Datacentres use 25% of US electricity
Will the EIA, the Department of Energy, or a DOE national laboratory report that datacentres used 25% or more of US electricity for a calendar year?
Chance by Q4 2031
<1%
Cumulative probability
- Latest forecast
| Year | Q1 | Q2 | Q3 | Q4 |
|---|---|---|---|---|
| 2026 | <1% | |||
| 2027 | <1% | <1% | <1% | <1% |
| 2028 | <1% | <1% | <1% | <1% |
| 2029 | <1% | <1% | <1% | <1% |
| 2030 | <1% | <1% | <1% | <1% |
| 2031 | <1% | <1% | <1% | <1% |
Model reasoning
Government estimates put US data centres at about 5% of national electricity in 2024. The Energy Department's own lab projects 9.5–15% by 2030 even in its high case. Reaching a quarter of all US power in a completed year by 2030 would take about two to five times the growth any official or industry forecast expects, so a qualifying report is extremely unlikely. Only an official misstatement or a radical redefinition of data-centre use would change this.
Current status
The question needs EIA, DOE or a DOE lab to report that data centres used at least 25% of total US electricity in a completed calendar year. Because of the observation-period rule, that year must be 2026 or later. The relevant years are therefore 2026–2030, with a report out by the end of 2031. Calendar year 2031 cannot be complete by then.
The authoritative series is far below 25%:
- LBNL, US Data Center Energy Usage Report: 2025 Update (LBNL-2001758), published 18 June 2026. It puts 2024 at 192 TWh, or 4.7% of US electricity, and treats 2024 as the last historical year. Its 2030 reference case is 649 TWh (11.8%). Its widest "compounded uncertainty" band for 2030 is 521–843 TWh (9.5–15.3%). Sources: https://www.osti.gov/biblio/3374245 and https://escholarship.org/uc/item/33m6w3x0 (PDF copy via RTO Insider).
- EIA. Total US consumption in 2025 was about 4.20 trillion kWh (https://www.eia.gov/energyexplained/electricity/use-of-electricity.php). Servers alone were about 7% of commercial-sector electricity in 2025 (Today in Energy, 19 May 2026, https://www.eia.gov/todayinenergy/detail.php?id=67704). In AEO2026, electric vehicles plus data-centre servers together make up only 10–25% of total demand even in 2050 (https://www.eia.gov/outlooks/aeo/narrative/). The October 2026 short-term outlook expects commercial-sector demand, which includes data centres, to grow 2.8% in 2027.
- Other projections, such as EPRI's 9–17% by 2030, are forecasts and are also well below 25% (E&E News, 22 June 2026; PolitiFact/WRAL fact-check, 22 July 2026).
What I computed (Python, from the LBNL and EIA figures above)
- Implied denominators: about 4,085 TWh in 2024 and about 5,500 TWh in 2030.
- 25% in 2030: holding non-data-centre load at the reference level, data centres would need about 1,617 TWh. That is 2.5 times the LBNL reference case, 1.9 times its compounded-high bound, and about 185 GW of average load. It implies roughly 43% annual growth from 2024, against 22.5% in the reference case.
- 25% in 2025: that would be about 1,050 TWh, against roughly 200–250 TWh actually used.
No published scenario from any credible source reaches 25% of US consumption in a completed year by 2030. Gas-turbine and grid build-out limits also make adding roughly 1,000 TWh a year beyond plan impossible by then.
Reference class and paths
Base rate. Official agency historical estimates jumping 2–5 times above the agency's own upper-bound projections within a few years: essentially zero.
Paths to YES are all non-physical:
- A typo or misstatement on energy.gov, eia.gov or a lab site that claims at least 25% for a completed year from 2026 on.
- A politically driven exaggeration in an official DOE document. The President has claimed AI "needs more energy than the entire country produces", but that is a statement about need, not a historical estimate, and it was not made by DOE.
- Resolver error on statistics the criteria exclude: commercial-sector shares (which could approach 25% around 2030), regional shares such as Virginia, shares of demand growth, or capacity shares.
Timing. Reporting lag also cuts the window. LBNL's 2024 numbers came out about 18 months later. EIA's mandatory data-centre survey, which follows the pilot launched in March 2026, may report 2026–2027 data, but at around 6–9%.
Forecast
I used a tiny hazard: about 0.005% in Q4 2026 (no eligible completed year yet), rising from 0.01% to 0.025% per quarter by 2031 as more years and publications become eligible. That gives a cumulative probability of about 0.33% by the end of 2031.
Strongest case against: an AI build-out far beyond forecasts combined with an official body adopting a broad definition. Even so, the arithmetic gap of 2–5 times makes this negligible, so the residual probability mostly reflects error or misresolution risk.
Resolution criteria
Question. By the last day of each quarter, has the EIA, the Department of Energy, or a DOE national laboratory reported that datacentres used 25% or more of total US electricity for a calendar year?
Definitions.
- Reporting bodies: the US Energy Information Administration (EIA), the US Department of Energy (DOE), or any institution officially designated a DOE national laboratory at publication.
- Datacentres: facilities housing computing, storage, and networking equipment, including hyperscale, colocation, enterprise facilities, and server rooms. Count electricity consumed by IT equipment and associated cooling, power conversion, and other facility overhead, using the reporting body's national datacentre estimate. No attribution specifically to AI is required. Do not add separately reported cryptocurrency mining or telecommunications-network consumption to a datacentre total.
- Share of US electricity: annual datacentre electricity consumption divided by total annual US electricity consumption across all sectors, for the same calendar year and national geographic coverage. A share of commercial-sector electricity, electricity-demand growth, peak demand, generating capacity, or primary energy does not qualify. Use the reporting body's national consumption denominator; do not substitute net generation or retail sales alone unless the body explicitly uses it as its measure of total US electricity consumption.
- Calendar year: January 1 through December 31 of a completed year. Historical model-based estimates and preliminary estimates count; projections for an unfinished or future year do not.
Resolves YES if a qualifying publication reports a national share of at least 25%, or supplies a datacentre consumption total and a matching national consumption denominator that yield a share of at least 25%. Both figures must be supplied or expressly identified as the basis of the comparison by the qualifying reporting body; independently combining unrelated series does not count.
Use the published central or best estimate where available. If only a range is reported, its lower bound must be at least 25%. A published numeric share of 25%, including one labelled approximate, qualifies unless more precise figures in the same publication establish a share below 25%; do not round a smaller share upward. Revisions may first establish YES, but later revisions do not reverse an established YES.
The Resolution window and Observation period conventions apply: the publication must fall within the resolution window, and the calendar year observed must end after the opening date.
Resolves NO / Exclusions. Resolve NO for a horizon without a qualifying publication by that horizon. Exclude forecasts, scenario results, annualised partial-year consumption, regional shares, upper uncertainty bounds alone, and statements that datacentres could or will reach 25%. Exclude third-party estimates merely quoted, linked, or hosted by a reporting body unless that body explicitly adopts them as its historical estimate. DOE funding alone does not make a non-laboratory author's report eligible.
Resolving source. Official EIA or DOE reports, data tables, analyses, or releases on eia.gov or energy.gov; and DOE-national-laboratory-authored reports, datasets, papers, or releases published on the laboratory's official website or in an official repository such as DOE's OSTI. Relevant report series include Lawrence Berkeley National Laboratory's 2024 United States Data Center Energy Usage Report (LBNL-2001637) and United States Data Center Energy Usage Report: 2025 Update (LBNL-2001758, OSTI record), and their successors. These named reporting bodies are the exclusive resolving sources, narrower than Public confirmation (A2). Apply Resolution date (A3) to the qualifying publication, not the consumption year.