China unveils first domestic GPU and brain-inspired chip hybrid inference system for AI models

At the 2026 China Compute Conference in Langfang, Hebei, from September 11 to 13, a hybrid inference system combining domestic GPU and brain-inspired chip technology was officially launched. Developed by China Mobile Cloud, the South Lake Academy of China Electronics Technology, Beijing Lingxi Technology, Shanghai Tianshuzhixing, Tsinghua University, and Peking University, the system is the first of its kind in China. According to Deepseek V4 tests, it offers more than double the cost-effectiveness of similar domestic GPU clusters and reduces operational costs by over 40%. The system is designed for high-real-time applications such as token factories, AI code generation, and multi-agent collaboration, while also supporting secure industries like finance, security, and telecommunications. The system’s core strategy, as stated by the developers, is to 'divide and conquer, enhance collaboration' by assigning attention calculations to GPUs and delay-sensitive modules to brain-inspired chips. This approach, combined with a custom model compiler and unified inference engine, overcomes traditional GPU inference limitations.