Green Computing: AI for Energy-Efficient Data Centers

Abstract

Modern hyperscale data centers represent the backbone of global digital infrastructure, supporting intensive workloads spanning cloud services, deep learning model training, and real-time big data analytics. However, their exponential proliferation has driven electricity consumption to unprecedented levels, accounting for approximately 1.5% to 2% of worldwide electrical energy demand and emitting significant carbon equivalents. Traditional heuristic-driven resource management frameworks and static threshold cooling protocols lack the dynamic adaptability required to handle highly variable, non-stationary workloads. This review presents a comprehensive, structured evaluation ofArtificial Intelligence (AI) and Machine Learning (ML) techniques engineered for energy-efficient data center operations. We dissect the multi-layered spectrum ofAI interventions—from Long Short-Term Memory (LSTM) workload demand forecasters and Deep Reinforcement Learning (DRL) resource orchestrators to AI-driven thermodynamic cooling optimization and carbon-aware spatial/temporal workload shifting. Furthermore, this study establishes a rigorous PRISMA-guided review methodology, presents an eight-layer conceptual system architecture, evaluates microclimate telemetry, and synthesizes critical research gaps to outline actionable pathways for sustainable, net-zero cloud computing architectures.