Context
Two weeks ago, The New Utility tracked how the hyperscalers (with Google leading the charge) were quietly becoming generation landlords, buying the economics of power plants without taking on the operating headaches. The regulated incumbent answered.
NextEra paid $66.8 billion for Dominion.
That is the biggest U.S. utility deal since the deregulation wave of the 1990s, and the price is not one a traditional utility analyst would recognize. Dominion's standalone story (Virginia rate-base growth, offshore wind capital spending, a gradual exit from coal) would not have justified what NextEra paid. But that is in the past.
What drives the premium is the AI-load position underneath: 70 gigawatts of generation pipeline, 24.7 gigawatts of power demand already committed to the Northern Virginia data center alley, plus the interconnection queue positions, transmission rights, and regulatory relationships that come with operating there.
NextEra did not buy a utility. It bought a position inside the most contested grid in the AI buildout, wrapped in the regulated structure that lets that position be financed cheaply, at the cost of capital a regulator allows on the rate base rather than the cost of capital a private developer pays.
That regulated structure is the key. The New Utility described hyperscalers moving up a control spectrum, from buying power under long-term contracts (PPAs) toward owning equity in the developer platforms that build the plants, and deliberately avoiding the rate base. The rate base is the pool of utility assets a state regulator allows the utility to charge customers to recover, plus a fixed return. Hyperscalers are avoiding it because once you are inside the rate base, the state Public Utility Commission (PUC) can decide how the cost of new wires and new power plants gets divided up among customers.
That fear was driven by an active cost-allocation debate. The North American grid reliability regulator (NERC) had issued an alert, FERC opened a federal rulemaking (RM26-4), the White House published a Ratepayer Protection Pledge, Colorado and Pennsylvania drafted model tariffs targeting large data center loads, and roughly 300 data center bills were introduced across 27 state legislatures. All of them pointed in the same direction: assign the cost of serving AI load to AI load. Hyperscalers responded by climbing the equity ladder.
The Rebundle is the opposite move. The rate base is now attractive again, not in spite of mandatory cost-allocation but because of it. If the AI customer is going to be assigned the cost of the wires no matter what, then owning the regulated entity that bills the customer, builds the wires, finances the build, and controls who gets connected and when is no longer a slow trade. It is the only way to put serious capital behind a forty-year demand curve and have a regulator backstop the price you charge to earn it back. The pre-1935 vertical-integration playbook (regulated monopoly plus generation plus transmission, all financed off the rate base) is being reassembled around AI demand, with a regulator's blessing.
Why now
Through 2025 and into the first half of 2026, long-term interest rates drifted from the low 3% range to the mid 4% range. That move quietly destroyed the math behind a lot of recent power deals. A 20-year PPA is worth less when rates rise, because future dollars get discounted harder. Every PPA in the pipeline lost value at the moment the AI buildout demanded more committed power, sooner.
Hyperscalers responded by climbing the equity ladder, moving up from buying contracted power to buying a piece of the developer building the plant, capturing more of the asset's economics to close the cost-of-capital gap.
Regulated utilities did the opposite. Rate-base financing, which had looked slow and politically constrained back when PPA arbitrage was easy, became the cheapest cost of capital available to anyone willing to live inside the regulatory wrapper. Same interest rate signal, two corporate strategies.
Same playbook in reverse
The hyperscaler's playbook for assembling a generation stack: long-dated rights to electrons, equity in the platform that develops the plants, queue position at the constrained nodes where transmission is scarce, upstream supply optionality on turbines and transformers, all structured so operating, regulatory, and decommissioning risk stays with somebody else.
The regulated utility's playbook is the mirror image. The same four pieces, wrapped in a structure that takes on operating, regulatory, and decommissioning risk and earns a regulated return for shouldering it.
These two stacks are not arguing about the same regulatory regime. They are answering different forcing functions inside it. Hyperscalers are running away from cost-allocation exposure. Regulated utilities are running toward AI demand as the load that finally justifies their next twenty years of capital spending. The point where both converge is the interconnection queue, the literal waiting list for new power plants and big customers to plug into the grid. The queue is what both strategies are buying: Dominion's 70 gigawatts of generation pipeline, NextEra's transmission rights, Virginia's interconnection backlog. And the queue is what state PUCs control.
GE Vernova's acquisition of Prolec GE is the same logic at the equipment layer. As I argued in The New Utility, locking up three years of transformer factory output is a more efficient way to solve supply access than buying into the manufacturer's equity. NextEra-Dominion is the analogous move one layer in. Owning the regulated platform that holds the queue beats fighting for queue position one project at a time.
Insull, revived
The New Utility argues that the hyperscaler form was not a vertically integrated utility, because hyperscalers refused operating responsibility and refused the rate base. That refusal is novel and made the historical analogy to Samuel Insull inapplicable.
“The Rebundle” is the Insull playbook…updated for AI. Between roughly 1907 and 1932, Insull built Commonwealth Edison and Middle West Utilities by combining three things:
A regulated monopoly franchise at the local level;
A holding-company structure that let him finance generation across the whole territory off a single consolidated balance sheet; and
A demand thesis, residential electrification, that justified building plants ahead of where the load curve actually was.
The model worked for a generation. It also produced the Public Utility Holding Company Act of 1935 (PUHCA), which broke the holding companies up, forced divestiture of non-contiguous operations, and created a federal regulator with authority over interstate utility ownership.
NextEra-Dominion deploys the same three ingredients against a new demand thesis. The franchise is the combined Dominion and NextEra service territory: Virginia, North Carolina, South Carolina, and Florida. The holding-company structure is NextEra Energy as parent, with regulated and unregulated subsidiaries financed against a consolidated balance sheet. The demand thesis is AI load: 24-hour, concentrated, and on a timeline that rate-base financing can finally meet, because hyperscalers are willing to underwrite the build with long-dated contracts and cost-allocation conditions.
Insull's load was residential electrification: diffuse, slow-growing, politically sensitive at the retail level. The Rebundle's load is concentrated industrial demand from a handful of AI counterparties. That difference matters for where the political reaction shows up. PUHCA arrived because the question of whether one entity should own the wires to your house became politically untenable. The Rebundle does not run that risk at the household level. It runs the antitrust risk one layer up, at the cloud-AI-load layer.
The relevant precedent is not PUHCA. It is the FTC's posture toward Microsoft, the DOJ's recent cloud market scrutiny, and the question of whether owning regulated generation that serves a small handful of hyperscaler counterparties is a horizontal concentration problem inside the AI market, rather than a vertical concentration problem inside the power market.
That is a slower clock than PUHCA was. PUHCA took roughly twenty-five years from Commonwealth Edison's franchise consolidation to federal breakup. The AI-layer antitrust reckoning, if it comes, will probably take less. But it will not arrive during the buildout currently underway.
The hyperscaler track did not pause
The Rebundle is a response, not a victory. While NextEra and Dominion negotiated, the unbundled hyperscaler track kept escalating. Constellation extended its Three Mile Island restart deal with Microsoft into a 20-year power offtake priced against 2024 fundamentals, locking the generator's economics to Microsoft's planning cadence rather than the merchant market's. Talen accepted Amazon's co-location structure at the Susquehanna nuclear plant, a setup the regional grid operator (PJM) had blocked, under terms that put scheduling authority and capacity rights inside Amazon's stack, and absorbed the resulting FERC challenge as a cost of doing business. Google's acquisition of Intersect Power's developer platform did the same thing one rung higher, transferring siting, queue strategy, and technology selection into Google's control layer without changing the title on any operating asset.
The pattern is consistent. The hyperscaler is not buying megawatts. It is absorbing the planning function — siting, queue position, technology choice, dispatch logic — through an escalating series of bilateral deals that leave the regulated wrapper in place while emptying it of operational authority. NextEra-Dominion reassembles the bundle at the level of the regulated holding company. It does not reach the layer where the planning function is being absorbed.
That is what makes the Rebundle a contested response. The corporate-form contest is not utility-versus-hyperscaler at the asset layer. It is whether owning the regulated wrapper, the rate base, and the queue is enough to take back the planning authority that has already migrated upstream. The Virginia SCC adjudicates a transaction. It does not adjudicate that deeper question. The deeper question gets answered transaction by transaction over the next eighteen months, and the bundle either holds or it doesn't depending on whether the planning function comes back inside the wrapper or stays where the bilateral instruments have already moved it.
The Virginia SCC is the gate
State environmental and public-utility police power is one of the strongest federalism shields in U.S. constitutional law. The Virginia State Corporation Commission’s (SCC) review is the first $30-billion-scale merger where that shield gets stress-tested under AI-load pressure. If the SCC imposes material conditions and they survive federal challenge, the shield holds, which will have a significant impact on future AI infrastructure and generation development in the U.S.
Even in the face of mounting public data center opposition, I expect Virginia's State Corporation Commission to approve NextEra-Dominion without bundle-breaking conditions by the end of 2027. Virginia is unlikely to require queue divestiture or transmission-protection covenants strong enough to strip the data center load economics out of the merger case. Any cost-allocation conditions are unlikely to be tougher than the White House Ratepayer Protection Pledge already on the table. Virginia is the data center capital of the world. A regulator that blocks the merger blocks the capital that funds the generation that serves the load that funds the tax base.
I also expect at least one more $30-billion-plus regulated utility consolidation to close in the same window, with AI demand named explicitly as the primary growth driver in the proxy materials.
If Virginia conditions the merger to break the bundle, NextEra-Dominion was one transaction read as a regime, and the unbundled hyperscaler form reasserts as the dominant corporate response. If no peer $30-billion-plus regulated deal closes with an AI-explicit growth thesis by end-2027, NextEra-Dominion is an outlier, a one-off premium paid by one strategic buyer rather than a structural shift.
The Virginia SCC docket is the fastest-resolving piece of this. The opposition brief filed by QTS and other data center operators on May 21, 2026 marks the start of formal adjudication. The SCC's standard merger review for a deal of this size runs roughly twelve to fifteen months, which puts a decision in Q4 2026 or Q1 2027.
The Grid-Silicon Order Outlook
The Rebundle tells us that the regulated utility is not exiting the AI buildout. It is reasserting its position in response to the same forces that produced the unbundled hyperscaler form. Which trajectory the regime takes from here turns on two axes: whether the global geopolitical architecture stays integrated or fragments, and whether the infrastructure build accelerates or stays constrained.
In this environment, Aroko analysis points to four scenarios worth watching:
1. Open Abundance: integrated and accelerated. The AI demand curve bends down. Silicon-side efficiency compresses energy per AI workload faster than agentic load multiplies, and the concentrated demand that justifies a $66.8 billion premium dissolves. The Rebundle becomes unnecessary, not because the grid clears but because the load it was solving for shrinks. Winners: commodity cloud, the chip vendors capturing the efficiency gains, and the inference layer that routes around centralized power. The signal would be a downward revision to hyperscaler capital spending guidance. In Q1 2026, the four largest hyperscalers guided to more than $300 billion combined, in the opposite direction.
2. Sovereign Stacks: fragmented and accelerated. Each geopolitical bloc builds its own energy-compute corridor. The U.S. runs on Permian and Marcellus gas plus a nuclear restart. The Rebundle becomes one form among several, competing with national-champion vehicles abroad: a European EDF analog, a Gulf G42-style integrator, a Japanese METI-coordinated utility-hyperscaler arrangement. The bundle holds at home but does not export.
3. Coordinated Scarcity: integrated and constrained. Bottlenecks force allocation. Gas is the swing fuel. The political fights are about methane regulation, stranded-asset risk, and who pays for new capacity. This is where the Rebundle lives. The regulated wrapper absorbs the cost-allocation fight, the rate base finances the build, AI load justifies the price. The arena to watch is the PUC docket (Virginia, Pennsylvania, Texas, Ohio) and the federal-versus-state contest over who assigns costs. Winners: integrated regulated utilities, the capital allocators inside them, and the equipment-OEM consolidations (GE Vernova-Prolec GE, Eaton, Hitachi Energy) hardening the supply layer.
4. Fortress Era: fragmented and constrained. Scarce resources, closed borders, gas used as a strategic weapon. The Rebundle survives as the U.S. domestic answer, but its surface area shrinks. The binding questions become LNG export policy, gas-turbine export controls, and which bloc fuels which fleet. The bundle becomes a national-security asset rather than an investment thesis. Winners: defense primes, sovereign-wealth gas holdings, LNG terminal operators, and turbine-component stockpilers.
The regime today sits in Coordinated Scarcity. NextEra-Dominion is a regulated bundle reforming under physical and capital constraint. Drift turns on two contests over the next three to five years. If federal preemption stalls in the 120th Congress and state PUCs remain the authoritative venue, the drift is toward Sovereign Stacks on that horizon. If AI politics fold into energy-security politics and gas-turbine export controls become binding, the drift is toward Fortress Era on the same window. Open Abundance would require resolving the queue, and nothing in current evidence supports that.
The Rebundle reframes the question. It is no longer "can we build enough generation." It is "who owns the wires, the queue, and the regulatory entity that decides who gets the megawatts and at what price." That ownership is consolidating in front of state regulators on a clock measured in quarters.
For capital allocators, the relevant exposure shifts with it: the regulated utility holding company that controls Virginia, the equipment OEMs whose order books are full through 2028, and the AI-aware capital sitting inside regulated-utility balance sheets. The argument is no longer whether AI breaks the grid. It is who profits when AI is priced into the rate base.
Watchlist
1. Virginia SCC merger docket. Procedural orders, intervenor motions, staff recommendations, and the final decision on NextEra-Dominion. Confirms or breaks the bundling thesis.
2. Peer $30B+ regulated utility consolidation. Any announced acquisition or merger above $30 billion where AI demand is named as a primary growth driver in proxy materials. Confirms the Rebundle template versus outlier.
3. OEM supply-chain consolidation beyond GE Vernova-Prolec GE. Any further equipment-OEM combination tightening the gas-turbine, transformer, switchgear, or high-voltage equipment supply stack. Hardens the Rebundle's supply layer.
4. Federal preemption legislation in the 120th Congress. Any bill stripping state PUC cost-assignment authority over data center load that advances out of committee. Threatens the Rebundle's forcing function.
5. Replacement AI executive order or reissued federal corridor authority. Any reissued order restoring federal corridor authority over data center siting and interconnection, or successor executive action achieving the same effect through DOE or FERC delegation. The April 2026 walkback removed the fastest preemption route over state SCC authority; a replacement order would restore it.
6. FERC RM26-4 progression to a Notice of Proposed Rulemaking (NOPR). The federal large-load interconnection rulemaking advancing on its mid-2026 track. Determines whether large-load cost-assignment becomes a federal floor or remains state-by-state.
7. Hyperscaler response to the Dominion deal. Any new developer-platform equity transaction or unusual procurement structure announced by Microsoft, Amazon, Meta, or Oracle in the next twelve months. Tests whether the unbundled form retreats or doubles down against the rebundled competitor.
8. Antitrust posture on cloud-power adjacency. DOJ or FTC investigations or statements referencing concentrated regulated generation serving hyperscaler counterparties. First signal antitrust extends into the cloud-AI-load layer rather than the public-utility-monopoly layer.
Synthesis
NextEra is buying Dominion for $66.8 billion. The trade is a new kind of utility: one that owns the wires, controls who plugs into the grid, finances the build through customer bills, and locks in long-term contracts with the AI companies that need the power. Virginia's regulator decides whether to allow it — late 2026 or early 2027.
Three other moments hit the same window: federal regulators take the next formal step on data-center grid rules by mid-2026, and the next Congress takes up whether Washington can override the states after the midterms. Each tests a different leg of the thesis.
The underlying force is older than AI. Whenever one customer set gets big enough that new wires have to be spread across every bill, the country lands back on regulated monopolies — that is how Samuel Insull built Commonwealth Edison a century ago. The customer set today is five companies, not a city of households, which makes the politics faster and sharper. The deal also bets that AI demand keeps outrunning chip efficiency long enough to justify paying off forty years of investment on customer bills.
Three regulators could end up writing the rules. States have the constitutional authority but stop at the state line. Federal energy regulators have national reach but no settled view on a deal this big. The Justice Department has antitrust tools but nobody who understands power markets. No one has been picked — and that choice itself is being made in Virginia.
The call fails one of two ways. If Virginia guts the deal with conditions, this is one transaction misread as a regime. If no other $30 billion-plus utility merger closes by end-2027 with AI demand named as the reason, it is a one-off, not a template.
Evidence Base
This analysis draws on primary and secondary sources across six categories. Quantitative claims (deal value, GW pipeline, peak load, OEM backlog) are attributed to issuer disclosures and regulatory filings. Structural claims (the Rebundle thesis, the Insull comparison, the Virginia SCC adjudication framing) synthesize across deal documentation, prior letters, and public dockets. Forward-looking claims are flagged as forecasts with explicit falsification conditions.
Federal regulatory bodies: FERC (RM26-4 ANOPR docket; Talen-AWS Susquehanna co-location tariff docket); NERC (Level 3 reliability alert on data center load); White House (Ratepayer Protection Pledge, Federal Register, March 2026; AI executive order April 2026 walkback record); U.S. Treasury / Federal Reserve (H.15 yield curve series, 2025–H1 2026).
State regulatory agencies: Virginia State Corporation Commission (NextEra-Dominion merger docket; QTS-led opposition brief, May 21, 2026); Pennsylvania PUC (mandatory cost-allocation tariff order); Colorado PUC (model tariff); Hill County, Texas (data center moratorium, May 2026); Ohio Secretary of State (25-MW data center ballot initiative signature drive).
Grid operators and market data: PJM interconnection queue data and Northern Virginia transmission planning; ERCOT large-load queue.
Corporate disclosures: NextEra Energy Inc. and Dominion Energy Inc. merger announcement and accompanying proxy materials (May 19, 2026); GE Vernova and Prolec GE integration disclosures (May 15, 2026); Eaton, Siemens Energy, Hitachi Energy Q1 2026 order-book disclosures; Constellation Energy / Microsoft Three Mile Island restart agreement (Sept 2024) and twenty-year extension structure (2025–2026); Talen Energy / Amazon Web Services Susquehanna co-location announcement (March 2024) and subsequent FERC / PJM filings; Google / Intersect Power developer-platform acquisition (March 2026).
Trade and industry analysis: Utility Dive, S&P Global Market Intelligence, BloombergNEF data-center power coverage; Data Center Watch aggregate blocked-capacity tracking.
Historical reference: Public Utility Holding Company Act of 1935 legislative history; Samuel Insull and Commonwealth Edison / Middle West Utilities corporate history (1907–1932); FTC Microsoft posture documentation; DOJ cloud-services scrutiny record.
Endnotes
NextEra Energy Inc. and Dominion Energy Inc., merger announcement, May 19, 2026 (deal value: $66.8 billion; Dominion pipeline: 70 GW; Dominion peak load: 24.7 GW).
GE Vernova and Prolec GE, integration announcement, May 15, 2026.
Pennsylvania Public Utility Commission, mandatory cost-allocation tariff order, May 18, 2026.
Virginia State Corporation Commission, NextEra-Dominion merger docket, opposition brief filed by QTS and intervenor data center operators, May 21, 2026.
Hill County, Texas, commissioners' court, data center moratorium vote (3-2), May 12, 2026.
Ohio Secretary of State, 25-megawatt data center ballot initiative signature drive, week of May 15, 2026.
Data Center Watch, aggregate blocked-capacity tracking, May 12, 2026 ($64 billion blocked nationally).
North American Electric Reliability Corporation, Level 3 reliability alert on data center load, mandatory response deadline summer 2026.
Federal Energy Regulatory Commission, Advance Notice of Proposed Rulemaking RM26-4, large-load interconnection, advancing to NOPR mid-2026.
Eaton Corporation, Q1 2026 quarterly disclosure (order book $14.5 billion, +44% year-over-year).
Siemens Energy AG, Q1 2026 quarterly disclosure (order book €146 billion).
GE Vernova Inc., Q1 2026 quarterly disclosure (orders +71% year-over-year).
Prolec GE, Q1 2026 backlog ($5 billion, three-to-four-year factory booking).
White House, Ratepayer Protection Pledge, Federal Register, March 2026.
Public Utility Holding Company Act of 1935 (49 Stat. 803); legislative record and PUHCA 2005 repeal context under the Energy Policy Act of 2005.
Samuel Insull, Commonwealth Edison (founded 1907) and Middle West Utilities (incorporated 1912), corporate history and pre-PUHCA holding-company structure.
The Subnational Sovereigns, Aroko, April 2026 (CALL-002, state federalism-shield claim).
The New Utility, Aroko, May 2026 (CALL-003, hyperscaler equity-coupled-offtake and OEM-acquisition bet).
Federal Trade Commission, Microsoft antitrust posture and cloud-services scrutiny record; Department of Justice, Antitrust Division, cloud-services inquiry.
Virginia State Corporation Commission, standard merger review procedure and twelve-to-fifteen-month adjudication timeline benchmarks (recent comparable dockets).
U.S. Treasury 10-year and 30-year yield trajectory, 2025 through first half of 2026 (drift from the low-3s into the mid-4s on the long end). Source: Federal Reserve H.15 release; U.S. Treasury daily yield curve rates.
White House AI executive order, April 2026 walkback. Original draft order included federal corridor authority over data center siting and grid interconnection; final issuance withdrew that authority following objections from a gas-state congressional bloc. Source: White House press record; Federal Register; congressional record.
Constellation Energy Corp. and Microsoft Corp., Three Mile Island restart power purchase agreement, twenty-year structure (September 2024 announcement; extended structure 2025–2026).
Talen Energy Corp. and Amazon Web Services, Susquehanna data center campus co-location arrangement (March 2024 announcement; FERC tariff filings 2024–2025; PJM rejection November 2024).
Google LLC and Intersect Power, developer-platform acquisition (March 2026 announcement; reported transaction value $4.75 billion).
About Aroko: Aroko provides strategic advisory and capital allocation intelligence at the intersection of energy transition, technology infrastructure, and geopolitical risk. Our analytical process combines proprietary evidence infrastructure with human-directed thesis formation. Every keystone claim is verified against primary sources, and all editorial judgment and capital allocation framing is conducted by Aroko’s team. The Letter is published biweekly for institutional allocators.
Methodology: AI-assisted evidence infrastructure · Human-directed thesis · Primary-source verified
Disclaimer: This article is for informational purposes only and does not constitute financial, investment, legal, or tax advice. The opinions expressed regarding macro trends and infrastructure investments are solely those of the authors. Past performance does not guarantee future results. Readers should consult with a qualified financial professional before making any investment decisions.