智算中心“狂飙”背后部分算力闲置 政策引导市场调节双轮驱动成破局关键

Group 1 - The emergence of large models like ChatGPT has rapidly increased the demand for intelligent computing, leading to a surge in the construction of intelligent computing centers by local governments, telecom operators, and internet companies [1] - The Ministry of Industry and Information Technology has issued the "Computing Power Interconnection Action Plan" to improve the efficiency and service level of public computing resources and promote high-quality development of computing power [1][8] - The total computing power in China is projected to reach 280 EFLOPS by 2024, with intelligent computing accounting for 32% of the total [2] Group 2 - There has been an increase in the termination of computing service contracts among listed companies, attributed to policy impacts and changes in the business environment [3] - The utilization rate of enterprise-level general computing centers is reported to be only 10% to 15%, with some intelligent computing centers having GPU utilization rates below 30% [4] - The mismatch between supply and demand for computing power is a significant issue, exacerbated by the shift in market demand from training to inference due to the emergence of products like DeepSeek [4][9] Group 3 - The selection of locations for intelligent computing centers is crucial for improving utilization rates, as it must consider the geographical preferences of talent and actual demand [5] - Industry representatives emphasize the need for deep engagement in local economic demands to transform AI technology into a growth engine for regional economies [6] - The operational capabilities of intelligent computing centers are critical, requiring a focus on network construction, scheduling integration, and system operation [4][7] Group 4 - The "Computing Power Internet" is a new infrastructure aimed at enhancing capabilities for computing applications and scheduling needs, with a goal to establish a standardized interconnection system by 2026 [8] - Innovation in chip technology and service ecosystems is essential to address structural contradictions in computing power supply and demand [9] - Companies are encouraged to integrate existing computing resources to avoid redundancy and optimize overall efficiency [7][9]