A new report from the UN Educational, Scientific and Cultural Organisation and University College London argues that modest changes to how large language models are built and used could slash energy use by up to 90 percent without degrading performance. The analysis notes that generative Artificial Intelligence tools are used by more than a billion people daily, with each prompt consuming roughly 0.34 watt-hours of electricity. That level of activity translates to around 310 gigawatt-hours annually, which the report states is comparable to the electricity use of more than three million people in a low-income country.
The researchers highlight three primary strategies to curb the environmental footprint of generative Artificial Intelligence. First, they advocate for smaller, task-specific models in place of massive, general-purpose systems, finding that targeted models can match accuracy while dramatically lowering energy draw, with reductions of up to 90 percent in some cases. Second, they recommend shorter, more concise prompts and responses, stating that clearer, leaner exchanges could cut energy consumption by at least half without changing model architecture. Third, they point to model compression techniques such as quantisation, which the report says can deliver energy savings of up to 44 percent while maintaining output accuracy.
The report also stresses that compact, resource-efficient Artificial Intelligence can improve global access. In regions with constrained infrastructure, energy and water, smaller models can broaden participation and capability. Citing the International Telecommunication Union, the report notes that only 5 percent of Artificial Intelligence talent in Africa currently has access to the computing power needed for developing or using generative systems, underscoring the inclusivity potential of a “smarter, smaller” approach.
In related developments, environmental and community groups are urging the UK Government to block funding for new Artificial Intelligence data centers proposed to be powered by Drax, operator of the UK’s largest power station in Selby, Yorkshire. A joint letter from 63 organisations, including Friends of the Earth EWNI and Greenpeace, was sent to Secretary of State for Science, Innovation and Technology Peter Kyle, warning against public subsidies that could support Drax’s biomass operations. Drax, alongside the North Yorkshire Combined Authority and the University of York, has applied for designation as an “Artificial Intelligence Growth Zone” under a Department for Science, Innovation and Technology initiative, a move that could unlock additional public investment in biomass-powered Artificial Intelligence infrastructure.
Campaigners say the plan conflicts with recent Government measures, including legislation passed last month to limit Drax’s operations to a 27 percent load factor by 2027 and redesigned subsidies. Critics warn an Artificial Intelligence campus could enable Drax to circumvent restrictions and increase biomass burning. While Drax argues carbon capture and storage would make the project clean, the technology for woody biomass remains unproven at scale. The article notes recent layoffs at the Drax-backed firm C-Capture and references scientific warnings about the viability of bioenergy with carbon capture and storage. In 2024, Drax accounted for around 3.5 percent of the UK’s total emissions and burned 7.3 million tonnes of wood.
