# The Great Inversion: Cheap Supply, Scarce Connection

### Paper II: The diagnosis

*Companion papers: I. How We Built Institutions for the Wrong Grid (history) and III. Completing the Market (the reforms).*

---

## America's new bottleneck

America has more generating capacity waiting in line than it currently operates. Some 2,061 gigawatts sits in interconnection queues against roughly 1,400 gigawatts in service, much of it financed, engineered, and sited [1]. The typical project waits more than five years. Historically, about 13 percent of that capacity ever connects.

Meanwhile a hyperscale data-center campus that would pay a premium for power today cannot buy it, gas turbine order books stretch toward five years [2], PJM's capacity charges have risen from $2.2 billion to more than $16 billion a year [3][4], and retail electricity prices are climbing fast enough to decide governors' races.

Scarce electrons no longer explain any of this. America increasingly has the ability to produce cheap electricity, but lacks an equally scalable institutional system for connecting and coordinating it. Some of that scarcity is physical and no reform repeals it: transformers, transmission corridors, HVDC equipment, skilled crews, and buildable land are all physically constrained. But in many high-growth regions, institutional access to physical capacity has become a binding constraint alongside the physical constraints themselves: permission to connect, to compete, and to be paid what a resource is worth, where and when it is worth it. The bottleneck moved from the power plant to the institutions, and the institutions have not caught up.

That is the thesis of this series: the monopoly has narrowed. Wires remain a natural monopoly. Generation does not. Storage does not. Demand flexibility does not. Interconnection and grid data should not. Yet the rules governing all of them still assume they do.

One framework organizes everything that follows:

| Level | The idea |
|---|---|
| Big idea | **The Great Inversion**: supply got cheap, connection got scarce |
| Core thesis | The natural monopoly has narrowed |
| Observable symptom | Scarce permission: queues, caps, churning tariffs |
| Policy objective | Complete the market |

## The inversion

Every assumption the grid's institutions were built on has flipped.

| | **1980** | **2026** |
|---|---|---|
| Generation | Scarce, central, utility-built | Much cheaper in many technologies; increasingly built by non-utility developers |
| Storage | Impossible | Commercially viable and increasingly system-integrated |
| Demand | Passive, predictable | Active, growing, self-supplying, flexible |
| Transmission capacity | Ample | Scarce |
| Connection to the grid | Trivial | The binding constraint |
| Who builds | The utility | A queue of thousands of independent developers |
| Institutions | Matched the technology | Changed unevenly; many legacy incentives remain |

Utility-scale solar's levelized cost fell 84 percent between 2009 and 2026, from $359/MWh to a $40-98 range, and battery packs fell further still, 93 percent since 2010 to $108/kWh, with stationary-storage packs at $70 [5]; solar became the first generation technology in a century that pencils at household scale. Batteries reduce one of the industry's historic constraints, the need to balance supply and demand continuously, by allowing energy to be shifted across time. They do not eliminate the need for firm capacity, transmission, or distribution infrastructure. An electron bought at noon now sells at 7 p.m. Demand returned after two decades of flat load, led by data centers whose binding variable is time to power. When the conventional answer, new gas turbines, quotes a five-year delivery date [2], the fastest source of new capacity is increasingly the resource that can be deployed within the existing network’s available headroom, and the second fastest is whatever is already built and waiting in the queue.

Cheap supply, scarce connection. Hold that inversion in mind; every failure below is a symptom of institutions still organized around the old column.

### The state of the grid

Each row moves the same direction: the technology and capital got cheaper and more abundant while connection, institutional friction, and customer costs got worse.

| Metric | Then | Now (2026) | Source |
|---|---|---|---|
| Utility-scale solar LCOE | $359/MWh (2009) | $40-98/MWh unsubsidized, avg $69 | Lazard 2026 [5] |
| Battery pack price | ≈$1,475/kWh (2010, real 2025$) | $108/kWh; stationary packs $70 | BloombergNEF 2025 [5] |
| Active interconnection queue | A fraction of the fleet (mid-2000s) | 2,061 GW, against ~1,400 GW in service | LBNL [1] |
| Typical queue duration | Under 2 years (projects built mid-2000s) | ≈5 years | LBNL [1] |
| Historical queue completion | Majority connected (mid-2000s cohorts) | 13% of capacity; 75% withdrawn | LBNL [1] |
| Gas turbine lead time | ≈2 years (2019) | ≈5 years | S&P Global [2] |
| PJM annual capacity cost | $2.2B (2024/25 delivery year) | $16.4B (2028/29) | PJM [3][10] |
| Retail electricity prices | Tracked inflation (2013-2023) | Outpacing inflation since 2022; nominal residential +33% since 2019 | EIA; LBNL |

## The stakes are larger than the grid

For households, the inversion arrives as a paradox. American electricity prices have outpaced inflation in four periods: the early 1970s, driven by the oil shocks; the early 1980s, driven by nuclear cost overruns and utility bond yields near 17.5 percent; the mid-2000s, driven by natural gas; and now. In each of the first three the input got more expensive and the bill followed. This time the inputs got cheaper, with utility-scale solar down 84 percent and battery packs down 93 percent, which is what makes the current episode different in kind rather than in degree. The cheapest electricity ever generated is showing up as steadily rising bills, with nominal residential rates up 33 percent since 2019 and electric and gas utilities requesting a record $31 billion of rate increases in 2025, double the prior year. Cheap supply and expensive bills coexist because the institutions stand between them, and every failure documented below lands, through one line item or another, on an electric bill.

Electricity has quietly become industrial policy. AI training capacity is gated on powered land. Semiconductor fabs and reshored factories site wherever interconnection is fastest, which increasingly means abroad or behind a private fence. The Department of Energy has invoked emergency authority to keep aging plants running [4]. A country's ability to connect new supply and new demand to its grid now shapes its competitiveness in the same way its ports and highways once did.

Suppose a hyperscaler needs 500 MW. Route one, new gas turbines: order placed in 2026 against a roughly five-year manufacturing backlog [2], then construction; first power around 2031. Route two, solar plus storage: the equipment is purchasable today and the construction schedule runs 12 to 18 months, but a study-first interconnection queue adds more than five years, and more than eight in PJM [1][21]; first power around 2031 or later. Route three, the same solar-plus-storage project under Texas-style connect-and-manage timelines, which run two to three and a half years: first power around 2028 or 2029. Route four, a flexibility agreement that fits the load into headroom on the existing system, the structure Google signed in 2025 and has since scaled toward a contracted gigawatt [11]: service in months. The engineering delay is 12 to 18 months. Everything beyond that is institutional, which means the schedule for American reindustrialization is currently set by procedure rather than by physics. The question this series answers is why a nation that got very good at building cheap generation got very bad at plugging it in.

## Four alternative explanations fall short

Four candidate explanations for alternative diagnoses are not supported by the evidence.

**Generation is getting more expensive.** Partly true for firm capacity: heavy-duty gas turbine lead times have stretched toward five years, and GE Vernova's backlog grew from 83 GW to 100 GW in a single quarter [2]. The unfirmed side moved the other way: utility-scale solar's levelized cost fell 84 percent since 2009, battery packs 93 percent since 2010, and solar paired with storage cleared at $87 per MWh unsubsidized against $144 to $276 for a gas peaker [5]. A cost story that runs through generation must also explain why bills rose fastest on the delivery lines, where generation does not appear.

**The wires are starved of capital.** The spending record says otherwise. Transmission investment nearly tripled between 2003 and 2023, distribution rose about 160 percent and is now the largest capital category at roughly $66.5 billion a year, and congestion stayed high through all of it [26][27]. Capacity stayed scarce while spending set records, which moves the question from how much money to what the money buys.

**Load growth did it.** The timing fails. PJM's application-to-operation run stretched from under two years in 2008 to more than eight by 2025, and the queue crossed the size of the operating fleet before the AI buildout reached the meter [21]. New campuses raise the stakes; they arrived at the tail of the backlog, and 2,061 GW of willing supply is standing in it [1].

**Permitting and equipment are the bottleneck.** Transformers, corridors, and crews are scarce, and no rule repeals that. The control is Texas: drawing on the same national equipment market and the same labor pool, ERCOT brought 14.2 GW of solar and storage online in 2021 and 2022 against PJM's 5.6 [6]. When the same machines move through different institutions at almost triple the speed, the machines are not the variable.

**The institutions allocate badly.** This is the residual the other four leave standing, and the only account that explains three facts at once: identical equipment connecting at different speeds across a border, record network spending beside persistent congestion, and a queue that predates the load it is blamed on. The first four pressures are real, and every one of them gets more expensive to answer under rules that ration access and pay for capital. The five failures below make that claim specific.

## Five institutional failures

These failures are functional. Each failure is an old rule still operating: a design choice that was correct when written and is now the constraint. They are ordered as a mechanism: the payment formula is the motive, the gatekeeper role is the shield that protects it, and the last three are reinforced by the shielded motive. Each failure falls to the principle in Paper III that shares its number. A system can fail at access through a slow interconnection queue, through a capital-biased distribution planner, or through an opaque tariff, and the failure is the same whichever institution produces it. That is why the reforms in Paper III are organized as principles with a menu of mechanisms under each: the function is what must be fixed and the institutional mechanisms are the way the fix gets implemented.

The dysfunction is specific, and it is fair to say first what the incumbents are coping with. Utilities face wildfire liability, storm hardening, cybersecurity mandates, an aging workforce, and assets past design life, all at once, and much of their record capital spending addresses real needs. RTO planners absorbed the fastest demand-forecast revisions in fifty years. Commissions process dockets with a fraction of the staff of the companies before them. And every institution described below was rationally designed for the engineering and regulatory conditions of its era. The problem is persistence after the underlying assumptions changed. Much of the persistence is itself rational, because these organizations were optimized above all to avoid catastrophic risk, and risk-avoidant systems change last. Utility executives are optimizing correctly against the incentives regulators built over decades, and would optimize differently against different ones. The critique that follows is aimed at the rules these actors operate under, and it holds even where every individual actor behaves reasonably. That is what makes it an institutional diagnosis: the system produces these outcomes when everyone follows their incentives.

### Failure 1: Utilities earn more when they spend more

The insight is sixty years old and comes from regulatory economics. Averch and Johnson showed in 1962 that a firm earning a regulated return on invested capital has a structural incentive to favor capital when the allowed return is tied primarily to rate base [12]. Watch the incentive operate at a single decision. A distribution planner in 2026 faces a feeder that will overload within three years, and four options sit on the desk:

| Option | Ratepayer cost | Utility earnings today | Utility earnings under totex |
|---|---|---|---|
| New substation | ≈$40M capital | ≈10% return for ~40 years | Same allowance as any option |
| Reconductor / upgrade in place | A fraction of greenfield | Return on the smaller amount | Keeps most of the savings margin |
| Battery contract (third party) | ≈$2M/year | $0 | Shares the savings |
| DER aggregation / flexibility | Lowest, most uncertain | $0, plus lost throughput revenue | Shares the savings |

The planner is honest, the engineering analysis is competent, and the substation still gets proposed, because the institution's payment formula biased the outcome before the analysis began. The rightmost column is Paper III's Principle 1, and it is the entire point: change what the same planner's employer earns, and the same desk produces different proposals. At transmission voltage the identical table applies. From the earnings perspective, a reconductoring project that doubles a line's capacity for a fifth of the cost of a new line appears as four-fifths of an asset that never got built [8].

**The spending is visible in the functionalized data, and so is what it did not buy.** Utility costs are reported by function on FERC Form 1 and aggregated in the Energy Information Administration's series, which means this argument can be checked. Between 2003 and 2023, distribution spending rose about 160 percent and transmission spending nearly tripled [26]. Investor-owned capital spending reached roughly $208 billion in 2025, up from about $140 billion in 2020, with distribution the single largest category at about $66.5 billion, generation at $62.4 billion, and transmission at $37 billion [27]. Over the same two decades in which distribution and transmission spending grew at those rates, congestion stayed high and interconnection got slower, which is the pattern a system that pays for capital, not outcomes, would be expected to produce.

**And bills are rising for more than one reason.** Eight forces move an electric bill, and all eight are now pushing upward, generation included as firm-capacity costs climb. This program directly targets three of them. Network spending for growth and capacity is rising sharply, and Principles 1, 3, and 4 aim directly at it. Capacity and reliability charges are rising sharply, and Principle 5 reassigns them to the loads that cause them. The cost of delay, study, and withdrawal is rising, and Principles 2 and 4 remove it. The remaining five are largely untouched. The cost of generating power is set by the market and needs no docket. Wildfire mitigation, storm hardening, and the replacement of assets past design life are all rising, and total-expenditure regulation funds them in full, which is the point of separating hazard spending from discretionary build. The cost of financing that spending and the scarcity of equipment and skilled labor are both rising, and reform touches them only indirectly, by deploying less capital and reusing more of what exists. The three forces of focus here also compound: capital earns a return for forty years, capacity charges recur annually, and delay multiplies both.

Two qualifications. Generation capital is rising quickly again on load growth and now stands at roughly 30 percent of the total, its highest share in more than a decade, so delivery is not the whole story of the last two years even though it is the whole story of the last two decades. And some of the delivery increase answers real needs, since storm hardening, wildfire mitigation, and the replacement of assets past design life all land in that line. What the functionalized data cannot explain away is the ranking. Distribution is the largest capital category in the industry and the least examined of the three. Compared with transmission it has far less federal economic oversight, no comparable nationwide regional-planning framework, and highly uneven competitive-procurement requirements, while generation in restructured regions faces market discipline. It is simultaneously the biggest place utilities spend and the least examined.

The return on capital is an important structural reason, but not the only one. Prudence review rewards tangible assets over contracted services. Operating expense carries disallowance risk that capital does not. Capital produces predictable earnings that analysts model easily, while a portfolio of flexibility contracts does not. A substation is easier to defend at a public meeting than a procurement strategy. And a utility that under-builds and then fails carries a liability that a utility that over-builds never does. A framework that pays for outcomes has to displace all of that, not merely the return.

This is a claim about incentives, and the incentives were written by commissions and legislatures, so responsibility runs through them. The evidence that the incentive dominates behavior is the pattern itself: enthusiasm for self-built infrastructure of every kind, resistance to third-party resources that compete with rate base, and slow-walked transparency on hosting capacity and interconnection automation. Where the incentive flips, behavior flips with it; Paper III presents the British and Hawaiian evidence.

The cheaper option works when someone is paid to choose it. Consolidated Edison's Brooklyn-Queens Demand Management program avoided or deferred a planned infrastructure investment whose original value was roughly $1 billion, using approximately $200 million of demand-side and distributed resources (52 MW of demand reduction plus 17 MW of distributed investment), and it happened because New York's regulators built a specific earnings mechanism to reward the deferral [18]. The accounting has been contested, and fairly: critics calculated a full ten-year deferral cost closer to $855 million once traditional spending inside the program is counted, so the honest claim is that the alternative was materially cheaper, not that a billion dollars vanished [18]. Arizona Public Service installed a 2 MW, 8 MWh battery at Punkin Center rather than rebuilding roughly 20 miles of rural distribution line, deferring the rebuild by three to six years at about half the cost. The system was later placed on standby and disconnected in 2019 after an unrelated battery-fire incident elsewhere on the system, which makes it an honest illustration of both deferral economics and early-storage operational risk. Sterling, Massachusetts, whose municipal utility still operates its 2 MW battery, reports peak-charge savings of roughly $400,000 a year, an estimate analyzed by Sandia National Laboratories [18]. The point of these cases cuts both ways. Non-wires alternatives are proven, and each one required either an unusual regulatory carve-out or a utility outside the investor-owned incentive structure. Under standard cost-of-service ratemaking, BQDM-style outcomes are exceptions someone had to engineer, which is the diagnosis restated as evidence.

The political economy compounds the ratemaking. The regulated firm holds the models, the data, the recoverable legal budget, and the career pathways; the public is represented by intervenors on shoestring budgets. Stigler formalized the pattern in 1971 [13], and Ohio demonstrated the extreme case in practice, where a roughly $60 million scheme bought a legislative bailout and sent the speaker of the Ohio House to federal prison [14]. Capture that blunt is rare. The everyday version, an information asymmetry that no one needs to corrupt, is universal.

### Failure 2: The competition administers the rules

The rule is structural: for a battery, a solar array, or a flexibility aggregator seeking to connect, the incumbent utility is both the competition and the administrator of access. It runs the queue, sets the study assumptions, holds the hosting-capacity and load data every access decision rests on, and earns a regulated return on the capital alternative to whatever is applying. The referee has a position in the game. Nothing here requires bad faith, which is what makes it a rule rather than a scandal: an honest engineer applying conservative assumptions to data only her employer can see, on a timeline only her employer controls, produces the same outcome as an obstructive one.

The country has run this experiment once and recorded the result. In 1996, FERC found that utilities administering access to their own transmission systems discriminated in favor of their own generation, and Order 888 answered with open-access tariffs, functional separation of the transmission function, and a public system for posting available capacity [25]. The finding was not that utility planners were dishonest; it was that self-administered access by an interested party could not be made neutral by conduct rules alone. That finding has never been extended to distribution, where the same structure persists: hosting-capacity data published late, partially, or only under commission order; study queues without binding clocks; and interconnection cost estimates that can move after a project has committed capital, produced by the party that benefits when the answer is a network upgrade. Where regulators have acted, the remedy is always the same: order the incumbent to publish what it knows.

This failure is the enforcement layer for the other four. Serial permission (Failure 4) persists because the gatekeeper administers it. The capital bias (Failure 1) persists because the party choosing between wires and services owns the analysis. Fragile compensation (Failure 3) persists because the data that would justify durable terms is held by the counterparty. A reform program that changed the other four rules and left this one standing would be re-litigated inside the gatekeeper's own processes, one docket at a time.

### Failure 3: Distributed resources are compensated by tariffs that can be rewritten, not by contracts that can be financed

Distributed and community solar, behind-the-meter batteries, and flexible demand interconnect in months and sit where distribution constraints bind. Their compensation is a patchwork of net-metering regimes and successor tariffs redesigned every few years, often retroactively, usually as the outcome of a political fight between the utility and the solar industry; California alone has cycled through three regimes. FERC's Order 2222, which was supposed to open wholesale markets to aggregated distributed resources in 2020, has compliance timelines stretching toward the 2030s (ISO-NE late 2026, PJM 2028, MISO 2029, SPP 2030) [15]. Whatever the right compensation level, and reasonable analysts disagree about it, the pattern is self-defeating: the most scalable resources on the grid cannot be financed against rules that change with each rate case.

### Failure 4: Projects must ask permission one at a time

The rule is serial and discretionary: each project is studied one at a time, in the order it arrived, against the network as it stood that day, and connection is granted case by case. Rationing by process means the queue itself, not price and not physics, decides what gets built and when. That was workable at a dozen large plants a year; it is the binding constraint at two thousand gigawatts of applicants.

**What FERC Order 2023 changed.** Order 2023 replaced much of the legacy serial process with a first-ready, first-served cluster framework and stronger readiness requirements. This paper does not assume that nothing has changed. It argues that study reform alone does not solve network expansion, cost allocation, operating limits, and distribution-level access.

Queue length is not itself evidence of failure. A queue can be long because entry is cheap, a large share of it always was speculative, and changing interconnection costs can make an otherwise viable project uneconomic. What indicates a process that no longer scales is the combination: withdrawal rates above 70 percent, median durations beyond five years, restudies triggered by every departure, and completion rates that have fallen to 13 percent of capacity. Follow a utility-scale solar project through the median experience in Lawrence Berkeley National Laboratory's data [1]. The developer files an interconnection request and enters a study queue behind hundreds of others. A feasibility study arrives, then a system impact study, then a restudy when a neighboring project drops out and changes the math, then a cluster restudy under reformed procedures, then a network upgrade cost allocation that may exceed the cost of the project itself. Five years pass between request and decision; in PJM specifically, the average application-to-operation timeline stretched from under two years in 2008 to more than eight by 2025 [21]. Four in five projects exit before connecting, and every exit triggers restudies for everyone behind them.

Now run the same project in Texas. ERCOT studies only local reliability effects, connects the project in a fraction of the time, and manages congestion afterward through the energy market, with the developer bearing curtailment risk. The approach is called connect and manage, and it is the main reason ERCOT has added new generation and storage faster than any other U.S. system this decade, bringing 14.2 GW online in 2021 and 2022 against PJM's 5.6 GW despite PJM being more than twice its size [6]. Faster is not immune: ERCOT's own median storage timeline has drifted toward four years as its queue swelled, which argues for pairing speed with proactive planning. Similar physics and technologies, materially different interconnection rules and market arrangements. Texas is therefore a useful real-world comparison for interconnection speed, but not a complete control group for national market design.

The network itself is constrained by the same institutions. Interregional transfer capacity stagnated for two decades while cost-allocation disputes consumed planning cycles; FERC's Order 1920 (2024) had to compel the long-term planning that regional tariffs should have produced [7]. The cheapest expansions go unbuilt: analysis published in PNAS finds that reconductoring existing lines with advanced conductors can roughly double a corridor's capacity at a fraction of greenfield cost, with no new permitting, though results vary with conductor choice, thermal limits, and structural condition, and not every corridor qualifies [8]. Why utilities rarely propose it anyway is Failure 1.

### Failure 5: Reliability is bought as a commodity, not financed as an asset

The rule treats reliability as a commodity to be bought a year or a season at a time, while the thing being bought, a power plant, is a thirty-year asset that must be financed. A short-dated purchase cannot carry long-dated capital, so scarcity shows up as price spikes rather than as new steel. This failure also indicts an institution the restructuring era itself built; the diagnosis applies the same functional test to a 1990s market that it applies to a 1920s payment formula.

PJM's capacity auction, which is supposed to secure supply for 67 million people, cleared at $28.92/MW-day for 2024/25 and has since hit administrative price caps in four consecutive auctions. The pattern is not one market’s quirk: the institutions are shortening visibility instead of reconsidering it, with ISO New England moving to prompt seasonal auctions and New York procuring monthly, while Texas runs no capacity construct at all. In PJM: zonal caps in 2025/26 (BGE at $466.35, Dominion at $444.26, the RTO at $269.92) and the RTO-wide cap in each of the next three, at $329.17, $333.44, and $325 [3][4][9][10]. The December 2025 auction fell 6,623 MW short of the reliability requirement, the first in which the entire RTO including FRR areas fell short, and PJM's own simulation put the uncapped clearing price at $529.80 [9]; the July 2026 auction attracted about 525 MW of new resources [10]. Annual costs rose from $2.2 billion to $16.4 billion over the period. When an auction clears at a ceiling negotiated with a governor, while attracting almost no entry, it has stopped discovering prices and started administering them.

The failure compounded across choices: auctions delayed for years, severing price signals from investment timelines; accreditation reforms that repriced the fleet overnight; data-center demand underestimated for a decade, then repriced onto all customers at once; and a queue that blocked supply from answering the very signal the auction sent. The data centers make an awkward villain here. Most would rather contract directly for their own capacity, and some are trying: Google's 2025 agreements with Indiana Michigan Power and the Tennessee Valley Authority curtail machine-learning workloads during system peaks [11]. The institutions offered no standard product, so their demand landed in the auction as an undifferentiated forecast, and every household in thirteen states got the bill.

The evidence does not establish that the capacity market is the sole cause of inadequate entry. The argument is narrower: repeated price caps, high clearing prices, and limited new entry indicate that the existing mechanism is not reliably converting scarcity into a predictable investment signal.

### The absorption record

Five times since 1978, federal regulators changed the rules of this industry, and five times the industry metabolized the change. PURPA created independent generators and was fought to a draw over avoided cost. Order 888 opened the transmission system and stopped at the substation [25]. Order 1000 invited competition into transmission planning and was answered in state legislatures with rights of first refusal. Order 2222 ordered the wholesale door open to distributed resources; the compliance calendar now stretches toward 2030. Order 2023 reformed the queue itself, and its implementation raised the cost of standing in line, thinning the independent pipelines it was meant to clear. The pattern is not that nothing changed. It is that process changed and incentives did not, so every process change was administered by parties paid to blunt it. This is what Failures 1 and 2 predict, run five times. Why would we do the same thing again, expecting a different outcome?

**The counterfactual.** Suppose the settlement had continued past the substation in 2001: performance pay on the wires, a neutral administrator at the gate, access under published rules. Two decades of distribution capital would have faced a buy-or-build test before earning a return. The queue would have been engineered for thousands of applicants before two thousand gigawatts arrived to stand in it. Storage would have come to market against posted terms it could finance. The 2020s load wave would have landed on a functioning connection market, and the data centers would be customers in it, not defendants in a blame contest. The counterfactual proves timing, not magnitude; it cannot say what bills would be. What it can say is that every institution this program proposes had a working precedent by 2001, so the cost of the last two decades was a choice about incentives, not a limit of technology.

## The pattern underneath: institutions became the expensive input

Step back from the five failures and one economic shift explains them all. For a century, electricity was expensive to produce and cheap to coordinate: the commodity dominated the bill, and the institutional overhead of one utility, one commission, and one annual rate case was rounding error. The cost curves inverted that relationship. Producing a marginal megawatt-hour has never been cheaper. While cheap generation is not the same thing as cheap delivered and reliable electricity, the claim throughout this series is that the gap between the two is now more institutional than physical. The coordination costs, five years of studies, decade-long compliance dockets, serial rate cases, and stakeholder processes with hundreds of parties, can rival the economic value of the production cost differences they are supposed to manage; a project can spend more calendar time and risk capital on permission than on construction. Economists since Coase have taught that institutions exist to economize on transaction costs, and that when relative prices shift this far, the efficient institutional boundary moves [20]. The grid's institutional boundary has moved far less than the economic boundary since 1935. The five failures above are what it looks like when transaction-cost-heavy institutions govern a transaction-cost-light technology, and the reforms in Paper III are, in this framing, boundary corrections: rules and prices replacing case-by-case discretion wherever discretion's coordination cost now exceeds its value.

## The strongest objections, answered directly

A thesis this critical of incumbent institutions earns skeptical readers, and the four objections they raise most often deserve full answers.

**"The queue is mostly speculative."** Largely true, and it strengthens the diagnosis. Developers file multiple queue positions because a position has been a cheap option on scarce, uncertain grid access; when the study process is slow and its cost allocations are unpredictable, buying several lottery tickets is rational. The speculation is the institution's product. FERC's Order 2023 already raised deposits and readiness requirements, and the headline number has begun shrinking without any easing of the underlying scarcity. And the argument survives heavy discounting: even at the historical completion rate of about 13 percent, the current queue implies on the order of 270 GW of eventually built capacity [1], several years of additions at recent build rates, waiting on procedure. One further caution belongs on the record rather than in a critic's hands: 549 GW already holds a draft or executed interconnection agreement, which is the furthest institutional milestone a project reaches before construction, but withdrawal rates stay above 40 percent even after that stage, so an agreement is a cleared gate rather than a commitment to build [1]. Texas holds queue volumes that are large too; the difference is how fast positions convert to steel.

**"Connect-and-manage threatens reliability."** The record says otherwise, and the mechanism explains why. Interconnection studies protect against local overloads, which connect-and-manage handles by studying local effects and curtailing flows in operations; the developer, holding the curtailment risk, self-selects better sites. ERCOT has operated this way through nearly two decades of the country's fastest resource growth, Britain adopted connect-and-manage transmission access in 2010, and Australian networks connect rooftop fleets under real-time operating envelopes [6][16]. ERCOT's genuine reliability failure, Winter Storm Uri in 2021, traced to plant weatherization and fuel supply in the FERC-NERC inquiry, failures a five-year interconnection study does nothing to prevent [19]. The reliability question is real; the study queue is the wrong instrument for it, and Principle 5 in Paper III addresses the right one.

**"Utilities need real capital investment."** Agreed, and the reform is capital-neutral. Storm hardening, wildfire mitigation, cyber defenses, and replacement of assets past design life are exactly the spending a totex framework funds, because totex removes the accounting preference rather than the money. What changes is the choice at the margin the current formula distorts: greenfield versus reconductoring, substation versus battery contract, build versus buy. The British networks invested heavily under RIIO; they were paid for delivering outputs with that investment instead of for the investment itself; Paper III presents the record.

**"Why isn't the answer just public power?"** Municipal utilities do charge less. In 2024 the average residential bill was $123.78 at public power utilities against $139.42 at investor-owned utilities [22]. Cooperatives, which are also not-for-profit and also earn no return on rate base, averaged $149.18, mainly because they serve low-density rural territory. Cost follows structure and circumstance rather than ownership form. The municipal advantage itself comes largely from tax-exempt debt, exemption from income and property taxation, and statutory preference on federal hydropower at cost-based rates, which are transfers rather than efficiencies and do not scale, because municipalizing additional utilities creates no additional federal dams [22]. Ownership also leaves the binding constraint untouched, since a municipal utility inside PJM waits in the same interconnection queue and pays the same capacity charges, and on permission it is frequently worse: municipal utilities and cooperatives are exempt from state interconnection and net-metering rules in many states, so a restrictive distributed-energy policy faces no commission and no appeal [23]. The transition arithmetic compounds it, with roughly a dozen communities having municipalized in two decades, most attempts abandoned, and Boulder ending a ten-year, $28.7 million effort in 2020 without acquiring a system [24].

The deeper answer is that public power is evidence for this diagnosis. A utility that is not paid a return on invested capital does not exhibit the capital bias, which is exactly what Failure 1 claims, and it is why the non-wires examples in the case evidence above come disproportionately from carve-outs and public power. Sterling, Massachusetts deferred with a battery what an investor-owned utility needed a bespoke earnings mechanism to attempt. Ownership change is one way to escape the payment formula. Changing the payment formula is the general case: it applies to the three-quarters of load that investor-owned utilities serve, and public power can adopt it faster than anyone else, by board action and without a rate case.

**"Distributed and variable resources aren't dispatchable like a gas plant."** Correct as stated. Capacity markets already discount every resource to its effective load-carrying capability, so a megawatt of solar counts as a fraction of a megawatt where the data say it should. The discounts move with evidence: PJM's accreditation of demand response rose from 69 percent to 92 percent between consecutive auctions once availability rules tightened and winter performance was counted properly [9]. Grid-scale batteries are among the most dispatchable assets on any system, responding in under a second, and portfolio diversity does the rest. The dispatchability objection is an argument for accurate accreditation inside the reforms, which Principle 5's obligation accounting provides.

## What would prove us wrong

The claim in this paper is causal, which makes it testable. Six results would force a revision.

**If queue durations fall substantially without interconnection reform**, the process was never the binding constraint, and the backlog was a transient response to a demand shock and never an institutional failure.

**If distribution capital spending slows while cost-of-service ratemaking stays in place**, the capital bias is not what drives the spending, and the diagnosis in Failure 1 is wrong.

**If states adopting total-expenditure regulation show no divergence in delivery-cost growth from comparable states**, the mechanism does not work in an American context, whatever the British record shows.

**If system costs fall but the savings arrive by shifting cost onto inflexible or low-income customers**, the program has failed its distributional test, even with aggregate costs falling.

**If large loads decline flexibility terms even when offered service years earlier**, the central trade in the reform program is not one they want, and the flexible-headroom estimates are academic.

**If ERCOT's speed advantage disappears while it retains connect-and-manage**, the rule was not the cause of the difference, and Texas stops being evidence for anything here.

## What we do not know

No one has operated a full distribution-level market at scale, and the engineering of grids dominated by inverter-based resources remains an active frontier, with grid-forming standards still maturing; South Australia's operating record and ERCOT's storage integration are encouraging but not conclusive [16]. Estimates of flexible-load headroom depend on assumptions about curtailment tolerance that hyperscalers have only begun to test in production [17]. Institutional overhauls have failure modes too, and California's 2000-01 crisis shows what a badly sequenced restructuring costs. The reform program in Paper III is sequenced with that history in view.

Operating a reliable power system remains difficult, and nothing here suggests otherwise. What changed is the ranking: institutional friction has become the longest pole in the tent, the delay and cost that binds before the engineering does, and every year of Australian and Texan operating experience lengthens the gap.

What is settled: the queue numbers, the auction results, the capital bias, and the direction of the inversion. Cheap supply is not returning to scarcity; even after levelized costs rose roughly 10 percent in Lazard's 2026 series, on higher capital costs, interest rates, and import tariffs, unsubsidized solar and wind remain the cheapest new-build generation [5]. The central open policy question is how much of today's connection scarcity is physical, how much is institutional, and how quickly institutional reform can reduce the latter without compromising the former.

*Paper III states the five principles, evaluates the mechanisms under each with a stance and an evidence tier, names who wins and who loses, and lays out a 2027-2035 implementation path. The companion document What Success Looks Like describes ordinary operation in 2035, once the reforms hold.*

---

## Sources

[1] Joseph Rand et al., *Queued Up: 2026 Edition*, Lawrence Berkeley National Laboratory (June 2026; 2,061 GW active at year-end 2025, ~8,200 projects, median interconnection-request-to-operation duration above five years; emp.lbl.gov/queues).

[2] S&P Global Commodity Insights and trade press reporting (2024-26) on gas turbine order backlogs at GE Vernova, Siemens Energy, and Mitsubishi Power.

[3] Utility Dive, "PJM capacity prices set another record with 22% jump" (July 2025); PJM Independent Market Monitor analyses.

[4] PJM Inside Lines, "PJM Auction Procures 134,311 MW of Generation Resources" (July 22, 2025); Utility Dive coverage of DOE emergency orders (2025-26).

[5] Lazard, *Levelized Cost of Energy+*, 2026 edition (utility-scale solar $40-98/MWh, average $69, against $359 in 2009, an 84 percent decline); BloombergNEF, *2025 Lithium-Ion Battery Price Survey*, December 2025 ($108/kWh average, 93 percent below 2010's ≈$1,474/kWh in real 2025 dollars; stationary-storage packs $70/kWh).

[6] Tyler H. Norris, "Beyond FERC Order 2023: Considerations on Deep Interconnection Principle" (Nicholas Institute, Duke University, 2023).

[7] FERC Order No. 1920, "Building for the Future Through Electric Regional Transmission Planning and Cost Allocation" (2024).

[8] Emilia Chojkiewicz et al., "Accelerating transmission expansion by using advanced conductors in existing line corridors," *PNAS* (2024).

[9] PJM, "2027/2028 Base Residual Auction Report" (December 2025).

[10] PJM, "PJM Capacity Auction Procures 138,318 MW of Generation Resources" (July 14, 2026).

[11] Google announcements of demand-response agreements with Indiana Michigan Power and TVA (2025); Utility Dive and Canary Media coverage.

[12] Harvey Averch and Leland L. Johnson, "Behavior of the Firm Under Regulatory Constraint," *American Economic Review* 52:5 (1962).

[13] George J. Stigler, "The Theory of Economic Regulation," *Bell Journal of Economics and Management Science* 2:1 (1971).

[14] ProPublica and Ohio Capital Journal reporting on the FirstEnergy/HB6 scandal and the conviction of Speaker Larry Householder (2020-24).

[15] FERC Order No. 2222 (2020) and subsequent RTO compliance dockets, with accepted or proposed implementation dates of November 2026 (ISO-NE), February 2028 (PJM), June 2029 (MISO), and Q2 2030 (SPP); CAISO implemented November 2024 and NYISO April 2024.

[16] ARENA and AEMO, Project EDGE and Project Symphony reports; NERC guidance on inverter-based resources.

[17] Tyler H. Norris et al., *Rethinking Load Growth* (Nicholas Institute, Duke University, 2025).

[18] Consolidated Edison, Brooklyn-Queens Demand Management filings and NY PSC orders (Case 14-E-0302), with the program's net savings contested in contemporaneous analysis (Greentech Media, 2017); Arizona Public Service, Punkin Center battery deployment (2018) and subsequent NAATBatt presentation disclosing the 2019 standby and disconnection; Sandia National Laboratories (2017) economic analysis of the Sterling Municipal Light Department system, with Clean Energy Group project documentation.

[19] FERC, NERC, and Regional Entity joint inquiry, *The February 2021 Cold Weather Outages in Texas and the South Central United States* (2021).

[20] Ronald H. Coase, "The Nature of the Firm," *Economica* 4:16 (1937), pp. 386-405; Oliver E. Williamson, *The Economic Institutions of Capitalism* (Free Press, 1985).

[21] RMI, "PJM's Speed to Power Problem and How to Fix It" (May 2026), documenting PJM's average application-to-operation timeline rising from under two years in 2008 to more than eight years in 2025.

[22] American Public Power Association, 2024 average monthly residential bills by ownership type ($123.78 public power, $139.42 investor-owned, $149.18 cooperative), reported in trade coverage of APPA's affordability analysis; APPA materials on tax-exempt financing and federal hydropower preference; Cato Institute analysis of the tax and financing advantages of public power and the Power Marketing Administrations.

[23] U.S. EPA, *Energy and Environment Guide to Action*, chapter on interconnection and net metering standards, noting that municipal and cooperative utilities are exempt from state rules in some states; DSIRE program records for Florida and Pennsylvania.

[24] Brattle Group, *Electric Utility Municipalization: Key Statistics and Risk Considerations* (2025); Colorado Public Radio and Energy News Network reporting on Boulder's ten-year, $28.7 million effort and its conclusion in 2020; American Public Power Association statements on the number of successful municipalizations over two decades.

[25] FERC, Order No. 888, Promoting Wholesale Competition Through Open Access Non-discriminatory Transmission Services, 1996.

[26] Clean Air Task Force, "A data-driven look at rising U.S. electricity costs and policy solutions" (March 2026), reporting distribution spending growth of about 160 percent and transmission growth of nearly threefold between 2003 and 2023, and attributing part of the pattern to planning processes that reward pre-emptive local capital spending over larger regional projects.

[27] Edison Electric Institute, capital expenditure data for 2025 (roughly $207.9 billion of investor-owned utility capital spending, against about $139.8 billion in 2020; distribution approximately $66.5 billion, generation $62.4 billion, transmission $37 billion), as reported by POWER and Utility Dive; Lawrence Berkeley National Laboratory, *Disaggregating Future Retail Electricity Rate Growth* (2021), for the FERC Form 1 functionalized decomposition method.