GREEN AI IN THE CLOUD: STRATEGIES FOR CARBON-AWARE RESOURCE MANAGEMENT

As AI workloads proliferate in cloud environments, their growing energy consumption contributes significantlyto carbon emissions. This paper presents a comprehensive framework for carbon-aware resource managementin cloud platforms, integrating real-time emissions data, workload scheduling, and energy-efficient autoscalingpolicies. We propose a hybrid architecture combining carbon telemetry ingestion, predictive modeling ofworkload carbon intensity, and closed-loop controllers that optimize VM placement and serverless functionprovisioning for minimal carbon footprint. A prototype on AWS and Azure using real-world energy grid datademonstrates a 30% reduction in carbon emissions and 20% cost savings under mixed batch and streamingworkloads. We detail design patterns, algorithmic strategies, and evaluation metrics, offering a blueprint foreco-friendly AI deployments