On Tuesday, Nvidia announced it would incur charges totaling $5.5 billion following the U.S. government’s decision to restrict exports of its H20 artificial intelligence chip to China—a crucial market for the company’s products. The H20, Nvidia’s most advanced chip available in China, had become central to its strategy for maintaining a foothold in the country’s rapidly growing AI sector.
U.S. officials, aiming to maintain a competitive edge in the global AI race, have tightened export controls on high-performance chips to China. In response, Nvidia had developed chips designed to stay just within the legal export limits. Despite these efforts, the H20 is now subject to restrictions due to concerns it could be used in supercomputing systems.
Nvidia’s stock fell approximately 6% in after-hours trading following the announcement. According to a February Reuters report, major Chinese firms like Tencent, Alibaba, and ByteDance had been placing large orders for the H20, particularly to support startups like DeepSeek that focus on low-cost AI models. Although the H20 is less capable in training models compared to Nvidia’s international offerings, it excels in inference—an increasingly dominant segment of the AI chip market. Nvidia CEO Jensen Huang recently emphasized the company’s strong position in this emerging area.
However, the U.S. government believes the H20’s high-speed memory and interconnect capabilities pose risks if used in Chinese supercomputers. These features, despite the chip’s lower compute power, could support the development of systems that breach existing U.S. regulations. Export restrictions on chips intended for supercomputing in China have been in place since 2022.
The Institute for Progress, a nonpartisan think tank in Washington, D.C., supported the move, noting evidence that Chinese firms were already misusing the chips. “At least one of the buyers, Tencent, has already installed H20s in a facility used to train a large model—likely in violation of current controls,” the group noted. They also flagged DeepSeek’s supercomputer for potentially breaching the same rules.
Nvidia revealed that the U.S. government informed the company on April 9 that the H20 chip would now require a license for export to China. Five days later, it was clarified that this requirement would remain in effect indefinitely. It remains uncertain whether any export licenses will be approved.
The $5.5 billion charge reflects inventory write-downs, purchase commitments, and related reserves tied to the H20 chip. Nvidia declined to comment further on the matter, and the U.S. Department of Commerce has not yet responded to inquiries.
This development comes just a day after Nvidia announced plans to invest up to $500 billion in AI server production in the U.S. over the next four years, in collaboration with partners like TSMC—a move aligned with the Trump administration’s emphasis on strengthening domestic manufacturing.
