It forecasts droughts, tracks water leaks, and manages entire electrical networks. Yet, behind this impressive toolkit lies a paradox: while artificial intelligence can make our societies more efficient, can it truly make them sustainable without curbing their appetite?
AI for the environment through massive data analysis
In a field, a camera identifies a weed and initiates targeted weeding. Elsewhere, a model detects signs of flooding or a heatwave. These applications are no longer science fiction; climate AI analyzes vast amounts of data, far beyond human capabilities.
The IPCC already highlights its role in climate modeling and forecasting extreme events. Some systems adjust irrigation or detect invisible leaks, while others balance electricity networks. In each case, small percentages saved can preserve thousands of tons of resources.
The rebound effect negates efficiency gains and stimulates overall consumption
However, a well-known economic mechanism accompanies these advancements. When a technology reduces costs, it makes a service more accessible, leading to rapid demand increases. This is known as the rebound effect, or the Jevons Paradox. Greater efficiency does not always result in reduced global consumption.
Aviation illustrates this phenomenon well. Modern airplanes consume less fuel per passenger. Nevertheless, declining prices and increased flight frequency have skyrocketed traffic. The net effect thus depends less on the efficiency of a device than on the total number of journeys.
The same mechanism could apply to artificial intelligence. An optimized factory produces less waste but manufactures more. A more efficient delivery system becomes cheaper and more frequent. Without a focus on reduction, improving efficiency can accelerate consumption rather than slow it down.
Data centers reveal the material and energy footprint of AI
AI may seem immaterial, made up of data and calculations. However, every query passes through servers located in vast data centers. These infrastructures consume significant amounts of electricity, require cooling, and utilize water, copper, and critical metals.
According to the International Energy Agency, these centers consumed about 1.5% of global electricity in 2024. Their consumption could reach 945 terawatt-hours by 2030. This increase is largely driven by the rapid development of AI, outpacing the overall growth in electricity demand.
AI as a tool to aid transition without directing societal choices
This footprint does not condemn artificial intelligence. It can accelerate the discovery of low-carbon materials, enhance battery design, and improve building efficiency. Researchers are already identifying various applications where well-targeted models support climate transition.
However, every benefit must be weighed against the resources used, including chip manufacturing, server construction, and energy consumption. The United Nations Environment Programme emphasizes this holistic approach. The real impact of AI often remains obscured behind simple interfaces.
Moreover, no algorithm will decide to reduce flights or consume less. It can propose scenarios and anticipate their effects, but the choice of moderation remains human. It depends on political, economic, and cultural decisions. The question persists: are societies ready to change?



