In an attempt to stay ahead in the global artificial intelligence race, the US government has been placing limits on what AI chips can be sold to countries like China. The US is skeptical that the latest chip is used in powerful Chinese supercomputers, which may pose a significant national security risk.
Nvidia created the H20 chip to follow the US rules and regulations. However, it still helped the Chinese technology giants, such as ByteDance (which owns TikTok), Tencent, and Alibaba. These companies were using H20 chip to build and run AI systems.
While H20 is not Nvidia's most advanced chip, it works well for a process called "inference," where AI gives answers to users. Inference is becoming one of the most important uses for AI chips today. The US government feels it is too powerful as it connects quickly to memory and other computing parts, which is crucial for building supercomputers. The US has had restrictions on selling chips for use in Chinese supercomputers since 2022.
A US Commerce Department spokesperson said that new regulations require companies to get a license to sell advanced chips like AMD's MI308 and H20 to China. These licenses may not be granted easily. Nvidia was informed about these new permanent rules on April 14. The company didn't say much more publicly.
According to experts from a Washington DC think tank, these restrictions are important as Chinese companies are already using these chips in their supercomputers. They mentioned that Tencent and a startup called DeepSeek may already be breaking the old rules by using H20s to train powerful AI systems.
Because of this ban, Nvidia is facing a whopping $5.5 billion loss, which includes losses due to unsold inventory, cancelled orders, and other costs related to H20 chips. Following the announcement, the company's stock price dropped by 6% and eventually AMD's stock also dropped 7%, as one of its chips is affected, too.
Nvidia is still moving forward despite this setback and recently, it announced that it is planning to build up to $500 billion worth of AI server infrastructure in the US over the next four years. This latest plan of the company aligns with efforts by the Trump administration to support the local manufacturing and reduce dependence on foreign tech markets.
