China’s leading AI firms remain dependent on Nvidia chips despite US sanctions and national self-sufficiency goals. High costs and technical complexities currently prevent these companies from switching to domestic hardware.
Nvidia’s CUDA software platform serves as the primary obstacle due to its status as the global industry standard. Migrating to domestic alternatives like Huawei’s Compute Architecture for Neural Networks (CANN) requires extensive and expensive code rewriting.
This software entrenchment secures Nvidia’s market position. Consequently, China faces significant delays in establishing an independent AI hardware and software ecosystem.