Thermal Management Innovations for Energy-Efficient A.I. Data Centres

¹ Department of Mechanical Engineering, Covenant University, Ota, Ogun State, Nigeria
² Department of Mechanical Engineering, University of Alberta, Alberta, Canada.

Abstract

This Paper examines innovative thermal management strategies to improve the energy efficiency of AI data centres. Data centres currently consume about 415 TWh of electricity annually, roughly 1.5% of global demand, with projections reaching 945 TWh by 2030 due to the growing needs of high-performance AI. Cooling accounts for up to half of this energy consumption and presents a significant efficiency challenge. Traditional air cooling is insufficient for high-density systems, whereas direct liquid cooling and immersion cooling offer superior heat dissipation, supporting ultra-dense AI and high-performance computing setups. Using scenario modelling, statistical analysis, and benchmarking, the study evaluates these advanced cooling approaches. Results indicate significant cooling energy savings (15-40%), improved Power Usage Effectiveness (PUE), and a reduced environmental impact. The analysis confirms that advanced cooling, particularly immersion cooling, substantially lowers both average and peak cooling energy compared with air-cooling systems. Cross-validation with Lawrence Berkeley National Laboratory (LBNL) benchmarks and Open Compute Project (OCP) guidelines further supports the reliability of these findings, highlighting next-generation cooling as essential for sustainable, AI-driven data centres