Nvidia's AI chips have been a key focus of U.S. export controls as U.S. officials have moved to keep the most advanced chips from being sold to China as the U.S. tries to keep ahead in the AI race. After those controls were implemented, Nvidia began designing chips that would come as close as possible to U.S. limits.
Nvidia shares were down about 6% in after-hours trading.
A U.S. Commerce Department spokesperson said late on Tuesday that it was issuing new licensing requirements for exports of chips including Nvidia's H20, AMD's (AMD.O) MI308 and their equivalents.
"The Commerce Department is committed to acting on the President's directive to safeguard our national and economic security," said the spokesperson of the department that oversees U.S. export controls.
AMD did not immediately respond to a request for comment. Its shares were down 7% in after-hours trading.
For Nvidia, the H20 is its most advanced chip available for sale in China and is central to its efforts to stay engaged with China's booming AI industry. Chinese companies including Tencent (0700.HK), Alibaba (9988.HK) and TikTok parent ByteDance had been ramping up orders for H20 chips due to booming demand for low-cost AI models from startup DeepSeek, Reuters reported in February.
While the H20 chip is not as fast at training AI models as Nvidia's chips for sale outside China, it is competitive with some of those chips at a step known as inference, where AI models serve up answers to users. Inference is fast becoming the biggest part of the AI chip market. Nvidia CEO Jensen Huang last month argued that Nvidia is well positioned to dominate that shift.
But Nvidia said on Tuesday that the U.S. government was restricting H20 sales to China because of the risk the chips could be used in a supercomputer. While the H20 has lower computing capabilities than other Nvidia chips, its ability to connect to memory chips and other computing chips at high speeds is still high.
Those memory and connectivity aspects could make the H20 useful in building supercomputers in China, and the U.S. has placed restrictions on selling chips for use in supercomputers in China since 2022. The Institute for Progress, a nonpartisan think tank in Washington, D.C., on Tuesday argued for restricting H20 chips, writing that Chinese firms were likely already building such systems.
"At least one of the buyers, Tencent, has already installed H20s in a facility used to train a large model, very likely in breach of existing controls restricting the usage of chips in supercomputers exceeding certain thresholds. DeepSeek’s supercomputer used to train their V3 model is also likely in breach of the same restrictions," the group wrote.
Nvidia said on Tuesday that the U.S. government informed it on April 9 that the H20 chip would require a license to be exported to China and on April 14 it told Nvidia those rules would be in place indefinitely.
It is unclear how many, if any, of those licenses the U.S. government might grant.
Nvidia declined to comment beyond its filing.
The $5.5 billion in charges are associated with H20 products for inventory, purchase commitments and related reserves, Nvidia said.
The news comes after Nvidia said on Monday it was planning to build AI servers worth as much as $500 billion in the U.S. over the next four years with help from partners such as TSMC (2330.TW), in step with the Trump administration's push for local manufacturing.
