CCWise: Carbon–Cost Aware Regional LLM Orchestration for Next-Gen Sustainable AI

This paper presents a comprehensive orchestration for evaluating the sustainability of Large Language Models (LLMs) lifecycle by integrating carbon emissions, energy consumption, and cost-efficiency metrics across diverse geographic regions. We introduce two novel indices—Carbon-Cost Tradeoff Index (CCTI) and Green Cost Efficiency (GCE)—to quantify the environmental and economic trade-offs inherent in token generation of LLM deployment. Through extensive experimental analysis, including Pareto assessment of cost versus carbon footprint, we reveal the substantial impact of regional grid carbon intensity and model architecture on operational sustainability. Our findings highlight that smaller, region-optimized models consistently achieve superior carbon-cost performance, whereas deployments in carbon-intensive grids exhibit pronounced inefficiencies.