Roughly 600,000 Nvidia H200 chips sit in Hong Kong awaiting a green light for mainland China, Reuters reported in an exclusive on November 5. Two people familiar said Beijing withholds approval for large-scale imports even as Chinese technology firms want the hardware.
The regulator handling industrial policy and the one handling economic planning both ignored questions from the wire service, as did Nvidia and the four Chinese buyers named in the report: Alibaba, ByteDance, DeepSeek and Tencent. No government has confirmed the figure.
Small shipments have been getting through
Some chips already reached China. One of three sources said several major Chinese internet companies each obtained about 10,000 H200s over recent months, and small volumes have arrived via intermediaries and indirect routes.
Gray market prices show how tight supply has become. By late October a single mainland quote for a server holding eight H200 cards ran past 6 million yuan, close to $900,000 and near triple the going rate a year before. Machines built around Nvidia’s newer Blackwell parts, which carry tighter export limits, carried tags above 20 million yuan each.
Why the count matters to labs
American restrictions cannot make the H200 obsolete, since no domestic part yet matches what Nvidia’s software stack can do, and that pairing is what labs rely on for training. Earlier reporting said Beijing wanted firms to lean on domestic silicon for inference while permitting limited H200 volume for training runs.
A Financial Times figure from August put the Hong Kong stock near 500,000. Different outlets and sources produced the two counts, so they track direction rather than measure the same thing.
Hong Kong was also described as short on data center capacity, so clearing the backlog helps only where power and floor space exist.
