Year: 2025
Type: article
Abstract: The rapid expansion of AI training workloads has significantly increased the energy demands and carbon footprint of cloud data centers.This paper proposes EcoCloud, an AI-driven sustainable cloud architecture designed to mitigate environmental impact by integrating three key innovations: green energy usage, carbon-aware computing, and serverless design.EcoCloud dynamically schedules AI training jobs based on real-time carbon intensity forecasts and regional renewable energy availability, optimizing workload placement across globally distributed data centers.By employing a serverless, microservices-based execution model, EcoCloud minimizes idle resource consumption and improves energy proportionality.The architecture also includes a carbon-intelligent orchestrator, a green energy manager, and checkpoint-based job migration to enable flexible, low-carbon AI training.Through technical analysis, case studies, and performance metrics, this paper demonstrates how EcoCloud can significantly reduce CO₂ emissions up to 80% in some scenarios without sacrificing model accuracy or SLA adherence.Challenges and future directions are discussed, emphasizing the importance of policy alignment, accurate carbon forecasting, and energy-aware AI design.
Author: Jyoti Kunal shah
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4410119915