DeepSeek V4 Runs Natively on Huawei Ascend Chips — China Demonstrates End-to-End AI Stack Without NVIDIA

Fortune·medium signal

DeepSeek V4 is compatible with Huawei Ascend A2, A3, and 950 chips, with V4-Flash partially trained on Ascend 950 clusters. V4-Pro costs $1.74/1M input tokens; V4-Flash costs $0.14/1M input — both cheapest in class. The new Hybrid Attention Architecture enables 1M token context. This represents China's first credible end-to-end sovereign AI stack.