NVIDIA H200: About 600,000 Chips Reportedly Held in Hong Kong by Beijing
Beijing has withheld approval for large-scale NVIDIA H200 imports, leaving about 600,000 chips in Hong Kong, Reuters reported; NVIDIA's 10-Q shows the licences yielded under 1% of data-center revenue.
Chips & Power DeskEdited by Ege Erten
2 min readAI-assisted
The H200 NVL, an H200 variant from NVIDIA's product page. The chips reported stored in Hong Kong are H200 accelerators; this render does not show those units.
Image: NVIDIA
Source · Press image from the NVIDIA website, used for news reporting.
About 600,000 NVIDIA H200 accelerators are being stored in Hong Kong because Beijing has not approved large-scale imports into mainland China, Reuters reported on Nov. 5, citing two people familiar with the matter, according to The Next Web's account of the report. NVIDIA, China's industry ministry and its main economic planning agency did not respond to Reuters' requests for comment.
Reuters' sources put the Hong Kong stockpile at about 600,000 H200s against a possible approval total below 200,000; the figures have different bases and are unconfirmed.
Illustration (AI-generated)
According to Reuters, as relayed by The Next Web, Beijing wants Alibaba, ByteDance, Tencent and DeepSeek to use Chinese-made chips instead, especially for inference, the work of running trained models. Several large Chinese internet companies have each received about 10,000 H200s in recent months, one source told Reuters. The total number of foreign chips China approves could end up below 200,000, Reuters reported.
Washington has already opened its side: the US began reviewing H200 export licences for China case by case in January, after President Donald Trump allowed limited sales in December 2025. NVIDIA's own filing shows the licences have not turned into sales. In its quarterly report for the period ended July 26, 2026, NVIDIA said the US government granted licences to ship small amounts of H200 to specific China-based customers, but that the Chinese government restricted those sales and it has been unable to sell all the products it is licensed for.
What the Filing and the Grey Market Show
NVIDIA booked a $0.4 billion charge in the first half of fiscal 2027 for excess H200 inventory and purchase obligations, as demand for the part diminished. It has since shipped a fraction of the allowed volume, which accounted for less than 1% of its data-center revenue in the latest quarter. The filing also says the licences require H200s to pass an inspection in the United States before shipment, which makes them subject to a 25% US tariff on import, and that NVIDIA has been unable to pass any of that tariff on to customers.
The squeeze shows in prices. Reuters' source said an eight-card H200 server was offered on China's unofficial market for more than 6 million yuan by the end of October, about three times the price a year earlier. Servers built on Blackwell chips, which the US does not allow to be sold to China, were offered for more than 20 million yuan each. For a Chinese data-center operator, that means usable AI capacity is set by what clears customs, not by how many accelerators sit in a warehouse.
This is a different problem from the one we covered earlier on Nov. 8, when the New York Times reported that H200 licences went to two firms with empty offices. That story is about who Washington licensed; this one is about whether Beijing lets the chips in.
What is not confirmed yet
Both the 600,000 count and the sub-200,000 approval ceiling rest on anonymous sources, and neither Beijing nor NVIDIA has confirmed them. They are also not measured on the same basis, so they should not be subtracted from each other. A search on Nov. 8 found no report of Beijing granting a large approval. The next datapoint is NVIDIA's next quarterly filing, which should show whether the H200 China line moves from under 1% of data-center revenue.
