China’s AI industry is reshaping its competitive model around domestic computing
power after DeepSeek released and open-sourced its DeepSeek V4 model with full
adaptation to Huawei Technologies Co. Ltd. Ascend chips, Xinhua reports. The
deployment marks the first full-stack rollout of a Chinese large language model
on domestic chips from training to inference, underscoring a shift toward
self-reliant AI infrastructure amid tighter external technology curbs.
The integration offers an alternative to CUDA and challenges doubts over whether
domestic chips can support frontier AI models, showing that algorithm-hardware
optimization can handle trillion-parameter inference workloads. Chinese
chipmakers including Cambricon Technologies Corp. Ltd., Moore Threads
Intelligent Technology (Beijing) Co. Ltd. and Hygon Information Technology Co.
Ltd. also announced “Day 0” compatibility with DeepSeek V4, signaling faster
coordination across China’s domestic AI computing supply chain.
