The United States government officially approved Nvidia’s export of its H200 artificial intelligence (AI) chip to China on Tuesday, January 13, but with several restrictions attached. This marks a significant adjustment in the U.S. policy on the export of advanced chips to China and has quickly sparked intense debate within the Washington political circles.
According to the final rule published by the Bureau of Industry and Security (BIS) under the U.S. Department of Commerce in the Federal Register, the export review of H200 and similar products to China and Macau will shift from the previous “presumptive denial” to a “case-by-case review” under specific conditions. However, the U.S. insists that this is not a total unrestricted release but a limited opening under strict supervision.
Under the new regulations, this high-end AI chip, which is considered second only to Nvidia’s most cutting-edge products, must undergo third-party testing and review before export to confirm that its actual computing power does not exceed the performance threshold set by the U.S. At the same time, the cumulative computing performance of H200 chips exported to China and Macau must not exceed 50% of Nvidia’s total shipments to the U.S. market during the same period.
Additionally, Nvidia must prove to the U.S. that the domestic market’s supply of H200 is sufficient and that the export to China will not hinder the normal delivery to U.S. customers or result in the diversion of advanced processing capacity originally serving the U.S. market for exports to China.
For the first time, the U.S. has incorporated multiple compliance requirements into the formal rules regarding the usage and safety aspects. Chinese customers must demonstrate “adequate security control measures” to prevent unauthorized remote access and explicitly pledge that the relevant chips will not be used for any military or sensitive purposes.
In terms of cloud services, the U.S. has also set up barriers. If the H200 is used in cloud or Infrastructure as a Service (IaaS) models, exporters must declare potential remote users and implement technical and managerial measures to prevent the transfer of AI model weights or training outcomes to restricted parties.
According to Reuters, orders for H200 chips from Chinese tech companies have exceeded 2 million, far surpassing Nvidia’s current inventory level of around 700,000 chips.
During the Consumer Electronics Show (CES) in Las Vegas last week, Nvidia CEO Jensen Huang stated that the company is rapidly ramping up the production capacity of H200 to meet the strong demand from China and global markets, with the hype even driving up the rental prices of H200 computing power in cloud data centers.
However, there is a high level of vigilance in the national security field about this matter. Saif Khan, who previously served as the White House National Security Council’s Director of Technology and National Security Affairs during the Biden administration, warned that the new regulations will significantly enhance China’s overall AI computing power in practice.
He pointed out that if around 2 million H200 chips flow into China, their computing power would be equivalent to the total computing power of a leading U.S. frontline AI company. Additionally, the government will face serious enforcement challenges in implementing requirements like “Know Your Customer” (KYC) and preventing Chinese cloud service providers from circumventing usage restrictions.
However, within the Trump administration, there are arguments supporting the release of advanced chip exports as a means to curb the development of Chinese domestic competitors, particularly heavily sanctioned companies like Huawei, to prevent them from narrowing the technological gap with Nvidia and AMD in cutting-edge chip design. David Sacks, in charge of artificial intelligence affairs at the White House, also supports this position.
President Donald Trump announced last month that he would allow the relevant sales to proceed and demanded that Nvidia pay a 25% tax to the U.S. government. He emphasized that the export of relevant chips would take place under the premise of ensuring the “continuously strong” national security and technological advantage of the United States.