House Panel Reveals 879,000 AI Chip Sales to China Were Approved
On 19 November 2026 the House Foreign Affairs Committee released export licence data showing the Commerce Department approved sales of more than 879,000 Nvidia H200 and comparable AMD AI chips to Chinese companies, out of 3.4 million requested between December and May. Approved licences are not deliveries: Nvidia says volumes were limited and that it has stopped shipping H200s to China.
On this page
What the committee released
On 19 November, House Foreign Affairs Committee chair Brian Mast used his authority under the Export Control Reform Act to make public licence application data for Nvidia and AMD AI chips bound for China. The headline figures: requests to sell 3.4 million AI chips, and approval of more than 879,000 of them by the Bureau of Industry and Security (BIS), the Commerce Department agency that licenses exports. The requests date from December through May.
The committee calls it the most detailed public disclosure in the history of that law. The act lets the panel release export data when it decides that withholding it is contrary to the national interest, and members voted to do so at a meeting on Thursday. Both parties backed the move: the committee's top Democrat, Gregory Meeks, said the public has a right to know how the decisions were made.
Who the buyers were
The approvals covered Nvidia's H200 and equivalent AMD chips. According to the committee, some went to companies the Pentagon lists as Chinese Military Companies, including Alibaba and Tencent. The released requests also name ByteDance, Tesla's Shanghai unit and Baidu.
The sharpest allegation concerns two little-known firms, Maginfra and New Matrix, which appear in Nvidia licence applications. Mast called both "fake companies" and suspected smuggling fronts, and said it would be of tremendous concern if BIS approved sales to them. Note the conditional: the committee's release describes requests that relied on possible smuggling fronts, not a confirmed approval for them. Neither firm replied to CNBC's requests for comment.
Why approved does not mean delivered
A licence is permission to sell, not a shipment. Nvidia told CNBC the review took more than half a year, that approvals came with strict conditions and limited volumes, and that it has since stopped shipping H200s to China. Analysts quoted by CNBC said sales stayed low largely because Beijing did not approve large-scale imports, and Nvidia has said H200 shipments to China were under 1% of data center revenue in the quarter that ended in July.
That is why Bernstein analyst Stacy Rasgon described 879,000 as "not that many" and noted that roughly one request in four was approved, which suggests a filter rather than a rubber stamp. BIS, for its part, said licensing decisions come out of consultation among several government agencies. Nvidia and AMD both said they comply with US export control law.
The sales were legal under the policy of the time. President Trump said last December that Nvidia could sell H200s to approved Chinese customers in return for a 25% cut for the US government, a decision that drew criticism from China hawks in both parties when the approvals followed.
What changes next, and why local AI users should care
The release lands during a turbulent stretch for Nvidia, after Jensen Huang's resignation and with lawmakers from both parties calling for stricter controls on advanced chip sales to China and other adversaries. Publishing the licence data is an oversight step, not a new rule, but it gives those lawmakers numbers to argue with.
For people who run models on their own hardware, the connection is indirect but real. Data center accelerators like the H200 are not what powers a laptop or a phone, yet the rules written for them decide which open weights models get trained, where, and by whom. Many of the strongest models you can download and run offline come from the same Chinese companies named in these licences. Tighter controls on their compute would shape the next generation of open models as surely as any release schedule.