Comparative analysis of annual energy consumption of air cooling and liquid cooling in data centers

With the widespread adoption of artificial intelligence and cloud computing, the energy consumption of data centers has risen dramatically, rendering liquid cooling technology a promising solution. In this study, based on four typical climatic conditions across China, three-dimensional server models and lumped cooling system models are established for both air cooling and full liquid cooling architectures. To ensure a fair comparison, all factors that may influence the comparative outcomes are taken into account, including identical server power consumption, identical ambient conditions, and identical cooling equipment. The results indicate that, under the climatic conditions of Xi'an, full liquid cooling can achieve 95.30% free cooling availability throughout the annual operating period. Through thermal resistance chain decomposition, the underlying mechanism is revealed. Compared with air cooling, the conductive and convective thermal resistances of the cold plate in the full liquid cooling system are reduced by over 60%, thereby permitting a higher coolant inlet temperature and substantially extending the duration of free cooling operation. At the component level, under mechanical refrigeration mode, the reduction in compressor power consumption accounts for 80.27% of the total PUE decrease achieved by full liquid cooling. Under natural cooling conditions, the elimination of server and air-conditioning fans contributes to 93.47% of the energy savings. Across four climate zones, the full liquid cooling system consistently exhibits lower PUE and higher SEER , with the advantages being particularly pronounced in hot climates. Moreover, when CPU power consumption increases by 22.67%, the pPUE of the full liquid cooling system rises by only 1%, demonstrating excellent load-following stability. This study integrates component-level thermo-fluid analysis with data-center-level annual energy assessment, providing a quantitative, multi-scale basis for the selection of cooling architectures.