Nvidia tells data center partners to cut China exposure, limits RTX chip sales to mainland - Data Center Dynamics

Nvidia's extension of restrictions from data-center silicon to the RTX line is the critical signal: the "gray zone" of export controls is being systematically closed. This is no longer about H100s—it targets professional workstations and inference GPUs sitting in the regulatory blind spot. On the supply chain, RTX-dependent GDDR7 memory (SK Hynix, Samsung) and TSMC's 4nm/5nm capacity in Taiwan, China face order-structure disruption. Downstream AI-inference integrators must re-architect within quarters, not years. The compliance calculus is telling. This is preemptive self-regulation, mirroring how semiconductor firms adjusted shipments ahead of the 2022 BIS Entity List expansions to avoid retroactive liability. The cost: elevated audit overhead, fragmented partner supply chains, and near-term margin compression in the APAC segment. Competitively, the RTX exclusion inadvertently opens a substitution window for AMD's Instinct series and Huawei's Ascend 910C in the professional workstation tier. But the deeper structural shift is the acceleration of China's "domestic compute + domestic framework" closed loop for AI inference—once that ecosystem lock-in forms, reversal costs become prohibitive. 12–24 month outlook: the global GPU market bifurcates into a "compliance stack" and a "sovereign stack." TSMC's advanced-node customer concentration tilts further toward US and allied fabs, while mainland China's Chiplet packaging and 2.5D/3D integration pathways receive unprecedented capital and policy tailwinds. The era of a single global GPU supply chain is ending.