DIGITAL LIFE

AI-related energy consumption now accounts for 20% of data center energy demands

A new study shows that AI-related energy consumption already accounts for 20% of data center energy demands. According to experts, the technology’s needs could double by the end of the year.

According to the study, published in the scientific journal Joule, this would correspond to almost half of all data center energy consumption worldwide, excluding consumption linked to Bitcoin mining. AI’s energy demands could even surpass those of Bitcoin mining by the end of the year.

Speaking to Wired, Alex de Vries-Gao, author of the study and founder of Digiconomist, says that this issue has become more urgent in recent years, due to the growing adoption of ChatGPT, as well as other language models that use large amounts of energy.

The development of AI is already having an impact on the climate goals of tech giants. In their latest sustainability reports, companies acknowledge that AI is the main driver of their energy consumption growth.

Just last month, a report from the International Energy Agency (IEA) revealed that data centers accounted for 1.5% of global energy use in 2024, close to 415 TWh, and this number is expected to continue growing. The IEA predicts that data center energy consumption will increase to over 900 TWh by the end of the decade.

In recent years, data center energy consumption has grown four times faster than global consumption. As Wired points out, investment in the sector has almost doubled since 2022, driven largely by data center expansions to meet AI demands.

But there is still much uncertainty about the true “weight” of AI in data center energy consumption. Furthermore, large technology companies tend to keep this information secret. To quantify the energy impact of AI, the study took a supply chain approach.

The high computational demands of AI create a natural “bottleneck” in the supply chain around hardware, says Alex de Vries-Gao. Taiwan Semiconductor Manufacturing Company (TSMC) is the largest producer of components for these systems. Companies such as Nvidia turn to TSMC for the production of their chips, and the producer also supplies services to Google and AMD.

According to the expert’s calculations, without a significant increase in current production capacity, AI could consume 82 TWh of electricity this year, equivalent to the annual energy consumption of Switzerland. If production capacity increases, as forecast by analysts, energy demands could grow at the same rate, accounting for almost half of all data center energy consumption by the end of the year.

mundophone