Oossa

Nvidia stops selling H200 AI chips in China after CEO departure

Acting CEO Colette Kress announced Nvidia will no longer ship its H200 chips to China, ending a controversial export loophole.

By Published by Oossa: 2 min read

Veli Yunus Ünal · Unsplash

Nvidia said it will stop selling its H200 AI accelerator chips to customers in China. The announcement came from acting CEO Colette Kress on Tuesday, just days after co‑founder Jensen Huang was removed as chief executive. The move ends a special arrangement that let the chips be shipped through the United States, where a 25% tariff was collected.

The decision was made alongside the creation of a new board‑level committee to watch over national‑security policy, export‑control compliance and global‑trade risks. Nvidia’s China sales have been a flash point for U.S. officials who want to limit Chinese access to cutting‑edge AI hardware. The company had argued that staying in the market helped keep U.S. technology competitive, but the new export rules announced on Monday now bar advanced U.S. chips from foreign data centers that serve Chinese customers remotely.

Why the change matters now

The H200 chips are among Nvidia’s most powerful AI accelerators, used for training large language models and other compute‑heavy workloads. Losing direct access could push Chinese labs to look for alternative suppliers or to set up training runs in data centers outside China that can certify they have no Chinese customers. Export‑control attorney Dan Fisher‑Owens warned that without such certification, many data‑center operators will be forced to buy competing hardware, potentially shrinking Nvidia’s share of the global AI market.

Analysts at Jefferies estimate that the new rules could cut Nvidia’s data‑center revenue by $44 billion to $62 billion next year. Nvidia’s own statement downplayed the impact on its top‑line, saying the backlog of demand for newer chips like the Vera Rubin series will keep revenue flowing for now. However, the company may lose future customers in Southeast Asian markets that often serve Chinese users, such as Singapore, Malaysia, Thailand and Indonesia.

Background on the H200 deal

The H200 arrangement was brokered in 2022 by Jensen Huang and then‑President Donald Trump. It allowed the chips to be exported to China under a special pathway that required U.S. shipment and a 25% tariff, a loophole that U.S. China‑hawk lawmakers criticized as too permissive. In April, Huang told podcaster Dwarkesh Patel that abandoning the Chinese market would be a disservice to the United States, arguing that competition was better than isolation.

Even before Nvidia’s latest move, Chinese authorities had largely refused to accept deliveries of the H200, preferring to develop a domestic chip industry. Researchers have found evidence that China’s leading AI models were trained on Nvidia’s Grace Blackwell chips, which are now prohibited from being exported to China. Those trainings likely took place in overseas data centers, and the new rules will now block further sales to any facility that cannot prove it does not serve Chinese customers.

Why it matters

For Chinese AI labs, the ban means they must either find non‑Nvidia hardware or set up training runs in overseas data centers that can certify no Chinese clients, which could delay projects and increase costs. For U.S. customers, the shift may keep Nvidia’s top‑line revenue stable in the short term but could open space for rival chip makers if export restrictions tighten further.

Was this article useful?
Share

Read next

Oossa · Newsletter

The week in AI, explained

Every Monday: the stories worth knowing, in plain language. Free, no spam.