U.S. Commercial Electricity Sales
Quarterly U.S. commercial electricity sales — macro proxy for whether AI compute is becoming a physical load
Key Insight
Commercial electricity sales are finally showing an AI signal in the macro data — Q2 2025 came in at 130.2 TWh, up +6.1% YoY, roughly 5x the pre-2023 trend rate of ~1% and the clearest quarter-over-quarter evidence that AI compute has crossed from line-item to load-driver.
130.2TWh
Trend YoY growth is +6.5%, accelerating by 1.6 pp/year over the last 5Y. Latest: +6.1%, 0.4 pp below trend, a 0.11σ deviation. The latest YoY reading is boosted by 0.7 pp due to an easy comparison base from Apr 1.
Level
YoY %
y = −1.8% + 1.6 pp/yr · t
Deviation from trend
Forecast
Projected value by forecast vintage (TWh)
Projected value (TWh)
| Forecast made in | Q3 '24 | Q4 '24 | Q1 '25 | Q2 '25 | Q3 '25 | Q4 '25 | Q1 '26 | Q2 '26 | Q3 '26 | MAPE |
|---|---|---|---|---|---|---|---|---|---|---|
| Q3 '24 | 147.5 | 124.0 | 122.9 | 127.7 | 153.7 | 132.3 | 131.0 | 136.2 | 161.0 | 36% |
| Q4 '24 | 126.8 | 124.2 | 129.1 | 155.7 | 134.2 | 133.1 | 138.6 | 166.2 | 15% | |
| Q1 '25 | 125.4 | 129.8 | 156.6 | 135.1 | 134.1 | 139.7 | 168.6 | 5% | ||
| Q2 '25 | 130.2 | 157.1 | 135.6 | 134.7 | 140.4 | 170.2 | ||||
| Q3 '25 | 157.1 | 135.6 | 134.7 | 140.4 | 170.2 |
YoY change forecast
| Forecast made in | Q3 '24 | Q4 '24 | Q1 '25 | Q2 '25 | Q3 '25 | Q4 '25 | Q1 '26 | Q2 '26 | Q3 '26 | MAPE |
|---|---|---|---|---|---|---|---|---|---|---|
| Q3 '24 | +5.3% | +3.7% | +3.9% | +4.0% | +4.2% | +4.3% | +4.5% | +4.6% | +4.8% | 36% |
| Q4 '24 | +6.1% | +5.0% | +5.3% | +5.6% | +5.9% | +6.2% | +6.5% | +6.7% | 15% | |
| Q1 '25 | +6.0% | +5.8% | +6.2% | +6.6% | +6.9% | +7.3% | +7.7% | 5% | ||
| Q2 '25 | +6.1% | +6.5% | +7.0% | +7.4% | +7.9% | +8.3% | ||||
| Q3 '25 | +6.5% | +7.0% | +7.4% | +7.9% | +8.3% |
Forecasts use ordinary least-squares linear regression fitted to the YoY change series over a rolling 3Y window. Each row shows a vintage — the forecast as it would have appeared at that point in time. Projected values apply the forecasted YoY change to the prior year's level, chaining forward where actuals are unavailable. MAPE measures forecast accuracy against realized values. These are mechanical trend extrapolations, not economic models.