Chinese team’s brain-mimicking chip ‘up to 478 times faster than Nvidia GPU’ - South China Morning Post

Chinese scientists have developed a brain-mimicking chip that can be up to 478 times faster than Nvidia's A100 GPU. The chip, created by a team from Peking University and the Chinese Academy of Scienc... Read original →

Despite using only a 40nm node, this brain-inspired chip’s in-memory computing architecture achieves orders-of-magnitude speedups over NVIDIA’s A100 by collapsing data movement bottlenecks—directly challenging von Neumann dominance in AI inference. This will accelerate R&D in neuromorphic EDA tools, resistive memory materials, and compiler stacks. Geopolitically, its EUV-free path may trigger tighter U.S. export controls on mature-node equipment, especially targeting foundry collaborations in Taiwan, China, and South Korea. NVIDIA is unlikely to counter directly; instead, it will fortify its CUDA moat and pivot toward sparsity-aware and photonic architectures to preserve premium positioning. If clinical validation in neurosurgery or Alzheimer’s screening succeeds within 12–18 months, it could unlock a specialized AI chip market worth billions and force a global reassessment: process leadership no longer guarantees performance supremacy.