Nvidia has stopped shipping licensed Hopper data-center accelerators to China. Interim CEO Colette Kress disclosed the halt on the November 17, 2026 earnings call and identified the H200 as the licensed chip in a CNBC interview that evening. For AI infrastructure buyers in China, earlier U.S. export approval offers no assurance of another delivery. Kress did not give a reason for the halt or say how long it would last. The Verge reported the development on November 18.
The Hopper shipments Nvidia’s forecast left out
Nvidia reported US$2.4 billion in Hopper data-center shipments to China in the fiscal quarter ended October 25. That figure measures shipments in dollars. Nvidia did not provide a unit count or say which H200 variants were shipped. The previous quarter's filing had already disclosed limited licensed shipments representing less than 1% of Nvidia's data-center revenue for that period.
Nvidia's August outlook for fiscal Q3 already assumed zero data-center compute revenue from China, even though it went on to report those Hopper shipments. Its Q4 outlook of US$130 billion, plus or minus 3%, carries the same assumption. That tells buyers how Nvidia built its forecast. The separate statement from Kress is what establishes that shipments have stopped.
U.S. licenses covered specific buyers
In January 2026, the U.S. Bureau of Industry and Security changed its licensing policy to review applications for H200-class chips case by case, subject to security and supply conditions. The criteria included independent testing in the United States and evidence that exports would not reduce manufacturing capacity available to U.S. customers. Nvidia's August filing says U.S. authorities started issuing licenses in February for small quantities of H200 products to specific China-based customers. Nvidia also said Chinese authorities restricted those sales and that it had been unable to sell all the products covered by its licenses. U.S. authorization opened a route for specific orders. Chinese restrictions and actual shipments still determined how much buyers could receive.
What Chinese data-center buyers can count on
The H200 is a data-center accelerator designed for workloads such as AI model training and inference. Its 141 GB of high-bandwidth memory gives a model and its working data more room on each accelerator. A capacity plan that depends on this hardware needs to distinguish systems already installed from units still on order. The earlier shipments show that deliveries took place, but give buyers no reliable timetable for more. For a new deployment that depends on H200 capacity, the relevant evidence would be a confirmed delivery commitment and the applicable approvals for that order.
Kress told CNBC that Nvidia's gaming, automotive and workstation shipments to China continued. That leaves other product lines in a different position from the licensed data-center chips. For the leadership change that preceded the call, see TechKili's coverage of Kress's interim appointment.