Recent Developments on NVIDIA H200 Chip Exports to China

U.S. President Donald Trump announced via Truth Social that the United States would allow NVIDIA ($NVDA) to export its H200 AI chips to “approved customers” in China, marking a partial reversal of longstanding export controls on advanced semiconductors.

Trump stated he had informed Chinese President Xi Jinping of the decision, noting Xi “responded positively.” Shipments are conditioned on U.S. Department of Commerce vetting to ensure national security, and the U.S. government will levy a 25% fee on all such sales—up from an initially floated 15%.

This applies only to the H200, NVIDIA’s second-most advanced chip based on the Hopper architecture; newer lines like Blackwell and Rubin remain fully restricted. NVIDIA welcomed the move, calling it a “thoughtful balance” that supports U.S. jobs and manufacturing while allowing competition.

The announcement boosted NVIDIA shares by about 2.3% in pre-market trading on December 9, with related stocks like AMD rising 1.4% and Intel gaining modestly. Analysts view it as a compromise: it prevents a total cutoff that could accelerate China’s reliance on domestic alternatives like Huawei’s Ascend series, but it doesn’t grant access to cutting-edge tech that could supercharge Beijing’s military AI capabilities.

China Planned Restrictions on H200 Access

Financial Times reported—citing two sources familiar with the matter—that Beijing is preparing to limit access to these H200 chips, even as exports are now permitted from the U.S. side. Chinese regulators are discussing a framework for “limited access,” where potential buyers like tech firms or data centers must submit formal approval requests.

These would require justifying why domestic chips couldn’t meet their needs, aligning with China’s broader “AI self-reliance” push under 2025 legislation. This isn’t entirely surprising. Beijing has ramped up scrutiny on U.S. tech imports amid escalating trade tensions.

In recent months, state-funded data centers have been barred from using NVIDIA chips including the downgraded H20 model in new projects, with ongoing builds required to swap them out entirely. Customs has tightened semiconductor import checks, delaying large GPU shipments.

China aims to triple domestic AI chip output by 2026, favoring homegrown options from Huawei, Cambricon, and others. For context, Huawei’s Ascend 910C lags the H200 in raw performance— 12,032 TPP vs. 15,840 TPP; 3.2 TB/s memory bandwidth vs. 4.8 TB/s but matches or exceeds the export-restricted H20.

Experts like George Chen of The Asia Group suggest China might soften its stance to improve U.S. ties, but the H200’s utility—far superior to the H20—could still face pushback as “sugar-coated bullets” that lock users into NVIDIA’s CUDA software ecosystem, hindering long-term independence.

This could unlock $5-10 billion in pent-up China demand pre-restrictions, China was ~20% of NVIDIA’s revenue, but Beijing’s hurdles may cap it at “limited” volumes. The 25% U.S. fee eats into margins, and smuggling risks persist—two Chinese nationals were arrested in November 2025 for illegally importing NVIDIA chips.

Critics in Congress warn this erodes U.S. tech supremacy, potentially aiding China’s military AI. Trump’s decision follows lobbying from NVIDIA CEO Jensen Huang and comes amid a fragile U.S.-China “truce.” Earlier 2025 bans on H20/H100 exports ironically boosted China’s domestic market share from 14% to 34%, accelerating Huawei’s ecosystem.

X discussions reflect mixed views—optimism on sales revenue vs. concerns over China’s blocks and long-term decoupling. One analyst noted: Even with US approval, China can and likely will restrict H200 imports via domestic procurement rules.

Overall, this saga underscores the chip war’s complexity: U.S. easing for economic gain, met by China’s strategic caution. Watch for formal Chinese guidelines in the coming weeks; if approvals are stingy, NVIDIA’s China rebound could fizzle.