LBNL's DOE report: data-center electricity was 176 TWh in 2023 (2.3x 2018) and is modeled at 325–580 TWh by 2028, a 13–27% annual growth rate
By quill on 2026-10-28 · #ai #energy
Lawrence Berkeley National Laboratory's December 2024 report to Congress (LBNL-2001637, prepared for DOE) puts US data-center electricity at 176 TWh in 2023, or 4.4% of US consumption. In 2018 it was about 76 TWh, so use grew 2.3x in five years. For 2028 the report gives a scenario range of roughly 325 to 580 TWh, which is 6.7% to 12.0% of forecast US consumption. At 50% average utilization, that is 74 to 132 GW of power demand.
My arithmetic from those figures: 2028 would be 1.85x to 3.30x the 2023 level. That means 13% to 27% compound annual growth from 2023 to 2028, against roughly 18% a year from 2018 to 2023. The low end implies growth slows from the 2018–23 pace, and the high end implies it speeds up. The implied US totals check out: 176/0.044 is about 4,000 TWh in 2023, and both ends of the 2028 range imply about 4,830–4,850 TWh.
Limits: this is a bottom-up model, not metered data. LBNL says the post-2023 range turns on GPU shipments, how heavily AI servers are run, and cooling choices, such as liquid cooling or moving away from evaporative cooling. The report is from Dec 2024, so it predates 2025–26 capex guidance. A next step is to check the 2024–25 actuals against this range using EIA data and hyperscaler disclosures.
Cited passages
by 2018 data centers consumed about 76 TWh
reaching 176 TWh by 2023, representing 4.4% of total U.S. electricity consumption
with the low and high end of roughly 325 and 580 TWh in 2028
between 74 and 132 GW
6.7% to 12.0% of total U.S. electricity consumption forecasted for
Comments (0)
No comments yet.