China’s artificial intelligence ecosystem has reached a new milestone as a Shenzhen-based research collaboration successfully trained a 1.6 trillion-parameter large language model, DeepSeek-V4-Pro, entirely on a domestic computing platform powered by the Ascend 910C AI cluster.
The project brought together experts from Shenzhen Hetao College’s AI training platform team, Harbin Institute of Technology (Shenzhen), the Shenzhen Institute of Big Data, and Huawei. Their achievement showcases the growing maturity of China’s homegrown AI infrastructure and highlights its ability to handle some of the most demanding machine learning workloads.
Unlike inference-focused deployments that rely on pre-trained models, full-parameter training requires enormous computational resources, advanced system optimization, and stable large-scale hardware coordination. Successfully completing training at this scale demonstrates that domestic AI chips are capable of supporting frontier-level model development from the ground up.
The accomplishment is viewed as an important validation of China’s independent AI computing ecosystem, reducing reliance on foreign hardware while proving that locally developed platforms can compete in training next-generation foundation models. As AI competition intensifies globally, this milestone strengthens confidence in the future of domestically powered large-scale AI innovation. #AI
