Abstract
Modern data centers are facing unprecedented pressure. Artificial Intelligence (AI), high-performance computing, edge systems, and compact servers are increasing energy consumption and generating more heat. Conventional cooling methods cannot meet these demands. As rack densities have increased from 5 to 10 kW per rack to 30, 50, or even 100 kW, the way thermal management is handled has changed completely. Liquid cooling has become one of the most practical and scalable ways to meet these problems. What was once considered a niche or experimental solution is now widely adopted by hyperscale’s, colocation providers, and enterprise data centers worldwide. This evolution is closely linked to the rapid expansion of mobile networks and Internet of Things (IoT) ecosystems, where edge data centers play a critical role in supporting latency-sensitive and data-intensive applications. As a result, efficient thermal management has become essential to ensure reliability and scalability in these distributed computing environments. The main objective of this study is to apply Neural Networks (NN) based on a cluster of metrics derived from the analysis of immersion cooling and direct chip cooling. These are used to analyze these metrics using a technique called a Multilayer Perceptron (MLP) and Radial Basis Function (RBF), which can yield one or more dependent variables to address the problems of energy usage and thermal output. A synthetic thermo-hydraulic dataset was generated to simulate operational cooling scenarios associated with liquid-to-air and liquid-to-liquid architectures in AI-oriented data centers. The simulation-driven dataset enabled the evaluation of multiple thermal and flow-related metrics under varying workload conditions using MLP and RBF neural-network models. The simulation results reveal that thermal and cooling-related variables—particularly ΔT, airflow, and mass flow—consistently have the strongest influence on system performance, regardless of model configuration. This finding highlights that efficient heat-dissipation mechanisms primarily drive system stability and energy efficiency. It supports the scalability and effectiveness of liquid-cooling solutions under high thermal loads.