Data centers: what the numbers actually tell us.

The useful distinction

All data-center demand is not AI demand. A small global share can still place a large demand on a particular grid. Both facts belong in the conversation.

Start with the denominator

The IEA estimated that data centers consumed 415 terawatt-hours of electricity in 2024, about 1.5% of the world’s total. That figure covers data centers broadly, including services beyond AI. Calling all of it “AI electricity” changes what the number means. Source

In its 2025 analysis, the IEA projected approximately 945 terawatt-hours by 2030 in its base case. This is a scenario, not a measurement of future demand. It is also a global estimate: it cannot tell a resident what a proposed facility will mean for their own community. Source

More efficient does not always mean less electricity

The IEA’s April 2026 update reported 17% growth in data-center electricity demand during 2025. It also described rapidly improving efficiency per AI task. Those findings can coexist when use expands and more demanding applications become common. Source

Efficiency is progress worth pursuing. It is not, by itself, evidence that total consumption is falling.

The local questions are real

The same IEA update says large, concentrated loads can require new generation and grid investment. Electricity prices do not inevitably rise with demand, but policy and infrastructure choices matter. It also describes both renewable-energy agreements and projects pursuing on-site natural gas. A single “clean” or “dirty” label misses those differences. Source

Berkeley Lab’s data-center work models electricity and water demand and examines cooling practices. Water deserves its own accounting: a power-demand figure alone cannot establish the water impact of a particular facility. Source

Ask questions a project can answer

Our interpretation: a more useful discussion starts with the specific facility and the assumptions behind its claims.

• How much electricity is requested, and when will the load arrive?

• Which grid upgrades are needed, and who pays for them?

• What cooling system and water supply will be used?

• Which benefits exist today, and which depend on future construction or technology?

There is room for optimism about better design and better planning. Neither removes the need to measure the costs where people live.