<p>Rapid advances in machine learning (ML) and artificial intelligence (AI) technologies have resulted in a considerable demand for energy and negative effects on the environment. The amount of energy consumed by AI models at each stage of their life cycle during training, inference, and infrastructure support the estimation of the associated carbon emissions and reliance on fossil fuels. For such assessment purposes, a structured approach is used to measure the amount of energy expenditure associated with the use of AI technologies on the basis of the eco2AI carbon tracker and empirical evidence of the sources of energy consumption in data centers. According to results, one large language model (LLM) alone requires more energy to be trained than several automobiles produce in their lifetime. Through a comparative analysis of the design of models from transformer-based LLMs to traditional ML techniques, a multidimensional framework for green AI development is proposed, and the differences in the environmental cost for models are measured. This study provides hardware efficiency, model optimization; carbon footprint measurement, carbon footprint aware workload scheduling, renewable energy friendly infrastructure, and sustainable AI governance are all included within the whole Sustainable AI Framework. Benchmark datasets and analysis case studies are used to illustrate the framework’s applicability. The introduced framework does not offer any new optimization algorithm but a structured methodology for the development of ecologically sustainable AI applications.</p>