AI's Hidden Challenge: Flexibility Over Electricity Demand

The world's growing demand for artificial intelligence (AI) is not primarily an electricity problem, but rather an inflexibility issue. According to the International Energy Agency (IEA), data centers are expected to account for only around 3% of worldwide electricity demand by 2030.

This might seem alarming, given that data centers consumed around 485 terawatt-hours of electricity globally in 2025 and are expected to rise to approximately 950 TWh by 2030. However, the problem lies not in the sheer numbers but in their concentrated nature. Data centers are typically built in large blocks, with extremely high demand for electricity in specific locations, often requiring several gigawatts of capacity.

The issue is further complicated by the fact that AI demand arrives in short timeframes, putting pressure on local grids to adapt quickly. Building more generation is part of the answer, but it's not a straightforward solution. Renewables, natural gas, and nuclear power will benefit from AI demand, but building enough capacity to meet every data center's maximum demand at all times would be unnecessarily expensive.