Academician Zheng Weimin: The core of the AI industry has shifted from model services to token services.

On March 28th, at a seminar organized by the Tech Trend Technology and Nine Source Intelligent Computing System Ecological Community at the Zhongguancun Forum, Academician Zheng Weimin stated that the future intelligent infrastructure should be reconstructed around Token as a service. This includes: 1. System-wide heterogeneous collaboration, distributing different computing tasks to GPU, CPU, memory, and SSD to break through the computing power bottleneck; 2. Storage-computing collaboration to achieve "computing in exchange for storage", greatly reducing redundant calculations and improving inference efficiency through technologies such as pre-caching KV Cache; 3. Intelligent scheduling oriented towards Service Level Objectives (SLO), accurately translating user business requirements into underlying resource decisions.