Jensen Huang's Stark Warning: China's 1M AI Workers vs. America's 20k (entropytown.com)

At a private Taipei dinner leaked to the press, Nvidia CEO Jensen Huang warned that U.S. semiconductor export controls are backfiring: rather than stopping China’s AI advance they’ve accelerated it. He cited steep revenue losses for Nvidia (advanced-China sales down from $17B to near zero, with ~$5.5B in charges and ~$15B in lost sales) and pointed to rapid Chinese mobilization — the AI workforce growing from under 10,000 in 2015 to ~52,000 in 2024, more than 30,000 active AI researchers, and 156 institutions each publishing 50+ AI papers in 2024 versus 37 in the U.S. He also highlighted Huawei’s Ascend program, claiming near-parity with Nvidia; independent tests show the Ascend 910C at roughly 60% of H100 inference performance, while industry estimates forecast hundreds of thousands to over a million Ascend dies shipped in 2025–26, helped by TSMC stockpiles and SMIC’s 7nm ramp—though high-bandwidth memory remains a near-term bottleneck (~2M stacks/year supporting ~250–300k high-end chips). For the AI/ML community this signals a likely bifurcation of the global stack: parallel hardware ecosystems, diverging developer toolchains (CUDA vs native Chinese stacks), and shifting talent flows as China becomes a net attractor of researchers. Projections that China could reach >1,100 EFLOPS by 2027, aggressive localization mandates and subsidies, plus open-source efforts like DeepSeek, suggest reduced U.S. leverage over supply chains and developers — raising strategic, research-collaboration, and deployment implications through at least 2027.