Texas Grid Watch

A daily numeric record of how the ERCOT grid is absorbing large constant load. This page does not predict, grade or reassure. It publishes measurements, the arithmetic done on them and the size of what nobody outside ERCOT can see.

The load factor is the mean demand across a day divided by that day's peak. On August 14th, 2026 it was 84.7%.

Peak megawatts is a weather story. Texas has broken its own demand record most summers since air conditioning arrived, and a data center barely shows up in it. The shape is the story. A data center draws at four in the morning close to what it draws at five in the afternoon. A grid taking on large constant load sees its overnight floor rise faster than its afternoon ceiling, and the load factor climbs. Weather lifts both ends together.

One day can't show that. A year can, which is why the series starts now.

ERCOT publishes a day ahead forecast, then publishes what the grid actually did. Both arrive in the same record, so the gap between them is arithmetic rather than opinion. Across the 5 days held, ERCOT's day ahead peak forecast missed the measured peak by 1,353 MW on average. That is 1.6% of peak. The worst miss so far was 2,503 MW.

This is published whichever way it falls. It is a measurement of somebody else's model rather than a grade. It is here because a page that checks its own inputs in public is worth more than one that asserts they are good.

Generation by fuel, integrated across the day from ERCOT's five minute telemetry. Storage is signed. It reads negative on a day the batteries took in more than they gave back, which is most days, because a round trip loses energy. Shares are computed against generation that was positive, so a charging battery is not published as a generator.

Two ERCOT feeds are read separately here, one for demand and one for generation by fuel. Everything generated inside the grid is consumed inside it, so net generation and measured load have to land on each other. The two rows above differ by 0.3%, which is about what direct current ties to the neighbouring grids and line losses should account for. It is the cross check that would catch either reader silently breaking. That is why both rows are printed rather than just the one that flatters.

The record holds 5 days, of which 5 are complete settled days and 0 are marked unverified. An unverified day carries no numbers at all rather than yesterday's. A gap that says it is a gap is honest. A gap filled with the last known figure is a fabrication nothing downstream can detect.

Once a day the settled previous day is read whole from ERCOT's public dashboard feeds. Measured demand hour by hour, the day ahead forecast ERCOT published for those hours, the capacity it had committed and generation by fuel. The raw responses are archived before anything parses them.

The record stores the full hourly series rather than only the summary. Every figure on this page can be recomputed from the open data, without refetching anything and without trusting the code that wrote it.

ERCOT keeps no archive of these feeds. Each is a rolling window. A day not collected is gone for good. So the collector runs on its own schedule, never as a step inside a routine that could fail for an unrelated reason.