Green AI in 2026: Power, water and the future of sustainable technology
The rapid rise of artificial intelligence (AI) presents a clear paradox. While AI can significantly strengthen climate resilience and improve environmental decision making, the process of developing and running these systems consumes considerable energy, generates carbon emissions and places pressure on water resources. These impacts must be managed if AI is to align with broader sustainability goals.
What is “Green AI”?
The Met Office has placed “Green AI” at the centre of its strategy. As it adopts new AI capabilities, it has committed to achieving carbon neutrality by 2030 and ensuring its digital infrastructure aligns with government sustainability standards.
Green AI refers to the effort to measure and reduce the environmental impact of AI across its entire lifecycle - from hardware and model development to training and real world use. Although the exact methodology for this accounting is still evolving, the principle is clear: organisations must understand and minimise the energy, emissions and environmental costs associated with AI.
The Met Office has already made progress by procuring zero carbon electricity across its main operational sites, resulting in significant yearly emissions reductions.
The scale of AI’s environmental footprint is now unmistakable. Global electricity demand from AI could grow more than tenfold and exceed the annual consumption of a country the size of Belgium as early as 2026. With figures of this magnitude, environmental accounting for AI is no longer optional. Organisations deploying AI need to measure and mitigate their energy, carbon and water use as standards develop. For the Met Office, the aim is to use AI to enhance weather and climate intelligence in ways that help society reduce waste and avoid cost - without creating new environmental burdens.
A growing focus on water
Water use is a critical and often overlooked part of AI’s footprint.
Data centres powering AI workloads consume increasing amounts of freshwater for cooling, driven by rising energy use and the density of modern chips. Some large facilities now use water volumes comparable to those consumed by towns.
The water footprint extends well beyond the data centre. Electricity generation, particularly from fossil fuels, requires substantial cooling water, and chip manufacturing relies heavily on ultrapure water for fabrication processes. This means AI carries a water cost long before a model is deployed.
There are, however, viable pathways to reduce water intensity. Direct to chip and immersion liquid cooling technologies can significantly cut water use compared with traditional evaporative cooling, although they require higher upfront investment and specialist fluids. On the energy side, shifting data centre power from fossil fuels to renewable sources dramatically reduces indirect water consumption, as coal and gas generation require large volumes of cooling water, unlike solar and wind.
Why this matters
AI’s value in forecasting and climate science is both substantial and immediate, but its deployment must sit within firm environmental boundaries to ensure it remains a net positive. The responsible approach is to pair commitments to low carbon power with rigorous, lifecycle aware AI practices and water efficient data centre strategies. This ensures that AI delivers benefits without shifting hidden environmental costs elsewhere.
Looking ahead, “Green AI” will be defined by transparent measurement, careful choices about model scale and accelerated adoption of water efficient cooling and renewable energy. Only then can the environmental promise of AI be realised alongside its technological potential.
