A Review of Thermal Management in Modern Data Centres: Water Usage Effectiveness and Heat Transfer Coefficients

Rapid growth in artificial intelligence, machine learning, and high-performance computing has substantially increased data centre rack power densities, resulting in higher heat generation and more demanding cooling requirements. As water remains widely used in many cooling systems, understanding the relationship between cooling technologies and water consumption is essential for improving cooling efficiency and sustainability. This paper presents a survey of reported water usage effectiveness (WUE) across 83 data centre entries, providing a combined dataset that links WUE with heat-rejection categories. The reported data shows that 23 of these data centres exceed 0.4 L/kWh, which is a sustainability target specified by the Climate Neutral Data Centre Pact for new data centres in water-stressed regions using potable water. Dry facilities employing closed-loop liquid cooling require essentially no water, while evaporative systems typically report water usage effectiveness values up to 2.5 L/kWh. Reported WUE is a facility-level operational metric, set by the proportion of the IT heat load rejected by evaporation, which depends on the heat-rejection topology, ambient wet-bulb conditions, and operating set points. A higher server-side heat transfer coefficient permits a higher coolant supply temperature for a given chip temperature limit, widening the range of ambient conditions under which heat can be rejected without evaporative assistance. Server-side heat transfer is therefore an enabling condition for low WUE rather than a determinant of it. One-dimensional heat transfer models are developed to estimate heat transfer coefficients for different server-level cooling mechanisms widely used for cooling servers within data centres, including air cooling, single-phase immersion cooling, direct liquid cooling, and two-phase immersion cooling. Air cooling, with the lowest heat transfer coefficient, remains widely used in small-scale facilities, whereas direct liquid cooling and two-phase immersion cooling achieve coefficients up to three orders of magnitude higher and are increasingly deployed in high-density installations. These coefficients are used to derive an equivalent evaporative water demand, an upper-bound estimate of the water that would be evaporated in rejecting the heat each mechanism removes; it shares the units of reported WUE but describes thermal capability rather than facility water consumption.