Source: Scmp
Despite years of U.S. export controls and state pressure to adopt domestic alternatives, Chinese AI research institutions continue standardizing on Nvidia's CUDA platform because the switching costs are prohibitive—rewriting codebases for Huawei's CANN requires months of engineering work and introduces performance unknowns. Technical lock-in compounds geopolitical fragmentation: even well-resourced Chinese labs find it economically rational to circumvent sanctions rather than abandon proven infrastructure, which perpetuates their dependence on American semiconductors.