AI doesn't just generate text or images. Behind the scenes, it mobilizes increasingly worrying amounts of energy. A recent study warns that Bitcoin's electricity consumption could be exceeded by the end of the year.
Digital technologies are transforming our daily lives, but they also have a hidden cost: electricity. Every click, every video, every online request relies on giant servers clustered in data centers. These facilities consume enormous amounts of energy, especially when running complex applications such as those based on artificial intelligence. Experts have recently become alarmed by a worrying trend linked to this rise in power. Until now, Bitcoin mining was considered the most energy-intensive digital activity. The process requires specialized machines that run non-stop to validate transactions on the blockchain. But according to a recent study published in the scientific journal Joule, artificial intelligence could surpass Joule's electricity consumption as early as late 2025. Researchers even estimate that it could represent up to half of the electricity used by all data centers in the world.
AI already consumes almost as much electricity as Bitcoin, and it's only just beginning
Behind every AI, ultra-powerful servers are running at full capacity. To train a model, huge volumes of data must be processed for days, even weeks. Even when used for simple tasks like writing a text or generating an image, it requires significant computing power. These activities, called “inference,” add up on a large scale. Not to mention the cooling requirements, which further increase the energy bill. Some data centers are already reaching consumption equivalent to that of entire cities.
This explosion in demand puts AI in direct competition with other energy uses, such as cryptocurrency mining. Some Bitcoin farm operators are even considering reorienting their infrastructure towards artificial intelligence services. This upheaval is forcing governments and energy providers to react. Because if the trend continues, it risks putting a strain on electricity grids and even compromising climate objectives. To avoid this, researchers are calling for the design of more economical models, while optimizing the performance of existing data centers.
0 Comments