Why are about 600,000 Nvidia H200 chips reportedly waiting in Hong Kong?

On Thursday, November 5, 2026, Reuters published an exclusive report saying that about 600,000 of Nvidia's H200 chips are sitting in Hong Kong, waiting to be shipped to mainland China. According to the two people Reuters spoke with, Beijing has not given the approval needed for large volumes to enter the country, even though Chinese technology companies want the chips.

This is a single news report based on unnamed sources, and the number is approximate. Nvidia, the Chinese ministries involved and the large Chinese tech companies named in the story did not respond to Reuters' requests for comment. What makes the report worth understanding is the situation it describes: a chip that the United States now lets Nvidia sell to China, held up by China's own reluctance to buy it.

What Reuters reports

The report does not say who owns the chips in Hong Kong, when they arrived or how they were licensed. Those are open questions, and they matter for reading the number.

What the H200 is, and why China still wants it

The H200 is an Nvidia GPU for data centers, the kind of chip used to train and run large AI models. It is no longer Nvidia's newest chip; servers built on the newer Blackwell generation remain under US export controls for China. Reuters notes that the H200 still has capabilities Chinese alternatives have not matched, especially for training AI models when paired with Nvidia's software.

That gap explains the prices. When legal supply is tight, buyers pay more through unofficial channels. Reuters' source puts an eight-chip H200 server at more than 6 million yuan, and servers with the newer Blackwell chips at more than 20 million yuan.

How the US side changed

Until late 2025, the United States barred H200 exports to China. Then the rules loosened:

The January rule came with strings. Exporters must certify, among other things, that US customers are not short of supply, that shipments to China stay at or below half of US volume for the product, and that buyers are checked to prevent unauthorized remote access. Nvidia's quarterly filing for the period ended July 26, 2026 adds that licensed H200s must pass an inspection in the United States before shipment and face a 25% tariff when they are brought into the US for that inspection, a cost Nvidia says it has not been able to pass on to customers.

Why the chips are stuck on the Chinese side

Nvidia's own filing already pointed to the problem. It says Chinese government restrictions limited its H200 sales, that it made only a fraction of the shipments its licences allow, and that those shipments were under 1% of its data center revenue in the quarter. It also took a $0.4 billion charge for excess H200 inventory and purchase commitments in the first half of its fiscal year.

The Reuters report fits that picture and puts a size on it. Its sources describe Beijing as reluctant to deepen reliance on US technology while Chinese companies need computing power for AI. Reuters also cites an earlier report by The Information, from July, that Chinese officials told companies including Alibaba, ByteDance and DeepSeek they could buy limited quantities for training, with total approvals possibly under 200,000 chips, and that Beijing wants domestic processors used for inference, the work of running trained models.

There is a tension in the numbers that neither source resolves. Nvidia's filing describes licences for small amounts, while 600,000 chips is a large stockpile. Without knowing who holds the chips and when they were shipped, the two figures cannot be reconciled. They may describe different periods, different sellers or chips that moved through channels the filing does not cover.

A worked example: what 600,000 chips means

Using Reuters' own server size helps make the number concrete. At eight GPUs per server, 600,000 chips would fill about 75,000 servers. If even the approvals The Information reported, under 200,000 chips, came through in full, that would cover about a third of the reported stockpile. The rest would stay in storage, be sold elsewhere or wait for a change in Beijing's position.

That arithmetic uses unofficial figures from anonymous sources, so treat it as an order of magnitude, not a count.

What would it take for these chips to reach mainland China?

On the US side, the path already exists: case-by-case licences with conditions, inspection and a tariff. The missing step, according to Reuters' sources, is Beijing's approval for large-scale imports. That decision weighs two goals that pull in opposite directions: Chinese companies want the fastest available chips to keep up in AI, and the government wants to build demand for domestic processors and reduce dependence on American technology.

Watch for three signals. First, any official statement from Chinese ministries or the companies named in the report, none of which had commented. Second, Nvidia's next quarterly results, which will show whether China revenue moved. Third, a change in either government's rules, since the H200 has already gone from banned to licensed in less than a year. The guide's article on how US export controls on AI chips work explains the full set of rules behind this story.