China scales AI without US chips

Artificial intelligence computing works like a giant pool of specialized processors handling many calculations at once. China has now put a 60,000-chip scientific AI cluster into operation in Zhengzhou using domestic accelerators rather than U.S. chips. (scmp.com) State broadcaster CCTV said the Zhengzhou core node entered operation on April 14, 2026, at the National Supercomputing Internet, after starting trial runs in early February with more than 30,000 domestically made accelerator chips. By mid-April, that count had doubled to 60,000. (scmp.com) China Daily identified the supplier as Dawning Information Industry, also known as Sugon, and said the system is dedicated to “AI for Science,” where models are used for work such as drug discovery, new materials and clean energy research. (chinadaily.com.cn) The project sits inside a part of AI that is less about chatbots and more about speeding up lab work and simulations. In practice, that means using AI systems to search huge chemical, biological and engineering datasets faster than conventional computing alone. (chinadaily.com.cn) Washington’s chip curbs were written to make advanced computing hardware harder for China to buy. The U.S. Commerce Department said on October 7, 2022 that the rules were meant to restrict China’s ability to obtain advanced chips, develop supercomputers and manufacture advanced semiconductors. (bis.gov) The controls expanded again in January 2025, when the Bureau of Industry and Security revised rules on advanced computing chips and added a new control on certain artificial intelligence model weights. The Trump administration later said it rescinded the January 15, 2025 AI Diffusion Rule while adding other chip-related restrictions. (federalregister.gov, bis.gov) Chinese state media cast the Zhengzhou system as proof that a domestic stack can run at national scale. CGTN said the cluster built an integrated local ecosystem for data, computing, models and applications, and that it would offer open services across research and industry. (news.cgtn.com) Outside analysts have argued that export controls can slow China more than stop it. A Council on Foreign Relations analysis published in December 2025 said China still faced an AI chip deficit even as domestic firms such as Huawei improved their alternatives to Nvidia. (cfr.org) The Zhengzhou build-out does not show that Chinese chips match the best U.S. hardware on every measure. It shows that, for at least one large scientific cluster, China assembled enough domestic accelerators in about two months to move from trial operation to full-scale deployment. (scmp.com, globaltimes.cn) That leaves the immediate test in Zhengzhou straightforward: whether a cluster built without American chips can keep feeding Chinese labs the computing power they need, at the scale Beijing says it has now reached. (globaltimes.cn)