Reuters reported on November 5, 2026, citing two people familiar with the matter, that about 600,000 NVIDIA H200 artificial intelligence chips are stored in Hong Kong awaiting shipment to mainland China, as Beijing withholds approval for large-scale imports despite demand from Chinese technology firms.
Three people familiar with the matter said China had allowed some H200s into the country in recent months. One said several major Chinese internet companies had each obtained about 10,000 units over the past few months. Three sources also described small shipments through intermediaries and indirect routes, without specifying volumes.
The report links Beijing’s import constraints to concerns about dependence on U.S. technology. China’s Ministry of Industry and Information Technology and the National Development and Reform Commission did not respond to requests for comment; neither did NVIDIA, Alibaba, ByteDance, DeepSeek or Tencent.
H200 Utility and Unofficial Prices
NVIDIA has newer AI processors, but the H200 remains useful for training AI models, especially when paired with NVIDIA’s software. The report describes capabilities that Chinese alternatives have yet to match, helping explain why Chinese companies continue pursuing the chips under tight supply conditions.
One person familiar with the matter said an eight-card H200 server was offered in mainland China for more than 6 million yuan by the end of October 2026, about three times its price a year earlier. Servers using newer Blackwell chips, which are subject to U.S. export controls, were offered for more than 20 million yuan each. Those offers illustrate the premium in an unofficial market shaped by constrained authorized supply and indirect purchasing routes.
U.S. Export License Conditions
On January 13, 2026, the U.S. Department of Commerce’s Bureau of Industry and Security issued a rule requiring case-by-case review of export-license applications for the H200, AMD MI325X and similar chips destined for China when specified security requirements are met. It followed President Donald Trump’s December 8, 2025 announcement that the United States would allow those products to be shipped to approved Chinese customers.
Applicants must show that the exports will not reduce global semiconductor production capacity available to U.S. customers. The Chinese purchaser must have export-compliance procedures that include customer screening, and the product must have undergone independent third-party testing in the United States to verify its performance and security.
NVIDIA’s Disclosed Commercial Constraints
In its quarterly report for the period ended July 26, 2026, NVIDIA disclosed that U.S. licenses beginning in February 2026 allowed small amounts of H200 products to be shipped to specific China-based customers. Chinese government restrictions nevertheless left it unable to sell all the products covered by those licenses. The company recorded a $0.4 billion charge for H200 excess inventory and purchase obligations during the first half of fiscal 2027 as demand diminished. It subsequently made only a fraction of the permitted shipments, which represented less than 1% of Data Center revenue in the quarter covered by the report.
The licenses require H200s to undergo inspection in the United States before shipment to customers. That process makes the chips subject to a 25% tariff when they are imported into the United States. NVIDIA said it had been unable to pass any of the tariff to customers and did not expect to do so if it becomes able to sell licensed products into China.
At the end of that quarter, NVIDIA assessed that it was effectively foreclosed from competing in China’s data-center computing market and could not offer a competitive product for wide distribution there without approval from both governments. It could still ship uncontrolled products to China, including gaming and workstation GPUs.
Demand for less restricted hardware has also surfaced in workstation products. Reuters reported on October 20, 2026 that ByteDance, Alibaba and Tencent had ordered more than 3 million RTX PRO 5500 graphics cards, with a combined potential value exceeding $35 billion. Those cards are designed for powerful professional computers rather than large data-center clusters, and Chinese companies primarily wanted them for AI inference, or running trained models. NVIDIA had told customers and partners it could not guarantee fulfillment because requested volumes exceeded planned production.
The records describe two separate gates on one supply chain: Washington licenses H200 exports under specified conditions, while Beijing controls whether large-volume imports can proceed. The Hong Kong stockpile sits between them, leaving Chinese AI developers with limited authorized deliveries and higher prices through indirect channels. Hardware in storage is not the same as deployable computing capacity, and NVIDIA’s filing shows the commercial cost of that distinction.
Credit: Source link
