NVIDIA takes $5.5 billion hit as US bans AI chip exports to China

[WORLD] NVIDIA announced on Tuesday that it would incur $5.5 billion in penalties after the US government restricted exports of its H20 artificial intelligence chip to China, a crucial market for one of its most popular chips.

The restrictions highlight the escalating tensions between the U.S. and China over advanced technology, particularly in AI and semiconductors. The Biden administration has steadily tightened export controls since 2022, aiming to curb China’s ability to develop cutting-edge AI for military and surveillance applications. NVIDIA, as the dominant player in AI chips, has found itself at the center of this geopolitical struggle, forcing the company to navigate complex trade policies while protecting its market share.

Nvidia's AI processors have been a major focus of US export controls, as officials have attempted to prevent the most advanced chips from being exported to China as the US attempts to stay ahead in the AI race. Following the implementation of these regulations, Nvidia began building processors that were as close to US limits as possible.

Nvidia shares fell roughly 6% in after-hours trade. The H20 is currently Nvidia's most sophisticated chip for sale in China, and it is critical to the company's aspirations to remain connected with China's thriving AI industry.

Analysts suggest that NVIDIA’s financial hit reflects not just lost sales but also the broader challenge of adapting to rapidly shifting regulations. The company had already developed downgraded versions of its flagship chips, such as the A800 and H800, to comply with earlier restrictions. However, the latest curbs on the H20 indicate that even these modified chips may no longer pass U.S. scrutiny, leaving NVIDIA with limited options in one of its largest markets.

Chinese corporations such as Tencent, Alibaba, and TikTok parent ByteDance have increased orders for H20 processors in response to rising demand for DeepSeek's low-cost AI models.

While the H20 chip is not as fast at training AI models as Nvidia's processors for sale outside China, it is comparable to some of those chips in the inference stage, where AI models provide responses to users. Inference is quickly becoming the largest segment of the AI chip market.

The inference market’s growth is driven by the widespread adoption of generative AI tools like ChatGPT, which require massive computing power to deliver real-time responses. NVIDIA had positioned the H20 as a cost-effective solution for Chinese firms looking to deploy AI applications at scale, but the new restrictions could force these companies to turn to domestic alternatives like Huawei’s Ascend chips or develop in-house solutions—potentially accelerating China’s push for self-sufficiency in semiconductors.

Last month, Nvidia CEO Jensen Huang said that the company is well-positioned to dominate this transformation. However, Nvidia announced on Tuesday that the US government is banning H20 sales to China due to concerns that the chips could be used in a supercomputer.

While the H20 has fewer computing capabilities than other Nvidia processors, its ability to connect to memory chips and other computing chips at high rates remains strong. These memory and connection features could make the H20 beneficial in the development of supercomputers in China, since the United States has set limits on the sale of chips for supercomputers since 2022.

The U.S. has long been concerned about China’s supercomputing ambitions, given their applications in nuclear simulations, hypersonic weapons research, and advanced surveillance systems. By targeting the H20, regulators are signaling that even chips with reduced performance could contribute to China’s military-civil fusion strategy—a policy that blurs the line between commercial and defense-related technological advancements.

The Institute for Progress, a nonpartisan research group in Washington, D.C., pushed on Tuesday for limiting H20 chips, claiming that Chinese manufacturers were likely already developing such systems.

"At least one of the bidders, Tencent, has already installed H20s in a facility intended to train a huge model, most likely in violation of existing restrictions prohibiting the use of chips in supercomputers that surpass specified criteria. DeepSeek's supercomputer, which was utilized to train their V3 model, is likewise likely to violate the same limits," the group wrote.

Nvidia stated on Tuesday that the US government informed it on April 9 that the H20 chip would require a license to be sold to China, and on April 14 that those requirements would remain in effect forever. It is uncertain how many, if any, of those licenses the US government may issue.

The indefinite nature of the restrictions raises questions about NVIDIA’s long-term strategy in China, which accounts for roughly 20% of its revenue. Some industry experts suggest the company may need to further diversify its supply chain or invest more heavily in markets like India and Southeast Asia to offset potential losses. Meanwhile, Chinese tech giants could face delays in AI deployment, though domestic chipmakers may see an opportunity to fill the gap.

Nvidia declined to comment beyond the filing. The US Department of Commerce, which supervises US export regulations, did not immediately respond to a request for comment. Nvidia stated the $5.5 billion in costs are for H20 product inventory, procurement agreements, and related reserves.

The announcement comes as Nvidia announced on Monday that it plans to create AI servers worth up to $500 billion in the United States over the next four years with the support of partners like as TSMC, in line with the Trump administration's goal for domestic manufacturing.