践行深度用云，大模型混合云，十大创新技术

Investment Rating - The report does not explicitly state an investment rating for the industry Core Insights - The report emphasizes the importance of innovative technologies in the AI and cloud computing sectors, particularly focusing on enhancing efficiency and performance in AI model training and deployment Summary by Sections Diversity Computing Scheduling - The report discusses the challenges of heterogeneous computing resource management, highlighting the need for unified scheduling of diverse computing resources such as CPU, GPU, and NPU to improve efficiency [17][22][23] - It introduces Huawei Cloud's diversity computing scheduling framework, which enhances distributed AI task scheduling and resource utilization through various innovative optimizations [22][23] Cloud-Edge Collaboration - The report outlines the increasing demand for real-time inference in industrial applications and the need for efficient deployment and operation of AI models [37][39] - Huawei's hybrid cloud solution supports centralized training and edge inference, enabling continuous model iteration and adaptation to changing environments [39][41] AI-Native Storage - The report identifies storage as a key bottleneck in AI model training efficiency, particularly with large-scale training clusters [51][55] - Huawei Cloud's AI-Native storage architecture addresses these challenges by providing high-performance data access and rapid checkpoint saving and recovery [52][55][63] Enhanced AI Network - The report highlights the significant communication overhead in AI model training, which can account for up to 40% of the training time [65][66] - It discusses the development of a lossless high-bandwidth network to optimize communication efficiency and reduce bottlenecks during training [69][78] Operator Acceleration - The report emphasizes the need for efficient tools and methodologies to enhance model performance and reduce the development threshold for operators [80][85] - Huawei's CANN heterogeneous computing framework aims to maximize hardware capabilities and streamline operator development processes [85][92] Full-Link Data Engineering - The report addresses the importance of data quality in AI model performance and the challenges in data acquisition and processing [97][100] - Huawei Cloud introduces a comprehensive data engineering framework with innovative tools to improve data quality and integration for AI model training [101][105]