The Energy-Water-Carbon Nexus of Liquid-Cooled Data Centers: A Comprehensive Evaluation of Different Cooling Topologies

Diverse facility-side cooling topologies are being proposed to support liquid cooling systems in data centers, driven by the rapid rise in artificial intelligence workloads. However, there is a lack of systematic evaluation regarding how these different topologies perform in terms of energy, water, and carbon across diverse climates and operational requirements. To address this gap, we first summarized six commonly used cooling topologies for liquid-cooled data centers.

Then, two comprehensive evaluation metrics are proposed. One is a resource-based metric that considers Power Usage Effectiveness (PUE), Water Usage Effectiveness (WUE), Carbon Usage Effectiveness (CUE), and thermal compliance with ASHRAE standards. The other uses a cost-based metric that accounts for the expenses associated with energy, water, and carbon consumption.

To enable the evaluation, a Modelica-based virtual testbed was developed for the six studied facility-side cooling topologies. Then those topologies are evaluated across major U.S. data center hubs in Ashburn, VA, Dallas-Fort Worth, TX, Atlanta, GA, and Phoenix, AZ. The analysis revealed that no single topology universally dominates all performance metrics; instead, the optimal choice depends on the climate zone and the relative weighting of resource and cost metrics.

Under an equally weighted resource metric, dry cooling with chillers achieves the highest composite scores (0.90–0.95) across all regions due to eliminating water use. However, water-cooled topologies yield the lowest operating costs (0.100-0.135 $/kWh IT), exposing a trade-off where dry systems are resource-efficient but demand a substantial financial premium due to the current electricity pricing. Finally, transitioning to next-generation high-temperature servers universally improves energy and water efficiency, making the chiller-less wet topology the financially dominant choice across all studied U.S. data center hubs.