U.S. export controls reshape China chip market

AI chips are turning into two different markets. One is built around Nvidia’s software and U.S. supply chains. The other is being forced together inside China with Huawei chips, local foundries, and whatever tools Washington still allows through. This week made that split a lot clearer. The U.S. moved to stop some equipment shipments to Hua Hong, and a U.S. senator demanded answers about whether Nvidia’s more powerful H200 chips are reaching China anyway. (usnews.com) ### What changed this week? The immediate news is about two pressure points. First, the Commerce Department ordered several equipment suppliers to halt certain shipments to Hua Hong, China’s second-largest chipmaker, for two facilities U.S. officials think could make the country’s most advan(usnews.com)utnick and Nvidia CEO Jensen Huang seemed to clash. (usnews.com) ### Why does Hua Hong matter? Because export controls are no longer just about the chip itself. They are about the factory that could make the next domestic substitute. Reuters reported in March that Hua Hong Group had developed manufacturing technology usable for AI chips, with Huali Microelectronics preparing a 7-nanometer process in Shanghai. If that ramps, China is no longer relying on one advanced domestic foundry. It starts building redundancy. (money.usnews.com) ### Why is Nvidia stuck in the middle? Nvidia spent years trying to stay in China by designing downgraded chips that fit under U.S. rules. The H20 was the key example — basically the legal minimum viable Nvidia AI chip for that market. But in April 2025 the U.S. told Nvidia that even H20 exports to China would need a license indefinitely, and Nvidia said the move would trigger a $5.5 billion charge. That was the moment the workaround stopped looking durable. (techcrunch.com) ### So is Huawei the main winner? Increasingly, yes. Reuters reported this week that demand for Huawei’s Ascend 950 chips surged after DeepSeek’s V4 model launched on Huawei hardware, with big Chinese tech firms scrambling to secure supply. That matters more than patriotic branding. Buyers are signaling that local chips are becoming usable enough for serious model training and deployment, especially when Nvidia access is uncertain. (msn.com) ### Why is the H200 fight important? Because it shows the rules are still being contested at the top end. Coons asked for hard numbers — how many H200 chips got export licenses, how many shipped, and how many more Commerce plans to approve. Lutnick had told senators on April 22 that no such chips had been so(msn.com) still probing for openings. (cnbc.com) ### What does this do to software teams? It breaks the old assumption that one accelerator stack can serve everyone. In the U.S. and allied markets, teams optimize for CUDA and Nvidia’s ecosystem. In China, they increasingly need code that can run on Ascend hardware or on domestically built systems with different compiler, networking, and memory constraints. Think of it li(cnbc.com)and becomes survival gear. (usnews.com) ### Is this just a temporary squeeze? Probably not. The pattern since 2022 has been ratchet, not reset — tighter chip rules, tighter tool rules, then pressure on the fabs that might close the gap. China’s response has been the mirror image: build local substitutes faster, even if they are less efficient at first. That feedback loop is what turns a trade restriction into a market redesign. (usnews.com) ### Bottom line This is no longer just a story about whether Nvidia loses a few sales in China. It is about whether the global AI hardware market stays one market at all. Right now, the answer looks more and more like no. (usnews.com)