Every query sent to an AI consumes energy, but exactly how much? Researchers have revealed some surprising figures. And the environmental issue is becoming increasingly difficult to ignore.
Artificial intelligence is becoming everywhere: in search engines, voice assistants, work tools, and even on smartphones. But behind the promise of time savings and productivity, another worrying reality lies: the energy impact. Because to function, AI needs powerful servers, themselves housed in giant data centers that consume an impressive amount of electricity.
AI does not work like a traditional application. It processes enormous volumes of data and relies on very demanding calculations. For each request sent to a tool like ChatGPT, hundreds of machines are called upon to produce a response. This requires not only computing power, but also a significant amount of energy to keep these systems at the right temperature. The more complex the model, the more expensive it is to run. And while it's difficult to precisely quantify each interaction, some studies suggest the energy equivalent of a lit light bulb or a half-liter of water per query, taking cooling into account.
A simple query to AI can consume more than you might think
According to researchers at MIT and Moody's, the most advanced models can consume several kilowatt-hours just for initial training. These data centers, often called "clouds," are actually increasingly gigantic physical infrastructures, built to cope with the growing demand for AI. As usage becomes more widespread, these installations consume more than other digital services like streaming or social media, and could well surpass Bitcoin.
To limit their impact, some companies are focusing on software optimization or the use of renewable energy. Others are developing smaller, more specialized models capable of meeting specific needs with fewer resources. For their part, users can also take a few simple steps: limit repetitive requests, favor less energy-intensive tools, or avoid using AI for simple tasks. On a large scale, these individual choices can significantly reduce the ecological footprint of this technology.
0 Comments