Every year, the International Energy Agency offers projections for global energy demand based on different scenarios. And this year, artificial intelligence is playing a role in the agency’s electricity demand projections.
Electricity demand is growing faster than expected, concluded the agency. And although data centers are playing a role, they are not the primary driver.
The World Energy Outlook estimates that total mid-century electricity demand could range from 50,000 terawatt-hours under a base case scenario, to 66,000 TWh in a net-zero emissions scenario that limits global warming to 1.5 degrees Celsius. The latter path, the report warns, is narrowing.
The base case scenario is 6% — or 2,200 terawatt-hours — higher than last year’s. The report notes that each year between 2023 and 2035, growth is expected to average nearly 1,000 TWh, “equivalent to adding another Japan to global electricity consumption each year.”
“In energy history, we’ve witnessed the Age of Coal and the Age of Oil,” said IEA executive director Fatih Birol in a statement. “And we’re now moving at speed into the Age of Electricity, which will define the global energy system going forward and increasingly be based on clean sources of electricity.”
The rapid advancement of AI is certainly contributing, but the agency notes it’s only one element of the spike in demand. Industrial consumption, electric mobility, and the rise of cooling are equally, if not more, responsible.
Space cooling via air conditioning, for instance, represents a much bigger share of electricity demand than AI does, as “the combination of rising incomes and increasing global temperatures generate more than 1,200 terawatt hours of extra global demand for cooling by 2035.” That’s “an amount greater than the entire Middle East’s electricity use today.”
The surge in adoption of electric vehicles, meanwhile, is predicted to be even more impactful.
Tentative predictions
Predictions for data centers and AI use are much harder to model because of rapid technology change. Experts agree that projections are difficult.
“With established technology companies and AI start-ups making major investments, a sharp rise in electricity consumption by data centers looks inevitable,” the report says. “But the relatively early stage of this new phase of growth and sparse data availability mean that any projections are bound to be tentative.”
Electricity demand growth by 2030 could vary by as much as 170 TWh. One element of the uncertainty the supply chain bottlenecks facing data center development; both the scarcity of chips and the need for better grid infrastructure are major industry concerns.
Additionally, the IEA’s outlook notes that contrasting policies add uncertainty to the predictions. While local policies support “the development of domestic value chains for chips and data centers,” federal governments have imposed trade restrictions that complicate expansion.
The concentration problem
Data centers may “account for a relatively small share of overall electricity demand growth to 2030,” but the main problem is that they tend to demand it all in the same place.
Currently, there are roughly 11,000 data centers in the world. And due to their location needs, data centers “are often spatially concentrated,” the IEA found; the ideal is somewhere with access to cheap electricity, fiber optic networks, and cooler climates that bring down the cost of air conditioning.
The clustering can be a cause of local tension, with jurisdictions taking steps to prevent further development (a trend that is also true for utility-scale renewables). And in places where there’s high demand for data centers, power infrastructure constraints can become “alarming,” especially in the United States.
Both factors can contribute to the general uncertainty the IEA maintains when it comes to the speed of AI growth.
