How much electricity will data centers actually use?
The global and US forecasts, side by side — and why credible forecasters disagree by a factor of three.
Global data-center electricity — the IEA outlook
- AI's slice of data-center power is projected to climb from 5–15% recently to 35–50% by 2030.
- Inference dominates. It accounts for the majority of a model's lifetime energy — over 90% by some operator accounts. Usage, not training, is the lever.
- Jevons paradox. Per-query efficiency keeps improving (Gemini fell ~33× in a year), but cheaper inference drives more usage, so total load still rises.
Forecasts disagree — a lot
Third-party projections of global data-center electricity vary widely by forecaster, year, and scenario. They measure the same thing, so they are comparable; the gap between them is honest uncertainty, not error.
US only, 2030
The spread is the whole point: central estimates run about 2.8× from low to high.
- In capacity terms: BloombergNEF sees US data-center power reaching ~106 GW by 2035, up from ~25 GW in 2024 — 8.6% of all US electricity, more than double today's 3.5%. BloombergNEF — US data-center power demand ~106 GW by 2035 (first-party insight, Dec 2025)
- Why the spread: forecasts hinge on how much announced pipeline actually gets built and powered — interconnection queues are heavily speculative — plus efficiency gains and utilisation assumptions.
Forecasters: IEA, Energy and AI (2025) + Key Questions update (2026) · BloombergNEF (BNEF) — data-center power-demand research & forecasts · Gartner — data-center electricity to double by 2030 (~980 TWh) · S&P Global / 451 Research — global data-center demand ~1,587 TWh by 2030 · EPRI — Powering Intelligence 2026 (US Low/Medium/High scenarios) · Lawrence Berkeley National Lab, 2024 US Data Center Energy Usage Report — 325–580 TWh (6.7–12% of US electricity) by 2028, up from 176 TWh (4.4%) in 2023 · World Resources Institute — US 2030 forecasts span 206–970 TWh
If a developer cites a single forecast at your hearing, ask which one and why. There is no consensus number, and the choice between 350 TWh and 970 TWh is doing a lot of work in any argument about whether the grid can absorb a new campus.
The rural migration
Pew Research Center analysed where new US facilities are going (April 2026). The finding that matters for community advocacy: the build-out has moved to rural counties, most of which have never hosted a data center and have no zoning precedent for one.
- Regional drivers: the South and Midwest are capturing three-quarters of all planned US developments. The South alone accounts for 48% of upcoming sites.
- Growth speed: planned developments represent a 62% increase in total facilities for the South and 64% for the Midwest, relative to current counts.
- Tight clusters: 90% of all operating and planned sites are within 5 miles of another. The first approval in a county is rarely the last.
Top states by planned & operating facilities
| State | Operating | Planned | Total |
|---|---|---|---|
| Virginia | 398 | 287 | 685 |
| Texas | 296 | 170 | 466 |
| Illinois | 139 | 123 | 262 |
| Georgia | 94 | 141 | 235 |
| Ohio | 166 | 57 | 223 |
| Arizona | 98 | 86 | 184 |
| Pennsylvania | 78 | 51 | 129 |
| North Carolina | 72 | 41 | 113 |
| Iowa | 64 | 41 | 105 |
| Indiana | 38 | 54 | 92 |
Pew Research Center, April 2026; facility counts via DataCenterMap. Counts, not megawatts — a small network room counts the same as a gigawatt campus. Pew Research Center — Most new data centers in the U.S. are coming to rural areas (April 13, 2026)
Why this lands on you: a county that has never permitted a data center has no template — no zoning overlay, no noise ordinance, no water study, and often no staff who have read one of these applications before. That is the gap the toolkit is built for.
What this means where you live
National forecasts don't decide anything. Your state's grid, rates, and regulator do.