
Despite strong demand from Chinese technology companies, Beijing has yet to approve large-scale imports, reflecting China's conflict between reliance on U.S. technology and independent research and development.
AI-generated summary
Starting in 2022, the United States will implement export controls on advanced AI chips and manufacturing equipment to China. Chinese companies are seeking to obtain high-performance computing power through middlemen to train AI models.
Reuters quoted two people familiar with the matter as saying that approximately 600,000 NVIDIA advanced artificial intelligence chips are currently stored in Hong Kong, waiting to be shipped to mainland China. Despite demand from Chinese technology companies, Beijing has yet to approve large-scale imports of these chips.
The inventory size of this batch of H200 chips has not been previously reported. It reflects Beijing's reluctance to deepen its reliance on U.S. technology while Chinese companies need Nvidia's advanced computing power to meet surging demand for artificial intelligence in the world's second-largest economy.
China has allowed some of these chips to enter the country in recent months, three people familiar with the matter said.
One of the people familiar with the matter said that in the past few months, several large Chinese Internet companies have each obtained about 10,000 H200 chips.
People familiar with the matter said that a small amount of chips have arrived in China through middlemen and indirect channels, but the specific quantity was not specified.
However, these people pointed out that Beijing has not yet approved large-scale imports into mainland China.
China's Ministry of Industry and Information Technology and National Development and Reform Commission did not respond to requests for comment. Nvidia and Chinese technology companies Alibaba, ByteDance, DeepSeek and Tencent also did not respond.
Although the H200 is not Nvidia's latest AI processor, its performance is still unmatched by Chinese alternatives and is very useful when training AI models - especially when used in conjunction with Nvidia's software.
Since 2022, U.S. export controls have sought to limit China’s access to advanced Nvidia chips and the equipment needed to make them.
With limited authorized supply, Chinese buyers have turned to middlemen or purchased through indirect channels, pushing up prices in the unofficial market.
According to a person familiar with the matter, as of the end of October, a server equipped with eight H200 chips in mainland China sold for more than 6 million yuan (about 900,000 US dollars), which is about three times the price a year ago. The person familiar with the matter added that servers equipped with Nvidia's newer generation Blackwell chips (which are also subject to U.S. export controls) are priced at more than 20 million yuan per unit.
The H200 chip comes amid broader trade and technology tensions between Washington and Beijing.
The United States once banned the export of H200 chips to China. Last year, the Trump administration allowed Nvidia to sell the chip to approved Chinese customers. This policy shift comes in the wake of U.S.-China trade negotiations.
At the same time, Beijing has also imposed restrictions on H200 imports, reflecting China's concerns about overreliance on American technology.
Technology media "The Information" reported in July that relevant Chinese authorities have informed companies including Alibaba, ByteDance and DeepSeek that they may be allowed to purchase a limited number of H200 chips for training AI models. According to reports, the total number of approved chips may be less than 200,000. In addition, Beijing hopes that companies will give priority to using Chinese domestic processors in the reasoning process. Inference refers to the task of running artificial intelligence models.
Reuters reported last month that ByteDance, Alibaba and Tencent had ordered more than 3 million Nvidia RTX Pro 5500 graphics cards. This is a weak product that meets export compliance requirements and has a total price tag of more than $35 billion.

Hua Hong Grace Semiconductor is scaling up its silicon photonics and power management chip production to capitalize on the rising demand for artificial intelligence infrastructure.

Huawei has unveiled its Qinyun HM650 series desktop computers, the first to run on its proprietary HarmonyOS and feature 100 per cent domestically sourced components, including the Kirin X90 processor, targeting enterprise and government sectors.

Lithuania is strengthening its position in the global semiconductor and AI hardware supply chain as laser manufacturer Litilit opens a new factory in Vilnius, despite ongoing trade and diplomatic tensions with China regarding Taiwan.

The US Federal Communications Commission has voted to stop recognizing Chinese testing laboratories and certification bodies for electronic products, effective December 1, 2028, citing a reliance on Chinese firms for 82% of US equipment-authorization applications.
The 2026 World Internet Conference Wuzhen Summit concluded on November 5 in Wuzhen, Zhejiang. This summit brought together about 1,800 guests from 138 countries and regions around the world, released a number of results such as the "Working Together to Build a Global Open Artificial Intelligence Action Initiative", and discussed artificial intelligence governance and cyberspace cooperation.
The 2026 World Internet Conference Wuzhen Summit will be held in Wuzhen, Zhejiang from November 2 to 5. With the theme of "Open Source, Openness, Co-construction and Sharing", this summit attracted 671 companies from 54 countries and regions to participate in the exhibition, focusing on the digital divide, artificial intelligence governance and international technological cooperation.