Texas is the clearest picture of the data center power problem, because Texas counts it in public. In April 2026, ERCOT told the Texas House State Affairs Committee it was assessing about 410 GW of large-load interconnection requests, roughly 87 percent of them data centers. By August 2026, Governor Abbott cited about 474 GW, about 90 percent data centers. The grid those requests would join has a record peak of roughly 90 GW. The queue is more than five times the grid. Only 5.8 GW of it has actually energized.
That gap is the story. A queue that large is not a demand forecast. It is a pile of options, and the next two months start pricing them.
How a queue gets to five times the grid
Holding a queue position was cheap, so developers filed early and everywhere. The same project could be filed with multiple utilities in multiple states, negotiated in one place, and left sitting in the rest. Camus Energy CEO Astrid Atkinson estimates requests run five to ten times actual construction, an experience-based estimate, not a systematic measurement. ERCOT itself discounts the queue by about half in planning.
The stage data supports the skepticism. About 71.5 percent of the ERCOT large-load queue had not even submitted an interconnection study. The research firm SemiAnalysis, screening projects for secured land, a defined power solution, and approved permits, concluded that 311 of the 410 GW in the April count are phantom: requests without the site control or financing to ever become real. Roughly three quarters of the demand signal one grid operator is being asked to plan against does not exist.
This is not only a Texas problem. Lawrence Berkeley National Laboratory counted 2,061 GW of generation and storage seeking interconnection at the end of 2025, and only 13 percent of the capacity requested between 2000 and 2020 had reached commercial operation by then. The median project that came online in 2025 took 61 months from request to operation. Load requests ride on top of that same congested system.
The filter arrives
The response, in Texas and nationally, is to make a queue position cost something. FERC acted first. On June 18, 2026, the Commission issued six Section 206 show-cause orders, one to each regional grid operator (PJM EL26-67, SPP EL26-68, NYISO EL26-69, MISO EL26-70, CAISO EL26-71, ISO-NE EL26-72), directing each to justify its large-load tariff or propose revisions. FERC preliminarily defined a large load as a single site of 50 MW or more connecting above 69 kV and not part of a co-located arrangement. Tariff revisions under Section 205 are due November 16, 2026, with PJM, MISO and SPP responses due November 20.
Texas moved on its own track. Senate Bill 6, signed in June 2025, requires large-load customers to demonstrate site control, self-fund grid upgrades, disclose duplicate interconnection requests, and accept curtailment during grid stress. The Public Utility Commission's March 2026 draft rule would require any new load of 75 MW or more to prove site control and sign an intermediate agreement with the transmission provider before ERCOT even begins a study, alongside a proposed 50,000 dollar per MW interconnection fee. A final rule is expected by December 31, 2026. ERCOT's Batch Zero, the first study cohort for loads of 75 MW and above under the process approved June 18, 2026, was then paused on August 3, 2026, when the governor directed an audit of data center load. ERCOT has said it will not authorize data center or crypto load at or above 75 MW to energize until it finishes a project-by-project verification of megawatts, site control, financial security, and ownership.
The other operators are converging on the same logic. MISO's October 2, 2026 filing proposes a 120-day large-load study and asks FERC to rule by December 2, 2026, so the first study period can open early next year; the catch is a 15-year pairing requirement tying a specific plant to a specific site. MISO separately proposed a 90-day review for generation built to serve one load at the same substation that injects nothing onto the shared grid. PJM, whose 2028/2029 capacity auction cleared 6,831 MW short of the reliability requirement at the 325 dollar per MW-day cap, has proposed that new large loads without their own capacity be curtailed before existing customers in an emergency, starting June 1, 2027. SPP offers up to seven years of nonfirm service. ComEd charges 1 million dollars to apply for 50 MW or more, Ohio requires new data centers to pay for at least 85 percent of forecast usage, and Virginia is moving to 14-year lock-in contracts. Once raising a hand becomes placing a bet, the queue shrinks.
The workaround: build the power yourself
While the grid queue tightens, data center developers are routing around it. As of early 2026, developers had planned about 56 GW of on-site generation, roughly 30 percent of total planned data center capacity, according to a Reuters Events white paper. Texas leads with over 20.6 GW of planned behind-the-meter capacity, followed by New Mexico at 9.2 GW, Pennsylvania at 7.5 GW, and Utah at 6.0 GW. On-site generation is the fastest answer to a five-year interconnection wait, and it changes what a load queue even means: the largest new loads may never appear in it.
What separates real load from paper
After November, the useful question about any data center site is no longer how many megawatts were requested. It is four checks: does the site have site control, has it posted security, what study stage has it reached, and has its operator actually contracted for power. Those four checks are the difference between the 5.8 GW that energized and the hundreds of gigawatts that will not.
That is the problem Load Factor Demand is built for. Load maps 1,800+ data center sites against 38,000+ queue projects and answers, year by year, how much generation queued near a site is likely to come online, weighted by historical completion rates and counted at published capacity values, plus the transmission around it and the operators' power buying history from FERC filings. It is an indicative screen, not an interconnection study, and it treats capacity claims the way the queue should: disclosed capacity only, with the source shown.
The ghost load era is ending because the filters are arriving. The winners will be the teams that can tell, site by site, which megawatts are real before everyone else does.
See Load Factor Demand at https://loadfactorenergy.com/products/demand or book a demo at https://loadfactorenergy.com/book.
| Category | GW |
|---|---|
| Large-load requests, Aug 2026 | 474 |
| Record system peak, approx. | 90 |
| Energized so far | 5.8 |
Source: ERCOT figures as reported, April-August 2026
| Category | GW |
|---|---|
| Texas | 20.6 |
| New Mexico | 9.2 |
| Pennsylvania | 7.5 |
| Utah | 6 |
Source: Reuters Events white paper, as reported, early 2026
Sources
- ERCOT large-load figures, April 2026 presentation and August 2026 update, as reported: https://www.mondaq.com/unitedstates/water/1851968/texas-escalates-data-center-oversight-with-attorney-general-water-use-investigation
- ERCOT Batch Zero process, August 2026 pause, and verification requirements: https://hyperscapedigital.com/scorecard/explainers/ercot-batch-zero
- SemiAnalysis phantom-load screening (311 of 410 GW) and LBNL queue figures: https://www.stewartconsultinggtx.com/insights/binding-constraint.html
- FERC June 18, 2026 show-cause orders, dockets EL26-67 through EL26-72, and November 16 filing deadline: https://www.jdsupra.com/legalnews/ferc-issues-show-cause-orders-to-all-7760295/
- FERC orders, background and five reform areas: https://www.duanemorris.com/alerts/ferc_acts_advance_data_center_large_load_integration_six_rto_regions_0626.html
- MISO October 2, 2026 filing, 120-day study, and December 2 ruling request: https://parallaxnexus.com/energy/miso-asks-ferc-to-fast-track-200mw-loads-and-their-power
- On-site generation plans, 56 GW, by state: https://en.wedoany.com/shortnews/443851.html
- Load Factor Demand product, 1,800+ data center sites and supply outlook methodology: https://loadfactorenergy.com/products/demand