Data Center Demand Proxy
Over the 12 months through July 24, 2026, average electricity demand across PJM and ERCOT was 14.1% above its 2019 baseline. Demand across the rest of the lower-48 grid was 5.6% higher. The proxy's growth gap is therefore 8.6 percentage points in favor of the PJM+ERCOT basket. That is evidence of where broad load growth is concentrated—not evidence that data centers alone caused it.
What this proxy does—and does not—measure
EIA does not publish a daily meter for data centers. This index tests a narrower, falsifiable proposition: whether total demand in two broad systems central to the data-center buildout story is growing faster than demand elsewhere. It cannot tell us how many megawatts data centers use, separate AI facilities from other data centers, or assign the gap to any one cause.
Demand growth: PJM + ERCOT vs the rest
Each line is a trailing-365-day average indexed to its own 2019 calendar-year mean. That makes the comparison about growth rates, not the basket's absolute size.
The growth gap
Positive means the PJM+ERCOT basket has grown more since 2019; negative means the rest of the grid has. The latest reading is +8.6 percentage points.
Inside the basket
| System | 12-month avg | Since 2019 | Vs prior 12 months |
|---|---|---|---|
| PJM / Mid-Atlantic (MIDA) | 97.7 GW | +6.9% | +3.1% |
| ERCOT / Texas (TEX) | 56.6 GW | +29.3% | +3.7% |
Why traders and investors watch this
Utilities, independent power producers and grid-equipment makers have re-rated on expected data-center load — expectations that are physically checkable. The demand has to show up in PJM and ERCOT, and this gap is the daily scoreboard for whether it is. Because a data center announced today draws power years from now, the proxy deliberately lags the news flow: it confirms or refutes stories the stocks priced instantly, which is exactly what makes it useful. The falsification test is explicit — if the corridors stop outgrowing the rest of the grid for several quarters while announced capacity supposedly energizes, the thesis's physical foundation is weakening before it reaches earnings. Read it next to the valuation side of the same trade on the Bubble Tracker.
How this is computed—and what can move it
We sum EIA-930 adjusted hourly demand for the Mid-Atlantic region (PJM) and Texas (ERCOT), reduce complete days to an average, and compare their trailing-365-day mean with the equivalent mean for the other 11 lower-48 EIA regions. Both sides are normalized to 100 using their own 2019 calendar-year average. Days missing any one region's 24 hourly values are dropped rather than estimated; a window is published only when at least 350 complete days are present.
The long window suppresses most seasonal noise but is not a weather adjustment. Population growth, electrification, industrial activity, efficiency, rooftop solar, unusual weather, reporting changes and data centers can all move regional demand. PJM is especially broad: EIA-930 does not expose the Dominion zone separately in this dataset. Recent values also remain subject to EIA's roughly 30-day revision window.
Sources, methodology & freshnessLast updated 2026-07-24 · Open ↓Close ↑
thetrading.tools PRO
Talk to this data with your AI
PRO is an MCP server that plugs Claude, ChatGPT, Cursor or any AI agent straight into everything on this site — the live market scores, ~90 economic indicators with full histories, and every signal study. Ask a market question; your AI answers from our data, dated and sourced.
Cancel anytime
How Data Center Demand Proxy Works
- 1Two baskets, one gapThe basket is PJM (Mid-Atlantic — home of Data Center Alley) plus ERCOT (Texas); the control is every other lower-48 EIA region. Each basket's trailing-365-day average demand is indexed to its own 2019 calendar-year average, and the headline is the growth gap in percentage points.
- 2Why a proxy and not a meterNo free source meters data-center load directly — utilities disclose it selectively and inconsistently. Broad regional footprints are reproducible back to 2015 and hard to game; the trade-off is that the basket also contains everything else in those regions (weather, industry, population), which the page states up front.
- 3Trailing 365 days kills the weatherA full-year rolling window averages out seasons and heat waves. What is left is structural: if the corridors are persistently outgrowing the rest of the grid, something in them is adding load — and the dominant new load in both since 2022 is data centers.
- 4The components are shown separatelyPJM and ERCOT are different stories — Virginia's build-out vs Texas's (data centers plus crypto plus industry plus population). The page shows each component's index, YoY and growth since 2019 so the aggregate cannot hide a divergence.