eess.SYOct 23, 2026
SaveModeling Liquid-Cooled Data Centers in EnergyPlus
Organizations: Stanford University, Stanford, California, USA
Abstract
As AI workloads accelerate data center transition toward high-density racks with direct-to-chip (DTC) liquid cooling, existing simulation tools remain limited in evaluating whole-building and multi-year efficiency. To bridge this gap, we develop a native EnergyPlus object, ElectricEquipment:ITE:LiquidCooled, that models DTC liquid cooling for chilled-water and warm-water architectures as plant-loop components. It couples facility-water loops, rack-side secondary loops, coolant distribution units, and chip-junction temperatures within a single solver. Validation against an independent Modelica implementation across 21 operating conditions, including two full-year runs compared at one-minute resolution, yields loop-temperature RMSE at or below 0.021 K, well within the 0.1 K acceptance threshold. Our case study demonstrates that warm-water liquid cooling reduces annual PUE from 1.43 to 1.25, eliminates on-site water use, and cuts total facility energy by 37.9% relative to an air-cooled baseline. By embedding liquid-cooling logic into EnergyPlus, our work provides whole-building, long-term energy modeling for high-density AI infrastructure within an open-source framework.
Figures & tables
| Scale | Time scale | Representative tools | Role in the DC liquid-cooling workflow |
|---|---|---|---|
| Component | µs–s | ANSYS Fluent/Icepak ( ANSYS, Inc., 2026 ) , Flotherm ( Siemens Digital Industries Software, 2026b ) , Cadence Reality DC ( Cadence Design Systems, Inc., 2026 ) , OpenFOAM ( OpenCFD Ltd., 2026 ) | Design verification of rack layout and heat-transfer surfaces; single-timestep snapshots. Cannot resolve annual operation or building coupling. |
| Loop | s–days | Modelica (Buildings Library) ( Wetter et al., 2014 ) , Amesim ( Siemens Digital Industries Software, 2026a ) , Simscape Fluids ( The MathWorks, Inc., 2026 ) | Dynamic response of pumps, valves, CDUs, controllers; start-up transients and control tuning. Facility-scale annual horizons are impractical. |
| Building/facility | 1 min–years | EnergyPlus ( Crawley et al., 2001 ) , TRNSYS ( Solar Energy Laboratory, UW-Madison et al., 2026 ) , Modelica (Buildings Library) ( Wetter et al., 2014 ) , Spawn-of-EnergyPlus ( Wetter et al., 2024 ) | Integrated thermal, electrical, water, and carbon flows in one engine. The only scale at which facility-boundary efficiency metrics are physically defined and where compliance and regulatory frameworks (e.g., ASHRAE 90.4 ( ASHRAE, 2022 ) , LEED ( U.S. Green Building Council, 2026 ) , California Title 24 ( California Energy Commission, 2025 ) ) operate. |
| Grid | 5 min–decades | OpenDSS ( Electric Power Research Institute, 2026 ) , GridLAB-D ( Pacific Northwest National Laboratory, 2026 ) , PLEXOS ( Energy Exemplar, 2026 ) , HOMER Pro ( UL Solutions, 2026 ) | External context: tariffs, emissions factors, demand response, portfolio dispatch. Facility thermal behavior abstracted as a constant or scheduled load. |
Table 1. Data-center thermal simulation tools by modeling scale; this work targets the building/facility scale.
| Liquid Cooling Class | Facility Water Supply Temperature | |
|---|---|---|
| °C | °F | |
| W17 | 17 | 62.6 |
| W27 | 27 | 80.6 |
| W32 | 32 | 89.6 |
| W40 | 40 | 104 |
| W45 | 45 | 113 |
Table 2. Facility water supply temperatures by liquid cooling class, from the 2021 ASHRAE thermal guidelines for liquid cooling ( ASHRAE, 2024 ) . The minimum water temperature for all classes is 2°C (35.6°F).
| Field | Air | Liq. |
| Fan / airflow | ||
| Fan power fraction | ✓ | |
| Fan airflow curves | ✓ | |
| Fan power curve | ✓ | |
| Design entering air temperature | ✓ | |
| Supply air node | ✓ | |
Table 3. Fields for the air- and liquid-cooled objects in EnergyPlus 26.1. A checkmark marks applicability.
Figure 1. Screenshot of the developed ElectricEquipment:ITE:LiquidCooled object in EnergyPlus in the IDF editor.
Figure 2. Schematic of the ElectricEquipment:ITE:LiquidCooled object and its coupling to the FWS plant loop and the server zone.
Figure 3. Annual IT-load schedule: a daily step schedule with four discrete utilization levels (0.25, 0.50, 0.75, 1.00) arranged as repeating quarterly ramps. The CPU power scales linearly with the load fraction, reaching the 2,143 kW design value at full load.
| Parameter | Chilled Water (W17) | Warm Water (W45) |
|---|---|---|
| Common to the validation and case-study models | ||
| Data-hall floor area | 557.42 m 2 | |
| IT power density | 3 844.69 W m -2 | |
| Peak CPU electrical power | 2 143.10 kW | |
| Number of rack units | 71 | |
| UPS efficiency | 0.94 | |
Table 4. Model parameters for the two plant configurations. The upper block is common to the validation and case-study models; the lower block lists what changes between them. Spanning values are shared by both plants; split entries differ.
| Variable | Chilled water (W17) | Warm water (W45) | ||||
|---|---|---|---|---|---|---|
| RMSE | MAE | Max | RMSE | MAE | Max | |
| Facility-water supply temperature | 0.000 | 0.000 | 0.000 | 0.008 | 0.003 | 1.362 |
| Facility-water return temperature | 0.004 | 0.003 | 0.005 | 0.008 | 0.004 | 1.363 |
| Rack-coolant supply temperature | 0.000 | 0.000 | 0.000 | 0.008 | 0.003 | 1.362 |
| Rack-coolant return temperature | 0.021 | 0.019 | 0.030 | 0.008 | 0.003 | 1.362 |
| Facility-water mass flow rate (kg s -1 ) | 0.002 | 0.002 | 0.002 | 0.001 | 0.001 | 0.001 |
Table 5. Full-year agreement between the EnergyPlus liquid-cooled object and the Modelica baseline for the warm-water and chilled-water plants (1-min intervals, first hour excluded, 525,541 samples per variable). Variables that agree to numerical precision in both plants are omitted.
Figure 4. EnergyPlus and Modelica loop temperatures for the chilled-water (top) and warm-water (bottom) plants over days 209–214, with IT utilization labeled. The two traces are indistinguishable. Lower panels: EnergyPlus outdoor air temperature and cooling power, shown for context and not part of the comparison.
Figure 5. Detailed view of the temperature residual spike during an isolated load step, illustrating the transient discrepancy between EnergyPlus and Modelica due to control-modeling differences.
| Variable | FWS temperature | Rack-coolant flow | CDU approach | FWS flow | ||||
|---|---|---|---|---|---|---|---|---|
| 12 to 45 °C | 10 to 41 K | 4 to 10 K | 0.5–2 design | |||||
| 9 conditions | 4 conditions | 4 conditions | 2 conditions | |||||
| RMSE | Max | RMSE | Max | RMSE | Max | RMSE | Max | |
| Facility-water supply temperature | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 |
| Facility-water return temperature | 0.004 | 0.006 | 0.003 | 0.005 | 0.003 | 0.005 | 0.007 | 0.011 |
| Rack-coolant supply temperature | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 |
Table 6. Agreement across the 19 swept conditions, reported for the same variables and in the same order as Table 5 . Each entry is the worst value over the conditions of that group (settled samples, 480 per condition; K unless noted). Variables that agree to numerical precision in every condition are omitted, as in Table 5 .
Figure 6. Schematic comparison of the three cooling architectures modeled in EnergyPlus for the case study. The air-cooled baseline relies on standard object types, whereas both the chilled-water and warm-water systems utilize our newly developed liquid-cooling object.
Figure 7. Annual PUE, WUE, EUE, facility energy, and cooling overhead of the air-cooled baseline and the chilled-water (12 °C) and warm-water (45 °C) liquid-cooled configurations in Tampa, FL.