Research conducted by Alex de Vries, founder of Digiconomist, warns that the widespread use of artificial intelligence (AI)-based tools will lead to a significant increase in energy consumption in data centres. In the coming years, the energy demand for this technology may exceed the annual consumption recorded in countries such as Argentina and Sweden.
Globally, according to the École Supérieure de Technique at the University of Quebec in Canada, data centers consume between 2% and 3% of global energy, with a growth trend of 50% annually.
The training and inference phases of AI models have become the focus of various researches on the sustainability of these resources. Both processes consume large amounts of energy.
The Organization for Economic Co-operation and Development (OECD) estimates that major technology companies allocate between 7% and 10% of their total spending on computing infrastructure for AI applications. Between 3 and 4.5% goes into training machine learning models, while up to 4.5% is about putting what you learn into practice.
De Vries’ analysis shows that AI’s footprint in energy will grow significantly over the next four years as tools evolve and become more commonly used.
“The result of making these tools more efficient and accessible may be that we simply allow more applications for them and more people to use them,” he said. The researcher predicts that in 2027, global electricity consumption associated with AI systems will increase between 85 and 134 terawatt-hours per year. This amount is similar to the annual electricity consumption in the Netherlands, Argentina or Sweden.
The widespread use of artificial intelligence is a problem for the planet
Companies like Meta or Microsoft are integrating AI systems into all their commonly used applications. This trend will generate a significant increase in energy demand. Vries uses Google and its search engine to exemplify this trend. By its estimates, the company processes up to 9 billion web queries per day. If each of these searches were solved by artificial intelligence algorithms, Mountain View will consume 29.2 terawatt-hours per year, which is equivalent to Ireland’s annual electricity consumption.
“The potential growth shows that we have to take into account what we use AI for. It consumes a lot of energy, so we don’t want to use it for all kinds of things that we don’t really need,” Freis said.
The International Energy Agency expects data centers to be responsible for 1% of the total greenhouse gas emissions generated in the world. The University of Massachusetts Amherst revealed that training an AI model generates the same emissions as five cars during their entire life cycle.
“Beeraholic. Friend of animals everywhere. Evil web scholar. Zombie maven.”