Every grid operator runs on a forecast. It has to decide hours ahead which power plants to commit, and years ahead what transmission to build. The forecast behind those two decisions is a different animal in each case, and in 2026 the long-range version has become the most argued-over number in the power industry. This explainer walks through how a grid forecast is made, where US operators publish theirs, and what the latest ones say. Every figure links to the operator or agency that published it.
Key takeaways
- Short-term forecasts are hourly (or finer), run out a few days, and are driven mostly by the weather forecast. ERCOT, PJM and CAISO all publish theirs openly.
- Long-term forecasts add economics, premise counts and large-load requests. PJM projects its summer peak growing an average 3.6% a year over the next 10 years (PJM 2026 Load Forecast Report).
- NERC's 2025 Long-Term Reliability Assessment expects aggregate summer peak demand to rise by over 224 GW over ten years (NERC 2025 LTRA).
- Large-load requests are the hard part. ERCOT itself says its preliminary 2032 forecast is probably too high (ERCOT, Apr 15 2026).
Short-term grid forecasts: hours to a week
| Operator | Product | Horizon and granularity | Source |
|---|---|---|---|
| ERCOT | Seven-Day Load Forecast (NP3-561-CD) | Hourly, by weather zone and system total, current day plus six days | ERCOT data product |
| ERCOT | Mid-Term Load Forecast | Hourly, next 168 hours, from current weather forecasts per weather zone | ERCOT load forecast |
| PJM | 7-day load forecast | Hourly by area, updated twice an hour | PJM Data Miner feed |
| CAISO | Today's Outlook demand | Day-ahead forecast as 1-hour averages; hour-ahead as 5-minute averages | CAISO Today's Outlook |
What goes into a short-term forecast
PJM's Manual 19 lists the usual ingredients (PJM Manual 19, rev. 39):
- Weather: temperature, humidity, wind speed and cloud cover. In summer PJM combines temperature and humidity into a temperature-humidity index.
- Calendar: day of week, month, holidays and Daylight Saving Time.
The model learns how load has historically responded to those inputs. It then feeds in the weather forecast. That is why a bad temperature forecast becomes a bad load forecast almost immediately.
How accurate are they?
ERCOT publishes monthly mean absolute percentage error (MAPE) for its 3-hour-ahead, 6-hour-ahead and day-ahead forecasts (ERCOT forecast metrics). We are not quoting a single accuracy number here: the published workbooks do not label every forecast horizon clearly enough to say which figure is the day-ahead error.
Long-term grid forecasts: years to decades
Long-range forecasts drive transmission plans, capacity markets and reliability studies, so the method changes:
- Economics and premises. ERCOT models hourly load as a function of the number of premises, weather and calendar variables. Its county-level economic and demographic inputs come from Moody's, and it uses actual weather from 2008 through 2024 (ERCOT 2025 LTLF report).
- Large-load adjustments. Operators add data centers and other big new loads on top of the econometric forecast, and they discount what customers request. In its 2025 report, ERCOT found average peak consumption per data center site was 49.8% of the requested MW (same report).
What the latest long-term forecasts say
- PJM (January 2026). Summer peak is forecast at 222,106 MW in 2036, a 10-year increase of 65,733 MW, reaching 253,077 MW in 2046. Growth averages 3.6% a year over 10 years and 2.4% over 20. Most zones are adjusted for data center load growth. The near-term forecast was trimmed from the 2025 report: 2,564 MW (1.6%) lower for 2026 (PJM 2026 Load Forecast Report).
- ERCOT (April 2026). The preliminary long-term forecast shows about 367,790 MW of demand by 2032. ERCOT says, "we believe this forecast to be higher than expected future load growth" (ERCOT release).
- NERC (2025 LTRA). Aggregate summer peak demand is forecast to rise by over 224 GW, 69% more than the prior year's 10-year projection of 132 GW (NERC 2025 LTRA).
- EIA (September 2026). US electricity sales are expected to total 4,135 billion kWh in 2026, up almost 2% from 2025, and 4,211 billion kWh in 2027 (EIA STEO release).
A 2026 reality check in Texas
ERCOT told the Texas PUC that summer 2026 peak load would likely fall between 90,500 MW and 98,000 MW. That compares with roughly 112,000 MW in its preliminary long-term forecast (ERCOT filing, PUCT 58777). The actual record came on July 22, 2026, when hourly peak load reached 91.1 GW (EIA Today in Energy). That landed inside ERCOT's range and far below the long-term number.
The lesson is not that forecasts are wrong. A weather-driven short-term forecast and a request-driven long-term forecast answer different questions, and they should be read that way.
Why a grid forecast matters outside the control room
- Prices. Day-ahead forecasts shape which plants are committed, and so real-time prices. See current prices on the ERCOT, PJM and CAISO live pages.
- Transmission costs in Texas. Large customers try to forecast the four summer peaks that set their transmission charges — see ERCOT 4CP explained.
- Large-load siting. Long-term forecasts decide where transmission gets built. That is why data center interconnection queues get so much scrutiny — see data center power outage.
The ZoningLabs live grid status pages show each operator's current load and forecast peak from the operator's public data, with an as-of timestamp. The operator's own site is always the authoritative grid forecast.
FAQ
What is a grid forecast?
A grid forecast is an operator's projection of electricity demand (load). Short-term versions run hourly for the next day to week and are driven by weather. Long-term versions project peak demand years ahead, using economic growth and large new loads such as data centers.
Where can I see the ERCOT grid forecast?
ERCOT publishes its seven-day hourly forecast by weather zone (report NP3-561-CD) and a 168-hour mid-term forecast on ercot.com. The ZoningLabs ERCOT page shows the forecast peak alongside current load, with a timestamp.
How far ahead does PJM forecast load?
PJM's short-term feed covers the next seven days, hourly by area, updated twice an hour. Its annual long-term load forecast report projects summer and winter peaks 10 to 20 years out.
Why do long-term grid forecasts keep changing?
Mostly because of large-load requests. Operators must decide how much of the requested data center and industrial demand will actually show up. ERCOT's 2025 report found data center sites peaked at about half of what they requested, and its 2026 preliminary forecast came with a warning that it was likely too high.
Is weather the biggest driver of a day-ahead grid forecast?
Usually, yes. Temperature, humidity, wind speed and cloud cover are core model inputs in PJM's Manual 19, alongside calendar effects such as weekdays and holidays.
About ZoningLabs Team
Grid data
Expert in real estate analysis and zoning regulations. Passionate about helping property professionals make data-driven decisions.