A U.S. official has stated that the United States intends to implement measures to prevent American chip manufacturers from selling semiconductors to China that evade government restrictions. This is part of the upcoming actions by the Biden administration to further restrict exports of AI chips.
This illustration picture, taken on February 17, 2023, displays semiconductor chips on a printed circuit board.
These new regulations, being reported by Reuters for the first time, will be integrated into the extensive restrictions on the export of advanced chips and chip manufacturing equipment to China that were initially introduced last October. While it is anticipated that these updates will be unveiled this week, it’s worth noting that timelines for such actions can sometimes experience delays.
According to the anonymous official, the forthcoming regulations will prohibit certain AI chips that narrowly miss meeting the current technical criteria, and companies will be required to report shipments of other AI chips.
The U.S. Department of Commerce, responsible for export controls, chose not to provide a comment on the matter.
This recent tightening of restrictions on technology exports to China occurs as the United States seeks to improve its challenging relationship with the world’s largest economy. Several high-ranking members of the Biden administration have engaged with their Chinese counterparts in recent months, and the latest set of rules has the potential to complicate diplomatic efforts.
The Biden administration has stated that it implemented these export restrictions with the aim of preventing U.S. chips and equipment from bolstering China’s military capabilities. Beijing, on the other hand, has accused the United States of exploiting export controls to stifle Chinese companies. These restrictions represented a significant shift in U.S.-China tech policy.
The Chinese embassy in Washington has not immediately responded to a request for comment.
Last year, government-imposed restrictions prevented Nvidia (NVDA.O), the world’s most valuable chipmaker, from delivering two of its most advanced AI chips to Chinese customers. These chips have become the industry standard for developing chatbots and other AI systems.
Nvidia quickly responded to the U.S. export controls by introducing new variants of its AI chips tailored for the Chinese market. These variants, such as the H800, offer computing power equivalent to their more powerful but restricted H100 chip in certain settings. However, there are limitations in some key performance aspects, according to a specification sheet seen by Reuters.
The U.S. is now planning to introduce updated guidelines for AI chips that will extend restrictions to certain advanced datacenter AI chips not currently covered, according to the U.S. official. While the official did not specify which additional chips would be affected, sources have suggested that Nvidia’s H800 is among the targets.
Nvidia, headquartered in Santa Clara, California, did not respond immediately to a request for comment. In June, the company’s chief financial officer mentioned that if the H800 and a related chip called the A800 were restricted, they did not anticipate an immediate material impact on their financial results.
The new restrictions will exempt chips intended for consumer products like laptops. However, companies will be required to inform the Commerce Department when fulfilling orders for the most powerful consumer chips to ensure they are not being used in ways that pose a threat to national security.
To prevent the export of AI chips that the U.S. considers too powerful to China, the U.S. intends to remove one of the parameters it has used to limit exports of specific AI data center chips – the “bandwidth parameter.” Removing this parameter activates another guideline, broadening the scope of chips affected. This change is likely to reduce the speed at which AI chips can communicate with each other. Slowing down this communication speed can make AI development more challenging and costly, especially when training large AI models that require multiple chips to work together.
The U.S. also plans to introduce a “performance density” parameter to prevent future workarounds. However, the official did not provide further details.
The updated rules are designed to cover AI chips as technology evolves. Companies will be required to notify the government about semiconductors that almost meet the guidelines before they are shipped to China. The government will then assess on a case-by-case basis whether they pose a national security risk, but they can be shipped unless instructed otherwise.
The revisions to the rules may also close a loophole that allows Chinese companies to access American AI chips through Chinese units located overseas, as reported by Reuters last week.
It is not anticipated that the rules will impose restrictions on access to U.S. cloud computing services or those of U.S. allies. However, the U.S. will seek feedback on the potential risks associated with such access and how they could be addressed, as indicated by the official.
The Biden administration had informed Beijing about its intentions to update the controversial rules earlier in October, with the aim of stabilizing relations between the two superpowers.
[Reporting by Karen Freifeld; additional reporting by Max A. Cherney; Editing by Chris Sanders, Lisa Shumaker, Deepa Babington, and Shri Navaratnam]
