Electricity demand for data centers expected to more than double by 2030, boosted by AI

Already energy-intensive, data centers have redoubled their appetite with the recent development of generative AI, which requires colossal computing capacities to process the information accumulated in gigantic databases.

In 2024, the electricity needs of these infrastructures represented only about 1,5% of global consumption (415 TWh), but they have already increased by 12% per year over the past five years. And it's not over yet.

"Global data center electricity demand is expected to more than double by 2030 to about 945 terawatt hours (TWh), slightly more" than Japan consumes today, according to the IEA.

By this date, data centers would consume a little less than 3% of global electricity, a still modest share but which masks much more significant local effects.

Unevenly distributed throughout the world and concentrated in a few regions of a country, and often near cities, they raise a number of challenges: energy supply, water consumption for cooling, pressure on the electricity network, etc.

A single 100-megawatt data center can consume as much electricity as 100.000 households annually, and tomorrow, the largest centers under construction today will consume 20 times more, the equivalent of the consumption of 2 million households, according to the IEA.

In its report, the first devoted to AI, the OECD's energy agency also highlights that "uncertainties" hang over this consumption, which will depend on "the speed of adoption of AI," progress in energy efficiency and "the possibility of resolving bottlenecks" in the energy sector.

Coal, in the lead today

This thirst for electricity is "particularly" marked in certain countries, such as the United States, where "nearly half of the growth in electricity demand (...) between now and 2030 will be driven by data centers," IEA Executive Director Fatih Birol stressed on Thursday.

Together, the United States, Europe, and China now account for approximately 85% of data center consumption. The first challenge, therefore, is finding affordable and abundant electricity.

"By 2030, renewable energy will cover nearly half of the increase in global electricity demand for data centers, followed closely by natural gas and coal," with nuclear gaining share after 2030, according to the IEA.

"The cheapest way to meet this growing demand will be to accelerate the deployment of renewable energy sources such as wind and solar, supplemented by battery storage," said Euan Graham, an analyst at Ember.

But today, 30% of this electricity comes from coal, an ultra-polluting fuel whose extraction US President Donald Trump promises to "boost" in order to meet the ambitions of AI.

In this situation, the race for data centers will inevitably increase emissions linked to electricity consumption, from 180 million tons of CO2 today to 300 million tons by 2035, a minimal share, however, on the scale of global emissions estimated in 2024 (41,6 billion tons), the IEA cautions.

They will remain "less than 1,5% of total emissions from the energy sector" over the period, but they are among those "increasing the fastest," the Agency specifies.

According to the IEA, these emissions could be offset by emissions savings three to five times greater, thanks to efficiency gains and innovations driven by AI in the economy (energy, transport, buildings, industry), the IEA suggests. Cautiously, however, it points out that "the (widespread) adoption of AI is not guaranteed and could be offset by rebound effects and increased fossil fuel consumption," if their costs fall.

In other words, AI is "not a miracle solution" for the energy transition and a "proactive policy" remains necessary.