Integrated Thermal and Energy Optimization of a Solar-Powered 500 MW AI Data Center in Central North Texas
TL;DR
The results demonstrate that coordinated optimization of cooling, heat exchangers, renewable generation, battery storage, and grid interaction can provide a practical pathway toward energy-efficient, water-conscious, grid-resilient, and lower-carbon AI data centers.
Abstract
AI data centers raise concerns about electricity consumption and costs, water shortages, and air pollution, as AI, machine learning, cloud computing, and high-performance computing rapidly increase electricity demand and thermal loads. This study develops an integrated framework for a large-scale, solar-assisted AI data center in Granbury, Texas, combining high-efficiency liquid cooling, heat-exchanger optimization, solar photovoltaic (PV) generation, and grid electricity. A MATLAB-based thermal and energy model was developed to evaluate GPU workloads, IT power consumption, cooling demand, coolant and air temperatures, pumping and fan power, Power Usage Effectiveness (PUE), and water consumption. Heat-exchanger performance was evaluated as a function of coolant velocity and channel gap, and single-objective optimization using Sequential Quadratic Programming (SQP), Genetic Algorithm (GA), and Particle Swarm Optimization (PSO) was compared with multi-objective optimization using Non-dominated Sorting Genetic Algorithm II (NSGA-II). The baseline 1.30 MW IT system achieved an average IT load of approximately 910.26 kW, daily facility energy consumption of 28,590.96 kWh, and a PUE of approximately 1.31 while maintaining acceptable coolant and air temperatures. Heat-exchanger analysis revealed a strong trade-off between thermal compactness and hydraulic efficiency. The minimum-pumping configuration occurred at a coolant velocity of 2.0 m/s and a 20 mm gap, requiring 0.0508 MW of pumping power and 1218.3 kWh/day, but also requiring 43,170 heat-exchanger modules. SQP, GA, and PSO produced similar system-level energy performance but different hydraulic configurations, whereas NSGA-II provided a broader set of Pareto-optimal solutions balancing cooling electricity, water flow, heat-exchanger area, infrastructure requirements, and solar capacity. A modular PV configuration with 69 rows, a 30° tilt angle, and a ground coverage ratio of approximately 0.655 was also developed for integration with a 500 MW-scale data-center campus. The results demonstrate that coordinated optimization of cooling, heat exchangers, renewable generation, battery storage, and grid interaction can provide a practical pathway toward energy-efficient, water-conscious, grid-resilient, and lower-carbon AI data centers.