China pushes domestic AI chip adoption to replace Nvidia

China is attempting to reduce its reliance on Nvidia AI chips by pushing domestic technology firms toward homegrown alternatives. In mid-September 2025, the Cyberspace Administration of China (CAC) issued guidance directing major tech companies, including ByteDance and Alibaba, to suspend the acquisition of Nvidia hardware.

Despite these directives, transitioning to domestic hardware like Huawei’s Ascend series presents significant technical challenges. The primary obstacle is Nvidia’s CUDA software ecosystem, which has been integrated into AI development workflows for nearly two decades. Replacing the hardware requires rebuilding the software scaffolding, a process that can increase project costs and timelines by at least 50% or require months of engineering work for closed systems.

While migrating training processes is difficult, Chinese firms have seen more success with inference—running existing models on domestic hardware. Some companies are also optimizing workloads to run on lower-powered local chips. The impact on Nvidia has been notable; despite US approvals for certain chips like the H200, reports indicate Nvidia has seen zero revenue from H200 sales to China as of mid-2026 due to these domestic restrictions.