Data centers are at the heart of today’s digital infrastructure, supporting everything from cloud
computing to enterprise applications. However, their rapid growth has led to increasing energy
consumption and a substantial carbon footprint. This research proposes an AI-based framework for
real-time energy optimization in data centers, aiming to significantly reduce power usage and associated
carbon emissions. The framework integrates machine learning for workload prediction, decision-tree
algorithms for thermal-aware workload scheduling, and deep reinforcement learning for intelligent
cooling control. Implemented and validated in a simulated environment using CloudSim and realtime data emulation, the system demonstrated a total energy saving of 22.5% and a carbon emission
reduction of approximately 972 kg CO₂e over a one-week test cycle. Comparative analysis against
baseline methods confirmed significant improvements in server utilization, cooling efficiency, and IT
power consumption. The results illustrate that AI can be a transformative tool for sustainable data
center operations, offering a practical pathway toward greener and more efficient digital ecosystems.
