The United States will take steps to prevent American chipmakers from selling semiconductors to China that violate government restrictions, a U.S. official said, as part of the Biden administration’s upcoming moves to block further AI chip exports. as
The new rules, details of which are being reported by Reuters for the first time, will be added to US restrictions on shipments of advanced chips and chip-making equipment to China that were unveiled last October. Updates are expected this week, other people familiar with the matter said, although such timetables often slip.
The official, who spoke on condition of anonymity, said the new rules would block some AI chips that fall under existing technical parameters while requiring companies to report shipments of others.
A spokesman for the US Commerce Department, which oversees export controls, declined to comment.
The latest crackdown on technology exports to China coincides with US efforts to thaw the strained relationship between the world’s two largest economies. Several senior members of the Biden administration have met with their Chinese counterparts in recent months, and the latest round of rules threatens to complicate diplomatic efforts.
The Biden administration has said it designed the export ban to prevent U.S. chips and equipment from bolstering China’s military. Beijing has accused the US of abusing export controls to stifle Chinese companies. The sanctions marked a historic shift in US China tech policy.
The Chinese embassy in Washington did not immediately respond to a request for comment.
Last year, government restrictions prevented Nvidia, the world’s most valuable chipmaker, from shipping two of its latest AI chips to Chinese consumers, chips that have become the industry standard for building chatbots and other AI systems. .
But Nvidia soon released new variants for the Chinese market that were less sophisticated and got around US export controls. One, dubbed the H800, has as much computing power as the company’s more powerful but blocky H100 chip in some settings used in AI work. Still, some key performance aspects are limited, according to a specification sheet seen by Reuters.
The U.S. now plans to introduce new guidelines for AI chips that will limit some advanced data center AI chips that aren’t currently captured, the U.S. official said.
Although the official declined to say which additional chips would be effectively banned, Nvidia’s H800 is a semiconductor sources have suggested the administration wants to block.
Santa Clara, Calif.-based Nvidia did not immediately respond to a request for comment. In June, the company’s chief financial officer said he did not expect “an immediate material impact on our financial results” if the H800 and a related chip called the A800 were phased out.
Chips for consumer products such as laptops will be exempted from the new restrictions, the official said. But companies will be required to tell the Commerce Department when they are filling orders for the most powerful consumer chips to ensure they are not being used in ways that threaten national security, the official said. in accordance.
To keep up with U.S. perceptions of AI chips as too powerful from China, the official said the U.S. plans to remove one of the parameters — the “bandwidth parameter” — it has imposed on exports of some AI data center chips. used to limit By removing this parameter, another guideline is revealed, which widens the range of chips. This will mean that the speed at which AI chips talk to each other will be reduced.
This is important because training the largest AI models is impossible on a single chip and requires many chips to be wired together. If one slows down the speed at which they communicate, it makes AI development more difficult and expensive.
The U.S. also plans to introduce a “performance density” parameter to help prevent future operations, the official said, but declined to elaborate.
The updated rules aim to cover AI chips as the technology evolves. The official said the U.S. will require companies to notify the government of semiconductors with performance well below guidelines before they are shipped to China. The government will decide on a case-by-case basis whether they pose a threat to national security, but they can be sent unless the chip maker is told otherwise.
An October 2022 update to the rules could also close a loophole that gives Chinese companies access to U.S. artificial intelligence chips through Chinese units located overseas, as Reuters reported last week.
The official said the rules are not expected to include restrictions on access to U.S. cloud computing services, or allies, but the U.S. will seek comments on the risks of such access and ways to address them.
The Biden administration told Beijing of its plans to update the controversial rules this month, Reuters reported in early October as part of a policy to strengthen ties between the superpowers.