Is AI a climate change villain, whose insatiable hunger for electricity will roast the planet? Or will it help predict where the next climate disaster will strike and what to do about it?
Author
- Francesco Grillo
Academic Fellow, Department of Social and Political Sciences, Bocconi University
In order to weigh these questions up, a good starting point is with AI’s energy consumption.
The 2015 Paris agreement was a deal among almost 200 countries to keep “the increase in the global average temperature to well below 2°C above pre-industrial levels” and pursue efforts “to limit the temperature increase to 1.5°C above pre-industrial levels”.
Training and deploying AI models takes place in large data centres which use a lot of power. A typical AI-focused data centre consumes as much electricity as 100,000 households. That’s why energy use is at the heart of this question about how AI will affect climate change .
Data centre use rises
But AI’s energy consumption is increasing fast. According to the International Energy Agency , the power used to train large language models (LLMs) and to run them has risen roughly tenfold since ChatGPT was launched in November 2022. And AI data centres of the future will consume far more electricity than they do today, perhaps 20 times more.
Use of AI has grown very fast, and is still growing. Sundar Pichai, the CEO of Alphabet which is Google’s parent company, recently said that Google’s monthly usage of its AI products has increased sevenfold to 3.2 quadrillion tokens (the units of data AI models read and process) since last year.
However, power consumption per AI task is going down . So efficiency is growing. The problem is that our demand for it is also growing. Economists know this as Jevons’s paradox : where technological gains can wipe out efficiencies as they create more use.
AI consumes about 155 terawatt-hours of electricity a year, based on IEA figures, and a third of the energy used by data centres overall. That is a lot. But it is still only about 0.5% of the world’s electricity , and around 0.1% of its total energy use (when we also consider oil, gas and carbon).
However, other factors are also significant. Data centres cluster, and in places where they do, such as Ireland and northern Virginia, they already strain local grids .
Can AI help with climate change?
But what about AI’s potential to help with climate change? In 2025, the European Centre for Medium-Range Weather Forecasts put an AI model into use, AIFS , which produces forecasts that improve on traditional physics-based models. It cannot tell you whether rain will fall on your street 30 days from now, but it can assess the probabilities of extreme weather events with much more geographical detail and offer more time to plan ahead.
This massively improves the possibility for insurers to assess risks as well as for government departments to put plans in place to avoid, or mitigate, disasters. It also significantly uses less energy than other physics-based weather forecasting systems.
The bigger prize lies beyond forecasting. The Intergovernmental Panel on Climate Change, the scientific pillar of the UN’s climate machinery, exists “to give governments at all levels the information they need to make policy” . Its reports take years, and not everybody seems convinced that its conclusions are fully independent .
AI could be a powerful lever for the IPCC to better serve its mission . A public AI model (similar to the AIFS) could potentially reduce biases, if used well. And it could do more than forecasting. It could suggest actions as well as describe risks; cover adaptation as well as mitigation; and reach not only government departments but city councils, farmers and households. This would turn global science into a decision-supporting system about which crop to plant or whether to insulate a roof as well as the policy decisions on how to manage energy transition .
That will not happen by itself. We need incentives to make the use of AI more energy efficient.
Here lies the most striking contrast between the two sides of the argument over AI and climate change. The world has a governance mechanism for climate change, however imperfect: the Conference of the Parties (COP), which has met almost every year since 1995. But AI has little more than voluntary codes, occasional summits and a UN scientific panel created only last year.
Could AI’s energy use be less than feared?
A new paper by Vision, the thinktank I direct, has tried to measure both sides of the ledger for AI on climate change. Its findings will feed into Vision’s fifth conference on the global governance of climate change , to be held in Venice from October 15 to 17.
The work suggests that AI’s appetite for energy could be smaller than feared, and its potential is larger. But concludes that neither will matter much unless we learn to govern it.
Whether AI warms or cools the planet could depend less on its chips than on the institutions we build around AI, and how we use it.
