What price must be paid to develop a national AI policy? As France and Europe elevate it to a strategic priority, the Competition Authority publishes a study on its energy and environmental cost, following its first opinion on generative AI published in the summer of 2024.
And the competition watchdog is sounding the alarm: the electricity consumption of data centers could account for 4% of national French consumption by 2035, while the sector’s giants are already securing their energy supplies, potentially skewing the competitive playing field.
Electric power consumption could double by 2030
The numbers speak for themselves. Data centers currently represent roughly 1.5% of global electricity consumption, but their local impact proves far more significant. The Competition Authority expects their consumption to at least double by 2030, reaching 945 TWh worldwide.
In France, the situation is equally worrying. Data center consumption, estimated at 10 TWh at the start of the 2020s, could rise to between 12 and 20 TWh in 2030, then between 19 and 28 TWh in 2035, representing nearly 4% of national electricity consumption. In response to this surge, major American players are already forging partnerships to secure decarbonized energy supplies, whether from renewables or nuclear power.
Natural resources under pressure
Beyond electricity, the environmental footprint of AI mobilizes substantial resources along the value chain: water, rare metals, and land. According to Arcep, the volume of water withdrawn by data centers reached 0.6 million cubic meters in 2023, with a sustained annual growth of 17% in 2022 and 19% in 2023. In addition to this direct consumption, there are indirect withdrawals linked to electricity production, estimated at more than 5.2 million cubic meters per year.
Several players in the digital sector have announced substantial increases in their greenhouse gas emissions, ranging from 30% to 50%, resulting notably from the rising energy use of their data centers.
Three major competitive issues
The Authority identifies three types of competitive issues arising from this situation.
First issue: access to energy and control of its cost. Sector players face grid connection difficulties and uncertainties about energy prices. Electricity accounts for between 30% and 50% of a data center’s operating costs, and the energy landscape is undergoing a transformation, notably with the termination of the ARENH mechanism on 31 December 2025.
The new dual system is being rolled out with the universal nuclear levy and long-term allocation contracts backed by nuclear production proposed by EDF. The Authority remains vigilant: it will ensure that the privileged position of the largest players does not enable them to secure energy supplies on advantageous terms to the detriment of smaller players. It also recalls that EDF must not engage in anti-competitive behavior such as discrimination, refusal to supply, or market lock-in.
Second issue: frugality as a competitive parameter. In light of AI’s environmental impact, the concept of frugality is developing, defined as the optimized consumption of resources to minimize environmental footprint. Demanders show growing interest in leaner tools, while several players develop smaller models or publicly share their environmental footprint.
The Authority believes that frugality can stimulate competitive dynamics by affecting price, quality, and innovation. It can particularly enable smaller firms to compete with large incumbents. However, it warns against several risks: adopting misleading frugality claims, failing to report environmental footprints where there is demand, or restricting innovation in frugality.
Third issue: standardization of environmental footprints. The Authority notes that companies disclose little about the environmental impact of their solutions, that there is no shared methodology, and that measurements taken are difficult to compare.
Several tools have been developed, such as the general framework for eco-design of digital services co-published by Arcep and Arcom, or the general framework for frugal AI led by AFNOR. Other tools focus on measuring energy and carbon footprints, while some players propose the creation of an environmental rating to compare different models.
But again, the Authority warns of several problems: the adoption of tools that are not grounded in a robust methodology, standardization that could deprive certain actors of its benefits, behaviors hindering standardization, or exchanges between competitors involving commercially sensitive information.
It insists particularly on the need for reliable, transparent data on the energy and environmental footprint of AI. Such transparency would allow frugality to play its full role as a competitive parameter. It also emphasizes ensuring that access to zones suitable for deploying data centers and energy, particularly nuclear-origin electricity, does not confer an advantage to certain players alone.
