Session: K10-02: Heat transfer equipment

Paper Number: 195181

195181 - Performance Analysis of Dry Coolers for Megawatt-Scale Data Centers Across the United States 

Abstract: 

The rapid increase in GPU power densities for AI workloads is driving the emergence of MW-scale racks and GW-scale data centers. These massive power inputs are converted into heat, which is removed from the chip and rack via direct-to-chip liquid cooling and ultimately rejected to the environment. Conventional facility-level cooling technologies, such as evaporative cooling towers, are very water-intensive. This results in significant on-site water consumption that makes them increasingly unsustainable, especially in water-scarce regions where they add to the strain on local water resources. Cooling towers are also accompanied by chillers, which contribute to the overall power requirement, thereby increasing the power usage effectiveness (PUE) and cost. The recent push towards a higher temperature coolant loop on the chip- and rack-level (~45 °C) enables the use of dry cooling as an as an alternative to evaporative cooling towers and chillers, but its performance in different climate regions needs to be analyzed.

In this work, we evaluate the performance of dry coolers with and without adiabatic assist across different climate regions in the U.S. throughout the year. Dry coolers are finned-tube air-cooled heat exchangers that eliminate water use at the expense of increased power consumption under dry mode operation, which is limited by sensible heat rejection at the ambient dry-bulb temperature. Under adiabatic assist (wet mode), a small amount of water is evaporated through wetted media to pre-cool the ambient air (latent heat) by approaching the wet-bulb temperature, thereby enhancing heat rejection to the ambient. A thermofluids model is developed to evaluate the performance of a 3 MW dry cooler using the effectiveness-NTU method, as well as to quantify the resulting water and energy consumption. The results show that the dry cooler can reject the target thermal load without any water use (dry mode operation) in winter months (from January to April and October through December) across most locations. However, adiabatic assist (wet mode operation) is necessary during summer months. A case study is performed to compare dry cooler performance in Tucson, Arizona (hot-dry) and Atlanta, Georgia (hot-humid), which shows a maximum water usage effectiveness (WUE) of 1670 L/h in July for Arizona, compared to approximately 500 L/h for Atlanta. Overall, the results indicate that water use is seasonal, and that significant water savings (low WUE) can be achieved with closed-loop dry cooling relative to evaporative cooling towers while meeting approach temperature constraints. Finally, the need for mechanical refrigeration (chiller) is also evaluated in different locations as a function of varying coolant loop temperatures to provide a comprehensive picture of facility-level cooling in AI data centers.

Presenting Author: Akanksha Menon Georgia Tech

Presenting Author Biography: Dr. Akanksha Menon is an Assistant Professor in the Woodruff School of Mechanical Engineering at the Georgia Institute of Technology, where she directs the Water - Energy Research Lab. Her research interests lie at the intersection of thermal systems and functional materials, with the overarching goal of developing sustainable technologies for clean water and energy. Dr. Menon is a recipient of the NSF CAREER Award and ACS-PRF Doctoral Young Investigator Award. Dr. Menon was awarded the ASME Pi Tau Sigma Gold Medal in 2023 and the ASTFE Early Career Researcher Award in 2025. She has also been featured in Mechanical Engineering magazine’s Watch List 2025 and by the U.S. Department of Energy in their Women @ Energy initiative.

Authors:

Ricardo Cruzado Valladares Georgia Institute of Technology
Akanksha Menon Georgia Tech