Introduction: The Sidewalk Repossession — An Analogy of Interconnected Defaults

In mid-2022, in the months after regional COVID-19 restrictions lifted across California, a daily walk up a suburban hill offered a recurring fixture of the landscape: a high-end luxury sports car parked, permanently it seemed, in the same curb space. It never moved. Its paint gathered a film of coastal dust; its tires slowly settled into the asphalt. Then one morning, without ceremony, the vehicle was winched onto a flatbed operated by a regional recovery agent. Pressed against the driver-side window was a brightly colored, high-tack legal notice: “SEIZED: Vehicle Repossessed Due to Default on Financial Obligations.” The car had looked like wealth. It had, in fact, been a liability wearing the costume of an asset — a physical object whose right to occupy space depended entirely on an invisible lattice of contracts, payment schedules, and enforcement mechanisms that its owner had allowed to lapse.

[Retail Auto Buyer] ──(Retail Credit Contract)──> [Auto Dealership / Finance Co]

                                                           │

                                              (Default triggers contract right)

                                                           │

                                                           ▼

[Public Right-of-Way] <──(Physical Seizure)── [Licensed Repossession Agent]

Figure 1. The mechanical execution of a contractual default: a multi-party web of operational interconnections.

This mechanical execution of a contractual default reveals a multi-party web of operational interconnections. The original transaction relied on a credit facility extended by a financial institution to a retail buyer. The contract granted the lender a security interest — the legal right to seize the physical asset upon default. The recovery agent acted as a transactional mechanism to enforce a legal timeline. Not one participant in this chain cared about the car as a car: its horsepower, its design lineage, its cultural cachet. Every participant cared about the car as a bundle of enforceable rights whose value survived only so long as its obligations were performed.

This sidewalk repossession offers a precise structural match to the operational realities of the modern AI power grid. A sports car sitting idle on a street, without the underlying capital to maintain its position, is repossessed by the network’s stakeholders. Similarly, in the AI economy, a developer cannot simply sit on raw real estate or an unutilized grid allocation. If a data center developer fails to meet strict readiness milestones, fails to post required security deposits, or defaults on infrastructure payment schedules, the regional transmission organization (RTO) or utility exercises its right to terminate the agreement and strip the developer of its position. The highly coveted megawatt allocation is repossessed by the grid and handed to the next entity waiting in line. This is not a metaphor; it is now written into tariffs. AEP Ohio’s data center tariff, approved by the Public Utilities Commission of Ohio in July 2025, imposes minimum monthly payments on 85 percent of contracted capacity for up to twelve years, backed by collateral requirements and exit penalties equal to three years of minimum charges — a repossession regime for electrons, drafted with the same cold precision as any auto finance contract [32, 33, 34].

The deeper lesson of the anecdote, and the organizing thesis of this paper, is that in mature capital systems the physical object is never the asset. The asset is the right — the enforceable, dated, transferable, seizable right. In 2022 that observation applied to a car on a curb. By 2026 it applies to the entire physical substrate of artificial intelligence. A gleaming data center shell filled with next-generation accelerators, but lacking a legally enforceable date of energization, is the sports car with the lapsed payments: an impressive object waiting for its flatbed.

Defining “Interconnection Capitalism”

The transition from traditional industrial structures to what this paper terms Interconnection Capitalism is defined by three structural realities, each of which will be developed, evidenced, and stress-tested across the six sections that follow.

First, rights to energization have become primary capital. Value has shifted decisively from the physical asset — the data center shell, the server rack, even the GPU — to the regulatory right of energization: the executed interconnection agreement, the confirmed study, the firm delivery date. Real estate professionals have coined a term for land that carries such rights: “powered land,” parcels with confirmed grid connections or contracted electricity that command premiums utterly disconnected from their acreage, topography, or zoning, and that institutional investors now underwrite as infrastructure rather than as dirt [39, 41, 42]. The presence or absence of deliverable power can double or halve a parcel’s market value, and in constrained markets the multiple is far larger.

Second, the grid operator has become the market arbiter. RTOs and independent system operators such as PJM Interconnection, ERCOT, MISO, and SPP — administrative bodies designed a generation ago to referee wholesale electricity markets among a modest population of large, centralized power plants — now function as gatekeepers of the single most contested industrial resource of the decade. PJM alone has processed more than 170,000 megawatts of new generation requests since 2023, with roughly 30,000 megawatts still in its transition queue entering 2026 [1]. Its reopened 2026 queue cycle drew applications from 811 projects totaling 220 gigawatts, while its own 2026 Long-Term Load Forecast projects summer peak demand rising from 160 gigawatts in 2025 to 253 gigawatts by 2046 — a 58 percent increase driven primarily by data centers [2]. In ERCOT, an almost surreal 198 gigawatts of large load applied for interconnection in the first quarter of 2026 alone, with 86 gigawatts of new load requests under active review — a volume roughly equal to the entire current peak load of the Texas grid [4]. Institutions built to study transformers are now, functionally, allocating the future of the technology industry.

Third, technical steps have been financialized. Grid studies, transformer production slots, queue positions, surplus interconnection service, and behind-the-meter configurations are bought, sold, hoarded, collateralized, and litigated by corporate entities seeking a competitive moat in the AI ecosystem. Every administrative milestone in the interconnection process has acquired a price, a secondary market, and a risk profile. The megawatt has become the base currency of AI real estate; the queue position has become its option contract; the energization date has become its bond maturity.

The scale of capital now denominated in this currency is difficult to overstate. In their most recent earnings reports through the first quarter of 2026, the four largest hyperscalers — Microsoft, Amazon, Alphabet, and Meta — guided to roughly $700 billion in combined capital expenditures for calendar 2026, nearly double their 2025 outlay [19]. Amazon reported $44.2 billion in quarterly capex as AWS grew 28 percent; Alphabet spent $35.7 billion in the quarter and raised full-year guidance toward $180–190 billion; Microsoft guided to approximately $190 billion for the calendar year; and Meta raised its 2026 envelope to $125–145 billion, citing higher component pricing and additional data center costs [19, 20]. Goldman Sachs now projects a combined $5.3 trillion of capex for the four largest hyperscalers between fiscal 2025 and fiscal 2030, across compute, data centers, and power [21]. Microsoft’s chief financial officer, Amy Hood, offered investors the single most important sentence of the earnings season — a sentence about electricity, not silicon:

“remain constrained at least through 2026.”

— Amy Hood, Chief Financial Officer, Microsoft, on capacity constraints, Q1 2026 earnings commentary [20]

When the most valuable software company in history tells the market that its binding constraint is not demand, not talent, and not capital, but deliverable power, the intellectual case for treating interconnection as the decisive asset class of the era has effectively been conceded by the market itself. What remains is to describe the system — its instruments, its pathologies, its valuation logic, and the regulatory counter-movement it has provoked. That is the work of this paper.

The Macro-Physical Context: How Large Is the Collision?

It is worth establishing, before the analysis narrows to queues and substations, the planetary scale of the collision between computational ambition and electrical reality — because the intensity of Interconnection Capitalism is a direct function of that scale. The International Energy Agency’s landmark Energy and AI report, the most comprehensive global treatment in the 2020–2026 literature, found that data centers consumed approximately 415 terawatt-hours in 2024 — about 1.5 percent of world electricity — after growing roughly 12 percent per year since 2017, more than four times faster than total electricity consumption; the agency’s base case projects that figure more than doubling to around 945 terawatt-hours by 2030, slightly more than the entire present electricity consumption of Japan, with AI-optimized facilities more than quadrupling their draw [17]. The IEA’s updated projections through 2026 hold that trajectory essentially intact — roughly 485 terawatt-hours in 2025 rising to about 950 by 2030 — while adding a caveat that reads as a one-sentence summary of this entire paper: bottlenecks across the value chain, not demand, are what now cap the more aggressive scenarios, despite booming investment and surging project pipelines [18]. Analysts at Brookings note that on some estimates data center consumption could approach 1,050 terawatt-hours as early as 2026, which — were data centers a country — would rank them the world’s fifth-largest electricity consumer, between Japan and Russia [44].

Three features of this demand shock make it uniquely destabilizing for grid institutions, and each maps directly onto a section of this paper. The first is geographic concentration. Unlike the diffuse load growth of electrification, AI demand lands in clusters: the IEA observes that a single AI-focused campus can draw as much power as an aluminum smelter while being far more geographically concentrated, with nearly half of U.S. data center capacity packed into five regional clusters; the United States and China together account for almost 80 percent of projected global growth to 2030 [17]. Concentrated demand means concentrated queues, concentrated substation scarcity, and concentrated ratepayer exposure — the preconditions of Sections 2 and 4. The second feature is temporal mismatch. Hyperscaler capital moves on quarterly cadences; transmission moves on decadal ones. The divergence between the two clocks is what creates the premium on every instrument this paper describes, from queue seniority to transformer slots. The third feature is financial asymmetry. Sequoia Capital’s David Cahn has calculated an annual revenue gap of roughly $600 billion between what hyperscalers spend on AI infrastructure and what the AI ecosystem generates in sales, a gap widening in 2026 as capital expenditure accelerates faster than revenue; Allianz Research measures the divergence between AI capex growth and revenue growth at roughly 46 percent — already exceeding the 32 percent divergence observed during the 2001 telecom cycle [22]. A capital formation of this magnitude, racing ahead of its own revenues into a physically constrained delivery system, does not merely stress the grid. It converts every scarce grid input into a speculative asset — which is precisely the phenomenon the remainder of this paper anatomizes.

Hyperscaler Grid Interconnection and Power Delivery Contracts

Before descending into the mechanics of Interconnection Capitalism, it is worth fixing in view the empirical anchor of the argument: the multi-gigawatt power delivery frameworks, infrastructure reservations, and co-location contracts executed between hyperscale technology firms and power suppliers between 2024 and early 2026. These are not press-release aspirations. They are executed, long-dated, legally enforceable instruments — power purchase agreements running twenty years, asset purchases, restart financings — and each one converts a technical relationship with the grid into a corporate asset with a delivery date. The following matrix consolidates the most consequential known frameworks as of July 2026, drawing on regulatory filings, company disclosures, and trade press reporting [23, 25, 26, 27, 28].

Tech CorporationSupplier / Balancing AuthorityFacility / ProjectPower TypeCapacityFinancial CommitmentLocationDelivery Timeline
MicrosoftConstellation / PJMCrane Clean Energy Center (former Three Mile Island Unit 1)Nuclear restart~0.835 GW20-year PPA; ~$1.6B restart programPennsylvaniaTargeted 2027 restart; PPA through late 2040s [25]
Amazon (AWS)Talen Energy / PJMSusquehanna campus (Cumulus)Nuclear co-location → front-of-meter PPA1.920 GW at full quantity$650M campus purchase + PPA (~$18B est. lifetime revenue to Talen)PennsylvaniaRamp through 2032; contract through 2042 [23]
MetaConstellation / MISOClinton Clean Energy CenterExisting nuclear (relicensing support)1.121 GW20-year PPAIllinoisJune 2027 commencement [28]
MetaVistra, TerraPower, Oklo / PJMMulti-asset nuclear framework (incl. Prometheus supercluster support)Existing nuclear uprates + advanced nuclear / SMRUp to 6.6 GWMulti-billion, undisclosed per-assetOhio, Pennsylvania, multi-sitePhased late 2026 – 2035 [26, 27]
Google (Alphabet)NextEra EnergyDuane Arnold restartNuclear restart~0.6 GWLong-term PPA (undisclosed)IowaAnnounced Oct. 2025; restart targeted ~2029 [26]
Google (Alphabet)Kairos PowerAdvanced reactor fleet frameworkSmall modular reactors (SMR)~0.5 GW fleet targetTechnology-development PPA (undisclosed)Multi-stateInitial deployment ~2030 [26]

Table 1. Executed hyperscaler power frameworks, 2024–Q1 2026. Capacity and dates reflect public disclosures; several timelines remain contingent on regulatory approvals and grid studies.

Two features of this matrix deserve emphasis before we proceed. The first is the sheer temporal asymmetry it encodes: capital committed in 2024–2026 against electrons that will not flow, in some cases, until the mid-2030s. Hyperscalers are not buying power; they are buying calendar positions — enforceable claims on future energization that competitors cannot easily replicate because the underlying assets (operating reactor licenses, restart-eligible plants, credible SMR pipelines) are close to fixed in supply. TerraPower’s chief executive framed Meta’s commitment in exactly these temporal terms:

“we must deploy gigawatts of advanced nuclear energy in the 2030s.”

— Chris Levesque, President and CEO, TerraPower, on the Meta framework agreement [27]

The second feature is jurisdictional: nearly every row of the matrix runs through PJM or its neighbors, which is to say through a single regulatory battlespace. The FERC’s November 2024 rejection of the amended Talen–Amazon interconnection service agreement — a 2-1 ruling that declined to expand behind-the-meter co-located load at Susquehanna from 300 to 480 megawatts — sent independent power producer equities tumbling and forced the entire industry to restructure around front-of-the-meter frameworks [24, 25]. Constellation’s chief executive told investors within days that the ruling was:

“not the final word on colocation.”

— Joseph Dominguez, President and CEO, Constellation Energy, Q3 2024 earnings call [25]

He was correct — but the final word, when it began to arrive in December 2025, came from regulators rather than markets, a development we take up in Section 6. The point, for now, is structural: the largest technology companies on earth have concluded that the scarce input to artificial intelligence is not intelligence at all. It is interconnection.

Section 1: From Land Banking to Megawatt Banking

1.1 The Valuation Collapse of Stranded Acres

Every economic era produces its own characteristic form of patient capital, and for most of the twentieth century, American real estate developers practiced the purest version of it: land banking. The strategy was almost meditative in its simplicity. A developer purchased strategic agricultural or unzoned acreage on the urban fringe — flat, cheap, well-drained, near a future highway alignment — and then simply waited, sometimes for decades, while commercial expansion, population growth, and municipal annexation performed the work of appreciation. The land itself did nothing. Its value derived from surface square footage, topography and soil mechanics, local zoning trajectories, and transport access. Time was the only input, and patience the only skill. The model rested on a silent assumption so universal that no one thought to state it: whenever development finally arrived, the utility would arrive with it. Power was infrastructure, and infrastructure followed demand. No land banker in 1985 underwrote the risk that electricity itself might become the scarce factor.

The AI economy has annihilated that assumption, and with it, the valuation logic of an entire asset class. Unpowered land — however flat, however well-located by every traditional metric — has suffered a structural asset impairment that the market is still learning to price. A 500-acre flat parcel situated directly over cross-continental fiber lines is worth effectively nothing to an AI developer if the local utility quotes an eight-to-ten-year timeline to bring transmission-level voltage to the site, and such quotes are no longer exotic: interconnection timelines in major markets now routinely stretch from three to seven years, constrained by transmission capacity, substation readiness, and queue systems designed for a far smaller world [16, 40]. Conversely, a 20-acre brownfield containing a retired heavy-industrial foundry with a 200-megawatt dual-fed substation commands an immense premium — not despite its contamination history and awkward geometry, but because its industrial past left behind the one inheritance that cannot be conjured on demand: deliverable power. The market has, in effect, repriced the entire American landscape along a single new axis, and the industry has given the favored end of that axis a name. “Powered land” — parcels with confirmed grid connections, executed interconnection rights, utility commitments, and supporting infrastructure secured before vertical construction begins — has become the most sought-after category in commercial real estate, with transactions in the 250-to-500-megawatt range now the institutional standard and sites capable of delivering power within 18 to 24 months commanding decisive premiums over otherwise identical parcels [40, 42].

┌────────────────────────────────────────┐     ┌────────────────────────────────────────┐

│          Historic Land Banking         │     │         Modern Megawatt Banking        │

├────────────────────────────────────────┤     ├────────────────────────────────────────┤

│ Value Driven By:                       │     │ Value Driven By:                       │

│ • Surface square footage               │     │ • Substation proximity                 │

│ • Topography & soil mechanics          │ ──> │ • Enforceable energization dates       │

│ • Local municipal zoning               │     │ • Confirmed system impact studies      │

│ • Highway / transport access           │     │ • Transformer manufacturing queue slot │

└────────────────────────────────────────┘     └────────────────────────────────────────┘

Figure 2. The migration of land value from surface attributes to energization attributes.

The institutional capital community has followed this migration with remarkable speed. Infrastructure funds, sovereign vehicles, and real estate investment managers that five years ago would have classified a substation as a due-diligence footnote now classify it as the asset itself. As Andrew McDaniel, founding partner at Meadow Partners, described the institutional appetite for grid-connected parcels:

“assets that can generate long-term, inflation-protected cash flows.”

— Andrew McDaniel, Founding Partner, Meadow Partners, on institutional demand for powered land [39]

Scarcity does the rest. Goldman Sachs Research projects that data center power demand will surge 165 percent by 2030 relative to 2023 levels — the equivalent of adding another top-ten power-consuming country to global demand — and estimates that meeting this growth will require approximately $720 billion in grid spending through the end of the decade [40]. Against that demand curve, the stock of genuinely powered land grows only as fast as substations can be built and transformers can be wound, which is to say: slowly, and at multi-year lead times. The result is a market in which megawatts, not acres, are banked.

1.2 The Megawatt as the Base Currency of AI Real Estate

Under Interconnection Capitalism, the transactional architecture of real estate itself has been rewritten to denominate value in energy delivery rather than physical space. Three instruments illustrate the shift.

The first is the power-first escrow. Real estate purchase agreements for data center developments are now routinely structured with explicit contingencies tied to grid approvals. Escrow accounts do not close upon clean title delivery; they close when the balancing authority executes a binding interconnection agreement, or when the utility issues a will-serve letter with defined capacity and dates. Legal practitioners advising on powered-land transactions emphasize that the diligence question is no longer “who owns the land” but “how committed is the committed power” — what maximum capacity the grid operator will actually supply, what network upgrades the delivery is contingent upon, and who bears the risk of delay, since grid operators have little commercial incentive to accept material liability for late connections [41]. Title has become the easy question; tenor is the hard one.

The second instrument is the arbitrage of grid allocations. Specialty infrastructure funds purchase distressed industrial properties — shuttered smelters, retired paper mills, decommissioned generation sites — not for their buildings, their equipment, or even their land, but solely to capture their legacy grid connection rights and interconnection headroom. These rights are then repackaged and flipped to hyperscalers at an exponential markup, treating megawatt capacity as a tradeable corporate asset severable, in economic substance if not always in legal form, from the dirt beneath it. The retired foundry, in this regime, is a mining claim: what is being extracted is the electrical easement of a vanished industrial economy.

The third instrument is the substitution logic that emerges when the currency runs short. When grid megawatts cannot be had on acceptable timelines, developers manufacture private ones — behind-the-meter gas turbines, on-site generation, bring-your-own-power configurations — explicitly as a bridge until interconnection arrives. Analysts at Ascend Analytics note that behind-the-meter gas can be a practical short-term solution when speed to market is the binding consideration, even though levelized cost analysis strongly favors front-of-the-meter grid connection over the long run [4]. One Texas developer, driven behind the meter by queue delays in Liberty County, distilled the entire repricing of the American industrial landscape into a single sentence:

“the power is much more valuable on the data center side.”

— Ashvin Tammabattula, CEO, BaRupOn, on reallocating power from manufacturing to AI infrastructure [43]

When an industrialist concludes that the highest and best use of an electron is to feed a GPU rather than a production line, the megawatt has completed its transformation from input to asset — from something a business consumes into something a business is. Section 2 examines the market where these assets are first minted: the interconnection queue itself.

Section 2: The Interconnection Queue as a Shadow Capital Market

2.1 From Engineering List to Options Exchange

An interconnection queue was never designed to be a market. In its original conception — codified in the early 2000s, when the archetypal applicant was a single large gas or coal plant seeking a transmission tie — the queue was an engineering worklist: a first-come, first-served ledger through which grid operators sequenced the studies needed to connect new facilities safely. The applicant paid modest fees, the operator studied thermal limits and voltage stability, and the list moved. The queue had no price, because a position in it conferred nothing except a place in an administrative sequence. That world is gone. Over the course of the 2010s and early 2020s, the queue metastasized into something its architects never imagined: the largest, least regulated, and most consequential shadow capital market in the American economy — a market in which the traded instrument is priority itself.

2.2 A Brief History of the Queue, 2020–2026

The transformation can be dated with unusual precision, because Lawrence Berkeley National Laboratory has counted it annually. At the start of the decade, queues were already swollen by the renewables boom; by the end of 2023 — supercharged by the Inflation Reduction Act’s incentives and the first wave of post-ChatGPT siting — active requests had reached their historic apex of nearly 2,600 gigawatts, roughly double the installed generating capacity of the entire United States, with solar, storage, and wind constituting some 95 percent of the waiting volume [7, 8]. Then the composition began to invert in ways that track the AI buildout almost perfectly. In 2024, total active capacity fell 12 percent — the first decline on record — driven not by efficiency but by a historic wave of withdrawals, including a record 112 gigawatts of solar and storage exiting under the pressure of political uncertainty, interest rates, tariffs, and permitting friction [7]. In 2025 the pattern deepened: total active volume fell another 10 percent to 2,061 gigawatts as more than 750 gigawatts withdrew, while — in the single most revealing datum of the period — active natural gas capacity surged 86 percent to 253 gigawatts, the dispatchable-fuel signature of data center developers demanding firm power on data center timelines [7, 8]. Meanwhile the operators themselves industrialized: FERC Order No. 2023’s cluster-study regime took effect across most regions in 2024; MISO, SPP, and PJM launched limited-term fast-track programs for reliability-critical resources; and grid operators began deploying automation, advanced computing, and — in a fine irony — artificial intelligence itself to accelerate the study of projects meant to power artificial intelligence [8]. Six years, in short, converted an administrative backlog into a structured, reformed, deposit-priced, and fiercely contested allocation system. What follows examines that system as what it has become: a market.

The numbers describe a system operating at civilizational scale. According to Lawrence Berkeley National Laboratory’s Queued Up series — the canonical empirical record of this phenomenon, published annually with data through the end of 2025 — the backlog of generator interconnection requests surged through the 2010s and early 2020s to a peak of nearly 2,600 gigawatts active at the end of 2023, more than double the entire installed generating capacity of the United States [7, 8]. As of the end of 2025, roughly 8,200 projects representing 1,312 gigawatts of generation and approximately 749 gigawatts of storage — 2,061 gigawatts in total — remained actively seeking connection, even after a historic wave of more than 750 gigawatts of withdrawals during 2025 [7, 8]. And this is only the generation side of the ledger. On the load side, the explosion is more recent and more violent: ERCOT’s 198 gigawatts of large-load applications in a single quarter of 2026 [4]; the more than 30 gigawatts of speculative data center demand that accumulated in AEP Ohio’s service territory — over 50 customers at over 90 sites — before a tariff imposed financial discipline [33, 35]; utilities within the PJM footprint collectively forecasting 55 gigawatts of new large load by 2030 and 100 gigawatts by 2037 [5].

2.3 Speculative Clogging and the “Phantom Queue”

Why did the queue swell so far beyond physical reality? Because rational actors treated it exactly as its incentive structure invited them to: as a cheap options market. Independent power developers and data center syndicates learned to file multiple speculative, often functionally identical connection requests for a single project across different grid coordinates — different nodes, different utilities, sometimes different RTOs — in order to discover, through the study process itself, which point of interconnection would require the lowest investment in network upgrades. The application fee was the option premium; the study result was the information purchased; the withdrawal was the option expiring worthless. Practitioners now openly describe this behavior in regulatory analysis: developers submit identical interconnection requests across multiple utility study queues to hedge against localized transmission constraints, producing aggregated load forecasts that include substantial speculative “phantom” load [5].

The result is the Phantom Queue: a paper pipeline whose gigawatts outnumber the deliverable capacity of the physical transmission system by a factor of five or more, and whose composition is epistemically opaque even to the operators who administer it. PJM’s own independent market monitor has repeatedly questioned the credibility of the long-range data center demand projections feeding the organization’s planning models and auction assumptions — projections that now sit at the center of a capacity-market convulsion in which the projected shortfall widened from roughly 209 megawatts in the 2026/2027 Base Residual Auction to more than 6.5 gigawatts in the 2027/2028 auction, and in which data centers were assessed to account for approximately 40 percent of capacity costs [3, 6]. The phantom is not harmless. Because grid planners cannot cleanly distinguish real projects from positional speculation, they must either overbuild against demand that may never materialize — socializing the cost — or underbuild against demand that does materialize — rationing the future. The queue’s information problem has become the grid’s capital allocation problem, and the grid’s capital allocation problem has become, through capacity auctions and rate cases, every ratepayer’s bill.

2.4 The Financial Mechanics of Queue Management

Regulators have responded to speculative clogging with a suite of financial instruments that — with almost poetic irony — complete rather than reverse the queue’s transformation into a capital market. Four mechanics define the current regime.

Milestone security deposits. FERC Order No. 2023, the landmark 2023 interconnection reform adopted in most regions the following year, replaced first-come-first-served serial studies with first-ready-first-served cluster studies, and armed the process with steep, escalating, increasingly non-refundable deposits at successive study phases, alongside site-control requirements and withdrawal penalties [8]. The explicit purpose was to burn the phantoms out of the queue by making optionality expensive. The practical effect, for well-capitalized players, was to convert queue positions into formally priced options: large developers now willingly tie up tens of millions of dollars in idle capital across multiple grid locations precisely because the deposits establish, for the first time, a defensible property-like interest in priority. What was once free and worthless became costly and valuable. Scarcity pricing did not drain the shadow market; it institutionalized it.

Network upgrade obligations as negotiable liabilities. When an RTO system impact study concludes that a new 500-megawatt data center load requires, say, a $150 million rebuild of a regional 345-kV transmission corridor, that obligation attaches to the project’s profile — and immediately begins to behave like any other financial liability. It can be negotiated, shared with electrically adjacent projects through cost-allocation mechanisms, restructured across study cycles, or shifted onto secondary buyers as part of a project sale. Sophisticated developers manage portfolios of upgrade exposure the way bond desks manage duration. Berkeley Lab’s cost analyses have documented how these network upgrade assignments — often revealed only late in the study process — function as the decisive economic variable in project survival [7, 8].

The churn of project withdrawals. The queue’s mortality statistics are extraordinary by the standards of any other capital market. Historically, only about 19 percent of projects — representing roughly 13 percent of capacity — that entered queues between 2000 and 2019 reached commercial operation, meaning that withdrawal, not completion, is the modal outcome of an interconnection request [7]. The typical project that did reach commercial operation in 2024 had spent an average of roughly 55 months — four and a half years — in the queue, a duration that continues to lengthen [7]. In 2024 alone, a record 112 gigawatts of solar and storage capacity withdrew; in 2025, total withdrawals exceeded 750 gigawatts as FERC Order 2023 reforms, readiness deposits, and market conditions forced speculative positions to liquidate [8]. Each large withdrawal is a small earthquake: because cluster studies allocate upgrade costs jointly, a sudden exit forces every remaining project in the cluster to be re-studied and re-priced, propagating volatility through the queue exactly as a large trader’s liquidation propagates through an order book.

Announced versus energizable capacity. Finally, the queue’s speculative structure has created a stark and strategically exploitable divide between corporate public relations and physical reality. A technology firm may announce a “5-gigawatt AI campus pipeline” — and the announcement is not exactly false; somewhere, applications exist — while the grid’s actual capacity to deliver that power on any specific date may be near zero, gated by substation construction, transformer lead times, and study backlogs. Market intelligence firm Sightline Climate’s tracking, widely reported in spring 2026, found that of roughly 12 gigawatts of U.S. data center capacity expected to come online in 2026, only about one-third was actually under active construction, with 30 to 50 percent of planned openings expected to be delayed or canceled outright [29]. In Interconnection Capitalism, the press release announces the option; the substation determines the exercise.

The intellectual conclusion of this section is uncomfortable but unavoidable. The interconnection queue now performs, badly and invisibly, the function that organized exchanges perform well and transparently: price discovery for a scarce, standardized, transferable right. The 2025–2026 reform wave — cluster studies, readiness deposits, large-load show-cause orders — represents regulators groping toward market design without admitting that a market is what they are designing. Until that admission is made, the queue will remain what it has become: a shadow exchange where the future of the AI economy trades at prices no one can see.

Section 3: The Financialization of Grid Readiness

3.1 Monetizing the Milestones of the Interconnection Process

If Section 2 described the minting of the raw asset — the queue position — this section describes its refinement. In classical project finance, a development passes through phases whose completion reduces risk and therefore raises value: permitting, engineering, procurement, construction. Interconnection Capitalism has taken that familiar logic and applied it, with unprecedented granularity and unprecedented sums, to the administrative anatomy of grid connection itself. Every discrete step required to connect a project to the grid — every study, every executed agreement, every equipment reservation — is now treated as an independent asset that can be valued, financed, insured, and transferred. The interconnection process has become a value chain, and the value chain has become a securitization pipeline.

┌───────────────────┐     ┌───────────────────┐     ┌───────────────────┐     ┌───────────────────┐

│ System Impact     │     │ Facility Study    │     │ Executed Inter-   │     │ High-Voltage      │

│ Study Approval    │ ──> │ Agreement         │ ──> │ connection Agmt.  │ ──> │ Transformer Slot  │

├───────────────────┤     ├───────────────────┤     ├───────────────────┤     ├───────────────────┤

│ Confirms thermal  │     │ Establishes firm  │     │ Grants durable    │     │ Secures physical  │

│ & voltage safety  │     │ engineering costs │     │ legal grid access │     │ hardware access   │

└───────────────────┘     └───────────────────┘     └───────────────────┘     └───────────────────┘

Figure 3. The interconnection milestone pipeline: each administrative gate is now an independently valued asset.

Valuing the feasibility and system impact study. The earliest tradeable milestone is informational. A project that has successfully completed its RTO engineering assessments — feasibility study, system impact study — has purchased something economically precious: the elimination of unknown upgrade exposure. Before the study, the project carries an unbounded contingent liability, since a late-stage restudy can reveal nine-figure network upgrade assignments that instantly destroy project economics; Berkeley Lab’s data show that roughly one-third of recent withdrawals occur at the facility study or interconnection agreement phase, precisely when true costs are finally revealed [7]. After the study, the liability is quantified and capped. Institutional investors pay a significant premium for that conversion of Knightian uncertainty into priced risk, and the premium is the study’s market value. In effect, the RTO’s engineering department has become an inadvertent rating agency: its work product does not merely permit construction — it re-rates the paper of everyone who holds a position behind it.

The executed interconnection agreement as senior instrument. Above the studies sits the interconnection agreement (IA) itself — the contract between transmission provider and developer that constitutes durable legal grid access. Berkeley Lab reports that as of the end of 2025, 549 gigawatts of capacity in U.S. queues already held a draft or executed IA without yet reaching commercial operations — a half-terawatt inventory of executed-but-unenergized legal rights, including 256 gigawatts of solar, 161 of storage, 76 of wind, and 45 of gas [7, 8]. That inventory is the closest thing Interconnection Capitalism has to an investment-grade bond ladder: rights of defined seniority, awaiting delivery, held on balance sheets, and — critically — transferable through project acquisition. When a hyperscaler or infrastructure fund buys a development-stage project, what it is buying, in economic substance, is the IA and its date.

3.2 Trading Substation Transformer Slots

The most physical — and currently the most binding — of the financialized milestones is the humblest: the transformer. Every data center campus requires massive step-down transformers to convert transmission-level voltage (345 kV, 500 kV) to distribution and facility levels, along with high-voltage switchgear, breakers, and grid-tie battery systems. These devices represent less than 10 percent of total data center cost, and something close to 100 percent of the current bottleneck [29]. The supply-demand imbalance has become the defining industrial shortage of the AI buildout. Since 2019, demand for generator step-up units has grown 274 percent and demand for substation power transformers has grown 116 percent, while Wood Mackenzie models a persistent 30 percent supply shortfall for power transformers across the national fleet [30]. Lead times that ran 24 to 30 months before 2020 now stretch to four and even five years for large units, according to analyses by PwC, Sightline Climate, and industry trackers, with medium-voltage switchgear effectively sold out through 2028 in many channels [29, 31].

The consequences for 2026 delivery have been brutal and public: of the roughly 12 to 16 gigawatts of U.S. data center capacity announced for the year, only about 5 gigawatts entered the year under active construction, with 30 to 50 percent of planned openings expected to slip or die — not for want of capital, chips, or demand, but for want of wound copper and grain-oriented electrical steel [29]. In such a market, a confirmed production slot at a global transformer manufacturer is no longer a procurement detail; it is a strategic asset with a secondary-market price. Hyperscalers now buy out entire production lines years in advance — reserving manufacturing capacity the way they reserve GPU allocations — and hold physical hardware delivery slots as bargaining chips in negotiations with utilities, who may find that the fastest way to energize their own grid upgrades is to cooperate with the customer who controls the equipment. The transformer order book has become a parallel interconnection queue: privately administered, entirely opaque, and in tight markets, sovereign over the official one.

3.3 Securitization of Grid Rights and the Powered-Land Capital Complex

The final stage of financialization is aggregation. Specialized infrastructure private equity funds now pool early-stage data center projects that hold secure queue positions, completed studies, or executed IAs, and structure these portfolios into financial vehicles whose underlying value is derived entirely from the legal right to receive power by a verified date. The powered-land investment thesis — articulated openly across the institutional real estate literature in 2025 and 2026 — is that secured, deliverable electrical capacity is a scarce, long-duration, inflation-protected cash-flow asset, and that the correct comparable is not land but regulated infrastructure [39, 42]. Investors are told, accurately, that underwriting now requires understanding utility capacity, interconnection queue status, transmission availability, and the regulatory environment governing energy development — that the exercise is, in the words of one market analysis, no longer purely a real estate problem but an infrastructure and energy problem, in which speed to power is itself the competitive advantage being purchased [42].

It is worth pausing on how strange this is by historical standards. A queue number issued by a nonprofit grid operator; a PDF study of thermal limits; a countersigned services agreement; a slot in a factory’s production schedule — none of these artifacts was designed to be an asset, and none is regulated as one. Yet together they now anchor portfolios worth tens of billions of dollars, secure credit facilities, and determine the valuation spreads between otherwise identical parcels of American land. Securities law spent a century learning to police markets in claims on future corporate cash flows. No comparable framework yet governs the market in claims on future electrons. Section 6 will show regulators beginning, belatedly, to notice. But first we must confront the pathology that arises when the new asset class is hoarded faster than it can be used: the stranded megawatt.

Section 4: The Stranded-Megawatt Problem

4.1 The Risk of Over-Allocation and Delayed Deployments

Every asset class that rewards hoarding eventually produces a hoarding crisis, and Interconnection Capitalism arrived at its own with remarkable speed. The mechanism is simple to state. Hyperscalers and developers, facing multi-year interconnection timelines and existential competitive pressure, rationally over-reserve: they lock up large power envelopes — a 1-gigawatt reservation here, a 500-megawatt allocation there — sized to aggressive long-term projections of AI training and inference demand, because under-reserving risks ceding the frontier while over-reserving merely costs money. But projections are projections. If the developer subsequently encounters GPU supply bottlenecks, transformer delays, financing friction, a strategic pivot, or simply a cooling of demand, it may deploy only 200 megawatts by its promised date, leaving 800 megawatts allocated, studied, planned-for — and idle. Those idle megawatts are not neutral. The utility has typically already committed capital against them: substations sited, transmission reinforced, capacity procured. Someone must carry the cost of infrastructure built for a customer who did not fully arrive, and the fight over who that someone is has become the central distributive conflict of the AI energy era.

┌────────────────────────────────────────────────────────────────────────┐

│                       Hyperscaler Over-Allocation                      │

│                Reserves 1.0 GW; Actually deploys 0.2 GW                │

└────────────────────────────────────┬───────────────────────────────────┘

                                     │

                  ┌──────────────────┴──────────────────┐

                  ▼                                     ▼

┌───────────────────────────────────┐ ┌───────────────────────────────────┐

│     The Ratepayer Equity Gap      │ │      RTO / Utility Remediation    │

├───────────────────────────────────┤ ├───────────────────────────────────┤

│ Capital costs for grid upgrades   │ │ • “Use-it-or-lose-it” clawbacks   │

│ are shifted onto regional consumer│ │ • Minimum-take demand charges     │

│ utility bills, sparking pushback. │ │ • Collateral, ramp audits, exit   │

│                                   │ │   penalties on unused capacity    │

└───────────────────────────────────┘ └───────────────────────────────────┘

Figure 4. The two downstream consequences of stranded-megawatt allocation: socialized cost and regulatory clawback.

4.2 The Ratepayer Equity Gap

The scholarly center of gravity on this question is Harvard Law School’s Electricity Law Initiative, whose March 2025 paper by Eliza Martin and Ari Peskoe — “Extracting Profits from the Public: How Utility Ratepayers Are Paying for Big Tech’s Power” — has become the most cited academic treatment of the cost-shifting problem in the 2020–2026 literature. Martin and Peskoe document at least three avenues through which the costs of serving data centers migrate onto the general public: special contracts between utilities and data center owners reviewed through opaque state processes; the socialization of transmission and generation buildout through rates that everyone pays; and co-location arrangements at existing power plants that can raise wholesale market costs for all consumers [9]. Their sharpest claim is institutional rather than arithmetical — that the very complexity of ratemaking is what makes the transfer possible, because the subjectivity of the process:

“conceals utility attempts to funnel revenue to their competitive lines of business.”

— Eliza Martin and Ari Peskoe, Harvard Law School, “Extracting Profits from the Public” (March 2025) [9]

The magnitudes at stake are material for households. Research cited in the Harvard analysis concluded that Virginia ratepayers could see electricity costs rise by $150 to $450 per year by 2040 if current rate structures persist and planned data centers are built [10]. In PJM, the independent market monitor attributed roughly 40 percent of the most recent capacity auction’s costs to data centers, existing and forecast, as capacity prices surged and Pennsylvania’s governor took the extraordinary step of filing a FERC complaint over auction outcomes [3, 6]. Peskoe himself has been careful to frame the objection not as hostility to the technology but as a demand for cost causation:

“we’re just asking who is going to pay for it.”

— Ari Peskoe, Director, Electricity Law Initiative, Harvard Law School [10]

The transmission mechanism from stranded allocation to household bill deserves explicit tracing, because it runs through the most technical and least publicly understood institution in American energy: the capacity auction. In PJM, load-serving entities must procure capacity commitments years in advance through the Base Residual Auction, and the price of those commitments is set by the marginal interaction of forecast demand — including forecast data center demand, phantom and real alike — with available supply. The recent record is a study in what happens when speculative gigawatts enter one side of that equation while retirements and queue delays constrain the other: an approximately 800 percent price spike in one auction cycle, followed by a further 22 percent increase to $329.17 per megawatt-day for the 2026/2027 delivery year, translating into estimated retail bill increases of 1.5 to 5 percent depending on the state [45]. The projected capacity shortfall then widened from roughly 209 megawatts in the 2026/2027 auction to more than 6.5 gigawatts for 2027/2028, with analysts warning of a gap that could reach 15 gigawatts by 2030 as load additions outpace generation [3, 4]. The political system responded in kind: Pennsylvania Governor Josh Shapiro filed a December FERC complaint arguing that load growth, interconnection delays, and auction design had combined to produce record power prices for his constituents [3]. The chain is now complete and visible end to end — a speculative reservation filed in a utility queue in Ohio or Virginia propagates through a load forecast, into an auction clearing price, onto a residential bill in Pittsburgh, and finally into a federal docket. Interconnection Capitalism, whatever else it is, has become a mechanism for converting corporate optionality into public cost.

Nor, in Peskoe’s analysis, do voluntary corporate commitments resolve the gap. Data center developers agreed in March 2026 to a White House “Ratepayer Protection Pledge,” committing to pay for all new power delivery infrastructure required for their projects. But utilities, he argues, are thwarting that commitment in practice by rolling billions of dollars of data-center-driven transmission upgrades into rates that everyone pays — justified by a FERC transmission pricing policy dating to 1994, built for an industry model with no relevance to the hyperscale buildout [11, 12]. The pledge, in this reading, is a press release; the tariff is the law; and until the tariff changes, the equity gap endures. The deeper point for our framework is that stranded and socialized megawatts are two faces of one phenomenon: capacity claimed under Interconnection Capitalism’s hoarding logic, paid for under regulated utility ratemaking’s socializing logic. The asset is private; the carrying cost is public.

4.3 Clawback Clauses and Non-Utilization Penalties: The AEP Ohio Template

The regulatory response — the repossession regime for electrons foreshadowed in our introduction — found its national template in Ohio. Confronting more than 30,000 megawatts of inquiries from over 50 customers at over 90 sites against a central-Ohio data center load that had grown from 100 megawatts in 2020 to roughly 600 megawatts in 2024, AEP Ohio imposed a 28-month moratorium on new data center connections and then, in May 2024, filed for a dedicated data center tariff [33, 35]. After a contested fourteen-month proceeding, the Public Utilities Commission of Ohio approved the framework on July 9, 2025. Its provisions read as a direct legal answer to every pathology catalogued in this paper: new data center customers above 25 megawatts must pay monthly for at least 85 percent of their contracted capacity regardless of actual usage; contracts run up to twelve years, comprising ramp-up periods and firm terms; collateral requirements screen for financial seriousness; and exit fees equal to three years of minimum charges make walking away from a reservation genuinely expensive [32, 33]. PUCO’s chair described the settlement as:

“a well-balanced package that safeguards non-data center customers.”

— Jenifer French, Chair, Public Utilities Commission of Ohio, July 9, 2025 order [34]

The empirical results were immediate and, for the theory advanced in this paper, close to a controlled experiment. Once reservations carried real carrying costs, AEP Ohio’s speculative queue collapsed from roughly 30 gigawatts to 13 gigawatts — the phantom load simply evaporated when priced [35]. The template is now propagating: Duke Energy Ohio filed a competing large-load tariff in December 2025; Dominion Energy Virginia, facing a 30-gigawatt interconnection backlog, implemented a cost-causation surcharge of $18 per kilowatt-month for new large loads above 10 megawatts; Indiana regulators approved large-load interconnection rules; and Wood Mackenzie projected in January 2026 that data-center-specific tariffs would be in force in at least a dozen states by year-end [35]. The hyperscaler community, through the Data Center Coalition, fought the Ohio framework as discriminatory and warned it would push investment toward friendlier states — an argument echoed by market-oriented critics who contend that take-or-pay mandates on the most mobile, price-elastic customers invert sound ratemaking principles [32]. That normative debate remains live. What is no longer debatable is the structural fact: the use-it-or-lose-it clawback has become the standard counterparty discipline of Interconnection Capitalism. The grid learned to repossess.

Section 5: The Interconnection Capital Stack

Financial markets abhor unpriced risk, and by 2026 they had evolved a structured hierarchy for assessing the value of AI infrastructure assets — a hierarchy this paper formalizes as the Interconnection Capital Stack. The organizing intuition is that a data center project is not one asset but a laminate of six, each layer corresponding to a regulatory, engineering, or procurement milestone, and each transition upward representing a discrete de-risking event that markets reward with an exponential, not linear, step in valuation. The stack does for energization what the venture capital stage model did for startups: it converts a continuous, chaotic development process into a discrete grammar of priced states. What follows is the six-layer model, ascending from dirt to electrons.

▲ [Level 6: ENERGIZATION] ────────── Fully operational asset; maximum financial valuation.

│ [Level 5: POWER CONTRACT] ──────── Executed corporate PPA secures cash flow and fuel supply.

│ [Level 4: EQUIPMENT ALLOCATION] ── Confirmed production slots for long-lead substations.

│ [Level 3: GRID STUDY APPROVAL] ─── System Impact / Facility studies cap upgrade cost risk.

│ [Level 2: QUEUE POSITION] ──────── Non-refundable deposits secure priced place in line.

│ [Level 1: LAND RIGHTS] ─────────── Raw real estate holding; lowest baseline asset value.

Figure 5. The Interconnection Capital Stack: valuation rises exponentially as projects ascend toward energization.

Level 1: Land Rights — The Foundation Layer

Definition: ownership or long-term lease control of raw acreage. Capital risk: highest in the stack. The property lacks approved zoning, environmental clearances, and — decisively — any claim on transmission infrastructure. Valuation model: baseline agricultural, industrial, or commercial market comparables. As Section 1 established, this layer has suffered structural impairment relative to the historical land-banking era: raw land near fiber but far from voltage is a stranded asset in waiting, and the market now prices it accordingly, with unpowered parcels facing heavy capital-expenditure burdens, uncertain timelines, and shrinking investor appetite relative to their powered counterparts [39, 42]. Land is where every project begins and where failed projects return.

Level 2: Queue Position — The Option Layer

Definition: an active, numbered entry in an RTO’s official interconnection register, secured under the post-Order-2023 regime by escalating readiness deposits and site-control demonstrations [8]. Capital risk: moderate to high. The position faces cancellation if subsequent engineering reveals prohibitive network costs, and cluster-study restudies triggered by neighboring withdrawals can reprice it overnight — recall that most requests historically die before commercial operation, and that 2025 alone saw more than 750 gigawatts of withdrawals [7, 8]. Valuation model: option pricing in all but name — value rises with queue seniority, cluster quality, node congestion economics, and remaining time to the study gates. This is the layer where the shadow market of Section 2 lives, and it is the cheapest ticket into the stack, which is precisely why regulators had to make it expensive.

Level 3: Grid Study Approval — The De-Risking Layer

Definition: successful completion of the RTO’s system impact study and facility study. Capital risk: moderate. This milestone defines exact engineering requirements and, crucially, caps maximum financial exposure for network upgrades — converting the unbounded contingent liability of Level 2 into a quantified line item. Valuation model: discounted cash flow incorporating the now-known upgrade costs and a materially reduced discount rate reflecting eliminated study risk. The one-third of withdrawals that occur at this gate demonstrate its function: it is where truth arrives, and truth is expensive for some and enormously valuable for the survivors [7]. Projects clearing Level 3 command a decisive premium from institutional investors precisely because the RTO’s engineers have, in effect, rated the paper.

Level 4: Equipment Allocation — The Procurement Layer

Definition: possession of legally binding production slots for long-lead equipment — step-down substation transformers, high-voltage switchgear, grid-tie battery systems. Capital risk: low to moderate; this layer mitigates the supply-chain delay that is currently the single most common cause of slipped energization dates. Valuation model: replacement-premium pricing off the secondary market for accelerated delivery slots — a market made brutally real by lead times of three to five years for large power transformers, switchgear effectively sold out through 2028, and a 2026 delivery cohort in which perhaps half of announced capacity will slip for equipment reasons alone [29, 30, 31]. In a shortage economy, Level 4 can temporarily outrank Level 3 in practical importance: a project with studies but no transformer has a date it cannot keep, while a project holding hardware has leverage over everyone who does not.

Level 5: Power Contract — The Commercialization Layer

Definition: an executed long-term power purchase agreement or energy delivery contract with a generation asset owner — the layer occupied by every row of Table 1, from Microsoft’s twenty-year Crane commitment to Amazon’s 1,920-megawatt Susquehanna framework through 2042 to Meta’s 6.6-gigawatt nuclear portfolio [23, 26, 27]. Capital risk: minimal and largely commercial. The layer locks in long-term energy costs, secures fuel-adjacent certainty, and delivers the clean-energy attributes required by corporate sustainability commitments. Valuation model: standard infrastructure finance — fixed long-term energy spreads and contracted generation cash flows, valued as quasi-bonds. It is at this layer that Interconnection Capitalism becomes visible to public equity markets: Talen’s disclosure of approximately $18 billion in expected lifetime revenue from its Amazon contract instantly repriced the entire independent-power sector [23].

Level 6: Energization — The Monetization Layer

Definition: the physical closing of high-voltage breakers, linking the live utility grid to operational AI server racks. Capital risk: operational and market only; every regulatory, engineering, and procurement hurdle has been cleared. Valuation model: maximum infrastructure multiples, as a functioning node of the global AI computing network. Level 6 is the stack’s telos and its scarcest state: recall that of 2026’s announced U.S. capacity, only a third entered the year under construction, and that the average project reaching this level had spent four and a half years climbing the stack [7, 29]. The exponential valuation curve across the six levels is thus not a metaphor but a survival distribution: each layer’s premium is the market’s payment for having outlived the attrition below it.

One theoretical refinement completes the model. The stack is not strictly sequential in practice — sophisticated players climb multiple layers in parallel, and the emerging flexibility literature suggests a lateral entrance to the upper stack that bypasses the queue entirely. Duke University research led by Tyler Norris demonstrated that existing U.S. power systems could accommodate roughly 98 gigawatts of new large load if that load curtails a mere 0.5 percent of annual consumption during peak stress hours, and a December 2025 follow-on study with Princeton’s ZERO Lab and Camus Energy found that flexible, bring-your-own-capacity data centers in PJM could cut net system cost increases by 96 percent while shortening the wait for grid power by three to five years [13, 15]. Flexibility, in stack terms, is a currency that purchases altitude: a load willing to bend can leapfrog layers that an inflexible load must buy. We return to this — the most hopeful finding in the entire 2020–2026 literature — in the pillars of Section 7.

Section 6: The Regulatory Counter-Revolution of 2025–2026

6.1 The State Rediscovers Its Grid

Markets that emerge in the shadows of administrative systems eventually force those systems to choose: ratify the market, or reclaim the territory. Between December 2025 and July 2026, American energy regulators made their choice with a speed and aggression that surprised nearly every observer of an agency ecosystem famous for its glacial pace. This section chronicles that counter-revolution — not as a digression from the theory of Interconnection Capitalism, but as its logical culmination. When queue positions, co-location rights, and energization dates became the most valuable industrial assets in the economy, the bodies with legal authority over them became, whether they wished it or not, the most important economic regulators in the country. In 2025 and 2026, they began acting like it.

The opening move came on December 18, 2025, when the Federal Energy Regulatory Commission issued its long-awaited order in the PJM co-location proceeding — the docket opened in February 2025, in the wake of the Talen–Amazon rejection, to determine whether PJM’s tariff adequately governed arrangements between generators and co-located loads such as data centers. FERC’s answer was unsparing: it found PJM’s existing tariff unjust and unreasonable for its failure to address co-location with clarity or consistency, and directed the grid operator to implement transparent, nondiscriminatory rules — clarifying the steps required to effectuate co-located load arrangements, establishing three entirely new categories of transmission service, creating new behind-the-meter generation rules, and requiring full accounting of behind-the-meter configurations in resource adequacy planning [36, 37]. The order was explicitly framed as an effort to establish transparent rules for serving AI-driven data centers and other large loads co-located with generation, with a requested effective framework taking shape by July 31, 2026 [5, 36]. Its timing carried its own message: the order landed on the heels of a PJM capacity auction that had closed 6,600 megawatts short of its reserve margin — a shortfall that transformed co-location from a commercial curiosity into a reliability imperative [37].

What FERC did for PJM in December, it extended to the entire federal jurisdiction in 2026. Acting on the Secretary of Energy’s October 2025 directive to initiate rulemaking on large-load interconnection under Section 403 of the Department of Energy Organization Act, the Commission issued tailored show-cause orders under Section 206 of the Federal Power Act to all six regional grid operators under its jurisdiction, directing each to justify or reform the rules governing how data centers, manufacturing facilities, and other large energy users connect to the grid [37, 38]. The agency described the initiative — informed by more than 3,500 pages of public comments — as among the most significant actions it has taken to modernize the nation’s electric markets, pairing accelerated large-load integration with what it called rigorous consumer safeguards, and building on the Southwest Power Pool’s newly approved High Impact Large Load study framework [38]. Read together with FERC Order No. 2023’s generator-side reforms, the arc of federal policy is unmistakable: the queue is being rebuilt, on both the supply and demand side, as a disciplined, deposit-backed, readiness-gated allocation system — which is to say, as a regulated market in the asset this paper has been describing all along.

Within PJM itself, governance strained visibly under the weight of the new asset class before hardening into policy. Stakeholder processes on data center interconnection rules collapsed without agreement in November 2025 — an impasse over load forecasting and cost allocation that itself testified to how much money now rides on every procedural word [5, 6]. In January 2026, the PJM Board of Managers broke the deadlock by decisional letter, setting out six principles for large-load integration whose contents read as a governance charter for Interconnection Capitalism: load forecasting improvements to exorcise the phantom queue; a “Bring Your Own Generation” expedited track that formally converts self-supplied capacity into accelerated interconnection — flexibility and additionality purchasing queue altitude, exactly as the capital stack of Section 5 predicts; a reliability backstop procurement; and a holistic review of PJM’s market design [5]. In the same January window, PJM simultaneously trimmed its near-term load forecast through 2032 on the strength of stricter data center vetting, even while raising its long-run trajectory — projecting summer peaks growing 3.6 percent annually to roughly 222 gigawatts by 2036 — a twin adjustment that captures the epistemic whiplash of planning under phantom load: the near future shrank as speculation was purged, while the far future grew as reality was admitted [6]. Governance, in other words, is learning to price the same distinction the market already prices: announced versus energizable.

6.2 The Limits of the Pledge: Voluntarism Meets the Tariff

The political branch moved in parallel, but with softer instruments. The White House convened data center developers around a Ratepayer Protection Pledge — commitments to pay for new power plants their facilities require, to fund delivery infrastructure, and to pay even if projected demand never materializes — and pursued executive action to accelerate federal permitting of data center infrastructure [11, 12]. The academic assessment of these voluntary instruments has been pointed. Harvard’s Peskoe, interviewed by the university’s climate institute in March 2026, observed that the pledge’s terms were not new — that:

“companies have been making similar promises for a while.”

— Ari Peskoe, Director, Electricity Law Initiative, Harvard Law School, Harvard Climate Brief interview (March 2026) [12]

The enforcement gap, in his analysis, is structural: states and public utility commissions — not the White House, and not the technology companies — control the tariff architecture through which infrastructure costs actually flow, and a 1994-vintage FERC transmission pricing policy continues to let utilities socialize data-center-driven upgrades across all customers regardless of what any pledge promises [11, 12]. The lesson generalizes into a principle of Interconnection Capitalism: in a regime where the asset is a regulatory right, only regulatory instruments can govern the asset. Pledges are denominated in reputation; tariffs are denominated in law; and the exchange rate between them is set by whoever writes the tariff. The state-level tariff wave documented in Section 4 — Ohio’s 85 percent minimum-take template, Virginia’s cost-causation surcharges, the dozen-state adoption curve projected by Wood Mackenzie — is therefore not a sideshow to federal policy but its necessary complement: the retail counterpart of FERC’s wholesale re-assertion [32, 35].

6.3 Flexibility as the Negotiated Settlement

If clawbacks are the counter-revolution’s stick, load flexibility is emerging as its negotiated peace. The intellectual foundation was laid at Duke University in February 2025, when Tyler Norris and colleagues published the most influential single finding of the period: that the existing U.S. power system — deliberately engineered to survive extreme peak swings — could absorb roughly 98 gigawatts of new large load, provided that load curtails only about 0.5 percent of its annual consumption during the grid’s most stressed hours [13]. Norris framed the implication for policymakers directly:

“load flexibility offers a promising near-term strategy for regulators and market participants.”

— Tyler Norris, Duke University Nicholas School of the Environment, lead author, “Rethinking Load Growth” (2025) [13]

The finding rests on an engineering fact that the public debate had largely missed — data centers do not, and should not, run at nameplate maximum around the clock. As Norris put it:

“You never actually run the chips and the servers to 100%.”

— Tyler Norris, Duke University, on data center load profiles [14]

Princeton’s Jesse Jenkins, whose ZERO Lab co-authored the December 2025 follow-on study quantifying flexible interconnection in PJM, framed the same insight as an indictment of the build-everything default:

“you only need it for a few hours a year.”

— Jesse Jenkins, Princeton University ZERO Lab, on peak-driven transmission overbuild [16]

The quantified stakes of that follow-on work bear repeating: flexible data centers pairing curtailable grid connections with bring-your-own-capacity resources contributed roughly $733 million per gigawatt toward the costs of their own incremental load, reduced net system cost increases by 96 percent relative to inflexible equivalents, kept grid power available more than 99 percent of hours, and — most importantly for the framework of this paper — shortened the wait for grid power by three to five years [15]. Within Interconnection Capitalism’s own valuation grammar, that last number is the whole argument: flexibility purchases queue seniority. By 2025–2026 the settlement was visibly forming in practice — Google signed demand-response agreements with Indiana Michigan Power and the Tennessee Valley Authority in what Norris called the first documented integration of AI data center flexibility into U.S. utility planning; the Southwest Power Pool proposed connecting flexible loads in as little as ninety days; and the Department of Energy and EPRI’s DCFlex initiative institutionalized the research agenda [13, 14, 15]. The counter-revolution’s deepest achievement may prove to be exactly this bargain: the state trades speed for flexibility, and the hyperscalers trade firmness for time. Interconnection Capitalism, having financialized the megawatt, is learning to financialize the negawatt.

Section 7: What Have We Learned? The Seven Pillars of Interconnection Capitalism

The evolution of the AI infrastructure market between 2020 and mid-2026 — from the first pandemic-era load inflections, through the post-ChatGPT capital supernova, to the regulatory counter-revolution now underway — can be synthesized into seven foundational pillars. The first five formalize the structural claims developed in Sections 1 through 5; the final two, added in light of the 2025–2026 evidence, capture the system’s emergent correctives.

Pillar 1: Megawatt dominance over real estate. Physical land has been reduced to a secondary consideration; the true value lies in securing a confirmed, high-voltage, dated grid connection. The powered-land premium, the collapse of unpowered fringe-acre values, and the rise of substation-first site selection all express a single repricing: location has been redefined electrically [39, 40, 42].

Pillar 2: Structural speculation in the queue. Interconnection queues operate as shadow options markets in which capital is deployed to control access to the grid — a market whose scale (2,061 gigawatts still active at the end of 2025, against a peak of nearly 2,600), whose mortality (the overwhelming majority of requests withdrawn), and whose phantom-load epistemics have made queue reform the central grid-policy project of the decade [5, 7, 8].

Pillar 3: Monetization of technical milestones. The administrative steps of grid integration — studies, agreements, equipment slots — have transformed into tradeable, financeable, securitizable assets, with the transformer order book functioning as a privately administered parallel queue during the 2024–2026 equipment shortage [29, 30, 31].

Pillar 4: Systemic exposure to idle power. Over-allocation creates stranded-megawatt liabilities that expose corporate developers to clawbacks and public ratepayers to socialized carrying costs — the equity gap documented by the Harvard Electricity Law Initiative and now contested in tariff dockets across at least a dozen states [9, 10, 35].

Pillar 5: Multi-tiered capital valuation. Projects advance through a strict Interconnection Capital Stack in which asset values climb exponentially at each de-risking layer from land rights to energization, with market pricing functioning as a survival distribution over the stack’s attrition [7, 23, 29].

Pillar 6: Flexibility as the new currency of priority. The 2025–2026 literature establishes that curtailability is convertible into queue seniority: loads willing to flex a fraction of one percent of annual consumption can access the grid years earlier and at radically lower system cost, making the negawatt the first genuinely deflationary instrument in Interconnection Capitalism [13, 15, 16].

Pillar 7: Regulatory re-assertion as the system’s equilibrium force. The December 2025 PJM co-location order, the 2026 six-RTO show-cause offensive, the DOE-directed large-load rulemaking, and the state minimum-take tariff wave collectively demonstrate that the market in energization rights will be ratified only on the state’s terms — transparent rules, consumer safeguards, and repossession clauses included [36, 37, 38].

Conclusion: The New Enclosure Movement of the Computing Age

The transition to Interconnection Capitalism represents a fundamental shift in how the technology industry builds competitive moats — and, like every prior shift of comparable magnitude, it rhymes with an older history. Early industrial corporations consolidated power by securing rights to the physical resources of their eras: railway corridors, mineral deposits, water rights, telecom spectrum. In each case the pattern was identical. A commons — land, ore, airwaves — was administratively enclosed; the enclosure created a new class of rights; the rights became assets; the assets became moats; and the moats, eventually, became the object of regulatory counter-assertion. The leaders of the AI economy are now enacting this cycle upon the last great industrial commons: the electrical grid. The 6.6 gigawatts of nuclear capacity Meta has framed through 2035, the 1,920 megawatts Amazon has contracted through 2042, the restart of reactors at Three Mile Island and Duane Arnold under twenty-year hyperscaler paper, the $700 billion in single-year capital expenditure guided by four companies against an ecosystem the International Energy Agency projects will roughly double its global electricity consumption to approximately 945–950 terawatt-hours by 2030 — these are the enclosure acts of the computing age, and the tariff dockets of Ohio, Virginia, and Washington are its anti-enclosure riots [17, 18, 19, 23, 26].

The introduction’s sidewalk repossession anecdote, viewed from the end of this analysis, discloses its full meaning. The luxury sports car was seized not because it lost its horsepower but because it lost its contractual standing; the physical object was always hostage to the rights beneath it. So too in the AI race. Any company that erects an advanced data center shell, or amasses a warehouse of next-generation GPUs, without locking in a legally enforceable date for power delivery, holds the sports car on the curb: a depreciating monument awaiting the flatbed. The evidence of 2026 has made this literal. Half of the year’s announced American data center capacity stands delayed or dying for want of transformers and interconnection, while capital expenditure guidance climbs toward figures exceeding the national defense budget — a divergence between announced ambition and energizable reality that is itself the signature market failure of the regime [20, 29].

And yet the system is not static. The deepest finding of the 2020–2026 literature is that Interconnection Capitalism contains, within its own valuation logic, the seeds of its correction. When priority became priced, phantoms fled the queue. When hoarding became expensive, reservations became honest. When flexibility became convertible into time, the most sophisticated players began trading firmness for speed — discovering that the cheapest megawatt is the one briefly not consumed. The regulatory counter-revolution has not abolished the market in energization rights; it has begun, haltingly, to civilize it, converting a shadow exchange into something that may yet resemble a governed commons with prices.

Under Interconnection Capitalism, the ultimate winner of the AI race will not be the company with the most elegant algorithm, nor even the one with the largest capital budget. It will be the enterprise — and, one should now add, the polity — that most successfully governs the high-voltage lines, substations, transformer order books, and connection queues required to power the future. The algorithm competes for benchmarks. The queue position competes for existence. In the economy this paper has described, existence is the scarcer prize.

Footnotes / Endnotes:

[1] PJM Interconnection (Inside Lines) — 2025 Year in Review: Planning Prepares for Burgeoning Electricity Demand. https://insidelines.pjm.com/2025-year-in-review-planning-prepares-for-burgeoning-electricity-demand/

[2] Engineering News-Record (ENR) — Grid Operator PJM Interconnection Aims Changes to Speed New Capacity Adds. https://www.enr.com/articles/63001-grid-operator-pjm-interconnection-aims-changes-to-speed-new-capacity-adds

[3] Shane Snider, Data Center Knowledge — PJM Monitor: AI Data Center Growth Reshaping Power Markets. https://www.datacenterknowledge.com/energy-power-supply/pjm-monitor-ai-data-center-growth-reshaping-power-markets

[4] Shalom Goffri, Ascend Analytics — Can US Interconnection Queues Survive Data Center-Driven Load Growth?. https://www.ascendanalytics.com/blog/large-load-interconnection-queues-data-center-grid-access

[5] White & Case LLP — PJM Proposes to Carve Out New Services for Co-Located Data Centers. https://www.whitecase.com/insight-alert/pjm-proposes-carve-out-new-services-co-located-data-centers

[6] Ethan Howland, Utility Dive — PJM Trims Near-Term Load Forecast on Stricter Data Center Vetting, Economic Outlook. https://www.utilitydive.com/news/pjm-interconnection-load-forecast-data-centers/809717/

[7] Lawrence Berkeley National Laboratory (J. Rand, J. Seel, et al.) — Queued Up: Characteristics of Power Plants Seeking Transmission Interconnection, 2026 Edition. https://emp.lbl.gov/queues

[8] Lawrence Berkeley National Laboratory — Backlog of Power Plants Seeking Transmission Grid Connection Eased Somewhat in 2025 Amidst High Withdrawals. https://emp.lbl.gov/news/backlog-power-plants-seeking-transmission-grid-connection-eased-somewhat-2025-amidst

[9] Eliza Martin & Ari Peskoe, Harvard Law School Electricity Law Initiative — Extracting Profits from the Public: How Utility Ratepayers Are Paying for Big Tech’s Power (March 2025). https://eelp.law.harvard.edu/wp-content/uploads/2025/03/Harvard-ELI-Extracting-Profits-from-the-Public.pdf

[10] Harvard Magazine — How AI Could Be Raising Your Energy Bill. https://www.harvardmagazine.com/2025/07/harvard-ai-increasing-energy-costs

[11] Ari Peskoe, Utility Dive (opinion) — An Outdated FERC Policy Is Undermining the White House’s Ratepayer Protection Pledge. https://www.utilitydive.com/news/ferc-transmission-policy-white-house-ratepayer-protection-peskoe/815438/

[12] Salata Institute for Climate and Sustainability, Harvard University — The Data Center Boom Is Colliding with the Grid’s Hardest Problems (interview with Ari Peskoe). https://salatainstitute.harvard.edu/data-centers-ai-artificial-intelligence-grid-permitting-transmission-electricity-energy

[13] Tyler Norris et al., Duke University / American Public Power Association — Study Examines Potential for Integration of Large Flexible Loads in U.S. Power Systems. https://www.publicpower.org/periodical/article/study-examines-potential-integration-large-flexible-loads-us-power-systems

[14] Canary Media — As Data Centers Go Up, North Carolina Weighs How to Meet Demand (featuring Tyler Norris). https://www.canarymedia.com/articles/utilities/north-carolina-duke-plans-data-center-demand

[15] Tyler Norris (Camus Energy, Princeton ZERO Lab, encoord) — Flexible Data Centers: A Faster, More Affordable Path to Power (December 2025). https://substack.com/@tylernorris/note/c-184540044

[16] Jesse Jenkins (Princeton University), via Yahoo Finance — The AI Boom Is Colliding with America’s Aging Power Grid. https://finance.yahoo.com/sectors/energy/articles/ai-boom-colliding-america-aging-142039344.html

[17] International Energy Agency (IEA) — Energy and AI — Executive Summary. https://www.iea.org/reports/energy-and-ai/executive-summary

[18] International Energy Agency (IEA) — Key Questions on Energy and AI — Executive Summary (updated projections). https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary

[19] Yahoo Finance — Hyperscalers Hit $700 Billion in 2026 AI Spending Plans. https://finance.yahoo.com/sectors/technology/articles/hyperscalers-hit-700-billion-2026-111243744.html

[20] 24/7 Wall St. — Google, SpaceX, Microsoft, and Amazon Are About to Spend “Twice the Entire U.S. Defense Budget” on AI. https://247wallst.com/investing/2026/06/19/google-spacex-microsoft-and-amazon-are-about-to-spend-twice-the-entire-u-s-defense-budget-on-ai/

[21] Goldman Sachs Research, via Quartz / Yahoo Finance — Meta, Microsoft, Amazon, and Alphabet Are About to Spend a Shocking Amount of Money to Dominate the AI Era. https://finance.yahoo.com/sectors/technology/article/meta-microsoft-amazon-and-alphabet-are-about-to-spend-a-shocking-amount-of-money-to-dominate-the-ai-era-115359575.html

[22] Jason Kirsch, Forbes — AI Spending Is Surging Faster Than Revenue — and Markets Are Repricing. https://www.forbes.com/sites/jasonkirsch/2026/06/02/the-ai-capex-to-revenue-gap-is-widening—and-markets-are-starting-to-notice/

[23] Ethan Howland, Utility Dive — Talen to Sell Amazon 1.9 GW from Susquehanna Nuclear Plant. https://www.utilitydive.com/news/talen-amazon-aws-susquehanna-nuclear-data-centert/750440/

[24] American Nuclear Society, Nuclear Newswire — FERC Rejects Interconnection Deal for Talen-Amazon Data Centers. https://www.ans.org/news/article-6534/ferc-rejects-interconnection-deal-for-talenamazon-data-centers/

[25] Ethan Howland, Utility Dive — FERC’s AWS, Talen Energy Ruling “Not the Final Word” on Nuclear, Data Center Colocation: Constellation CEO. https://www.utilitydive.com/news/ferc-amazon-talen-energy-ruling-nuclear-data-center-colocation-constellation/732016/

[26] Perkins Coie LLP — Nuclear Industry Kicks Off 2026 with Major Public and Private Sector Announcements. https://perkinscoie.com/insights/update/nuclear-industry-kicks-2026-major-public-and-private-sector-announcements-0

[27] Meta Platforms Newsroom — Meta Announces Nuclear Energy Projects, Unlocking Up to 6.6 GW to Power American Leadership in AI Innovation. https://about.fb.com/news/2026/01/meta-nuclear-energy-projects-power-american-ai-leadership/

[28] Utility Dive — Meta, Constellation Ink 20-Year Nuclear Power Deal to Support AI Goals. https://www.utilitydive.com/news/meta-constellation-illinois-clinton-nuclear-ppa-support-ai-goals/749992/

[29] Sightline Climate, via Yahoo Finance / Tom’s Hardware — Half of Planned US Data Center Builds Have Been Delayed or Canceled. https://finance.yahoo.com/sectors/technology/articles/half-planned-us-data-center-150928890.html

[30] POWER Magazine — Transformers in 2026: Shortage, Scramble, or Self-Inflicted Crisis?. https://www.powermag.com/transformers-in-2026-shortage-scramble-or-self-inflicted-crisis/

[31] pv magazine USA (citing PwC) — U.S. Transformer Market Faces Severe Supply Constraints as Lead Times Extend to Four Years. https://pv-magazine-usa.com/2026/05/11/u-s-transformer-market-faces-severe-supply-constraints-as-lead-times-extend-to-four-years/

[32] Data Center Frontier — Ohio Sets New Precedent: AEP’s Power Rules Shift Data Center Cost Burden. https://www.datacenterfrontier.com/energy/article/55304787/ohio-sets-new-precedent-aeps-power-rules-shift-data-center-cost-burden

[33] POWER Magazine — Regulator Approves AEP Ohio’s Landmark Data Center Tariff. https://www.powermag.com/regulator-approves-aep-ohios-landmark-data-center-tariff/

[34] Public Utilities Commission of Ohio — PUCO Orders AEP Ohio to Create Data Center Specific Tariff (July 9, 2025). https://puco.ohio.gov/news/puco-orders-aep-ohio-to-create-data-center-specific-tariff

[35] MGRID — AEP Ohio Data Center Tariff Sets National Precedent After 30 GW Interconnection Surge. https://mgrid.org/2025/12/03/aep-ohio-data-center-tariff-sets-national-precedent-after-30-gw-interconnection-surge/

[36] K&L Gates LLP — FERC Orders PJM to Reform Tariff for Co-Located Generation and Load. https://www.klgates.com/FERC-Orders-PJM-to-Reform-Tariff-for-Co-Located-Generation-and-Load-1-15-2026

[37] Baker Botts LLP — FERC Issues Order Providing Guidance for “Co-Locating” Power Plants with Data Centers within PJM. https://www.bakerbotts.com/thought-leadership/publications/2025/december/ferc-issues-order-providing-guidance-for-co-locating-power-plants-with-data-centers-within-pjm

[38] Federal Energy Regulatory Commission (FERC) — FERC Launches Aggressive Targeted Action to Speed Large Load Integration. https://www.ferc.gov/news-events/news/ferc-launches-aggressive-targeted-action-speed-large-load-integration

[39] Hiffman National — From Dirt to Data: Infrastructure and Real Estate Investors Unearth Opportunities in Powered Land. https://hiffman.com/from-dirt-to-data-infrastructure-and-real-estate-investors-unearth-opportunities-in-powered-land/

[40] Hanwha Data Centers (citing Goldman Sachs Research) — Power Availability: The New #1 in Data Center Site Selection. https://www.hanwhadatacenters.com/blog/power-availability-the-new-1-in-data-center-site-selection/

[41] Linklaters LLP — Powered Land: An Emerging Strategy. https://www.linklaters.com/en/insights/thought-leadership/powered-land/powered-land-an-emerging-strategy

[42] Datacenters.com — Powered Land Is Becoming the Most Valuable Asset in Data Center Investing. https://www.datacenters.com/news/powered-land-is-becoming-the-most-valuable-asset-in-data-center-investing

[43] Data Center Knowledge — Interconnection Delays Push Texas Data Center Behind the Meter. https://www.datacenterknowledge.com/build-design/unconventional-texas-data-center-explores-off-grid-power

[44] Brookings Institution — Global Energy Demands within the AI Regulatory Landscape. https://www.brookings.edu/articles/global-energy-demands-within-the-ai-regulatory-landscape/

[45] POWWR — How Data Centers Are Reshaping PJM’s Energy Market. https://www.powwr.com/blog/how-data-centers-are-reshaping-pjms-energy-market