On March 20, 2026, at the AI Solutions Launch Event tailored for small and medium-sized enterprises, Huawei Cloud proudly introduced its innovative Flexible Intelligent Computing Operating System, FlexNPU. Leveraging advanced technologies such as PD (Predictive Dispatching) dynamic co-scheduling and online-offline co-scheduling, this system significantly boosts the efficiency of computing power utilization within inference pools. Consequently, it achieves a higher Token throughput without necessitating additional hardware investments.

FlexNPU is designed to offer unparalleled sharing capabilities, flexibility, and high availability. It effectively tackles prevalent issues in the AI landscape, including the underutilization of large model inference, the inefficient exclusive use of computing power by smaller models, and the exorbitant costs incurred from recalculations following system failures. By doing so, FlexNPU facilitates a paradigm shift in AI computing power allocation, transitioning from a traditional 'resource-based model' to a more streamlined and cost-effective 'efficiency-based model'.