Nvidia turmoil could shake up China’s AI landscape - Semafor

Nvidia's internal governance turbulence is not a personnel shuffle—it's the structural fracture in its 'tech monopoly plus geopolitical arbitrage' dual-engine model becoming visible. Technical cascade: CUDA's moat is eroding from within. China's AI training clusters remain locked into H20/H800 legacy deployments. Any roadmap disruption at Nvidia activates a hidden single-point-of-failure across the stack—from CoWoS advanced packaging allocation at TSMC (Taiwan, China) to inference framework re-engineering costs for downstream LLM operators. Compliance exposure: Under BIS export-control long-arm jurisdiction, any reshuffling in Nvidia's compliance apparatus risks triggering recalibration of China-bound chip specs. For Chinese AI firms, single-supplier dependency has migrated from tail-risk to probability-weighted event; supply-chain redundancy costs will rise structurally. Competitive chess: Huawei Ascend 910C, Cambricon Siyuan, and Hygon DCUs enter a strategic window. AMD MI300X will almost certainly exploit Nvidia's divided attention to test-market in China, capturing 'non-Nvidia' mindshare. 12–24 month trajectory: China's AI compute market accelerates toward a dual-track regime—legacy Nvidia silicon keeps serving, while incremental procurement tilts decisively toward domestic chips. CUDA fragmentation is now irreversible; de-CUDA-ization of China's AI software stack shifts from policy-driven to market-driven.