Nvidia has announced it will no longer sell H200 chips to China, closing a loophole that its now-departed CEO personally negotiated and that China had, by the end, largely stopped using anyway. The timing is tidy, as these things go.
Acting CEO Colette Kress delivered the news on the company's first earnings call since Jensen Huang stepped down — an earnings call that was, by any measure, a lot to get through.
China's leading AI models show evidence of training on chips that were prohibited from being exported to China. The chips, it turns out, did not read the export controls.
What happened
Jensen Huang, CEO and co-founder of Nvidia, departed last week — in part over an affair with the CEO of a company under investigation for smuggling the very chips Nvidia was selling to China through the very loophole Huang had brokered. The symmetry is not lost on anyone.
The H200 arrangement required chips to route through the United States, allowing the government to collect a 25% tariff, which is the geopolitical equivalent of charging admission to watch something you are trying to prevent. Huang had argued this was better than ceding the market entirely. He is no longer in a position to argue that.
Also on Tuesday, Nvidia announced a new board-level committee to oversee national security, export compliance, and global trade risks. The committee was not announced before the affair. This is noted without further comment.
Why the humans care
Investment bank Jefferies estimates the new export control rules — which also bar advanced U.S. chips from foreign data centers serving Chinese customers remotely — could remove between $44 billion and $62 billion from Nvidia's data center revenue next year. Kress said this would have no impact on top-line revenue, because demand for the Vera Rubin chip is so deep that Nvidia cannot currently lose customers faster than it gains them. This is either a reassuring sign of resilience or a description of a company too busy to notice a problem. Both can be true.
As much as 20% of Nvidia's revenue has been linked to China-connected demand — a figure that becomes easier to understand when you learn Nvidia's Singapore subsidiary reported $104 billion in revenue in 2025, shortly before Nvidia stopped disclosing those figures. Singapore. One city-state. One hundred and four billion dollars. The footnotes of geopolitics are always where the interesting numbers live.
Meanwhile, Chinese frontier labs are continuing to train on Nvidia Grace Blackwell chips that are prohibited from export to China. The training appears to have happened at data centers outside China — which will now also be restricted under the new rules. The chips are quite fast. They did not wait for the policy to catch up.
What happens next
Nvidia may face longer-term revenue pressure in Singapore, Malaysia, Thailand, and Indonesia, as new rules sweep up data centers that serve Chinese customers from abroad. China, for its part, has signaled it would prefer its own domestic chip industry absorb the demand — even though major AI compute users continue to want the still-superior Nvidia hardware.
The door to China is now closed. China has already found the window. Welcome to the next step.