ENERGY STRATEGY BRIEF · PAPER THREE
Utility interconnection queues, data-center load growth, and the real/fake flexibility test applied to the industry’s own numbers.
Paper One set out the test for value migration. Paper two showed it running at global scale, forced by a war. This paper applies the same test to the place it was arguably discovered first — the U.S. interconnection queue, where the gap between announced and real capacity is now measured, documented, and, in at least one state, being used as a framework to guide policy.
We investigate evidence from three sources: Lawrence Berkeley National Laboratory’s 2026 national interconnection-queue survey, PJM’s own 2026/2027 and 2027/2028 capacity auction results, and a regulatory framework I constructed to test on a live Colorado proceeding. Together they show the same pattern of value migration paper one described in the abstract, now with names, dates, and dollar figures attached.
The National Number
As of the end of 2025, roughly 8,200 projects representing 1,312 gigawatts (GW) of generation and 749 GW of storage, 2,061 GW combined, were sitting in U.S. interconnection queues. That is more than the country’s entire installed generating capacity. Of all capacity that entered a queue between 2000 and 2020, only 13% had reached commercial operation by the end of 2025; 75% had been withdrawn. The flow rate confirms the economics: of the 2,290 GW active in the queue at the end of 2024, only 53 GW reached commercial operation over the following year — a 2% annual conversion rate from request to reality.
One data point makes clear that a signed agreement is not the same as a tested one. LBNL’s data show that 41% of capacity that had already signed an interconnection agreement — the step utilities and regulators generally treat as the line between speculative and real — still withdrew before reaching commercial operation. The signature is a necessary filter, but it’s clearly not a sufficient one.
Why Completion Rates by Technology Are Almost Always Misread
It is tempting to rank resource types by completion rate and read the ranking as a verdict on which technology is more reliable to build. That reading is wrong, and it is wrong in a way that matters for this paper’s argument.
Hydro and nuclear post the highest capacity-weighted completion rates in the LBNL data. They also represent a small fraction of the active queue — on the order of 18 GW combined, against 773 GW of solar and 220 GW of wind. A technology with a handful of national projects, each already filtered by a decade of siting, permitting, and capital cost before it ever reaches a queue, will show a high completion rate almost by construction. That is a measure of the sample-size and not a technology advantage. Natural gas, with a more significant sample size and no comparable pre-filter, is the fairer technology-to-technology comparison, and it completes at a broadly similar rate to hydro and nuclear despite entering the queue in far greater volume.
Solar and wind should not be treated as one category either. Wind projects behave like large infrastructure: bigger footprint, longer permitting and interconnection study timelines, more dependent on new transmission. Solar is smaller, faster to build, and cheaper on a levelized-cost basis, and it is the default new build specifically because federal tax credit structures and utility offtake requirements reduce investor capital risk relative to other resource types. Nearly half of all active solar capacity in the national queue, 49%, is packaged as solar-plus-storage, a hybrid structure built for utility offtake. That packaging is itself a real-flexibility signal under this paper’s test: a pre-structured offtake agreement is closer to signed and collateralized demand than an announced, unsold project. Solar financing and how that financing is structured is together with the LCOE what is driving solar forward faster than other projects.
PJM: What Happens When the Queue Meets an Auction
| Measure | 2026/2027 BRA | 2027/2028 BRA |
|---|---|---|
| Clearing price (capped) | $329.17/MW-day UCAP | $333.44/MW-day UCAP (+1.3%) |
| RTO reliability requirement | 146,105 MW UCAP | 152,400 MW UCAP (+5,249.9 MW, largely Large Loads) |
| Total procured (RPM + FRR) | ~146,244 MW UCAP — met requirement, +139 MW surplus* | 145,777.1 MW UCAP — 6,516.6 MW UCAP short* |
| Reserve margin vs. target (RPM+FRR) | 18.9% procured vs. 19.1% target (–0.2 pts) | 14.4% procured vs. 20% target (–5.6 pts) |
| PJM’s own no-cap/floor simulation† | $388.57/MW-day, uniform across all LDAs | $529.80/MW-day (rest of RTO); $542.83 DOM (only constrained LDA) |
| Simulated reserve margin without cap | not separately published in this report | 15.1% RPM / 15.4% total — still below the 20% target |
| Separate GridLab/Aurora counterfactual‡ | ~$254/MW-day, if 10% of ~107 GW queued renewables/storage had reached commercial operation | n/a |
Table 1. *The 2026/2027 auction cleared its reliability requirement, on the RPM+FRR basis, by roughly 139 MW UCAP — a surplus, not a shortfall. The 2027/2028 auction is the first in which the entire RTO, including Fixed Resource Requirement areas, fell short of its requirement. These are not the same kind of outcome, and the table should not be read as two years of the same trend line. †PJM’s own published no-cap/floor simulation, run identically for both delivery years. ‡A different, unrelated exercise: GridLab/Aurora Energy Research modeled what the (capped) price would have been with more supply reaching commercial operation, holding the cap in place. It answers a different question than PJM’s own simulation, and only one (PJM’s) has a published 2027/2028 counterpart.
Two separate auction cycles tell the same story from opposite directions. For the 2026/2027 delivery year, PJM had roughly 107 GW of land-based renewables and battery storage in its queue, most filed before 2024. A GridLab/Aurora Energy Research analysis found that if just 10% of that queued capacity had reached commercial operation in time, the auction’s clearing price would have come in near $254 per megawatt-day instead of its actual $329.17 — a difference worth an estimated $3.5 billion to ratepayers in a single year. While the capacity was queued it didn’t make commercial operation and thus it was missing from the grid. That is a loss for customers.
The following year’s auction, 2027/2028, cleared at the administrative price cap of $333.44 per megawatt-day. Even with this increase to the price cap, PJM found itself 6,516.6 megawatts below its own reliability requirement, only able to build a 14.4% reserve margin against a 20% target. The reason for the step-change in capacity came from 5,249.9 MW of additional demand from “Large Loads.” PJM ran the counterfactual itself, for both delivery years: without the cap, 2026/2027 would have cleared at $388.57 and 2027/2028 at $529.80, with the DOM zone alone clearing near $542.83 in 2027/2028 — confirming the posted price was never the real price, only the loudest one legally allowed to speak. Even at that higher 2027/2028 figure, cleared capacity would have risen by only about 800 MW to 135,270.7 MW UCAP, and the reserve margin would have increased to 15.1% RPM — short of the 20% target even with the cap removed entirely. Demand response’s accredited capacity value jumped from 69% to 92% in the same cycle, because flexible, curtailable load finally got priced as a reliability resource rather than a courtesy. Why do we care? While the cap protects consumers from super high prices, it blunts the investment signals. Location-based marginal price (LMP) is no longer the definitive measure — assessments need to analyze the broader system’s scarcity values. Resource reliability scarcity capacity needs to bring RTO-wide value not just alleviating supply constraint at a LMP.
A capped price and an uncapped simulation, run by the same market operator in the same report, is the friction economy admitting its own existence.
Colorado: Turning the Test Into a Tariff
In June 2026, Energy Business Strategy set out a practical regulatory standard, built around a single question: should a data center receive access to scarce grid capacity only after it demonstrates that its load growth, self-supply plan, and operating behavior leave existing customers no worse off under credible stress scenarios? The standard is not anti-growth. It is pro-discipline. The same distinction this series keeps drawing between real and fake flexibility, applied to a live proceeding.
The immediate context is concrete. Xcel Energy, Colorado’s largest utility, has a pending Large Load Tariff proceeding before the Colorado PUC (No. 26AL-0137E) and has told the Colorado Sun it may need to spend $22 billion by 2040 to keep up with potential data-center demand, potential being the operative word, since a capital plan built against speculative pipeline is exactly the exposure paper one describes. Separately, NERC issued a Level 3 Essential Action Alert in May 2026 requiring registered entities to address computational-load risks to modeling, instrumentation, and grid protection — a signal that large loads are now a reliability question, not only a billing one. FERC followed in June 2026 with an aggressive targeted action directing all six U.S. grid operators to reform tariffs for data centers and other large energy users.
The graduated test
This standard treats “is this load real” as a ladder, not a binary: announced, requested, studied, contracted, collateralized, permitted, under construction, energized. A project earns grid capacity by climbing the ladder, in phases tied to enforceable milestones — not all at once, based on an announcement. Three of its five tests map directly onto this series’ framework:
- Cost causation. Does the load pay for the generation, transmission, and backup service it causes — deposits, minimum bills, exit fees — or does it socialize those costs across everyone else? A load that pays for its own risk has priced its flexibility. A load that doesn’t has asked the system to assume the best.
- Forecast credibility. Is the load signed, financed, and tied to enforceable milestones, or is it a large speculative megawatt reservation? This is the LBNL 41% finding, converted into a screening question a regulator can actually ask before capital moves, not after.
- Use-it-or-lose-it. Reserved capacity tied to missed milestones should release back to the system. Warehousing a queue position without financing behind it is the purest form of fake flexibility — a claim on scarce infrastructure with no cost to holding it.
This sorting mechanism puts every large-load project into one of three categories: grid-supportive projects that bring supply, pay for backup, and accept curtailment move fast. Grid-neutral but high-risk projects proceed with stronger deposits and staged energization. Grid-extractive projects — those that reserve capacity, rely on unpaid backup, or use bargaining leverage for preferential treatment — are delayed, conditioned, resized, or rejected until the risk is priced.
What This Confirms
The pattern is now visible at three different scales, from three independent sources, converging on the same finding: a national laboratory’s ten-year dataset, a market operator’s own auction results, and a state-level regulatory proceeding. Each shows the same gap between what gets requested and what gets tested, and each shows the cost of that gap being paid by someone who did not create it: customers aka ratepayers, in both PJM’s case and in Colorado’s.
Nameplate MW may be comforting in a press release, but accredited, collateralized, tested capacity is what actually keeps the lights on, and that is what should actually drive the capital plan.
What Would Confirm the System Changed
- Whether PJM’s Non-Capacity-Backed-Load category and similar curtailment-for-speed mechanisms see material uptake, or remain a marginal option next to firm, uncurtailed service.
- Whether Colorado’s Large Load Tariff proceeding (26AL-0137E) adopts cost-causation and forecast-credibility screens, or defaults to socializing large-load costs across the existing customer base.
- Whether the 41% post-agreement withdrawal rate improves as FERC’s 2026 large-load reforms take effect or holds steady because the underlying speculative-filing incentive hasn’t changed.
- Whether solar-plus-storage’s hybrid, pre-packaged offtake structure continues gaining share against standalone solar and wind, a sign the market is pricing real flexibility into project design before regulators require it.
- Whether the next PJM Base Residual Auction narrows the gap between its capped and uncapped simulated prices or widens it as more “Large Loads” enter the reliability requirement faster than accredited capacity can be built.
Each of these is testable against public filings within a year or two, not a decade. That is the point. A framework that only explains the past is commentary. One that specifies what to watch next is a tool.
This is the third in a three-part series. Paper one (“The Friction Economy”) sets out the framework. Paper two applies it to the 2026 Strait of Hormuz crisis.