Chinese chipmakers integrate DeepSeek V4 across domestic hardware

Chinese semiconductor manufacturers including Huawei and Cambricon are rapidly integrating DeepSeek's V4 LLM across domestic platforms. Huawei adapted V4 to its Ascend 950PR chip, reportedly achieving 2.87x single-card inference performance vs NVIDIA's H20 and a 60% improvement in multimodal generation efficiency at lower deployment cost.