China has allowed Nvidia's latest artificial intelligence chips into the mainland, a move aimed at narrowing the technology gap amid intensifying AI competition between the United States and China.
The Financial Times reported on the 19th local time, citing sources, that ByteDance and Tencent had each brought in 10,000 H200 chips over the past several weeks, and that other Chinese technology firms could soon receive approval to import the H200 on a similar scale.
The H200 lags Nvidia's most advanced chips by at least two generations, but it is the most powerful U.S.-made chip cleared for export to China.
Confirmation of actual H200 shipments into China this year is unusual. Last year, U.S. President Donald Trump signaled an easing of export controls in a post on Truth Social, saying Washington would allow Nvidia to ship the H200 to approved customers in China and other countries on the condition that it did not undermine national security. China, however, has continued to restrict imports of U.S.-made AI chips as it seeks to nurture domestic chipmakers such as Huawei.
What changed the situation is China's shortage of advanced chipmaking equipment. The United States has restricted exports of extreme ultraviolet (EUV) lithography machines needed for semiconductor manufacturing, and this year it is considering a full ban on exports of deep ultraviolet (DUV) lithography machines as well. As this made it harder to meet demand for AI development, Beijing is seen as having ultimately allowed U.S.-made chips into the country. Nvidia chips are considered indispensable, particularly in training. China, however, is increasing its use of domestic semiconductors for AI inference tasks that generate real-time responses.
Chinese authorities are said to require that most U.S.-made chips be kept outside the mainland. While recently allowing a small volume of H200 chips into the country, they have directed companies to transport the chips to Hong Kong for use. The United States is reported to have authorized Chinese companies to buy up to 100,000 H200 chips each.
Still, it remains uncertain whether China's shift will allow companies to make broad use of the H200. Bringing in the chips is difficult because Hong Kong does not have enough data centers to house them. An executive at a semiconductor company told the FT that it was a dilemma, with everyone needing the chips but struggling to find ways to use them in Hong Kong.
Meanwhile, Nvidia's partners appear to be joining the easing trend. Partners such as Lenovo, which combine Nvidia chips with central processing units (CPUs) and networking equipment to sell as AI servers, have recently told Chinese customers they can resume orders for H200-equipped products. The FT reported that Nvidia is holding about 500,000 H200 chips in reserve for Chinese customers.
