Nvidia Pulls H200 Chips From China as Post-Huang Era Begins
Nvidia will stop selling H200 chips in China, acting CEO Colette Kress said Tuesday, ending a loophole brokered by Jensen Huang and President Trump as new export rules bite.
By Grace Kim
3 min read

What's News
- Nvidia will stop selling H200 chips in China, acting CEO Colette Kress announced Tuesday
- Jefferies forecasts new export rules could cut $44–62 billion from Nvidia's 2026 data center revenue
- Nvidia's Singapore subsidiary posted $104 billion in revenue in 2025 before disclosure stopped
- New rules issued Monday bar advanced chips from foreign data centers serving Chinese customers remotely
- Critics estimate up to 20% of Nvidia revenue links to China-related demand
Nvidia will stop selling its H200 chips in China, acting CEO Colette Kress announced Tuesday, closing a controversial loophole in the U.S. ban on advanced chip exports to the country. The decision lands one week after co-founder Jensen Huang stepped down as CEO — and on the same day Nvidia formed a board-level committee to oversee national security policies, export control compliance, and global trade risks.
What was the H200 arrangement?
The deal Kress just killed was personal. Huang brokered it directly with President Donald Trump. Under the structure, H200 chips shipped to China had to pass through the U.S., letting the government collect a 25% tariff on every unit. It was a workaround that kept Nvidia inside the world's second-largest AI market while technically honoring export restrictions.
China hawks hated it. They wanted to deny Chinese labs computing power and preserve the American lead in AI. Huang pushed back for years. In April, he told podcaster Dwarkesh Patel: "to concede that market for the United States technology industry is a disservice to our country."
That argument ended with his departure.
How much money is on the line?
The numbers get big fast. New export rules issued Monday bar advanced U.S. chips from foreign data centers that serve Chinese customers remotely. Investment bank Jefferies forecasts those rules could trim between $44 billion and $62 billion from Nvidia's data center revenue next year.
Kress told investors not to panic. The backlog of demand for advanced chips like Vera Rubin, she said, means the new rules will have no impact on the company's top-line revenue.
The longer-term exposure spreads beyond China. Nvidia may lose potential customers in Singapore, Malaysia, Thailand, and Indonesia. Critics of the company's China dependence claim as much as 20% of its revenue is tied to demand linked to the country. Nvidia's Singapore subsidiary reported $104 billion in annual revenue in 2025 — the last year the company disclosed that figure.
In short: near-term revenue looks safe, but the addressable market is shrinking at the edges.
Did China even want the chips anymore?
Not really. Even before Nvidia decided to stop shipping H200s, China had largely refused to accept delivery. Jacob Feldgoise, a researcher at Georgetown's Center for Security and Emerging Technology, said the government prefers to generate demand for its domestic chip industry — even though major AI compute users still want Nvidia's superior equipment.
The twist: Chinese frontier labs keep getting Nvidia silicon anyway. All of China's leading AI models show evidence of having been trained on Nvidia Grace Blackwell chips, which are prohibited from export to China. That training likely happened at data centers outside China.
What do the new rules actually block?
Monday's rules target exactly that workaround. Export control attorney Dan Fisher-Owens told TechCrunch that large data centers suited for AI training will now have to certify they aren't serving any Chinese customers if they want to buy Nvidia chips.
If a facility can't provide that certification, it faces a choice: drop its Chinese clients or drop Nvidia. Many may move to other chip suppliers, further eating into Nvidia's market share.
The certification requirement turns buyers into enforcement agents — and gives Nvidia's competitors an opening at every data center with Chinese customers on the books.
What happens next?
Nvidia enters the post-Huang era with a narrower map. Its acting CEO has pledged national security oversight at the board level, closed the H200 channel, and bet the revenue impact on a Vera Rubin backlog that stays hot. Whether that backlog survives a market where uncertified data centers drift toward rival chips is the question investors will watch through next year.
Original: dwarkesh.com
More from Grace Kim
Show full bio
Market editor covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.
2,301 articles