Does Artificial Intelligence Contribute To Carbon Emission? - A Systematic Review

Abstract

Artificial Intelligence (AI) is quickly revolutionizing industries. Industries such as healthcare, finance, manufacturing, and logistics have heavily adopted AI in recent years. The expansion, though, comes with substantial environmental consequences. This systematic review of peer-reviewed research from 2015 to 2025 discusses the contribution of AI to carbon emissions. It discusses the direct effects, like energy-consuming model training, inference, and data centre use-and indirect ones, such as e-waste, water consumption, and resource extraction. The results show that large AI models release tens of thousands of tons of CO₂ equivalent (tCO₂eq), with inference dominating training emissions more and more. Corporate-scale AI applications and generative models are particularly energy-hungry, using as much as 4600 times more energy than normal models. While some studies point to the ability of AI to optimize energy systems and minimize emissions, the trend so far shows a net-positive carbon footprint. Additionally, the absence of standardized reporting of emissions and regulations complicates mitigation even further. This Sustainable AI development needs lifecycle carbon accounting, legal frameworks, and carbon-aware innovation. Without these interventions, the environmental liabilities of AI may undercut its promise as a means to meet global climate objectives.