Sustainable Green Computing and Carbon-Aware Artificial Intelligence

Abstract

The energy footprint of large-scale computing has become a pressing concern as artificial intelligence workloads expand. Training a single frontier language model can consume thousands of megawatt-hours, and global data-centre electricity demand is projected to rise sharply over the next decade. This paper surveys the field of sustainable computing with a focus on artificial intelligence: efficiency at the algorithm, model, and system levels; carbon-aware scheduling that exploits regional and temporal variation in grid carbon intensity; hardware efficiency improvements; and reporting frameworks. We discuss the policy landscape, including the EU AI Act and U.S. Executive Order on AI, and the role of standardised carbon accounting. Open challenges in measuring and reducing both operational and embodied emissions of AI systems are highlighted, alongside the trade-offs between efficiency, capability, and equity.