ERCOT's telemetry crossed an all-time demand record on a week without record heat

ERCOT's telemetry read 91,089 MW on July 22, 2026, above the 85,508 MW record that had stood since August 10, 2023. The record was not a single spike. July 20 came within 231 MW at 85,277 MW, July 21 reached 87,403 MW, and demand then held within 1.6% of the record for three more days: six consecutive days between 1.6% below and 6.5% above it (hero chart), climbing to that plateau from 69,005 MW on July 14. Every 2026 reading stays preliminary until ERCOT's Demand and Energy Report confirms it.

The week was hot, and no reference station reached a record. ERCOT's four largest metro areas, Dallas Fort Worth, Houston, San Antonio and Austin, plus Midland at the centre of Permian load growth, all stayed 3 to 9 degrees Fahrenheit below their own July records, and 7 to 13 below their all-time highs. Houston came closest, at exactly 3 degrees. San Antonio did not reach 100 degrees until the peak day itself, about three weeks later than its typical first hundred.

That combination is what makes the week worth reading. Crossing a standing demand record while every reference station sat well below its own heat record is consistent with a higher load baseline underneath. How much higher is a separate question.

ERCOT served roughly 16.3 GW more at its 2026 peak than at its entire 2019 annual peak of 74,820 MW. The peak has grown 2.85% a year compounded since 2019, roughly double the 1.4% a year it averaged between the 2010 record of 65,776 MW and the 2019 record.

Three zones carry two thirds of it. ERCOT's own hourly weather-zone load data puts the coincident increase since 2019 at about 20,235 MW system-wide (Grid Flexibility analysis). The DFW zone carries 27%, Houston 22%, and South Central 17%. Far West carries 15% on a base small enough that it amounts to a 75% increase, and the North zone, which takes in the Panhandle, grew 77%.

Far West is the Permian Basin, where ERCOT's own account of growth runs to oil-and-gas electrification, crypto mining, and data centers. The North zone's 77% is the steepest rate of any zone, off the smallest base: about 1.0 GW added on a 2019 base of 1.4 GW. Our data centers analysis itemizes ERCOT's forward large-load pipeline by load type.

Start with what ERCOT has already measured. Its January 2026 System Operations Update counts about 400 operational non-crypto data centers of 1 MW or larger, drawing 3.1 GW to 3.6 GW on an average day, up 15.5% in a year. The large loads it tracks through interconnection grew from 2.6 GW energized in 2022 to 5.9 GW in 2026.

Its forecast is a different measure entirely. ERCOT's 2025 long-term forecast puts peak growth from 2026 to 2031 at about 49.9 GW, of which large load is 46.8 GW: 2.1 GW of large flexible load, 15.8 GW under signed and financially committed contracts, and 28.8 GW resting on officer letters with no signed interconnection agreement.

A third number dwarfs both, and it is the one that ends up in headlines. ERCOT publishes three counts of its large-load pipeline. Its June request list holds about 474 GW through 2032 and 2033, roughly 90% of it data centers, and 88% of that submitted no studies or is still under ERCOT review. Its Batch Zero screen in July sorted a slightly larger pile, about 498 GW, and identified roughly 205 GW with a qualifying study and dynamic model behind it. Its energized series, which sums each site's own highest recorded draw rather than a simultaneous total, reads 5.9 GW.

Most of that pipeline is unstudied. In the request list, 284.3 GW submitted no studies and another 135.5 GW is still under ERCOT review. In the Batch Zero screen, 274 GW came out for having no qualifying study and 20 GW for no dynamic model. The counts do not sum into each other: the screen examined more megawatts than the request list held, so the difference between them is not attrition.

Real-time prices at ERCOT's trading hubs averaged $34.30/MWh across the eight days around the record, July 18 to 25, 2026. Across the eight days around ERCOT's previous record, August 6 to 13, 2023, the hubs averaged $198.56 (Grid Flexibility analysis).

The averages hide the more interesting fact. The median hub observation was nearly identical in the two weeks, $26.61 in 2026 and $27.47 in 2023, while the 95th percentile differed by about sixteen times, $76.73 against $1,238.09. Both are unweighted distributions of hub-price observations rather than the cost of an hour.

That pattern is visible across the record series.

On all eight of ERCOT's all-time peak days since the 2010 move to nodal pricing, the median hub price lands between $19 and $71. The 95th percentile ranges from $93 in 2016 to $3,192 in 2023, a factor of 34. The typical price on a record day is relatively stable across fifteen years; the tail is where record days differ from each other.

2026 sits with the quiet end of that record. Its 95th percentile of $170 is higher than 2016's $93 and 2015's $137, and a fraction of 2018's $602 or 2022's $1,384. What separates it from those mid-2010s days is the load underneath: about 20 GW more than 2016 carried at a comparable price.

Across hourly distributions pooling the roughly 1,114 settlement points ERCOT tracks, the daily median of hourly medians stayed in a $21 to $38/MWh band, including on the two highest-load days. The highest hourly nodal P95 reached $362.23 on July 22 and was back near $47 by July 25, while daily peak demand remained within 1.6% of the record.

Across the eight days of the event window, three produced any interval with a settlement point above $1,000/MWh: July 21, 22 and 23, together 41 node-intervals out of 854,438. At its worst, on July 22, that reached 0.18% of settlement points in a single hour.

Those hours cluster in the evening, roughly 18:00 to 22:00 CT, after solar rolls off while cooling load is still high. July 22 was the sharpest, reaching a single-node maximum of $2,344.25. July 23 is the exception, with its highest prices before dawn. A 9.9 MW battery cleared $2,046/MWh in hour 19 on July 23: Carrizo Springs BESS in Dimmit County, 9.9 MWh of storage (EIA-860M). Every one of those prints was local and brief: a handful of settlement points for a handful of intervals, with the system-wide picture staying flat throughout (see “No emergency, and no notices”). Separating how much of each print was congestion, losses, energy or reserve opportunity cost would take a component decomposition this case does not perform.

ERCOT declared no Energy Emergency Alert during the window, and its Emergency Response Service never deployed. ERS is a procured emergency product that can only be called once ERCOT enters an EEA, so its silence records that ERCOT never reached that condition.

The stronger finding sits one level up. ERCOT's Control Room Operations Messages log, which covers the window continuously, records no formal grid notice of any type: no Operating Condition Notice, Advisory, Watch or Emergency Notice under ERCOT's Nodal Operating Guide, no public-facing Weather Watch, and no Energy Emergency Alert. What it carries instead is routine operating traffic, including Non-Spinning Reserve deployments on the record evening. An ERCOT spokesperson described the grid as operating under normal conditions in contemporaneous news coverage.

Setting an all-time record has usually come with ERCOT saying something. Of the 14 records since 2000, 11 had an emergency alert or a formal notice either issued that day or already running over it, and 10 had one issued on the day itself: a reserve-capacity Operating Condition Notice in 2005, 2010, 2015 and 2019; a reserve Advisory in 2009, 2010, 2016, 2018 and 2019; a system-wide extreme-heat OCN in 2003; an Energy Emergency Alert in effect at the record hour in 2011; and a transmission Watch in 2023. The oldest record, August 31, 2000, predates all four notice types.

Two of the ten ran the same sequence sixteen years apart. On August 23, 2010 ERCOT issued an OCN at 04:59 CT projecting a reserve capacity shortage for hours ending 16:00 through 18:00, added an Advisory at 13:47 as responsive reserve fell below 3,000 MW, and cancelled both that evening. On August 12, 2019 it issued an OCN at 05:00 CT projecting a shortage for hours ending 14:00 through 18:00, the record landed in hour ending 17:00, the Advisory followed at 13:50, and the next day ERCOT declared its first Energy Emergency Alert since January 2014.

Three records had nothing issued: August 3, 2004, July 20, 2022 and July 22, 2026. A week-long extreme-heat notice covering July 20, 2022 was already running. The 2004 entry is disputed: ERCOT's own market monitor puts that year's peak below 2003 (Methodology and basis). July 22, 2026 is the only confirmed record with nothing issued, nothing in effect, and no emergency alert.

ERCOT's log does carry two entries that month: an Advisory on July 4 for a geomagnetic disturbance and a Watch on July 9 for an HRUC-timeline deviation, 18 and 13 days before the record and unrelated to the demand event.

Winter Storm Uri makes the point from the other direction. Served load peaked at 69,692 MW on February 14, 2021, below the then-standing record of 74,820 MW, then fell to 54,523 MW and 46,797 MW over the following two days as ERCOT shed load. Later winters have carried up to 80,525 MW without incident (Grid Flexibility analysis). Uri was a supply-side failure, and load fell because ERCOT cut it.

July 22 had three peaks and they landed in three different hours. Solar peaked at 33.5 GW at 12:39 CT. Demand peaked in the hour ending 18:00, with solar supplying 30.1 GW of it and the storage fleet drawing more power than it sent. Then at 20:24 CT, with solar down to 792 MW and system load below its daily high, ERCOT's ESR Integration Report recorded a 12,017 MW discharge, 14.11% of the load being served at that reading and the fleet's own all-time record. It had first crossed 1,000 MW on January 9, 2024, thirty months earlier.

For this event, the solar, demand and storage peaks were separated by hours.

Over the full month the battery fleet discharged 706 GWh and charged 900 GWh, and July 22 was its largest discharge day at 29.8 GWh against a 22.8 GWh daily average for the month (EIA-930 hourly net generation).

Two things hold solar below its nameplate: the heat itself, and curtailment.

The first is heat. On a clear 105 degree Fahrenheit afternoon the cells inside a utility-scale panel run near 150 degrees, and at that temperature the module ERCOT's own solar-profile study assumes gives up 16% to 19% of its DC output against its 77 degree lab rating.

Some of that never reaches the meter, for a reason built into how solar plants are sized. A plant carries roughly 30% more panel capacity than its inverters can convert, on ERCOT's own representative 1.30 ratio, because panels only hit their rating in ideal conditions and sizing the expensive equipment for a rare peak wastes it. Depending on irradiance and inverter state, that headroom can mask part of a module-level loss. Whether it masked anything on July 22, and how much, is not something ERCOT's public fleet trace can show.

The second is curtailment. That is a decision, not a property of the equipment. ERCOT calls a plant curtailed when its dispatch instruction holds output below what the plant reports it could produce. That is the number worth watching, because a curtailed megawatt is one a co-located battery or a flexible load could in principle be paid to take instead, where the location and the market allow it.

ERCOT publishes no curtailment figure. It has to be rebuilt from dispatch instructions, released 60 days after the operating day, which for the record week arrive between September 20 and 23, 2026. Until then no public source states how much solar ERCOT held back on July 22.

ERCOT publishes its supply side by the hour and in gigawatts. Its demand side sorts into tiers by how much of it reaches the public record, and only the benchmark produces an event-week number.

For scale, that benchmark is the 12,017 MW storage discharge above: front-of-meter, and reported publicly at 5-minute resolution on the day. Every category below it is reported later, in less detail, or not at all.

Capacity known, event output unknown.

What would test this fleet, and when:

The ERCOT and PUCT proceedings that will shape how the next week like this one goes, current as of July 2026.

One line per basis item. Every 2026 figure is preliminary real-time telemetry. Grid Flexibility analysis means computed from ERCOT-published data.