Amid the excitement surrounding artificial intelligence, a crucial figure has gone unnoticed: its electricity consumption. In France, data centers could consume 7.5% of the national energy by 2035. This trajectory could jeopardize our climate ambitions.
Data Centers in France Consume an Increasing Share of National Electricity Without a Clear Future Plan
We often talk about the remarkable advancements in AI, but less about its energy appetite. Yet, today, data centers consume 10 TWh in France, equivalent to 2.5 million households. This figure could quadruple by 2035. The Shift Project sounds the alarm: if no action is taken, data centers will account for 7.5% of the national electricity consumption.
This is not an outlandish projection. RTE, the network operator, estimates that an additional 18 TWh will be required just for data centers. In simple terms, without foresight, we will have to choose between powering a hospital, a train, or a ChatGPT request. France seems to be rushing towards this wall without clearly perceiving the danger.
Ireland, Overwhelmed by Its Data Centers, Shows What Not to Do in Digital Transition
If you think this is science fiction, look at Ireland. By betting entirely on favorable taxation to attract cloud giants, it has seen data center consumption soar from 5% to 18% of its total electricity demand in seven years. This rapid growth has destabilized the entire electrical network of the country.
The result? EirGrid has imposed a moratorium until 2028: no more connections for new sites around Dublin. Too late: some are developing gas power plants just for their servers, undermining climate goals. France is making similar strides toward a comparable scenario, seemingly unbothered by the concerns of project backers, eager to ride the AI wave.
AI Usage Is Exploding, Making Its Energy Footprint Heavier Than Ever
It was believed that AI consumed most of its energy during model training. False. Today, it is the daily usage, the inferences, that are skyrocketing electricity bills. The more we use these AIs, the more we need to optimize the models, which leads to more frequent usage. It’s a vicious cycle that is as subtle as it is massive.
Large models, chatbots, image and voice recognition, content generation… each use, no matter how trivial, has a very real energy cost. But it remains hidden: we don’t see the servers running at full capacity, the air conditioners operating day and night, or the data moving across continents. Behind the magic of the interface is an energy-intensive infrastructure that pulses relentlessly.
Plan, Choose, and Regulate AI Usage to Avoid Energy and Climate Crises
So what can be done? The Shift Project suggests a decarbonization roadmap specifically for data centers, currently submerged in the tertiary or industrial sectors. We need to quantify real needs, regulate, and optimize. This means promoting frugal AIs that are less demanding and more targeted. We also need to know how to limit unnecessary usage, including certain AI gadgets that provide no real societal benefit.
Above all, it is becoming crucial to incorporate data centers into national energy planning, alongside heavy industries. This is not a regressive discourse; it is a form of clarity.
Today, it is easier to secure funding for an AI project than for an energy conservation program. Yet, without this conservation, the risk is that we will have to choose between digital transition and climate transition. Before servers take priority over citizens.
