Shipments of Nvidia's H200 artificial intelligence chips to China and Hong Kong have started to resume, but still at a very reduced volume, according to what United States Under Secretary of Commerce for Industry and Security Jeffery Kessler said on Tuesday (14).
During a hearing in the U.S. Congress, Kessler stated that only a small quantity of chips was authorized for export.
"Very few shipments related to the licenses for the H200 and equivalent products have been carried out. It is a very small quantity of chips," said the Commerce Department representative.
The statement signals that export licenses have started to be granted again, which may represent progress for Nvidia in one of the world's largest artificial intelligence markets.
In recent months, the manufacturer had been adopting a cautious stance regarding the Chinese market. Since last year, the company has stopped including any revenue expectation from China in its financial projections. In May, CEO Jensen Huang even stated that investors should not expect relevant contributions from sales to the country.
Nvidia preferred not to comment on the statements by the American government.
The company has faced restrictions imposed by Washington since the worsening of the technological dispute between the United States and China. A large part of its AI accelerators came to depend on special authorization for export, due to the American government's concerns that these components could be used in military or strategic applications.
According to Kessler, each license request is analyzed individually. Interested companies need to comply with national security requirements and accept inspections to ensure that the chips are used in accordance with the established rules.
The under secretary also highlighted that some requests continue to be rejected by the U.S. government.
Although licenses are being granted on a limited basis, there is still no guarantee that China will allow large-scale imports of H200 chips. If access to Nvidia products remains restricted, Chinese companies should continue turning to local manufacturers to train artificial intelligence models, even if these alternatives are still considered less competitive in performance.
The H200 belongs to the Hopper architecture, an earlier generation of Nvidia accelerators. Meanwhile, companies in the United States are already using chips from the Blackwell family, considered more advanced and with significantly superior performance for workloads focused on artificial intelligence.