Chinese customs authorities have instructed border officials that Nvidia’s H200 artificial intelligence chips are not permitted to enter the country, according to three people briefed on the matter, in a move that sharpens tensions at the heart of the U.S.–China technology rivalry.
At the same time, Chinese government agencies summoned domestic technology companies to meetings this week, where officials delivered unusually blunt guidance discouraging them from purchasing the H200 chips unless deemed absolutely necessary, two of the people said, along with a third source familiar with the discussions.
“The language used was extremely strong — effectively a ban for now, even if it has not been formally declared as one,” one source said, adding that authorities left open the possibility the position could change depending on how broader geopolitical talks evolve.
The H200, Nvidia’s second-most powerful AI processor, has become a major flashpoint in the already fraught relationship between Washington and Beijing. While Chinese firms have expressed strong interest in acquiring the chip to support advanced artificial intelligence development, Beijing’s true intentions remain unclear.
Analysts say Chinese policymakers may be weighing several objectives at once: protecting domestic chipmakers, applying pressure in ongoing negotiations with the United States, or reserving the restrictions as leverage ahead of high-level diplomatic engagements.
The timing is particularly sensitive. The Trump administration approved exports of the H200 to China earlier this week under specific conditions, triggering renewed debate in Washington over whether allowing the sale of advanced AI hardware could erode U.S. strategic advantages or bolster China’s military capabilities.
China hawks in the United States have warned that even limited access to the H200 could accelerate Beijing’s progress in training large-scale AI models with potential defense applications. Those concerns have fueled bipartisan scrutiny of export controls first imposed in 2022 to curb China’s access to high-end semiconductors.
Despite the firm tone of the recent Chinese directives, sources said authorities did not specify whether the measures constitute a formal ban or a temporary administrative action. It also remains unclear whether the restrictions apply to existing orders or only to new purchases.
China’s General Administration of Customs, the Ministry of Industry and Information Technology, and the National Development and Reform Commission did not respond to requests for comment.
The Information reported that Chinese officials told some technology firms that purchases of the H200 would only be approved under exceptional circumstances, such as research and development projects conducted in partnership with universities. One source confirmed that exemptions are under discussion for academic institutions and select R&D initiatives.
Beijing’s approach suggests a calibrated strategy rather than an outright rejection of the chip, analysts say. By restricting commercial deployment while leaving room for limited exemptions, Chinese authorities may be attempting to balance short-term technological needs against longer-term industrial policy goals.
The move could also be linked to diplomacy. Analysts note that U.S. President Donald Trump is scheduled to visit Beijing in April for talks with Chinese President Xi Jinping, as both sides navigate a fragile truce on trade and technology.
“Beijing appears to be testing how much leverage it has ahead of the next round of negotiations,” said Reva Goujon, a geopolitical strategist at Rhodium Group. “It is pushing to see what concessions it can extract in exchange for loosening restrictions or approving licenses.”
For Nvidia, the stakes are substantial. Last year, Trump initially banned exports of the much weaker H20 chip before later approving limited sales. Beijing then effectively blocked those purchases from around August, prompting Nvidia Chief Executive Jensen Huang to say the company’s share of China’s AI chip market had fallen to zero.
The H200 represents a far more powerful offering. Delivering roughly six times the performance of the H20, it is considered highly attractive for training advanced AI systems. Chinese technology companies have reportedly placed orders for more than two million H200 chips, each priced at around $27,000 — far exceeding Nvidia’s estimated inventory of about 700,000 units.
Although Chinese firms have developed domestic alternatives such as Huawei’s Ascend 910C, industry experts widely regard Nvidia’s H200 as significantly more efficient and reliable for large-scale AI workloads.
Whether China or the United States stands to gain more from H200 sales remains a subject of debate. Re-entry into the Chinese market would deliver substantial revenue for Nvidia and the U.S. government, which would collect a 25% fee on approved exports.
Some U.S. officials, including White House AI adviser David Sacks, have argued that controlled exports could actually slow China’s domestic chip ambitions by reducing incentives for local firms to accelerate development of competing designs.
Others see the situation differently. “Beijing believes Washington is eager to sell AI chips to China,” said Chris McGuire, senior fellow for China and emerging technology at the Council on Foreign Relations. “That belief gives China confidence it can extract concessions by tightening approvals.”
For now, Chinese companies and global markets are left in limbo, waiting to see whether Beijing’s hard line is a negotiating tactic — or the start of a new phase in the global battle over artificial intelligence supremacy.
