities.

The committee's chair, Brian Mast, a Republican from Florida, said on releasing the data: "This is the most detailed public disclosure in the history of the Export Control Reform Act." The 2018 legislation expanded the Commerce Department's authority to restrict the export of sensitive technologies for national security reasons, and the chair's point is that the public has never before seen this level of detail about what was actually licensed. The disclosure was made under the chair's authorities under the Act to publicly reveal export-control licence application information regarding Nvidia and AMD's AI chip licences for China.

The political significance is that the disclosure cuts across the usual partisan lines. Export controls on AI chips to China have been a bipartisan project, pursued by successive administrations, and the committee that released the data is chaired by a Republican examining approvals granted by a Republican administration. The argument is not between parties but between two views of the same policy: that advanced AI chips are a strategic asset that should be denied to a geopolitical rival, and that they are a commercial product whose sale benefits American companies and the American economy.

California opens a probe into Nvidia

The federal disclosure was followed quickly by a state-level action. The BBC reported that California is targeting Nvidia in a probe into the China AI chip sales, and the probe adds a second front to the company's legal and political exposure.

California's interest is not abstract. Nvidia is headquartered in the state, and its chips are designed and, in many cases, sold from California. A state probe into whether the company's China sales complied with the relevant rules — and whether the company disclosed what it was required to disclose — is a different kind of risk from a congressional disclosure. A committee can embarrass a company; a state attorney general can bring an enforcement action.

The details of the California probe have not been fully reported, and this article does not invent them. What is known is that the probe exists, that it concerns the China chip sales, and that it was reported by the BBC on the same day as the committee's disclosure. The timing suggests that the committee's release has given state investigators a factual basis they did not previously have in public, and that Nvidia's China business is now under scrutiny from more than one direction at once.

The probe also lands at a moment of unusual vulnerability for the company. Jensen Huang has resigned as chief executive, and the company is in a leadership transition. A probe into the conduct of the business under the previous chief executive is a difficult inheritance for any successor, and it will complicate the new leadership's efforts to present Nvidia as a company focused on the future rather than on the controversies of the past.

The scandal framing: what Forbes says the disclosures expose

Forbes has framed the disclosures as a scandal that exposes a wider AI threat, and that framing deserves careful examination, because it goes beyond the question of whether the approvals were legal. The argument, as reported, is that the licensing data reveals a gap between the stated policy — denying China access to advanced AI hardware — and the practice, in which nearly 900,000 advanced chips were approved for Chinese companies in six months, including companies on the Pentagon's list.

The wider threat, on this reading, is not a single shipment or a single company. It is the cumulative effect of a licensing system that approves a quarter of what is requested, in a sector where the hardware in question is the limiting factor on AI capability. Advanced AI chips are not a commodity. They are the bottleneck: the thing that determines how fast a country can train frontier models, how many models it can run, and how quickly it can deploy AI across its economy and its military. If that bottleneck is leaking at a rate of nearly 900,000 chips in six months, then the policy is not doing what it was designed to do.

There is a counter-argument, and it is worth stating fairly. The chips approved were, in many cases, less advanced than the most powerful chips Nvidia sells, and the approvals were granted under a licensing system that is designed to distinguish between different uses, different customers and different levels of capability. The Commerce Department's job is not to block all sales to China; it is to block sales that would materially advance China's military AI capability while allowing sales that serve commercial purposes. On that view, 879,000 approvals out of 3.4 million requests is evidence that the system is working — that it is filtering, not failing.

The problem for that counter-argument is that the recipients include companies the Pentagon itself has designated as alleged military-backed entities. If the licensing system is meant to keep advanced chips away from the Chinese military, and it approves 150,000 chips for a company on the Pentagon's list, then either the list and the licensing system are using different definitions of risk, or the system is not applying the list's logic. The committee's disclosure does not resolve that tension; it exposes it, and that is why the Forbes framing has gained traction.

What the chips are for — and why the number matters

To understand why 879,000 chips is a significant number, it helps to understand what an advanced AI chip does. A single high-end AI chip can cost tens of thousands of dollars, and a cluster of thousands of them is what trains a frontier model. The largest AI training runs now use tens of thousands of chips working together for months. A company that receives 150,000 chips — the number approved for Tencent — has received enough hardware to train multiple frontier-scale models and to run them at scale.

That is the strategic significance of the approvals. They are not a marginal flow. They are a flow large enough to matter to the balance of AI capability between the United States and China, and large enough to raise the question of whether the export-control regime is achieving its stated purpose.

The timing also matters. The approvals cover December 2025 through May 2026, a period in which the AI race was intensifying and in which the United States was publicly tightening its controls. The disclosure shows that, during that same period, the licensing system was approving substantial quantities of advanced hardware for the very companies the controls were meant to constrain. That contradiction is the story, and it is a story that will not be resolved by a single committee release. It will be resolved, if it is resolved at all, by the policy changes that follow.

The Huang context: a resignation and a reckoning

The chip disclosure cannot be separated from the leadership change at Nvidia. Jensen Huang's resignation as chief executive, which we covered in our pieces linked above, removed the person who had built Nvidia into the most valuable company in the AI boom and who had also been the public face of its China strategy. Huang had argued consistently that restricting chip sales to China would harm American companies and push Chinese customers towards domestic alternatives, and he had lobbied against tighter controls. His resignation does not settle that argument, but it removes its most effective advocate at precisely the moment the licensing data has given its critics their strongest evidence.

The new leadership inherits a company that is simultaneously the most important supplier of AI hardware in the world and the subject of a federal disclosure and a state probe over its China sales. That is a difficult position, and it will shape the company's strategy in ways that go beyond the choice of a new chief executive. Nvidia's future depends on its ability to sell advanced chips to customers around the world, including in China, while also maintaining the confidence of the United States government that its sales do not undermine national security. The committee's disclosure has made that balancing act harder, because it has put the numbers in public.

What happens next

Several things will determine how this story develops. The first is whether the committee releases more data. The chair described the release as the most detailed public disclosure in the history of the Export Control Reform Act, which suggests that more detail may follow, and that the committee regards the disclosure as the beginning of an inquiry rather than its end.

The second is the California probe. State investigations move at their own pace, and the BBC's report does not say what stage the probe has reached or what it is likely to conclude. But a state probe into a company of Nvidia's size and importance is not a routine matter, and its outcome — whether it results in enforcement, a settlement, or nothing — will matter.

The third is the policy response. The disclosure will strengthen the hand of those in Washington who want tighter controls, and it will put pressure on the Commerce Department to explain how the approvals were granted and whether the licensing criteria will change. If the criteria do change, the effect on Nvidia and AMD's China revenues could be substantial, and the effect on Chinese AI development could be equally substantial.

The fourth is the commercial response. Chinese companies that have received approvals for nearly 900,000 chips now have hardware that will shape their AI capabilities for years. Whether they can obtain more, and on what terms, will depend on the policy that follows the disclosure.

The wider picture: controls, commerce and capability

The export-control regime for AI chips was built on a simple idea: that the United States could slow China's AI development by denying it the hardware on which AI depends. The committee's disclosure shows that the idea is harder to implement than it is to state. A licensing system that approves a quarter of what is requested, including for companies on the Pentagon's list, is a system that is making fine distinctions in a domain where the distinctions are hard to draw and harder to enforce.

The debate that follows will not be about whether 879,000 or nearly 900,000 is the right number. It will be about whether the United States wants to sell advanced AI hardware to China at all, and if it does, under what conditions and to whom. That is a question about the future of the AI industry, the future of the US–China relationship, and the future of the companies — Nvidia and AMD among them — that sit at the centre of both. The committee's disclosure has put the numbers on the table. The argument about what they mean is only beginning.

What We Could Not Verify

  • The difference between "nearly 900,000" and 879,000. Both figures are reported from the committee's release. The reporting we reviewed does not explain the difference, and we have not reconciled the two.
  • The precise chip models approved. The reporting describes advanced AI chips from Nvidia and AMD, and refers separately to the H200, but we have not verified the full breakdown by model.
  • Whether the approved chips have been shipped. Reuters reported that, as of May, about ten companies had been cleared to buy the H200 but no deliveries had taken place, and that several major Chinese internet companies had each since obtained about 10,000 H200 chips. The Commerce Department official Jeffrey Kessler testified in July that shipments had begun but there had been "very few". The full shipment picture has not been verified.
  • The scope of the California probe. The BBC reported that California is targeting Nvidia in a probe into the China chip sales. The specific allegations, the stage of the probe and its likely outcome have not been verified.
  • The Forbes "wider threat" argument in full. Forbes has warned that the disclosures expose a wider AI threat. The full argument, and the evidence behind it, were not verified in the reporting we reviewed.
  • Individual company responses. We have not verified responses from Nvidia, AMD, Alibaba, Tencent, Baidu or BYD to the committee's disclosure.

Sources: Reuters; Al Jazeera; BBC; Forbes; Channel News Asia; House Foreign Affairs Committee statement as reported on 19–20 November 2026.

Similar Posts