How to read data-center clusters through the interconnection queue.

Open the ERCOT map, turn on Interconnection queue and Data centers, then zoom out far enough to see county-scale concentrations. The queue layer aggregates projects at county centroids because the public ERCOT GIS report does not publish exact project geometry. That makes it excellent for finding regional concentrations and deliberately poor evidence for claiming that two facilities share a parcel, substation, or transmission path.

The first question is comparative: where is proposed generation or storage capacity concentrating, and where do independently sourced data-center records appear nearby? On the June 30, 2026 queue snapshot, the agent ranked Pecos County first by active capacity at 16,600.6 MW across 37 active projects. Milam followed at 7,069.3 MW across 20 projects and Reeves at 6,635.6 MW across 15. Queue megawatts are requested generation or storage capacity—not installed capacity, accredited capacity, a large-load request, or proof of transmission headroom.

Keep the map sparse enough to read. Zones establish the market geography. The queue and data-center layers establish the two candidate patterns. Transmission, substations, and generation come on only when the question moves from cluster detection to physical context.

ERCOT publishes the queue as periodic snapshots. Ask the Grid diffs consecutive publications into entries, exits, and field updates, while the timeline keeps the map on the same historical state. Moving the clock backward asks what was on file at that publication. Moving it forward toward a proposed completion date asks what the queue said was expected—not what was actually built.

The Pecos County card is explicit about the state used here: latest queue June 30, 2026. Compared with the prior snapshot, seven projects entered, none exited, and six were updated. Those changes are the investigatory lead. A large increase might be new projects, revised capacity, a status change, or several records moving together. The current total alone cannot tell you which.

Treat proposed completion as a queue field, not a delivery forecast. Treat ERCOT's Completed status as a feed status, not proof of commercial operation. The timeline is valuable precisely because it preserves those seams rather than smoothing them into a single present.

The county card turns the map cluster into a denominator. Pecos County contains 54 projects totaling 20,399 MW in the selected snapshot: 37 active projects totaling 16,600.6 MW and 17 marked Completed totaling 3,798.5 MW. The active expected-delivery stack is heavily back-ended—207 MW in 2026, 1,620.5 MW in 2027, 2,031.6 MW in 2028, 2,970.3 MW in 2029, and 9,771.2 MW in 2030. One active 207 MW project is already past its proposed date.

Now select a project from the county list. The Giant Arc-II is a 1,300 MW gas project from Powering Knowledge, LLC, queue ID 30INR0105, filed April 21, 2026 with a proposed completion date of April 10, 2030. Its named point of interconnection is a tap between the BOTTLEBRUSH and SOLSTICE 345 kV buses. The May 31 publication revised capacity from 1,335.18 MW to 1,300 MW.

That record is more useful than a marker because it exposes what is known and what is not. The point shown on the map remains the Pecos County centroid. The named point of interconnection is a queue-study fact. Neither should be substituted for exact site geometry, a load-service agreement, or available capacity at a nearby substation.

Turn on high-voltage transmission, substations, and generation around the selected county or project. These layers answer adjacency questions: which 345 kV corridors cross the area, which substations and plants are recorded nearby, and what other infrastructure occupies the same regional footprint? They do not answer loading, thermal headroom, upgrade cost, deliverability, or whether a data center can take service at a visible substation.

The disciplined next questions are specific. Which mapped substation corresponds to the named queue buses? Which transmission owner and study documents govern the tap? Which existing generators share the corridor? Which data-center points have corroborated identities rather than a single-source candidate? Which of those loads are grid-connected, and which plan to operate behind the meter or on a private grid?

This is where the map is most useful as an evidence index. It narrows the records to inspect and keeps each asset attached to its source. It is not a substitute power-flow model.

The map shows where records concentrate; web research explains what some of those records may represent. Ask the agent for named facilities, exact county, operating or planned status, reported power scale, the primary source, and the power arrangement. Also ask it to state what remains unverified. That final instruction matters because West Texas project names routinely blur city and county.

The correction that changes this example is simple: Microsoft's approximately 2 GW Pecos campus is in the city of Pecos, which is in Reeves County—not Pecos County. Microsoft's announcement also says the campus will launch with co-located behind-the-meter natural-gas generation. In Pecos County itself, Pacifico Energy describes GW Ranch as a private-grid campus able to deliver more than 5 GW, with first power planned for Q1 2027. Project Horizon describes a 559-acre Pecos County computing campus using on-property behind-the-meter gas generation, with a public-grid interconnection maintained for backup reliability.

Milam County supplies a different comparison. Riot says its Rockdale site has 700 MW of developed capacity and seven buildings; its January 2026 AMD lease began at 25 MW of critical IT load, with a path to 200 MW, while Riot intends to convert the site's full 700 MW gross capacity for data-center tenants. That is corroborated infrastructure in the same county as a large active generation queue, but it still does not establish that one caused the other.

A county does not have an ERCOT demand series, so the analysis must use a labeled regional proxy. For Pecos County, the agent used ERCOT's Far West weather-zone actual load for the 30 completed days from June 26 through July 25, 2026 Central. Across 720 hourly intervals, mean load was 7,632 MW, population standard deviation was 307.8 MW, and the range was 6,710.6 to 8,442.0 MW.

The five highest intervals clustered on July 24 and July 11. The maximum was 8,442.00 MW at the July 24 11:00 interval-end, followed by 8,392.03 MW at 12:00 and 8,388.44 MW at 10:00 that day; July 11 12:00 reached 8,371.08 MW, and July 24 09:00 reached 8,348.80 MW. Daily peaks landed at hour-ending 13, 14, or 15 on 18 of the 30 days, with the rest spread from 16 through 21.

The ramps are as important as the peaks. The largest absolute hourly changes included a 691.1 MW drop into July 22 09:00 and a 672.4 MW rise into July 23 04:00. For a prospective 1,300 MW computing load, this does not predict interconnection feasibility. It frames the regional operating regime that a developer should test against ramping, backup generation, curtailment, and demand-response assumptions.

For market conditions, the agent used HB_WEST as a regional pricing proxy—not a project-specific settlement point. Across 8,505 native five-minute SCED intervals in the same 30-day window, median real-time LMP was $24.93/MWh, p95 was $50.15/MWh, the maximum was $394.07/MWh, the minimum was negative $11.37/MWh, and population standard deviation was $21.47/MWh. Prices were negative in 690 intervals, or 8.11 percent of the sample.

The largest five-minute move was a $147.03/MWh jump to $235.17/MWh on July 20 at 15:15:21 Central. Several of the largest moves and highest daily maxima landed from July 20 through July 23, near the period containing the Far West series' largest hourly ramps and a July 24 peak. That is an association worth investigating, not a tested cause.

To advance the claim, inspect binding constraints, outages, and the transmission facilities implicated at the same intervals. A regional hub series cannot identify which line moved, and a nearby queue project cannot explain a price spike merely because it shares a county. This stop condition is part of the analysis, not a disclaimer added after it.

Interconnection queue plus data-center assets can answer where independently sourced patterns overlap, which counties carry the most active proposed supply, what one queue item says, and how that record changed between publications. Adding transmission, substations, generation, and the timeline can answer which physical records should be opened next and what was knowable at a given date. Adding the agent can rank, calculate, reconcile names against primary sources, and return the exact data contracts behind a load or price chart.

The same stack cannot, by itself, prove that a data center caused a generation project, that a visible substation has headroom, that requested queue megawatts will reach commercial operation, or that a regional load or hub-price move applies to one facility. Those require the large-load process, interconnection studies, facility ratings, constraint attribution, and—where available—private operating data.

That boundary is the high-level picture. The queue identifies proposed supply and its publication history. The asset layers place it in a physical and commercial neighborhood. Web research distinguishes the facilities and power arrangements behind the names. Demand and price history describe the operating regime. The answer becomes useful when all four remain connected and none is allowed to impersonate another.