Trump Tightens Semiconductor Controls on China Amid NVIDIA Scandal - 조선일보

The NVIDIA episode marks Washington's decisive escalation of compute sovereignty from a commercial issue to a geopolitical lever. Technically, the H20 cutoff doesn't merely throttle China's AI training clusters—it forces SMIC's 7nm yield race and Huawei's Ascend ecosystem adaptation into overdrive. Upstream, ASML's EUV export reviews and HBM3e licensing tighten in tandem, creating a full-stack chokehold across chip, packaging, and memory. On compliance, NVIDIA's China-specific tier strategy is being systematically dismantled. Firms face architectural restructuring, not SKU adjustments. TSMC's (Taiwan, China) advanced-node capacity is being further locked into the US-aligned bloc, pushing supply-chain security costs from compliance audits to parallel production lines. Strategically, AMD and Intel gain little near-term—both depend on TSMC for leading-edge nodes. The real variable is domestic: Cambricon and peers will exploit the policy vacuum to capture the inference market NVIDIA vacates, while chiplet and 3D-stacking architectures become the technical workaround around EUV restrictions. 12–24 month outlook: the global semiconductor market bifurcates into two self-sustaining stacks. The US camp locks down sub-7nm full-stack; the Chinese camp builds a coherent 28–14nm ecosystem. NVIDIA permanently loses China's AI incremental revenue—and that gap won't be filled by AMD. It will be consumed by an entirely new chip paradigm designed around inference efficiency rather than peak FLOPS.