AI energy consumption: what official sources say, and what a business can do

In its Energy and AI report, published in 2025, the International Energy Agency (IEA) estimates that data centres consumed around 415 terawatt-hours (TWh) of electricity in 2024, or about 1.5% of global consumption. In its base case scenario, this consumption would more than double to reach around 945 TWh in 2030.

The IEA identifies artificial intelligence as the most important factor in this growth, alongside demand for other digital services. These figures cover all data centres, not AI alone.

The update of the study by ADEME (France’s ecological transition agency) and Arcep, published in January 2025 and covering the year 2022, puts the carbon footprint of digital technology at 4.4% of France’s national carbon footprint, or 29.5 million tonnes of CO2 equivalent. Digital services account for 11% of French electricity consumption in the study, or 51.5 TWh, and 65 TWh when data centres located abroad that host French usage are included.

In this assessment, devices account for 50% of the carbon footprint of digital technology, data centres for 46% and networks for 4%, and the manufacture of equipment accounts for 60% of the footprint. ADEME itself points out an important limitation: the study does not take into account the development of generative AI.

Training a model draws on a great deal of computing power over a given period. According to the report Arcep published in May 2026 on the environmental challenges of generative AI, its impact depends mainly on the number of parameters in the model, the electricity mix and the hardware used, and both the number and the size of models have kept growing.

Use, known as inference, refers to each query sent to the model once it has been trained. Arcep notes that its impacts are less well documented but could become a significant issue as usage grows. Tests carried out with PEReN (the Pôle d’expertise de la régulation numérique, the French government’s digital regulation expertise centre) show that the largest models consume the most, but that models of very different sizes can consume as much as each other, and that some less energy-hungry models provide equally relevant answers.

Arcep highlights the lack of transparency among industry players. It cites a study by Luccioni et al. (2025): of 754 AI models analysed, deployed between 2010 and the first quarter of 2025, 84% had no environmental information disclosed as of May 2025. Published estimates therefore rely on assumptions, partial measurements or open models, and can differ from one source to another.

Where the electricity is produced also matters. Arcep states that on average in 2024, the Polish electricity mix had a carbon content 17.6 times higher than the French mix. The same query therefore does not have the same climate impact depending on the country where the data centre is located. Be wary of figures quoted without a date, a scope or a method.

The first lever is to use AI only where it provides a real service. A tool that handles calls that would otherwise be lost or would tie up a member of staff has a clear justification; generating content that nobody will read does not. The second lever is the choice of model: PEReN’s work cited by Arcep shows that some less energy-hungry models provide answers as relevant as those of large models.

Question your suppliers too: where the processing is hosted, which models are used, what environmental information is available. Finally, do not forget devices, which account for half of the carbon footprint of digital technology according to ADEME; for digital technology as a whole, the manufacturing phase accounts for 60% of that footprint. Extending the life of computers and phones acts on this share, with or without AI.

Alyas uses AI for one specific task: answering your business’s phone using the information you provide, transferring calls when necessary and summarising every call in your client area. Its data is hosted in the European Union. The ten-day free trial lets you check whether this service meets a real need in your business.