“冷却”之于数据中心为何如此重要?
高工锂电·2026-01-27 11:21

Core Viewpoint - Efficient cooling is an effective approach to reduce the Power Usage Effectiveness (PUE) of data centers [2] Group 1: Energy Consumption and Cooling - The energy consumption of data centers is closely related to cooling, as stable electricity is required to maintain efficient chip operations, and heat generation must be effectively dissipated to prevent performance degradation [3] - PUE is a key indicator of data center operational efficiency, calculated as total energy consumption divided by IT equipment energy consumption, with a target of PUE < 1.3, ideally approaching 1.2 [3] - Cooling is often a major energy consumer in data centers, with early developments seeing PUE values exceeding 2, primarily due to cooling processes [4] Group 2: Paths to Reduce PUE - There are two main paths to reduce PUE: utilizing external cooling resources and implementing efficient internal cooling systems [5] - External cooling can be achieved through natural cooling from outdoor air in cold climates, as demonstrated by Kolos's data center in Norway, which aims to reduce energy costs by 60% due to its cold climate and hydropower resources [6] - China's Tibet Ningxuan data center is noted as the highest elevation data center globally, and there are proposals to build data centers in space to enhance cooling efficiency [7] Group 3: Internal Cooling Solutions - The current mainstream internal cooling solutions are air cooling and liquid cooling [8] - Liquid cooling is gaining traction as the next-generation mainstream solution, with its penetration rate increasing from 10% to 30% due to its superior heat dissipation capabilities [9] - The main types of liquid cooling technologies include cold plate liquid cooling, immersion cooling, and spray cooling, with companies specializing in data center cooling and traditional HVAC manufacturers involved in this space [10] Group 4: Future Trends - As AI demands continue to grow, liquid cooling is expected to replace air cooling as the primary method for reducing PUE in data centers [12]