US electricity generation grows 10% in a year
Will EIA report US net electricity generation for a calendar year at least 10% above the previous 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
US electricity generation grew about 2.8% in 2025, and EIA expects about 2.6% in 2026 and 1.5% in 2027. A 10% jump in one year has not happened since 1959, and even the most aggressive data-centre forecasts imply under 4% a year. I put the chance of a qualifying year by 2031 at well under 1%. The thing that would most change this is evidence of data-centre construction several times faster than today's pace by 2029–2030.
Where things stand
Latest Electric Power Monthly, Table 1.1 (released Sept 24, 2026, data through July 2026; https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=table_1_01). In thousand MWh, Total Generation at Utility Scale Facilities was 4,308,634 for 2024 and 4,429,502 for 2025, a rise of 2.81%. Generation for January to July 2026 was 2,629,899, against 2,578,682 for the same months of 2025, a rise of 1.99%. For 2026 to come in 10% above 2025 it would need 4,872,452. That means August to December 2026 would have to run 21.2% above the same months of 2025. That will not happen.
EIA's October 2026 Short-Term Energy Outlook (modeling finished Oct 1, 2026; STEO_m.xlsx, Table 7a, series TSEOTWH, which I summed from the monthly values) forecasts generation of 4,544.6 TWh in 2026 (+2.6%) and 4,613.3 TWh in 2027 (+1.5%). For 2027 to reach the threshold it would need about 454 TWh more than 2026, which is roughly 52 GW of extra load running around the clock. That is about seven times EIA's forecast increase.
Timing. The December figures, and with them the first preliminary annual total, come out in the late-February issue. So the years that can be tested inside the window are 2026 through 2030, published around February 2027 through February 2031. The 2026-12-31 horizon can only resolve NO, because no tested year will have ended.
Base rate
I used EIA's Monthly Energy Review Table 7.2a annual net generation for 1949–2025 (https://www.eia.gov/totalenergy/data/browser/csv.php?tbl=T07.02A).
- Growth of 10% or more happened only in 1950, 1951, 1953, 1955 and 1959.
- Since 1974 the largest rise is 9.6% in 1989. That jump came from a change in coverage, when non-utility generators were added, not from real growth.
- Since 1990 the largest rise is 4.4%, the 2010 rebound after the recession. Average growth was 1.1% a year, with a standard deviation of about 2.1 percentage points.
- The largest fall is 4.1%, in 2009.
A 10% year would sit about four standard deviations above the modern average.
Even the most aggressive outlooks stay far short of 10% in any year:
- Grid Strategies: 166 GW of peak demand growth from 2025 to 2030 in utilities' own forecasts, about 3.7% a year, and it calls some of these forecasts inflated (gridstrategiesllc.com, Nov 2025).
- EPRI: total demand growth of 2–3.6% a year for 2025–2030.
- Rabobank: consumption up about 20% by 2030 (Jan 2026).
- BloombergNEF: data centres at 12% of US consumption by 2030 (RTO Insider, Jul 22, 2026).
- EIA's 2026 Annual Energy Outlook: 0.9–1.6% a year.
On the ground, about 12 GW of data centres were expected to come online in the US in 2026, on the Sightline figure SemiAnalysis cited on Jun 18, 2026; SemiAnalysis itself thinks more is under construction.
Paths to YES
- A demand surge. This would take something like 60–90 GW of new data centre or industrial load connected within one or two years around 2029–2030, which is several times current build rates. Shortages of gas turbines, transformers, grid connections and labour make that very unlikely, and some data centres generate their own power off the grid. I put this at about 0.3% across 2028–2030.
- Collapse and rebound. A fall of 9% or more in 2027–2029, from a grid-scale geomagnetic storm, an attack or a depression, followed by a recovery of 10% or more the next year, with EIA still publishing on schedule. Neither COVID (−2.9%) nor 2009 (−4.1%) came close. About 0.1–0.2%.
- An EIA data error, or a change in coverage within the same column. About 0.1%.
Forecast
I set the per-quarter chance at 0.01% in ordinary quarters. In the first quarter of each year, when the new annual total appears, it rises with the year being tested: 0.03% (2026 data), 0.04% (2027), 0.08% (2028), 0.12% (2029) and 0.18% (2030). Combined, that gives about 0.6% by the end of 2031.
Strongest case against: a genuine AI-driven takeoff in which data centre and industrial build-out doubles every year through 2030. Even then, my rough calculation gives annual growth of about 6–8% unless additions reach roughly 70 GW or more a year, so it should not move the forecast much. A future outlook from EIA or a grid planner showing any single year above 6–7% growth would be the signal to revise upward.
Resolution criteria
Question. By the last day of each quarter, has EIA reported US net electricity generation for a completed calendar year at least 10% above the previous calendar year?
Definitions.
- EIA: the US Energy Information Administration.
- US net electricity generation: the annual total in EIA's Electric Power Monthly, Table 1.1, “Net Generation by Energy Source: Total (All Sectors),” column “Total Generation at Utility Scale Facilities.” This covers all energy sources and all producer sectors, including commercial and industrial generators, in the 50 states and District of Columbia. It excludes the separately reported estimated small-scale solar generation. Net generation is electricity generated less electricity consumed in operating the generating facilities; it is not generation net of international electricity trade.
- Calendar year: January 1 through December 31. Under the Observation period convention, the year being tested must be 2026 or later; its comparison year is the immediately preceding calendar year.
- At least 10% above: annual generation in year Y divided by annual generation in year Y−1 is at least 1.10. Calculate from the published annual totals at their greatest available precision in the qualifying release, without rounding the calculated percentage. Use both years from the same release or data vintage.
Resolves YES if an EIA publication or dated EIA data release reports annual totals meeting the threshold. Preliminary estimates for a completed year and final figures both count. A revision that first brings a qualifying year over the threshold counts on the revision's publication date. A later downward revision does not reverse YES. No attribution to AI, datacentres, or any particular cause is required.
Resolves NO for a quarter if no qualifying release has been published by its end. Forecasts, projections, annualized monthly figures, year-to-date comparisons, non-calendar rolling twelve-month periods, regional totals, individual fuel totals, gross generation, electricity sales or consumption, and generating capacity do not count. A rounded statement of “10%” does not count if the accompanying annual totals imply less than 10%. Republishing or revising only pre-2026 annual observations does not count.
Resolving source. EIA only: Electric Power Monthly, Table 1.1, its downloadable tables, or the identical annual series in EIA's Electricity Data Browser. A successor EIA table counts only if it preserves this coverage, or supplies comparable totals for both years on this basis. Use the dated EIA release or an archived release to establish the data vintage. The Resolution window and Observation period conventions apply; resolve on publication date under Resolution date (A3), not on the end date of the measured year. This named-source requirement is narrower than Public confirmation (A2).