The U.S. government has announced a new set of rules regulating the export of AI chips from Nvidia and AMD to China, according to Bloomberg. The new rules allow export requests to be reviewed on a case-by-case basis, marking a shift from the previous policy that denied all exports to China.
Under the new regulations, exporting companies must ensure that shipments to China do not cause domestic shortages, prioritizing U.S. companies’ needs first. Companies seeking export approval must demonstrate that production for Chinese customers will not directly affect production for U.S. clients.
The rules also allow Nvidia to export its high-performance H200 chips, which were previously banned, after the company had to produce a slightly weaker version called H20 for China. Additionally, the regulations require the importing company to undergo a full security review, complete customer due diligence, and subject the chips to testing through external firms to ensure compliance with specifications.
Other restrictions include limiting shipments to Chinese companies to no more than 50% of total shipments intended for U.S. companies. Nvidia and AMD are also preparing lists of partner companies, particularly those in Belarus, China, Cuba, Iran, Macau, North Korea, Russia, and Venezuela. AMD confirmed its full compliance with the new U.S. Department of Commerce rules, according to Bloomberg.
However, a new Reuters report indicated that Chinese customs have stated imports of Nvidia’s H200 chips are not allowed, urging local companies to use Huawei chips or other domestically developed alternatives.
This announcement coincides with Nvidia unveiling its AI supercomputers under the “Rubin” brand at CES 2026.
Media coverage shows that the new rules have not significantly impacted Nvidia or AMD’s stock prices, reflecting investor skepticism about their effectiveness. Reports also note that Chinese companies may bypass these complex procedures, as they are developing domestic AI chips capable of running advanced models and continue to use older generations effectively. The DeepSeek model serves as a clear example of this approach.
