Numerical coupling of energy efficiency and thermal performance for cold plate cooling optimization in high-density compute AI data centers

The rapid growth of AI-driven, high-density data centers has pushed conventional air cooling to its operational limits, creating an urgent need for more efficient thermal management solutions. This study develops a coupled numerical framework that integrates CFD-based thermal analysis at the component level with system- and building-level energy performance evaluation to optimize cold plate cooling systems for a 30 MW-class data center. A total of 49 matrix cases were simulated using a k–ε turbulence model, varying coolant supply temperature (S-Class) and flow rate to assess the thermal stability of high-power chips and the associated pressure drop. These CFD results were then translated into the sizing of key cooling system components, including the Technology Cooling System (TCS), Facility Water System (FWS), and Condenser Water System (CWS), from which PUE_cooling was calculated. The findings show that higher flow rates enhance chip temperature stability but increase coolant pump power due to greater pressure drop, requiring a balance between thermal safety and energy efficiency. At the system level, all liquid cooling cases outperformed the conventional air-cooled baseline (PUE = 1.60). Optimized operating conditions achieved PUE_cooling values below 1.1, representing significant efficiency gains. This work demonstrates the novelty of numerically coupling component-level thermal performance with system-level energy analysis for large-scale AI data centers. The methodology provides practical design insights for identifying operating ranges that ensure both thermal safety of high-power chips and energy-efficient cooling, offering a scalable and sustainable solution for next-generation data center operations.