China’s ‘New Way’ of Breaking Into NVIDIA’s CUDA Moat Isn’t by Building a Replica: It’s by Changing the Way We See Hardware

Chinese semiconductor leaders are urging China to move away from NVIDIA CUDA dependence by pursuing software-defined chips.

Wei Shaojun argued that building a direct CUDA clone is too costly and that compiler-driven reconfigurable hardware is a better long-term path.

The idea is to shift more intelligence into software and reduce reliance on Western AI software stacks.

But the approach remains technically difficult and is more viable today for specialized workloads than for replacing GPUs outright.