【财经分析】算力需求趋向“提质” 有效供给有待破“三关”

Group 1 - The core viewpoint of the article emphasizes that the competition in the AI industry is not about speed but about stability and long-term sustainability, with a significant shift from conceptual hype to value reassessment in the capital market by the end of 2025 [1] - The demand for computing power has been rapidly increasing due to the comprehensive explosion of AI technology, yet the computing power leasing industry is experiencing a situation of "increased revenue but decreased profit" [1][5] - In the third quarter of 2025, 59 out of 99 listed companies in the computing power leasing sector reported a year-on-year decline in gross profit margin, indicating a challenging market environment [1] Group 2 - The rapid development of AI applications is stimulating the construction of computing power centers, with significant government policies supporting enterprise applications and the cultivation of vertical models [4][5] - As of November 1, 2025, a total of 611 generative AI services have been registered in China, with an average of 31 new registrations per month, reflecting a 55% increase compared to the previous year [4] - The overall scale of computing power in China has been growing at an annual rate of about 30% since the 14th Five-Year Plan, with expectations of over 40% growth in 2025 driven by AI advancements [5] Group 3 - The operational challenges faced by computing power centers include high operational costs, market demand segmentation, and the need for strategic regional adjustments [2][6] - Many computing power centers in western regions are underutilized due to distance and latency issues, leading to a re-evaluation of their value as demand shifts towards eastern regions [7] - The high energy consumption of computing power centers, comparable to that of small steel plants, poses significant compliance costs and operational challenges, especially with increasing regulatory scrutiny on green development [6][7] Group 4 - The market is shifting from a focus on training chips to inference chips, with higher demands for computing power efficiency and cost-effectiveness, making targeted supply a competitive focus [8] - Large enterprises are the main players in computing power construction due to the substantial investment and operational costs, making it difficult for small and medium-sized enterprises to access necessary resources [8] - The development of domestic chips is crucial for the computing power sector, with a need for collaboration across the industry to enhance the ecosystem and support the growth of domestic AI chips [9]