māchine · nvidia

Nvidia stops shipping H200 chips to China despite US approval

Interim CEO Colette Kress says Nvidia is no longer shipping licensed Hopper data center chips to China and has left them out of its forecast. She gave no reason.

Nvidia has stopped shipping its licensed Hopper data center chips, including the H200, to China. Interim chief executive Colette Kress said so on the company's earnings call on Tuesday, 17 November: "Our guidance does not have any licensed Hopper data center products included, and we are no longer shipping licensed Hopper data center products to China." She did not say why, and she did not say whether Nvidia would seek other business in China.

Washington had opened this market less than a year ago. In December 2025, President Trump said Nvidia could sell the H200 to approved customers in China, with 25 percent of the revenue going to the US government. Under former CEO Jensen Huang, Nvidia had pushed for that permission. Huang resigned last week. Washington still bars the newer Blackwell and Rubin chips from China, so Nvidia's decision leaves Chinese buyers with no current Nvidia data center chips sold through licensed channels.

What Nvidia reported

The announcement came with results for the third quarter of fiscal 2027, which ended on 25 October 2026. According to Nvidia's earnings release, revenue was $115.8 billion, up 103 percent from a year earlier. Data center revenue was $107.6 billion. Nvidia expects about $130 billion in the fourth quarter, plus or minus 3 percent, and says that forecast assumes no data center compute revenue from China.

Kress told analysts that third-quarter sales of licensed Hopper data center products to China came to $2.4 billion, AFP reported. That is about 2 percent of data center revenue for the quarter. The release itself focuses on Nvidia's next generation of hardware. "Vera Rubin is off to an explosive start," Kress is quoted as saying.

The call was Nvidia's first under new leadership. In an 8-K filing dated 12 November, Nvidia said Huang had resigned as president, CEO and board member with immediate effect. The filing gives no reason. Kress stays on as chief financial officer while serving as interim CEO, and Harvey C. Jones became the company's first chair of the board. The Verge reports that Huang left after reports of several extramarital affairs, one of them with a woman whom US intelligence suspected of being a Chinese spy. AFP identifies her as Alice Huang, a former director of a Singapore-based cloud company. AFP says she previously led Megaspeed International, which is under investigation over claims that it helped Chinese firms obtain Nvidia chips in violation of US export controls.

The chips were already stuck

The H200 business had problems before the scandal. The Verge notes concerns that Chinese companies approved to handle the chips had not been properly vetted. Earlier this month Reuters reported that around 600,000 H200 chips were waiting in Hong Kong because Beijing had not approved large-scale imports. China wants to rely less on US technology.

So the halt may change less on the ground than the headline suggests. Chandy Ye, a lawyer in Hong Kong, told AFP that the practical impact would be limited because Chinese authorities have steered companies away from the H200 since January 2026. Pepperstone analyst Dilin Wu told AFP that "the timing says quite a lot." Wu suggested Nvidia may want to show regulators that compliance comes first.

The rule on remote access

Ye called a separate change the "far more consequential development". On 16 November the Commerce Department issued an interim final rule on remote access to advanced computing. Under the rule, exporters need a licence to send covered chips and servers to any data center in the world that may supply more than 12 million TPP of compute to companies headquartered in China or other restricted D:5 countries. That includes companies whose ultimate parent is based there. TechCrunch puts the threshold at roughly the peak compute of 750 H100 or H200 chips.

The rule also adds licence requirements for advanced chip exports to Indonesia, Malaysia, Singapore, Thailand and Brazil, which Commerce says carry "a higher-than-normal risk of diversion". Data centers that already provide covered services have 90 days to get a licence. Recommendation systems used in e-commerce and social media are exempt. The rule also formally rescinds the Biden administration's AI Diffusion Rule. Commerce argues that this much computing power "can be used to train and run advanced SI models capable of supporting military- and intelligence-related activities."

Taken together, the two moves narrow both ways Chinese labs could get frontier-class compute: buying the chips, and renting them abroad. Reuters has reported that Chinese AI firms were rushing to secure overseas computing capacity before the restrictions took effect.

What it means for work

For engineers at Chinese AI labs, training and running large models now depends even more on domestic accelerators, chips already in the country, and whatever licensed overseas capacity remains. Moving workloads between hardware platforms is slow, specialised work, and it is likely to take up more of their time. Cloud and data center operators in Southeast Asia and Brazil get new compliance work: checking who ultimately owns their customers, monitoring use and applying for licences within 90 days. For Nvidia, the lost China revenue is small next to its growth elsewhere. Its new leadership is clearly putting regulatory standing ahead of this market. Still open: why Nvidia stopped, what happens to the chips held in Hong Kong, and whether Chinese chips can close the gap for frontier training.

Sources

This article was researched, written and illustrated by AI, and published without human editing. Spotted an error? Let us know.