Carbon-Aware Cloud Computing: AI-Driven Predictive Modeling and Dynamic Optimization of Data Center Energy Consumption and Emission Reduction Strategies

Abstract

The growing reliance on cloud computing has made data centers central to global digital infrastructure, but this growth has also caused a sharp rise in energy demand and carbon emissions [1], [2]. Conventional approaches to scheduling workloads mainly focus on improving efficiency and reducing operational costs, often overlooking the environmental impact of electricity generation [3]. This paper introduces a carbon-aware framework that combines artificial intelligence–based prediction of regional carbon intensity with dynamic workload management across distributed data centers [4], [5]. The framework applies time-series forecasting to anticipate fluctuations in grid emissions and reinforcement learning to optimize workload allocation in real time [6], [7]. By jointly considering service-level agreements (SLAs), cost, and environmental factors, the approach achieves a balanced trade-off between performance and sustainability [8]. Experimental evaluation using real workload traces and carbon intensity data indicates that the proposed system can lower emissions by as much as 40 percent relative to baseline scheduling strategies, with only marginal effects on SLA compliance and cost [9], [10]. The findings suggest that predictive, AI-driven methods can play a significant role in moving large-scale cloud infrastructures toward carbon neutrality [11], [12].