【热点赛道速递】AIDC估值逻辑为啥变了？

Core Viewpoint - The data center industry is undergoing a profound restructuring driven by advancements in artificial intelligence technology and the ongoing national computing power strategy, shifting from a "real estate logic" centered on cabinet rental to a "computing power logic" focused on computing power delivery [1][4][5]. Group 1: Industry Reconstruction - The underlying logic of the industry is fundamentally changing, with data centers evolving from traditional Internet Data Centers (IDCs) to AI Data Centers (AIDCs), which are fundamentally different entities [1][18]. - AIDCs focus on high-density GPU computing clusters rather than cabinet quantity and rack rate, necessitating systemic upgrades in power supply, cooling technology, and network architecture [1][3]. - The demand for computing power is becoming the core engine driving industry growth, with China's intelligent computing power expected to reach approximately 10,373 EFLOPS by 2025, growing at a rate significantly higher than traditional cloud computing [1][22][26]. Group 2: Key Factors - Electricity is defined as the "first principle" of data centers, with grid access capacity becoming the decisive factor for new projects due to long expansion cycles and complex approvals [2][33]. - Liquid cooling technology has transitioned from an optional choice to a necessity, with its penetration rate in new AIDC projects expected to increase significantly, while older air-cooled data centers face accelerated asset depreciation [2][37]. - The demand for low-latency networks driven by AI inference is reshaping the value of data center locations, with high-grade data centers in major cities like Beijing, Shanghai, and Shenzhen maintaining premium pricing [2][43]. Group 3: Industry Chain Analysis - The AI wave is driving a collaborative evolution across the upstream hardware, midstream operations, and downstream applications, with high technical barriers and significant demand elasticity in the upstream segment [3][48]. - The future winners in the industry will be "computing power service providers" that efficiently convert underlying computing power into customer business value, shifting profit functions from "cabinet quantity × rental rate" to "deliverable computing power × energy efficiency × service added value" [3][74]. - Key considerations for evaluating data center projects and companies include real electricity access capability, liquid cooling deployment conditions, service focus on training or inference scenarios, and whether the business model is cabinet rental or computing power sales [3][81]. Group 4: Market Scale and Growth - The transition from IDC to AIDC is reshaping the growth curve of computing power demand, with AI computing power expected to grow at an annual rate of 50% [16][20]. - By 2028, China's intelligent computing power is projected to achieve approximately 168% growth from 2025 levels, indicating new development opportunities for data centers [22][26]. - The demand structure is shifting from primarily training to real-time inference scenarios, which is reshaping the spatial distribution and profitability models of data centers [1][26].