AI data-center energy coverage is now crowded with large numbers, but the most credible 2026 picture still comes from a short list of institutional sources. The IEA provides the global demand baseline, Berkeley Lab’s DOE-backed work provides the U.S. planning view, and the real takeaway is that local grid concentration matters as much as the headline global total.
Top-line AI data-center energy statistics
| Metric | Latest published figure | Source year | Why it matters |
|---|---|---|---|
| Global data-center electricity use | About 415 TWh, or roughly 1.5% of global electricity consumption | 2025 report using 2024 data | Sets the global baseline and shows that data centers are still a manageable share globally, but growing quickly |
| U.S. data-center electricity use | 176 TWh in 2023, equal to about 4.4% of total U.S. electricity use | 2025 Berkeley Lab summary of the 2024 DOE-backed report | Shows why the U.S. planning conversation is more urgent than the global average alone suggests |
| Projected U.S. data-center electricity use | Roughly 325 to 580 TWh by 2028 | 2025 Berkeley Lab summary of the 2024 DOE-backed report | Shows the size of the potential grid-planning range rather than a single deterministic outcome |
| IEA estimate for U.S. data-center electricity use | Around 180 TWh in 2024, with demand expected to rise by about 240 TWh by 2030 | 2025 | Shows that recent U.S. growth has continued even after the 2023 Berkeley baseline |
Methodology note
The “2026” in this page title is the Rewiredz update year, not a claim that every figure was published in 2026. Each figure below is labeled with its real source year, and the table intentionally separates global and U.S. numbers so they are not blended into one misleading headline.
What the current statistics actually show
The global numbers are big enough to matter, but the operational risk remains regional and local. The IEA explicitly notes that data centers are geographically concentrated. Berkeley Lab’s U.S. work shows the same pattern from a grid-planning angle: the challenge is not simply total annual electricity use, but how quickly large new loads show up in specific regions with finite interconnection, transmission, and reserve headroom.
Why source labeling matters here
This topic is especially vulnerable to sloppy extrapolation. Global 2024 demand, U.S. 2023 demand, and U.S. 2028 or 2030 projections are not interchangeable. Keeping source years visible helps separate current consumption from forecast ranges and keeps AI headlines tied to the right planning horizon.
Related Rewiredz reading
- Read the broader Rewiredz explainer first.
- Compare the demand side with the grid-constraint side directly.
- Connect the numbers back to the wider energy-trilemma context.