Nvidia’s H100 and H200 GPUs Are Now Banned from China — and the AI Chip War Just Got Serious
The US just banned Nvidia H100 and H200 sales to China, costing Nvidia an estimated $3B annually and pushing Chinese AI labs further toward domestic chips from Huawei and Alibaba.
The US government just made its most aggressive semiconductor move yet. Effective today, February 26, 2026, the Department of Commerce has formally added Nvidia’s H100 and H200 GPUs to the Entity List, prohibiting US companies — and their foreign subsidiaries using US-controlled technology — from selling these chips to Chinese entities. This isn’t a warning shot. It’s a direct hit on the hardware stack that powers most serious AI development worldwide.
The H100 has been the de facto standard for training large language models since 2022. With 80GB of HBM3 memory and 1,456 GB/s of bandwidth, it’s the chip that trained most of the models you’ve actually heard of. Its successor, the H200, pushes that further with 141GB of HBM3e memory. Both are now off the table for Chinese buyers. For Nvidia, the pain is concrete: Greater China brought in approximately $9.7 billion in revenue during FY2024, and analysts estimate the H100/H200 lines account for a significant slice of that. A $3 billion annual hit is the number circulating in financial circles.
How Did We Get Here?
This didn’t come out of nowhere. The US has been tightening the semiconductor screws on China since August 2022, when the Commerce Department first flagged H100 and A100 exports. October 2022 brought formal restrictions. Then came wave after wave of Entity List additions through 2023 and 2024. Today’s action is the logical endpoint of a policy that has been building for nearly four years — except now there’s no ambiguity and no workarounds. The H100 and H200 are explicitly, formally prohibited.
The national security argument from Washington is consistent: advanced AI chips can train military AI systems, power surveillance infrastructure, and accelerate weapons development programs. Whether you find that convincing or alarmist likely depends on which side of the Pacific you’re on.
Nvidia Loses, But Who Gains?
Jensen Huang has publicly positioned Nvidia as supportive of national security goals while quietly watching billions in revenue disappear. The company had already been selling a downgraded China-specific variant (the H800) after the 2022 restrictions, so some of this pain was anticipated. But today’s ban closes the remaining door. There is no compliant Nvidia workaround left.
The beneficiaries are obvious: Huawei’s Ascend series and Alibaba’s in-house AI chip development have been racing to fill exactly this vacuum. Huawei’s Ascend 910B has been benchmarked against the H100, and while it doesn’t match it on raw performance, it’s credible enough for inference workloads and increasingly for training. Alibaba, meanwhile, has been quietly developing silicon alongside its Qwen model series. China’s domestic chip ecosystem is nowhere near as mature as Nvidia’s, but export controls are the best subsidy Beijing never had to pay for.
The Bigger Picture: Two AI Ecosystems
The real consequence of this ban isn’t a short-term slowdown in Chinese AI development — it’s the acceleration of a bifurcated global AI hardware stack. American AI runs on Nvidia (and increasingly AMD and custom silicon from Google and Meta). Chinese AI is being forced onto a different architecture entirely. Over time, that divergence shapes which models get built, how fast, and for what purposes.
For the rest of the world — European labs, Middle Eastern sovereign AI projects, Asian research institutions — the question of which ecosystem to build on just got more complicated. US export controls don’t only hit China. They create uncertainty for any non-US entity that might end up on a list someday.
What’s Next?
Nvidia’s stock took an immediate hit on the news, which is the market’s way of saying it believes the revenue loss is real even if it was partly priced in. The Commerce Department is expected to expand the Entity List further in coming months, with Nvidia’s next-generation Blackwell architecture already being discussed in policy circles as a potential future restriction target. Meanwhile, Chinese AI labs are not sitting still — expect announcements from Huawei and Alibaba on next-generation domestic chips within the year. The US drew a hard line today. China’s response will be hardware, not diplomacy.
