Texas Has 474 Gigawatts of Data Center Requests. Four Out of Five Are Fiction. It Does Not Matter.

• 

  Steel transmission towers carrying high-voltage lines across a canyon under clear sky

The interconnection queues that supposedly prove America's power crisis are mostly speculative paperwork. Grid operators now expect roughly one gigawatt in five to energize. Deflate the numbers anyway and Texas still has to build a second Texas.

In May 2026, ERCOT's Large Load Working Group recorded the observed peak consumption of all large loads on the Texas grid at approximately 3,900 megawatts. That is every crypto mine, every hyperscale campus, every industrial electrolyzer big enough to register, running at their combined high-water mark, adding up to less power than a single mid-sized city.

Ten weeks later, Governor Greg Abbott's office cited 474 gigawatts of large-load interconnection requests sitting in ERCOT's pipeline.

The ratio is roughly 120 to one.

That gap is the single most misunderstood number in American infrastructure. It gets reported as evidence of a historic power shortage, and it is quoted in earnings calls, legislative hearings and utility rate filings as though it describes demand. It does not. It describes intentions, most of which will evaporate.

It helps to establish the actual size of the sector before interpreting the queues. Lawrence Berkeley National Laboratory, in the study Congress mandated under the Energy Act of 2020, measured US data center consumption at 176 terawatt-hours in 2023, or 4.4 percent of national electricity, up from roughly 76 terawatt-hours in 2018 and 58 terawatt-hours in 2014 (18). Its forward figures, the ones most often quoted as fact, are scenarios rather than forecasts: 325 to 580 terawatt-hours by 2028, or 6.7 to 12 percent of national electricity, implying total data center power demand somewhere between 74 and 132 gigawatts at an assumed 50 percent utilization. The observed number is a measurement. The 2028 range is a modeled band roughly twice as wide as the entire sector was in 2023.

The interesting part is what happens when you take the queue numbers seriously and deflate them to what grid operators themselves expect to survive. The shortage does not disappear. It gets worse in a more specific way, and the bill has already started arriving.

What an interconnection queue actually measures

A generation interconnection queue has been a well-understood object for two decades. A developer submits a project, pays a deposit, sits through impact studies, and either signs an agreement or withdraws. Attrition is high, roughly 70 to 80 percent historically, and everyone in the industry prices that in.

Load interconnection queues are new, and until recently they carried almost no equivalent discipline. Historically there was no standardized transparency requirement for large loads at all, because for a century nobody needed one. Industrial customers arrived slowly, one at a time, and utilities served them under obligation-to-serve rules that assumed the customer was real.

Hyperscale computing broke that assumption in a specific way. A data center developer selecting a site does not commit to one grid. It submits requests to several utilities across several states simultaneously, holds the positions while it negotiates power price, tax treatment and fiber, then drops the losers. The same 500 megawatts can appear in four queues at once. Nothing about that is fraudulent. It is rational behavior when a queue position costs essentially nothing and the option value of holding one is enormous.

The result is a set of numbers that measure interest, duplicated across jurisdictions, with no mechanism to net out the duplicates. Karl Rábago, a former Texas Public Utility Commission member, has described the incentive plainly: when holding a queue position is cheaper than forgoing one, developers will hold them without limit.

Utilities have started saying this out loud, and the disclosures are the most useful data in the sector.

Exelon offers the cleanest example. Between year-end 2025 and the second quarter of 2026, its high-probability data center load fell by roughly 40 percent, from about 18 gigawatts to about 11 (3). Its total interconnection pipeline dropped from roughly 43 gigawatts to about 25. Chief Financial Officer Jeanne Jones attributed the decline to the utility's improved ability to screen speculative projects through transmission service agreements. Separately, Exelon has indicated that only about 22 percent of its 65 gigawatt pipeline through 2040 is likely to materialize.

Georgia Power's filings tell the same story retrospectively. Since its 2023 integrated resource plan update, the utility has removed 33 data center projects totaling 11,332 megawatts from its models, roughly 65 percent of the load that had been announced (4). About a quarter of every data center project that entered its planning model was later scrubbed.

ERCOT itself expects approximately 100 gigawatts of its roughly 450 gigawatt queue to survive its new Batch Zero screening process, which the Public Utility Commission of Texas approved on June 18, 2026.

Three independent sources, three different methods, and all three land near the same figure. Call it 20 to 22 percent.

The map

The table below ranks the major US grids and utility territories by large-load interconnection requests, converts each to a multiple of that territory's reference peak demand, and then applies a 20 percent conversion rate to estimate what might plausibly energize.

The conversion-adjusted column is our own computation, derived by applying the 20 percent survival rate that ERCOT, Exelon and Georgia Power's disclosures independently converge on. It is an estimate, not a forecast, and it is offered as a corrective to headline queue figures rather than as a prediction of any individual project.

Grid or territory	States	Large-load requests (GW)	As of	Reference peak (GW)	Requests ÷ peak	Adjusted at 20% (GW)	Adjusted load as % of peak
ERCOT	Texas	474	Aug 2026	86	5.5x	94.8	+110%
Dominion Energy Virginia	Virginia	70	Feb 2026	24.7	2.8x	14.0	+57%
Georgia Power	Georgia	50.9	Q3 2025	see note	not computed	10.2	not computed
PJM (RTO-wide)	13 states and DC	60	Jan 2026	160	0.38x	12.0	+8%
SPP	14 states	26.4	2026	54	0.49x	5.3	+10%
MISO	15 states	18	2026	see note	not computed	3.6	not computed
NYISO	New York	12	Dec 2025	see note	not computed	2.4	not computed
TVA	7 states	11	Jul 2025	38	0.29x	2.2	+6%
PG&E	Northern California	12.7	Q2 2026	see note	not computed	2.5	not computed
ISO-NE	6 states	0.3	Mar 2026	see note	~0	0.06	~0%

Sources: ERCOT, Dominion Energy Virginia SCC filings, Georgia Power IRP, PJM 2026 Long-Term Load Forecast, SPP, NERC, NYISO, TVA, PG&E, ISO-NE. Conversion-adjusted figures computed by Analytics.loan. Reference peak is each territory's record or most recent system peak. Rows marked "see note" cover territories where the operator does not publish a load queue at a granularity that maps cleanly to a single peak-demand figure, generally because the footprint spans multiple states or because load interconnection is handled by member utilities rather than the grid operator.

Two rows carry the entire story.

Texas. Apply the deflator that ERCOT's own screening process implies, and 474 gigawatts of requests becomes roughly 95 gigawatts of plausible new load. ERCOT's record peak is approximately 86 gigawatts. The surviving fraction of the queue, the part everyone agrees is real, would more than double the largest load the Texas grid has ever carried. Not over a generation. Over roughly a decade.

That is the finding. The queue is fiction, and it does not matter, because the non-fictional remainder is still an existential build.

Virginia. Dominion Energy Virginia's queue of large-load delivery-point requests reached approximately 70 gigawatts by February 2026. Its all-time system peak, set on January 23, 2025, was 24,678 megawatts. Deflated to 14 gigawatts, the surviving load would still expand peak demand by 57 percent in a territory that already hosts the densest concentration of data centers on earth. Dominion has contracted for 47.2 gigawatts beyond current need, but only about 25 gigawatts carries an assigned connection date through 2031.

Everything below those two rows is a materially smaller problem. PJM as a whole, SPP and TVA all land in single-digit to low-double-digit percentage increases once deflated, which is fast growth by utility standards but not a structural break. ISO-NE, the outlier in the other direction, had roughly a couple hundred megawatts in formal study as of March 2026, consisting of one 200 megawatt data center in northeast Massachusetts and one 85 megawatt electrification project in Connecticut, contributing precisely zero to its upcoming seasonal peaks.

Growth rate is a separate question from absolute scale, and on that measure New York deserves attention it rarely gets. NYISO went from six large-load proposals totaling roughly one gigawatt in 2022 to 48 proposals totaling approximately 12 gigawatts as of December 31, 2025. More than two-thirds of that capacity entered the queue during 2025 alone. The absolute number is small next to Texas. The trajectory is not, and it explains why New York moved to a statewide moratorium before any other state did.

California is the instructive counterexample, because its numbers move in both directions. Pacific Gas and Electric reported a data center pipeline of 7.3 gigawatts at year-end 2025, then 5.4 gigawatts in the first quarter of 2026, then 12.7 gigawatts in the second. Of that final figure, 490 megawatts carried executed interconnection agreements, which is under 4 percent. A pipeline that swings by more than half in two quarters is not measuring demand. It is measuring how many developers happened to be shopping that quarter. CAISO, for its part, has stated that its own planning assessments do not indicate a systemic generation adequacy shortfall comparable to those in other regions.

The concentration is the point. This is not a national power crisis distributed evenly across fifty states. It is an acute crisis in two jurisdictions, a rapid but small-based acceleration in a third, a manageable trend in five or six, and a non-event in New England.

Reliability regulators are reading it the same way, with one important caveat. NERC's 2025 Long-Term Reliability Assessment, released in January 2026, flags 13 of 23 assessment areas at elevated or high risk over the next five years, with MISO, PJM, ERCOT, WECC-Northwest, WECC-Basin and SERC-Central in the highest category. It projects bulk-system summer peak demand growing 224 gigawatts over ten years, a figure 69 percent above the prior year's projection, and names new data centers as the largest single contributor. The caveat is that the assessment has been challenged from the other direction: a March 2026 Grid Strategies analysis prepared for several environmental organizations argues the LTRA pairs an aggressive demand forecast with conservative generation and inter-regional transfer assumptions, and specifically overstates how much data center load will actually connect. Both documents cannot be right about the connection rate, and the reader should note that the criticism runs in the same direction as the utility disclosures above.

The sorting has begun

Through 2024 and most of 2025, no grid operator had a reliable way to distinguish a funded hyperscale campus from a landowner with a spreadsheet. That changed quickly.

ERCOT's Batch Zero, approved by the PUCT in June 2026, screens the accumulated backlog before it enters formal study. SPP's High Impact Large Load process, known as HILL, received FERC approval on January 14, 2026, and the Commission has since pointed to it as a template. PJM tightened its own vetting enough that its near-term forecast actually fell, trimming summer 2026 large-load expectations by 0.7 percent (8), a small number that represents a large methodological shift: the first time a major grid operator revised data center load downward.

MISO filed large-load reliability requirements with FERC in August 2026, alongside a new process defining large and computational loads as a distinct interconnection category. And on June 18, 2026, FERC issued orders to show cause to essentially every RTO and ISO in the country, including CAISO and ISO-NE, directing each to justify or reform its large-load interconnection tariff.

Texas went furthest through legislation. Senate Bill 6, effective June 2025, directs the PUCT to impose curtailment protocols on new large loads above 75 megawatts interconnecting after December 31, 2025, requiring them to shed load during firm load-shed events. It also mandates financial security, demonstrated site control, and, critically, disclosure of duplicate service requests filed with other utilities. The implementing rule, 16 TAC 25.194, was proposed on March 12, 2026 with a $50,000 per megawatt security deposit, reduced from an original $100,000 per megawatt, and final adoption is expected by the end of 2026. A companion co-location and curtailment rule, 16 TAC 25.205, was adopted in March 2026 and is already being enforced.

Duplicate-request disclosure is the provision that matters most for anyone reading these numbers. It is the first regulatory mechanism in the country designed to net out the double-counting that inflates every queue in every market.

The cost is already real

While the industry argues about which gigawatts are genuine, the money has moved.

PJM's independent market monitor, Monitoring Analytics, attributes $29.4 billion of capacity market cost increases to data center load, approximately 46 percent of the $63.6 billion in total capacity charges across the last four base residual auctions. Its president, Joseph Bowring, published the figure in July 2026. The mechanism is straightforward: capacity auctions clear against forecast peak demand, forecasts incorporate large-load projections, and PJM's 2025/2026 auction cleared at $269.92 per megawatt-day against $28.92 the prior year, an increase of roughly ten times.

That clearing price is paid by every customer in the territory, including the roughly four-fifths of projected data center load that will never plug in.

At the retail level, federal Energy Information Administration data show average residential electricity prices rising 94 percent in Washington D.C., 74 percent in Maryland, 73 percent in Maine and 58 percent in New York between March 2021 and March 2026. Data centers are not the sole cause of those increases, and the honest reading is that they are one driver among several including gas prices, storm hardening and transmission investment. But the geography is suggestive: the sharpest increases cluster in and around PJM.

Dominion's own 2025 integrated resource plan makes the arithmetic explicit. If data centers do not pay their allocated share, the utility projects average residential monthly bills rising from $159 today to between $255 and $308 by 2035, and as high as $381 by 2045.

The regulatory response has converged on a single instrument: the take-or-pay large-load tariff.

The Public Utilities Commission of Ohio approved AEP Ohio's version on July 9, 2025, effective later that month. New or expanded data centers above 25 megawatts must pay for at least 85 percent of contracted capacity regardless of consumption, on 12-year terms, with collateral requirements and exit penalties. AEP's FY2025 annual report describes similar structures across its other jurisdictions with contract terms running as long as 20 years and take-or-pay minimums reaching 90 percent, approved or pending in Virginia, West Virginia and Indiana (25). TVA took up a dedicated data center rate at its August 2026 board meeting, framing it explicitly as protection against rate pressure on other customers.

Georgia shows what happens when the generation gets approved before the contracts do. The Georgia Public Service Commission approved roughly 9,885 megawatts of new generation to serve large loads in December 2025, but commission public interest advocacy staff testified that only about 1,900 megawatts of that build was backed by executed contracts under the state's new large-load rules. The remaining 8,000 megawatts is being constructed against expectation, and if the expectation does not convert, the cost of the plants does not disappear. It gets allocated.

Not everyone accepts the premise. The Institute for Energy Research found essentially no statistical correlation between a state's data center concentration and its electricity price level, with the ten highest-concentration states averaging 14.46 cents per kilowatt-hour against 14.39 cents elsewhere, and argues the causation runs backward: data centers seek out cheap power rather than creating expensive power (26). A 2026 Charles River Associates study prepared for the Edison Electric Institute reached a broadly similar conclusion. On the other side, the Institute for Energy Economics and Financial Analysis attributed 63 percent of PJM's 2025 to 2026 price increase directly to data center load growth.

The dispute is genuine and both sides are measuring real things. At the individual facility or ZIP code level, the price signal is muted. At the system and capacity market level, the cost is large and documented. A reader deciding whether to believe this should note that the $29.4 billion figure comes from PJM's own independent monitor, not an advocacy group.

The bottleneck moved

Suppose every queue were purged of duplicates tomorrow and every remaining project were genuine and financed. Most still could not be energized on schedule, because the constraint has migrated from generation capacity to the physical equipment required to deliver it.

Gas turbines are the tightest. GE Vernova's combined gas power equipment backlog and slot reservation agreements reached 116 gigawatts in the second quarter of 2026, up from 100 gigawatts in the first quarter and 83 gigawatts at year-end 2025. Chief Executive Scott Strazik confirmed on the July 2026 earnings call that the company is now booking reservations for 2031 deliveries. Siemens Energy carries a firm backlog near 69 gigawatts with lead times of three years or more. Mitsubishi Heavy Industries holds roughly 35 gigawatts for large-frame units and is sold out into 2028. Turbine prices have roughly tripled since 2019.

Large power transformers are close behind. Standard power transformers average approximately 128 weeks for delivery, generator step-up units approximately 144 weeks, and specialized orders extend past four years. Prices are up roughly 77 percent since 2019. The binding constraint is not capital and not even steel, though Cleveland-Cliffs' Butler Works in Pennsylvania remains the sole domestic producer of grain-oriented electrical steel. It is a small, specialized coil-winding workforce that cannot be scaled by writing a check.

Set those timelines against the asset being built. A data center campus can go from groundbreaking to commissioning in 18 to 36 months. New transmission takes seven to ten years. The equipment to energize it takes three to five.

That mismatch, not land, water or tax abatement, is what makes announced capacity undeliverable on announced timelines. It is also what is driving the behind-the-meter response: on-site gas generation, fuel cells, and nuclear power purchase agreements including the Three Mile Island and Palisades restarts and small modular reactor commitments from Microsoft, Google, Amazon and Meta. Very little of that nuclear capacity delivers before the early 2030s.

The political window is closing

The regulatory environment hardened faster in 2026 than most underwriting assumed.

New York became the first state with a statewide moratorium. Governor Kathy Hochul signed Executive Order No. 62 on July 14, 2026, pausing discretionary state environmental permits for data centers of 50 megawatts or more for one year while the state builds a regulatory framework, and directing the Public Service Commission to establish new electric and water rate classes. The legislature had passed a broader bill covering facilities above 20 megawatts in June 2026, which Hochul declined to sign in favor of the narrower order. She also committed to pursuing repeal of the state's data center sales tax exemptions.

Local restriction is spreading faster than state action. Data Center Watch counted at least 75 projects worth approximately $130 billion blocked or delayed in the first quarter of 2026 alone, roughly matching the total for all of 2025. A March 2026 Gallup survey found seven in ten Americans oppose data center construction in their own communities. More than 300 state-level data center bills were filed in the first six weeks of 2026, with statewide moratorium proposals introduced in 14 states.

The tax incentive retreat is the development most likely to reprice existing assets. Arizona enacted a three-year moratorium on its sales tax exemption running from July 2026 through June 2029. Illinois and Ohio paused new incentive agreements. North Carolina repealed its electricity tax exemption. Nebraska and Washington narrowed theirs. As of August 2026, four states had enacted rollbacks and at least nine more were considering them. Most apply only to new agreements, which protects existing operators and simultaneously raises the value of an already-executed abatement.

What this means for anyone underwriting the sector

Four conclusions follow from the data rather than from the narrative.

Announced capacity is not a demand signal. Any pro forma, land valuation or utility capital plan that treats queue volume as forecast demand is overstating by a factor of roughly five. The correct treatment is to apply a conversion rate near 20 percent and then ask whether the project pencils on the remainder.

Deflating the queue does not rescue Texas or Virginia. In every other territory the adjusted numbers are manageable. In those two they are not. Concentration risk in the sector is geographic and extreme, and it sits in the two jurisdictions that market themselves hardest on power availability.

Queue position is now a distinct, valuable and separately underwritable asset. With turbine and transformer lead times running past 2029, a site with an executed interconnection agreement and secured equipment slots is categorically different from a site with a request. That difference is not currently reflected in most land pricing.

Take-or-pay is the new baseline. An 85 to 90 percent minimum on contracted capacity over 12 to 20 years converts a variable operating cost into a fixed obligation that survives the tenant. Anyone underwriting data center real estate, or the utility credit behind it, is now underwriting a long-dated take-or-pay contract whether or not the deal documents call it that.

The queues will keep making headlines, and the headline number will keep rising. It is the wrong number. The one to watch is the survival rate: whether ERCOT's Batch Zero really does clear 100 gigawatts from 450, whether PG&E converts more than 490 megawatts of its 12.7 gigawatt pipeline into executed agreements, and whether turbine lead times ever fall back inside a decade.

Frequently asked questions

Which US state grid is the most oversubscribed by data center power requests?

Texas, by a wide margin. ERCOT was tracking approximately 474 gigawatts of large-load interconnection requests as of August 2026, roughly 5.5 times its record system peak of about 86 gigawatts, with roughly 87 percent of that volume attributed to data centers. Virginia is second: Dominion Energy Virginia held approximately 70 gigawatts of delivery-point requests against an all-time peak of 24,678 megawatts, a multiple of about 2.8. No other territory in the country approaches those ratios.

Will the data centers that companies announce actually get built?

Most will not, at least not where they were announced. Grid operators and utilities now converge on a conversion rate near 20 percent. ERCOT expects roughly 100 gigawatts of its approximately 450 gigawatt queue to survive screening. Exelon cut its high-probability data center load by about 40 percent in a single quarter in 2026 and estimates roughly 22 percent of its long-range pipeline will materialize. Georgia Power has removed about 65 percent of previously announced load from its planning models. The primary cause is not project failure but duplication: developers file the same project in several jurisdictions at once and drop the losers.

How much are data centers raising my electricity bill, and who actually pays?

At the wholesale level the effect is documented and large. PJM's independent market monitor attributes $29.4 billion of capacity market cost increases to data center load, about 46 percent of total capacity charges across the last four auctions. Those costs are recovered from all customers in the territory. At the retail level, residential prices rose 94 percent in Washington D.C., 74 percent in Maryland and 58 percent in New York between March 2021 and March 2026, though data centers are one driver among several. The counterargument, advanced by the Institute for Energy Research and a Charles River Associates study for the utility industry, is that states with heavy data center concentration do not show measurably higher prices, and that data centers locate where power is already cheap. States are increasingly settling the question through take-or-pay tariffs that force large loads to pay for capacity whether or not they consume it.

Why can't utilities just build more power plants to serve data centers?

Because the equipment is sold out. GE Vernova's gas power backlog reached 116 gigawatts by the second quarter of 2026 and it is now booking reservations for 2031 delivery. Siemens Energy and Mitsubishi Heavy Industries carry multi-year backlogs of their own, and turbine prices have roughly tripled since 2019. Large power transformers average about 128 weeks for delivery, with generator step-up units near 144 weeks and some specialized orders past four years. The deepest constraint is a small specialized coil-winding workforce rather than capital or raw material. A data center takes 18 to 36 months to build. The equipment to power it takes three to five years, and new transmission takes seven to ten.

Which states have banned, paused or pulled tax breaks from data centers in 2026?

New York enacted the first statewide moratorium through Executive Order No. 62, signed July 14, 2026, pausing discretionary state environmental permits for facilities of 50 megawatts or more for one year. On tax incentives, Arizona imposed a three-year moratorium on its sales tax exemption from July 2026 through June 2029, Illinois and Ohio paused new incentive agreements, North Carolina repealed its electricity tax exemption, and Nebraska and Washington narrowed theirs. As of August 2026, four states had enacted rollbacks with at least nine more considering action. Separately, more than 300 state-level data center bills were filed in the first six weeks of 2026, including statewide moratorium proposals in 14 states, and at least 75 projects worth approximately $130 billion were blocked or delayed at the local level in the first quarter of 2026 alone.

Sources:

1. ERCOT Large Load Working Group, meeting materials, May 21, 2026

2. Office of the Governor of Texas, letter, August 3, 2026

3. Exelon Corporation, second quarter 2026 earnings disclosure and executive commentary, July 2026

4. Georgia Power Company, Integrated Resource Plan filings and updates, Georgia Public Service Commission, 2025

5. Public Utility Commission of Texas, Batch Zero large load screening approval, June 18, 2026 Dominion Energy Virginia, filing in Virginia State Corporation Commission Case No. PUR-2026-00011, February 2026

6. ISO New England, Final Draft 2026 Large Load Forecast, March 27, 2026

7. PJM Interconnection, 2026 Long-Term Load Forecast, January 14, 2026

8. Texas Senate Bill 6, 89th Legislature, effective June 2025; Public Utility Commission of Texas proposed rule 16 TAC 25.194, March 12, 2026; adopted rule 16 TAC 25.205, March 26, 2026 Monitoring Analytics, independent market monitor for PJM, capacity market cost analysis, July 2026

9. US Energy Information Administration, average retail price of electricity by state, March 2021 through March 2026

10. Public Utilities Commission of Ohio, AEP Ohio data center tariff approval, July 9, 2025

11. GE Vernova, second quarter 2026 results and earnings call, July 2026

12. Wood Mackenzie, transformer supply chain survey, 2025

13. State of New York, Executive Order No. 62, July 14, 2026

14. Data Center Watch, project opposition tracking, first quarter 2026

15. North American Electric Reliability Corporation, 2025 Long-Term Reliability Assessment, January 29, 2026

16. Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report, December 2024, and 2025 update

17. Southwest Power Pool, High Impact Large Load process, Federal Energy Regulatory Commission approval, January 14, 2026

18. Federal Energy Regulatory Commission, orders to show cause to regional transmission organizations and independent system operators, June 18, 2026

19. Tennessee Valley Authority, large load service data and board materials, July 2025 and August 2026

20. New York Independent System Operator, large load interconnection reporting, December 31, 2025

21. Pacific Gas and Electric Company, data center pipeline disclosures, second quarter 2026 Midcontinent Independent System Operator, large load reliability requirements filing, Federal Energy Regulatory Commission, August 2026 (25) American Electric Power, FY2025 annual report on Form 10-K

22. Institute for Energy Research, analysis of data center concentration and state electricity prices, 2026

23. Charles River Associates, study prepared for the Edison Electric Institute, 2026

24. Institute for Energy Economics and Financial Analysis, PJM capacity price analysis, 2026

25. Gallup, survey on local data center development, March 2026

26. Siemens Energy and Mitsubishi Heavy Industries, reported gas turbine backlog disclosures, 2026