The International Energy Agency said AI-focused facilities grew even faster, after data centres used about 415 terawatt-hours in 2024, or 1.5% of global electricity demand.
It projects consumption could reach about 945 terawatt-hours by 2030, even as power per AI task falls, because wider use, new features and slow grid upgrades keep pushing demand higher.
The report says this is spurring research into brain-inspired efficiency, including sparse neural networks and timing-based hardware, as training large AI models can consume more than a million kilowatt-hours.
Can brain-inspired hardware scale fast enough to avert the looming AI-driven energy crisis?
Is the race to build brain-like AI ignoring its growing thirst for the world's water and resources?
With EU rules on AI energy use imminent, will tech giants be forced to reveal their true environmental costs?
Data Center Power Crunch: The Global Impact of AI-Driven Electricity Demand Doubling by 2030
Overview
In 2025, global electricity demand from data centres surged to unprecedented levels, mainly driven by the rapid expansion of tech companies focused on advanced computing and AI workloads. This spike led to significant price increases in electricity markets, such as a $9.3 billion rise in the PJM capacity market, which directly impacted consumers with higher monthly bills. The United States and China became the main contributors to this demand, with data centre energy use in the U.S. projected to nearly double by 2030. Overall, data centres now consume more energy than some entire nations, highlighting the scale and urgency of this trend.