利用率仅三成，“沉睡”的算力如何激活

Core Insights - The rapid development of artificial intelligence technology has led to a surge in demand for intelligent computing centers, but many of these centers are underutilized, with an average utilization rate of only 30% across the country [2][3][4] - The construction of intelligent computing centers has often been driven by blind expansion without a clear understanding of market needs, resulting in a significant amount of idle computing power [4][5] - The demand for computing power is evolving, with a shift from large-scale training needs to more fragmented and flexible requirements driven by smaller companies and AI startups [7][8] Group 1: Current State of Intelligent Computing Centers - As of November 2024, nearly 150 intelligent computing centers are operational in China, with around 400 more under construction or planned [2] - The total computing power in China is projected to reach 280 EFLOPS by the end of 2024, with intelligent computing power accounting for 90 EFLOPS, representing 32% of the total [3] - The demand for intelligent computing power in 2023 was estimated at 123.6 EFLOPS, while supply was only 57.9 EFLOPS, indicating a shortfall of over 50% [3] Group 2: Challenges in the Industry - Many intelligent computing centers, especially in smaller cities, have utilization rates below 30%, primarily due to blind construction driven by local government incentives and subsidies [4][5] - The lack of understanding among investors about the industry has led to a proliferation of centers that do not meet the complex demands of the market, resulting in oversupply of low-quality computing power [5][6] - High-performance computing centers often fail to meet the necessary power density requirements, with some centers operating below the minimum needed for effective AI training [6] Group 3: Evolving Demand and Market Opportunities - The shift from training to inference models in AI is changing the landscape of computing power demand, with smaller companies and startups increasingly seeking flexible and cost-effective solutions [7][8] - The emergence of a large number of small and medium-sized enterprises (SMEs) is expected to drive demand for computing resources, as they represent a significant portion of the market [8][9] - Initiatives like the computing power town in Hangzhou are creating platforms to match supply and demand, facilitating better resource allocation for SMEs [9] Group 4: Policy and Infrastructure Development - The Chinese government is actively working to enhance the efficiency of computing resource allocation through the establishment of unified computing service systems and platforms [11][12] - Policies such as the issuance of "computing power vouchers" are being implemented to lower barriers for SMEs and encourage broader participation in the computing ecosystem [12][13] - The development of integrated computing power trading platforms aims to streamline resource distribution and improve overall utilization rates [11][12]