China plans to let top AI firms buy limited amount of Nvidia H200 chips, the Information reports - Yahoo Finance

According to a report by The Information, China is planning to allow its leading AI firms to purchase a limited number of Nvidia H200 chips. The report, citing two sources with direct knowledge, indic... Read original →

China’s limited approval for H200 imports reflects a tactical retreat driven by acute compute shortages. Technically, it eases immediate AI training bottlenecks but delays domestic chip software stack maturity—once developers lock into CUDA, migration costs to local alternatives surge. Compliance-wise, firms now face dual volatility: U.S. export license reversals and China’s allocation quotas, heightening supply chain fragility. Competitively, Huawei Ascend and Cambricon may rush CUDA-compatible middleware to capture this window, while NVIDIA leverages the opening to embed its upcoming Vera architecture deeper into China’s residual AI infrastructure. Over the next 12–24 months, expect hybrid compute clusters: H200s for core model training, domestic chips for edge inference. Yet without breakthroughs in sub-7nm processes—blocked by EUV restrictions—China’s AI foundation will remain capped by external tech ceilings.