The International Energy Agency (IEA) recently published a major new report on AI and energy, with an emphasis on the electricity needs of data centres for training and running AI models including LLMs like ChatGPT. IEA projections span a range of scenarios from ‘Lift Off’ (unconstrained AI growth) to ‘Headwinds’ (fewer economically-viable AI use cases and grid connection constraints for new hyperscale data centres).
Yee Van Fan from the CircEUlar project and Charlie Wilson from the iDODDLE project have been working on similar projections of AI’s energy needs but using a different methodology that’s consistent with how global climate mitigation pathways are modelled. Encouragingly, our estimates to 2030 are consistent with the IEA’s when we consider both the potential impact of AI on GDP growth (macro-uncertainty) and the expansion of data centres (micro-certainty) (Figure). This will help us explicitly represent tech sector energy and emissions in global climate scenarios – the next stage of our work.
