China launches customs crackdown on Nvidia AI chips (www.ft.com)

Chinese customs has begun a targeted crackdown on shipments of high-end Nvidia AI accelerators, increasing inspections and enforcement around cross‑border movement of GPUs used for large-scale machine learning. The move follows heightened global export-control tensions and focuses on the Hopper‑class and similar datacenter GPUs that are the workhorses for training and running large language models. Companies importing, reselling or cloud‑hosting these cards are seeing longer clearance times, higher compliance costs and uncertainty about inventory flows. The significance is practical and strategic: these accelerators provide the FLOPs and memory bandwidth essential for training state‑of‑the‑art models, so tighter controls can create immediate bottlenecks for Chinese labs, startups and hyperscalers trying to scale models or offer GPU cloud services. Technically, constrained access pushes teams toward more aggressive model parallelism, quantization and distillation to stretch limited GPU resources, and accelerates investment in domestic alternatives and custom inference chips. The enforcement also raises the risk of grey‑market routing, higher prices for hourly GPU access, and a faster timeline for China’s local AI hardware ecosystem to mature—changes that will reshape where and how large models are developed and deployed.