KEY POINTS
- Nvidia will stop selling H200 chips in China, acting CEO Colette Kress said on Tuesday.
- New U.S. rules issued Monday bar advanced chips from foreign data centers that serve Chinese customers remotely.
- Jefferies estimated the rules could cut $44 billion to $62 billion from Nvidia's data center revenue next year, while Kress said there would be no top-line impact.
Nvidia will no longer sell its H200 AI chips in China, acting CEO Colette Kress said on Tuesday, according to TechCrunch. The decision ends the company's use of a controversial exception to the U.S. ban on exporting advanced chips to the country.
Kress spoke on Nvidia's first earnings call since co-founder and CEO Jensen Huang stepped down last week. On the same day, the company said it was forming a board-level committee to oversee national security policy, export control compliance and global trade risks.
The H200 arrangement had been negotiated by Huang and President Donald Trump. It required the chips to ship through the United States so the government could collect a 25% tariff. China hawks disliked the setup because they wanted to keep Chinese labs short of computing power. Huang had argued that the U.S. must compete in the Chinese market, telling podcaster Dwarkesh Patel in April that giving it up would be a disservice to the country.
The announcement lands alongside new rules issued Monday that prohibit advanced U.S. chips from going to foreign data centers that serve Chinese customers remotely. Investment bank Jefferies forecast that the rules could reduce Nvidia's data center revenue by $44 billion to $62 billion next year. Kress said the backlog of demand for chips such as Vera Rubin means the rules would not affect the company's top-line revenue.
Over the longer term, Nvidia could lose potential customers in Singapore, Malaysia, Thailand and Indonesia. Critics of the company's reliance on China say as much as 20% of its revenue is tied to demand linked to the country. They point to its Singapore subsidiary, which reported $104 billion in annual revenue in 2025 before Nvidia stopped disclosing those figures.
China had already largely declined to take delivery of H200 chips, TechCrunch reported. Jacob Feldgoise, a researcher at Georgetown's Center for Security and Emerging Technology, said Beijing would rather build demand for its domestic chip industry, even though major AI users still prefer Nvidia's more capable hardware.
Chinese labs have kept finding ways to reach Nvidia hardware. According to the report, all of China's leading AI models show signs of training on Nvidia Grace Blackwell chips, which cannot legally be exported to China. That training may have happened in data centers outside the country, which the new rules would now cut off from sales.
Export control attorney Dan Fisher-Owens told TechCrunch that large data centers suited to training AI models will have to certify that they do not serve any Chinese customers if they want to buy Nvidia chips. Operators unable to give that assurance may turn to other chip suppliers, which could further reduce Nvidia's market share.