Nearly 600,000 Nvidia H200 chips are stored in Hong Kong awaiting shipment to mainland China because Beijing has withheld approval for large-scale imports, Firstpost reported on 5 November, drawing on Reuters reporting. Chinese technology firms want the processors for AI work, but the authorities have permitted only smaller deliveries. The stockpile sits between two decisions: the United States has allowed sales to approved Chinese customers, while China has yet to authorise imports on this scale.
Key points
- Nearly 600,000 H200 chips are awaiting permission to move from Hong Kong to mainland China.
- Several Chinese companies have received shipments of 10,000 chips in recent months, while large-volume imports remain unapproved.
- By late October, an eight-card H200 server was offered for more than 6 million yuan in mainland China, about three times its price a year earlier.
- Chinese officials have indicated that selected firms could receive permission to buy limited quantities for model training.
Beijing allows smaller H200 shipments
The United States introduced export controls in 2022 restricting China’s access to advanced Nvidia chips and equipment used to make them. President Donald Trump’s administration gave Nvidia permission in 2025 to sell H200 chips to approved Chinese customers, after earlier US restrictions had barred those exports. That permission governs what Nvidia may supply under US rules. Chinese authorities have separately allowed some H200 deliveries into mainland China without approving large-scale imports.
Several Chinese companies received shipments of 10,000 H200 chips each in recent months, one person familiar with the matter said, according to Firstpost. Chinese officials also told companies including Alibaba, ByteDance and DeepSeek earlier in 2026 that they could be permitted to buy limited quantities for training AI models. The account put possible approvals at fewer than 200,000 chips and said Beijing wanted companies to give Chinese processors priority for inference, the work of running a trained model.
The H200 is older than Nvidia’s latest AI processor, but its capabilities and Nvidia’s software make it useful to Chinese companies training models, Firstpost reported. Beijing’s approach therefore leaves companies with access to some imported hardware for training while directing their inference purchases towards domestic processors. The smaller deliveries and the Hong Kong stockpile occupy different sides of that approach: permission for selected purchases has not extended to the chips awaiting shipment in bulk.
H200 servers exceed 6 million yuan
By late October, an eight-card H200 server was being offered in mainland China for more than 6 million yuan, about $900,000, according to a person familiar with the matter cited in the report. That was about three times its price a year earlier. Servers using Nvidia’s newer Blackwell chips, which are subject to US export controls, were being offered for more than 20 million yuan apiece. These were offers in an unofficial market, rather than prices charged by Nvidia for authorised shipments.
Chinese buyers have also used intermediaries and indirect routes to obtain chips while authorised supplies are constrained. Those routes sit alongside the limited H200 shipments permitted by Beijing, not in place of an approval for large-volume imports. The distinction matters for buyers seeking a substantial installation: individual deliveries of 10,000 chips have occurred, but the much larger quantity held in Hong Kong remains awaiting permission to enter mainland China.
The demand extends beyond the H200. Chinese groups ordered more than 3 million Nvidia cards and faced a supply squeeze, AI Affairs previously reported. Orders for those cards and offers for H200 servers describe different parts of the market, but both concern companies trying to secure Nvidia hardware while access to it is constrained. For H200 buyers, Beijing’s import decision determines whether chips already stored in Hong Kong can reach them through authorised large-scale shipments.
Nvidia’s H200 shipments and inventory charges
Nvidia disclosed that licensed H200 shipments made up less than 1% of its quarterly data centre revenue, Coinpaper reported. That figure concerns shipments made under licence, rather than the chips held in Hong Kong. In the first half of its fiscal year, Nvidia incurred approximately $400 million in charges tied to H200 stock and purchase commitments.
