NEXT-GENERATION ENERGY-EFFICIENT DATA CENTERS: TECHNOLOGIES, TRENDS, AND AI-DRIVEN OPTIMISATION FOR SUSTAINABLE COMPUTING

Abstract

The swift expansion of digital infrastructure has increased data centers' energy requirements, with cooling systems contributing significantly to operating costs. This study presents an AI-powered framework for optimising data centre cooling using supervised machine learning and deep learning models. A real-world telemetry dataset from Kaggle, representing chilled water setpoints, compressor frequencies, and flow rates, was preprocessed and engineered to derive proxy metrics including Cooling Load and Power Usage Effectiveness (PUE). Using RMSE, MAE, and R2, three prediction models, Random Forest, XGBoost, and Long Short-Term Memory (LSTM, were created and assessed. With RMSE = 0.075 and R 2 = 0.93, LSTM was the most accurate, demonstrating superior temporal relationships and workload management variance. Additional benefits of the model were identifying cooling inefficiencies and fewer latent compressor problems early, which helped simulate an improvement in PUE up to 0.12. The performance was compared with hyperscaler solutions by Google (DeepMind), Microsoft (Project Natick), and AWS (solar + ARM optimisation). These comparisons justified the potential industrial applicability of the solution provided. Such a framework can provide a scalable and interpretable route to the AI-enabled energy-efficient management of data centres.