Artificial intelligence: a more important carbon footprint than “all laptops”

“”Before each AI project carried by a company, before each prompt, you have to ask yourself the question: do you really need it?“During the presentation of its study on the environmental impact of digital worldthe Greenit collective has set the tone.

In a very long relationship which gives itself the heavy task of concretely encrypting the ecological footprint of this industry worldwide, a good place is left to artificial intelligence.

Ultra-growth figures

After learning that digital is responsible, globally, for the emission of 1.8 billion tonnes eq. CO2 – or 5.5 times the imprint of France, twice that of Canada or 100 times that of a poorer country, like Tanzania – the study claims that servers dedicated to artificial intelligence represent 4 % of this very big carbon cake.

This is roughly at the carbon footprint of a country like Chile. To stay in the IT sector, this is “Already more than all laptops“In terms of gross greenhouse gas emissions. These figures being more from the year 2023, the year one for Chatgpt, there is a safe bet that they are already largely underestimated.

“”In view of the upcoming growth projections of the uses of IA technologies, the proportion of the impacts of the IA servers is led to increase even faster in the coming years, with the corresponding environmental consequences“, Underlines the study. There is not only a question of greenhouse gases and greenhouse gases, but also of extraction of minerals and metals, to use abiotic resources (land rare) or even fine particle emissions.

AI does not replace uses, it adds up

In all, “The servers configured for AI consume more than 18 % of the electricity in computer centers, while they represent only 2 % of the amounts of servers“, Note Greenit. The fault of specialized fleas and GPU very, very energy delicious,”especially in their model training phase“. This phase is that which corresponds to the moment when the machines run at full speed to ingest all the data necessary for their operation.

And unfortunately, if you can believe that these uses replace others, modulating their global impacts a little, it would seem that no. “”In addition, the rise of generative AI is an additional use which does not replace the previous uses, but adds up. The same goes for servers type equipment, which are added to existing stocks“. To put it differently, there is no question of recycling the old server farms, AI as a request for new ones.

And beyond the direct emissions due to the construction of machines and their exploitation, AI also has visible edge effects. “”These figures do not take into account the equipment, which will sell in addition to the IA marketing principle“, points to a co -author of the study.

The essential debate on use

As the Ademe advised in his own reportGreenit militates for a revolution of our practices. “”We will have to refere our uses more frankly. Using AI to advance science or thinking about an existential problem is not the same as using it to choose your next shoe color“Advises one of the authors of the study. Because if the terminals are still the major polluters of digital, the trend could be reversed in the years to come.