Roughly six hundred thousand of Nvidia‘s advanced artificial intelligence chips are sitting in storage in Hong Kong awaiting shipment to mainland China, according to two people familiar with the matter, as Beijing continues to withhold approval for large-scale imports despite sustained demand from Chinese technology firms. The stockpile of H200 chips, a figure that has not previously been reported, reflects the delicate balancing act China’s government is performing between its desire to reduce reliance on American technology and the practical reality that domestic companies need Nvidia’s computing power to meet surging demand for artificial intelligence across the world’s second-largest economy. The H200 is not Nvidia’s newest AI processor, but it possesses capabilities that Chinese alternatives have yet to match, making it particularly useful for training AI models when combined with Nvidia’s software ecosystem.
China has allowed some of the chips to enter the country in recent months, according to three people familiar with the matter, and several major Chinese internet companies have each obtained about ten thousand H200 chips over the past few months. Small quantities have also been shipped through intermediaries and indirect routes, though the people did not specify how many. But Beijing has yet to approve large-volume imports into mainland China, leaving the bulk of the stockpile stranded in Hong Kong and creating a grey market that has driven prices sharply higher. Neither China’s Ministry of Industry and Information Technology nor the National Development and Reform Commission responded to requests for comment, and Nvidia, Alibaba, ByteDance, DeepSeek and Tencent all declined to comment.
The situation is a direct consequence of the export controls the United States has imposed since 2022, which sought to restrict China’s access to advanced Nvidia chips and the equipment needed to manufacture them. Those controls have constrained authorised supply, pushing Chinese buyers toward intermediaries and indirect routes and inflating prices in the unofficial market. By the end of October, an eight-card H200 server was being offered for more than six million yuan in mainland China, roughly three times its price a year earlier, according to one person familiar with the matter. Servers using Nvidia’s newer Blackwell chips, which also fall under US export controls, were being offered for more than twenty million yuan apiece. The price premiums tell a story about scarcity and desperation, but they also illustrate how difficult it is to keep advanced technology out of a market where the demand is overwhelming and the workarounds are plentiful.
What Beijing is weighing is not simply a question of trade policy but a broader strategic calculation about the future of China’s AI industry. On one hand, allowing unrestricted imports of Nvidia chips would accelerate the development of Chinese AI models and help domestic companies compete globally, particularly as the race to build more capable systems intensifies. On the other, permitting large volumes of American chips to flow into the country entrenches dependence on a foreign supplier that Washington has already shown itself willing to weaponise, and it risks crowding out domestic chipmakers who are still striving to match Nvidia’s performance. The trickle of chips that has been allowed in suggests a deliberate approach, one that provides enough computing power to keep Chinese AI development moving without fully opening the floodgates. Whether that middle path is sustainable as demand continues to grow is an open question.
For Nvidia, the stockpile represents both an opportunity and a risk. The company has consistently maintained that its chips are essential to AI development and that restricting their sale to China ultimately harms American competitiveness by pushing Chinese firms to develop their own alternatives. The H200, while not the company’s latest offering, remains a powerful piece of hardware, and the fact that Chinese companies are willing to pay three times the normal price through indirect channels demonstrates how much they value access. Yet the longer the chips sit in Hong Kong, the more uncertain their fate becomes, and the more pressure builds on both Washington and Beijing to clarify the rules of the road. For the Chinese companies waiting on those chips, the calculus is simpler. They need computing power to train the models that will define their products and services, and every month of delay is a month in which their competitors elsewhere in the world pull further ahead. The stockpile in Hong Kong is a symptom of a larger tension that shows no sign of resolving, and the question of who ultimately bears the cost of that tension remains unanswered.
