Nvidia spying scandal emboldens Washington’s China hawks - Semafor

The real damage isn't the chip itself—it's the narrative anchor it hands Washington. Once AI compute is reclassified from commercial commodity to strategic asset, export control logic shifts from 'generational gap protection' to 'sovereignty isolation,' and the policy floor becomes structurally irreversible. Technically, if H20 is fully banned, Chinese training clusters accelerate migration to Ascend 910C. But the true bottleneck migrates downstream: HBM memory (SK Hynix, Samsung) and CoWoS advanced packaging in Taiwan, China. The chokepoint shifts from GPU die to memory and packaging—creating new dependency vectors that no single ban can resolve. On compliance, multinational AI firms are forced into dual-stack architectures. R&D, validation, and ops costs inflate 20-35%. This is a permanent structural overhead, not a one-time hit. Competitively, AMD gains little (identical export constraints apply). The real winner is Huawei—being banned confers a 'political correctness premium' that converts domestic procurement from cost-benefit analysis to security-mandated necessity. 12-24 month outlook: the global AI chip market fractures from 'one ecosystem, two markets' into 'two closed-loop ecosystems.' Nvidia's revenue ceiling contracts from global to non-China markets; its valuation logic faces fundamental re-rating. The parallel stack—Ascend, Cambricon, Yangtze Memory—matures from 'backup' to 'default.'