Two years ago the scarce input for AI was chips. Increasingly it is electricity. The International Energy Agency puts data centres' electricity use at about 415 terawatt-hours in 2024, around 1.5 per cent of global consumption. It projects that figure will more than double to about 945 TWh by 2030, a little more than Japan uses today. AI is the main driver of the increase.
Data centre electricity demand
Global consumption, terawatt-hours (IEA Base Case)
The demand is concentrated
Three markets account for 85 per cent of data centre electricity use. Within those markets, capacity clusters tightly: nearly half of US capacity sits in five regional clusters. That concentration explains why grid connections, not chips, are now the constraint in several regions. The IEA expects the United States to use more electricity for data centres by 2030 than for all its energy-intensive manufacturing combined.
Where data centre electricity is used
Share of global data centre electricity consumption, 2024
Why this matters outside the energy sector
- AI unit costs will track power prices. Model the cost of an AI business case at production volume. Treat the price of inference as a variable, not a constant.
- Emissions targets need an AI line. If your sustainability commitments include cloud and IT emissions, an AI roll-out can move the number noticeably. Plan for it before the audit finds it.
- Location is a strategic choice again. For firms building or leasing capacity, grid access and power contracts now rank alongside land and latency in site selection.
- Efficiency is a competitive advantage. Smaller models, better prompts and fewer redundant calls reduce cost and energy together. They are worth measuring.
The AI business case that ignores the power bill is the one that fails in year two.
Sources
- International Energy Agency: Energy and AI, Executive summary (2025). iea.org