The Blind Spot of the Twin Transition: Regulating the Environmental Footprint Of Large Language Models in the UK and EU

Abstract

The European Union and the United Kingdom are simultaneously pursuing a Twin Transition towards digital transformation and climate neutrality, predicated on the assumption that these goals are synergistic. However, the exponential rise of Large Language Models (LLMs) exposes a critical tension in this narrative: the massive energy, water, and carbon demands of compute threaten to undermine net-zero targets. This article provides a comparative doctrinal analysis of the emerging regulatory landscapes, specifically juxtaposing the EU’s AI Act and Ecodesign frameworks against the UK’s pro-innovation White Paper approach and Environment Act 2021. It argues that a regulatory blind spot persists in both jurisdictions. Traditional environmental law remains fixated on physical industrial pollution, while nascent digital regulation prioritizes safety, bias, and privacy over sustainability. Consequently, the material environmental footprint of generative AI remains largely externalized. The article concludes that current procedural transparency mandates are insufficient to mitigate this growing impact. To resolve the Twin Transition paradox, legal frameworks must evolve to substantively integrate compute intensity into Environmental Impact Assessments and public procurement standards, treating digital infrastructure as a critical resource challenge rather than a purely virtual service.