人工智能优化算力布局

Core Insights - China Mobile has established the first nationwide computing power network integrating "general, intelligent, super, and quantum computing," with over 1 million servers and an intelligent computing capacity exceeding 43 EFLOPS, accounting for one-sixth of the national computing resources [2] - The demand for computing power is experiencing explosive growth, with projections indicating that by the end of 2024, China's total computing power will reach 280 EFLOPS, with intelligent computing power exceeding 90 EFLOPS, representing over 30% of the total [3] - The rapid development of artificial intelligence is driving a significant transformation in computing power demand and infrastructure, leading to new opportunities and requirements for computing networks [2][3] Group 1: Computing Power Expansion - The intelligent computing power in China is expected to grow by over 2.5 times in the next three years, with an annual compound growth rate of nearly 40% [5] - By 2025, China's intelligent computing power is projected to reach 1,037.3 EFLOPS, a 43% increase from 2024, and by 2026, it will double to 1,460.3 EFLOPS [3][5] - The structure of computing power is shifting, with inference computing power demand expected to surpass training computing power demand, indicating a new growth cycle for the industry [5][6] Group 2: Infrastructure Development - China Mobile has built 13 national and regional intelligent computing center nodes and is developing large-scale intelligent computing centers, including a national hub for integrated computing power scheduling [5][6] - The newly established "four-in-one" computing power scheduling platform can support over 100 million computing power calls daily, enhancing resource allocation efficiency [5][6] - The intelligent computing network aims to provide lower-cost computing services and create a new ecosystem for precise supply and demand matching [5][6] Group 3: Structural Optimization - The rapid advancement of artificial intelligence is prompting a reconfiguration of computing power layouts, focusing on large-scale clusters and optimizing resource allocation [6][7] - There is a growing consensus on the need for green and intelligent computing power solutions to maximize social computing applications and address current supply-side imbalances [6][7] - The shift from traditional resource-based services to task-based services and model-as-a-service (MaSS) is transforming the computing power service model [7] Group 4: Cost Reduction Strategies - Companies are advised to strategically plan their computing power needs to balance performance and cost, especially when deploying large AI models [8] - The selection of foundational models for AI applications is crucial, with recommendations for proprietary and private deployment to enhance data security and responsiveness [8] - The integration of various computing power types is essential for optimizing resource structures and improving utilization levels [8][9] Group 5: Policy and Future Directions - The National Data Bureau is committed to deepening the market-oriented allocation of data elements and reducing computing power costs for small and medium enterprises [9] - China Mobile plans to build large-scale "computing power factories" and expand intelligent computing centers significantly [9] - There is a focus on enhancing the quality of computing power supply and accelerating breakthroughs in core software and hardware technologies [9]