About 600,000 Nvidia AI chips are stored in Hong Kong, waiting for permission to enter mainland China. Reuters reported on 5 November that Beijing has held back approval for large-scale imports of the H200 processor, even though Chinese technology firms want the hardware. The stockpile shows how the trade in advanced processors has become a contest between two governments. Washington controls what Nvidia can sell, and Beijing controls what Chinese firms can buy. The Nvidia AI chips at the centre of the dispute are not the company’s newest, but Chinese firms still want them.
The H200 belongs to Nvidia’s Hopper generation. It was the standard processor for training and running AI software such as ChatGPT until Nvidia introduced its Blackwell line. The H200 is not the company’s newest processor, but it has abilities that Chinese alternatives have yet to match, and it works well with Nvidia’s software. That combination makes the chips useful for training AI models, which is why Chinese firms have sought them. The Nvidia AI chips held in Hong Kong show demand that has not been met.
The United States has restricted exports of advanced Nvidia chips to China since 2022, citing national security and the military uses of advanced AI. The H200 was barred at first. Late in 2025, the Trump administration gave Nvidia permission to sell the chips to approved Chinese customers. Under the new policy, the United States collects 25 percent of the revenue from those sales. The more advanced Blackwell and Rubin processors remain banned from export to China. The Nvidia AI chips that can be sold are therefore a generation behind the best Nvidia makes.
Nvidia AI chips and the Hong Kong stockpile
The scale of the stockpile had not been reported before Reuters described it. About 600,000 Nvidia AI chips are held in Hong Kong, according to two people familiar with the matter. China has allowed some of the chips to enter the country in recent months, but it has yet to approve large-volume imports. Nvidia told investors in August 2026 that it had sold only small quantities of H200 chips to Chinese customers because of opposition from Beijing, and actual sales were below the quantities American officials had authorised. The Information reported in July that Chinese officials had told companies including Alibaba, ByteDance and DeepSeek that they could receive permission to buy limited quantities of H200 chips for training AI models.
Nvidia chief executive Jensen Huang has criticised the restrictions. In interview footage released by Bloomberg on 9 November, he called the American policy on selling chips to China a mistake that is hard to unwind. He said he had done his best to tell policymakers in Washington that engaging with the Chinese market mattered, and that Chinese AI researchers outnumber those of any other nation. The remarks came as Nvidia’s filings recorded the cost of the curbs in stock the company cannot move.
Nvidia AI chips and the cost of the curbs
Nvidia’s filings record two China-related charges. A charge of 4.5 billion dollars followed the April 2025 licence rule for the less advanced H20 chip, and a further charge of 0.4 billion dollars was tied to the H200 in the first half of fiscal 2027, after the Chinese government restricted licensed sales. Together the charges come to 4.9 billion dollars. The H20 licences that followed from August 2025 allowed some sales, but the Chinese government restricted them, and Nvidia said it could not sell its H20 stock. The figures show what the curbs have cost the company in inventory it cannot shift.
American lawmakers now want to go further. Republican Senator Jim Banks and Democratic Senator Elizabeth Warren are seeking an 18-month restriction that would keep China at least two generations behind America’s most advanced AI chips. Their AI OVERWATCH Act has been included in the Senate’s fiscal 2027 defence authorisation legislation, though talks on the final package continue. A separate proposal would cap sales at 75,000 H200 chips per Chinese customer, less than half of what firms such as Alibaba and ByteDance privately told Nvidia they would like to buy. Neither proposal has become law.
Nvidia AI chips and what happens next
The question for Nvidia is whether any of the 600,000 Nvidia AI chips in Hong Kong will reach Chinese buyers. Nvidia said last week that it still earns no data centre revenue from China and does not know whether Beijing will allow imports, even if Washington gives permission. Washington must decide whether to tighten the rules further, and Beijing must decide whether its own chip makers can close the gap. The Nvidia AI chips in Hong Kong are the clearest sign of a market that both sides want to control, and neither side is ready to give up.