Green AI is a challenge that touches all of us at the Met Office. Whether we’re developing new technologies, supporting our customers, or helping deliver on our sustainability commitments, the way we use artificial intelligence (AI) has a direct impact on our mission and our values.
As we strive to be carbon neutral by 2030, it’s vital that every innovation - including AI - is considered through the lens of environmental responsibility. This article explores the paradox of Green AI: its potential to help tackle climate change, and the environmental costs that come with its use.
Why it matters to us
Our commitment to environmental sustainability is at the heart of what we do. As a lead contributor to the United Nations climate research, we know first-hand how greenhouse gas emissions are impacting the climate. This commitment is echoed at an organisational level in our pledge to be carbon neutral by 2030.
The Met Office has long been aware of the impact of our energy consumption and taken steps to align with initiatives such as ‘Greening government: ICT and digital services strategy 2020-2025’ (Howes 2020) and government-wide strategies and policies that embed digital sustainability.
‘Green IT’, also known as IT Greening or green computing, is the practice of designing, manufacturing, using, and disposing of information technology (IT) components in an environmentally friendly manner to reduce harmful impact to the environment. The use of AI now adds a new dimension that we need to consider.
The Met Office is also committed to pushing forward the frontiers of weather and climate science to ensure our science and services provide the intelligence needed to help society ‘stay safe and thrive’. This means that we may adopt new technologies such as AI.
The green AI paradox
We know that the environmental cost of AI is not inconsequential: global AI energy demand is projected to increase exponentially, reaching at least ten times its current level and exceeding the annual consumption of a small country, such as Belgium, by 2026 (Ren & Wierman, 2024).
At a more relatable level, processing a single ChatGPT query consumes ten times as much electricity as a Google search, and image-generating tasks are even more energy-intensive (Cherry, 2025).
This presents us with an uncomfortable paradox: AI has the potential to help us tackle climate change and increase our resilience to extreme weather (Bolón-Canedo et al., 2024; Rolnick et al., 2022), but its use could also exacerbate the situation through heightened energy usage, carbon emissions and water consumption (Elsworth et al., 2025).
Thankfully, we are not alone in feeling this discomfort and it’s an emerging field of research. The application of AI technology with an emphasis on energy consumption, CO₂ emissions and environmental sustainability is increasingly termed ‘Green AI’ (Alzoubi & Mishra, 2024).
The challenges of Green AI are multifaceted, and many aspects have yet to be decided. For example, what are the ‘measurement boundaries’? What should be included when calculating the environmental cost of AI use and how do we compare it with not using AI? When considering the environmental cost of AI should we include hardware, model development, training, and execution (inference)? What else, and where do you draw the line? (Alzoubi & Mishra, 2024; Elsworth et al., 2025).
No matter how challenging the situation may seem, we cannot afford to wait for perfect answers to complex questions. As the potential for AI to transform how we forecast the weather, model the climate, and deliver vital products and services becomes a reality, there is a growing recognition that these advances must be balanced with environmental responsibility.
At the Met Office, we are developing approaches to Green AI to complement, and build on our experience and ongoing work on Green IT.
Our supercomputer, or High-Performance Computer (HPC), takes in hundreds of thousands of weather observations from all over the world. It uses these observations as a starting point for running an atmospheric model. As of 2020, the electricity supply to our main operational locations was switched to zero carbon electricity. At the time, this included our supercomputing capability in Exeter and saved over 15,000 tonnes CO2 emissions a year.
Tensioned against energy usage (and carbon emissions), the capacity of the supercomputer enables us to improve our weather and climate forecasts. These forecasts help people make better-informed decisions, realising socio-economic benefits by avoiding costs, reducing waste and improving efficiency and effectiveness.
Ultimately, our supercomputer helps to shape a more sustainable world - from warning of extreme weather to helping communities plan for and reduce the impacts of climate change.
However, AI presents a new dimension and added complexity to the decisions we must now make about energy usage.
While our efforts in Green AI are still in their early stages, our commitment is clear: to harness the power of AI in ways that not only drive innovation but also respect and protect our planet.
Together, we can lead the way in creating solutions that are both cutting-edge and sustainable - ensuring that progress today does not come at the expense of tomorrow.
Find out more about the Met Office and AI.
References and further reading
Alzoubi, Y.I. and Mishra, A., 2024. Green artificial intelligence initiatives: Potentials and challenges. Journal of Cleaner Production, 468, p.143090.
Bolón-Canedo, V., Morán-Fernández, L., Cancela, B. and Alonso-Betanzos, A., 2024. A review of green artificial intelligence: Towards a more sustainable future. Neurocomputing, 599, p.128096.
Cherry, M 2025 accessed 01/12/25: https://annenberg.usc.edu/research/center-public-relations/usc-annenberg-relevance-report/ais-environmental-costs
Debus, C., Piraud, M., Streit, A., Theis, F. and Götz, M., 2023. Reporting electricity consumption is essential for sustainable AI. Nature Machine Intelligence, 5(11), pp.1176-1178.
Elsworth, C., Huang, K., Patterson, D., Schneider, I., Sedivy, R., Goodman, S., Townsend, B., Ranganathan, P., Dean, J., Vahdat, A. and Gomes, B., 2025. Measuring the environmental impact of delivering AI at Google Scale. arXiv preprint arXiv:2508.15734.
Howes, C., 2020. Greening government: ICT and digital services strategy 2020-2025. Department of Environment, Food, and Rural affairs, United Kingdom.
Ren, S. and Wierman, A., 2024. The uneven distribution of AI’s environmental impacts. Harvard Business Review.
Rolnick, D., Donti, P.L., Kaack, L.H., Kochanski, K., Lacoste, A., Sankaran, K., Ross, A.S., Milojevic-Dupont, N., Jaques, N., Waldman-Brown, A. and Luccioni, A.S., 2022. Tackling climate change with machine learning. ACM Computing Surveys (CSUR), 55(2), pp.1-96.