Chinese companies are ditching Nvidia’s advanced accelerators for domestic AI suppliers - Fortune

Chinese companies are increasingly shifting away from NVIDIA's advanced AI accelerators toward domestic silicon suppliers, reflecting intensifying US-China tech rivalry and Beijing's push for technolo... Read original →

U.S. export controls on advanced chips are forcing a structural overhaul of China’s AI compute stack. Technically, domestic accelerators—though constrained by lack of 3nm and EUV—are closing performance gaps via chiplet integration and software co-optimization, while memory bottlenecks are accelerating CXMT’s HBM development. Compliance risk has shifted from procurement to ecosystem adaptation, compelling firms to rebuild training frameworks and data center architectures—at 20%+ higher operational costs. Strategically, NVIDIA may route custom chips through foundries in Taiwan, China, while Huawei’s Ascend gains cloud provider traction through full-stack synergy. Within 18 months, localization will pivot from policy mandate to cost-performance validation; if SMIC achieves >50% yield at 7nm, China’s AI chip self-sufficiency could exceed 60% by 2027, redrawing the global AI infrastructure map.