You Are Winning the Data Center Fight

Nobody defends a victory they did not win.

On August 19, 2026, the Senate Republican campaign committee sent a private memo to a group of technology companies. It was titled “Ohio Data Center Risk.” Its argument was that Sherrod Brown had made data centers the de facto opponent in his race against Jon Husted, that it was working, and that no campaign or party organization could fix it. The AI companies would have to do that themselves, by explaining who benefits, who pays, and why a community should want one. Until they did, the memo said, the issue would keep dominating the race, and if Husted lost and data centers took the blame, politicians everywhere would notice and never go near the next one. [1]

Set aside the horse race. As a document, that memo is a party telling an industry that the machinery of electoral persuasion cannot produce public consent for the industry’s core physical asset, and that the industry is on its own. It is an admission of failure filed upward, from the institution that exists to win arguments to the institution that needs one won. The interesting thing about it is not the panic. It is that the panic is entirely about a Senate seat.

Something has come apart between the fight and the decision. The public objection to data centers is loud, sincere, cross-partisan and electorally effective. In the venue where the terms of the buildout are actually set, that objection does not appear. Something resembling one piece of it appears instead, argued by different people on different grounds, and it has been winning.

This essay is about that gap, and about a second thing that is harder to see and worse. The win in the second room was not produced by the fight in the first one. It has no parent among the people it protects. That is why it will be difficult to keep.

The One Grievance This Is About

The objection to data centers is at least six objections: cost, water withdrawal, noise, land conversion, tax abatement, and a general distrust of the people building them. Only the first is at issue here. And within cost, only one component of it: the direct shifting of a specific project’s infrastructure cost onto other customer classes.

Not generation scarcity. Not capacity prices. Not fuel markets, where new gas plants built for data center load buy on the same market as every existing plant. Not a transmission expansion whose forecast turns out wrong. All of those move a household’s bill and none is addressed below.

That one component is the part of the objection a rate proceeding is built to hear, which is exactly why it is the part that got answered. Nothing here claims more.

The Room Doesn’t Need Your Anger

Every governing decision has a room, and every room has rules about what can be said in it that are prior to, and more determinative than, any argument made inside it.

A zoning board will hear that a project is too loud, too large, too close to a school, out of character with the county. It will not hear that the applicant’s industry is overcapitalized. A legislature will hear almost anything, which is part of why it decides so little. A rate proceeding before a state utility commission will hear that a cost has been misallocated between customer classes, that a forecast is unsupported, that a capital expenditure is not used and useful. It will not hear that the applicant should not exist. Standing, evidence rules, statutory mandate and the professional habits of the participants combine to decide what an objection has to become before it can be answered at all.

Call the filter venue translation. If a grievance is going to be heard as a grievance, it must be restated in the language of whichever institution holds authority over the decision, and the restatement is not a formality. It is the price of being answered, and it is charged before anyone weighs the merits.

Two consequences follow, and this essay needs both.

The first is the obvious one. A grievance that cannot make the translation is not heard at all, however loud it gets. That is what happened to the demand that the buildout stop.

The second is stranger and is the harder half of the story. A result favorable to a class of people can be produced in a room that none of them entered, by a participant pursuing an interest that happens to point the same way. Nothing was translated, because nothing was submitted. The outcome arrives regardless. That is what happened to the cost objection.

Neither observation is new on its own. Schattschneider argued sixty years ago that the definition of the alternatives is the supreme instrument of power, and that who wins depends on which conflict gets to be the conflict. Political scientists since have described venue shopping, in which an organized interest picks the forum most likely to give it what it wants; the lawyers’ version of the same filter is standing doctrine. Derrick Bell’s interest convergence thesis described the second pattern, in which a subordinated group’s gains arrive when they happen to serve a dominant group’s interests.

What follows differs from all of them in one respect. Those accounts have a claimant: someone choosing a room, or a group whose advancement is at stake and who knows it. Here there is neither. The people protected did not select the venue, did not appear in it, do not know it exists, and cannot tell whether anyone inside was arguing on their behalf or merely arriving at a result they happen to like.

Three properties make that possible.

The room is not chosen by the person with the grievance. It is fixed by the structure of the decision, long before anyone objects. Anger about a data center is one thing; whether it is heard as a land use question, a rate question or a political question is four different proceedings with four different sets of people in them, and only some of them decide anything.

Volume and effect come apart. A grievance can be extremely loud in a venue that decides nothing and completely silent in the venue that decides everything, and there is no mechanism that corrects this. Loudness does not migrate. A hundred people at a county meeting do not become a filing.

And the room’s occupants are whoever the room’s rules admit, which need not include anyone who shares the grievance. When a favorable result comes out of such a room, the household has not won an argument. It has received a byproduct.

The Fight You Can See Is Losing

Look at what the visible fight has produced, on its own terms.

Its loudest demand was that the buildout stop, and measured against that demand it has mostly lost. Fourteen state legislatures introduced bills restricting new data center construction, and as of July 2026 none had been signed into law. [2] Maine came closest: LD 307 would have paused projects above twenty megawatts until late 2027, passed both chambers, and was vetoed by Governor Janet Mills in April 2026 over its effect on a single redevelopment project at a shuttered paper mill in Jay. [3] New York’s legislature passed a twenty megawatt moratorium the governor has still not signed; instead, on July 14, Kathy Hochul issued Executive Order 62, pausing state environmental permits for facilities of fifty megawatts or more for up to a year while agencies write standards. [4] That is the first statewide moratorium in the country, and it arrived as an executive workaround around the legislature’s own bill.

Local action has done better. Seattle adopted a one-year moratorium on projects above twenty megawatts in June, the largest American city to do so. Monterey Park, California banned them outright by ballot measure with eighty-eight percent of the vote. Texas paused new approvals in August pending audits by its utility commission and grid operator. [5] But a local moratorium more often relocates a project than prevents one, and developers have generally been willing to relocate.

Stopping the buildout was never the only demand. Plenty of the same organizing asked for cost protection, water limits, noise standards or a community benefits agreement, and some of that succeeded locally. But the demand that carried nationally, the one the polling measures and the memo is frightened of, was the categorical one.

Measured as electoral pressure, the fight has done considerably better. On November 4, 2025, Democrats Alicia Johnson and Peter Hubbard unseated Republican incumbents Tim Echols and Fitz Johnson on the Georgia Public Service Commission, each by roughly fifty-nine to forty-one, after the commission approved six Georgia Power rate increases since 2023 that added an estimated five hundred dollars a year to the average household bill. They were the first Democrats to win a non-federal statewide race in Georgia in nearly two decades. [6]

The polling is not ambiguous about direction, though the instruments disagree about magnitude in a way worth noticing. A Fox News survey of registered voters conducted July 17–20, 2026 found seventy percent opposed to a data center being built in their area and thirty percent in favor, with nearly eight in ten preferring slower construction to speed. Gallup found seventy-one percent opposed in March, forty-eight percent strongly. A POLITICO poll conducted by Public First the same month as the Fox survey found forty-one percent would oppose a data center within three miles of their home, up from twenty-eight percent in January. [7] [8] [9] Two of those instruments ask nearly the same question in the same month and come back thirty points apart, which should make anyone careful about the exact number and confident about the trend.

So the anger is real and seats are really changing hands. What has not happened is the thing the anger asked for. Construction continues. Announced capacity keeps arriving. The visible fight has generated an enormous quantity of political consequence and a very small quantity of stopped concrete.

The Fight You Can’t See Is Winning

Now look at the venue nobody is watching.

A tariff, in American utility practice, is the filed schedule of rates and conditions under which a utility serves a class of customer. Your residential rate is a tariff. It is approved by a state commission in a proceeding with a docket number, a hearing examiner, utility counsel, a consumer advocate and whatever intervenors bother to appear. A large load tariff creates a new class for very big customers, usually defined by a megawatt threshold, and attaches conditions to membership in it.

As of July 2026, twenty-four states had approved at least one, with six more pending. [10] A separate tracker counting individual utility tariffs rather than states put it at fifty-one approved and twenty-six proposed across thirty-six states as of March. [11] The conditions are unglamorous and substantial. Minimum take commitments, so a customer pays for contracted capacity whether or not it draws power. Upfront capital contributions, so the transmission a project requires is funded by that project rather than folded into a rate base and recovered from everyone. Collateral against abandonment. Exit fees. Curtailment obligations that put the data center’s load ahead of the household’s in a grid emergency rather than behind it.

New Jersey’s version, signed in July 2026, is the clearest to state. It directs the Board of Public Utilities to define a large data center customer for each utility at a threshold no greater than fifty megawatts, aggregating commonly owned or contiguous facilities so a campus cannot be subdivided below the line, and requires financial guarantees that the customer pay for at least eighty-five percent of the service it requests for at least ten years, plus upfront deposits toward new transmission. [12] Virginia’s GS-5 tariff requires data centers above twenty-five megawatts to sign fourteen-year contracts and post collateral of $1.5 million per megawatt. [11] Virginia also kept its sales tax exemption and added a consumption tax of $0.011 per kilowatt-hour on data center electricity, capped at $600 million a year, effective July 1, 2026. [13]

None of this stops a data center. All of it bears on who carries the cost of one.

Three qualifications, all of which matter later. The designs vary widely across twenty-four states and some will not do much. An approved tariff is a rule, not a result: New Jersey’s statute sets standards and then gives utilities a hundred and eighty days to apply them, so the state that got the most coverage has enacted an instruction to build the thing rather than the thing. And several of these are temporary by construction. Virginia’s consumption tax expires June 30, 2028 unless the legislature extends it. [13]

Still: in about half the country, the household holding a utility bill has been answered on the one part of its objection a commission is competent to hear. It did not ask in that language. It mostly does not know the answer was given.

Nobody Did This for You

The natural reading of the last section is that public anger got translated. That it entered the rate proceedings as cost causation and won there in a form it would not recognize. That reading is attractive, and it is wrong.

The template is specific enough to trace. A minimum billing obligation around eighty-five percent, a long contract term, collateral, an exit fee, a megawatt threshold for membership in the class. It did not originate in 2026 and it did not originate with anybody’s constituents.

In March 2023, AEP Ohio stopped connecting new data centers in central Ohio, a moratorium it imposed on its own authority without asking the commission first. In May 2024 it applied to the Public Utilities Commission of Ohio for a dedicated data center tariff covering customers above twenty-five megawatts. The proceeding ran fourteen months and was contested; more than a dozen parties filed testimony and the hyperscalers opposed it. On July 9, 2025, the commission approved a stipulation joined by AEP Ohio, commission staff, the Ohio Consumers’ Counsel, the Ohio Energy Group, the Ohio Manufacturers’ Association Energy Group and Industrial Energy Users-Ohio. It required qualifying data centers to pay for at least eighty-five percent of subscribed load for up to twelve years, with creditworthiness requirements, collateral and exit fees. In the same order, the commission directed AEP Ohio to file the tariffs and to lift its moratorium and connect the load as soon as possible. [14]

That last clause tells you what the utility was after. Not to stop anything. To be paid for infrastructure it would otherwise be left holding if the forecast proved optimistic, and then to build. The commission’s stated ground was cost causation: costs should be borne by the customers who cause them, a principle it would have applied in an empty room, because applying it is the job.

Now compare dates. Ohio moratorium, March 2023. Application, May 2024. Order, July 2025. The Georgia commissioners lost in November 2025. The moratorium bills, the local bans, the polling collapse, the party memo and the New Jersey statute are all 2026. New Jersey’s eighty-five percent guarantee for ten years closely resembles the Ohio settlement, arriving two years downstream. That is similarity and sequence rather than proven descent, the same design could be arrived at independently by any utility facing the same stranded-asset problem, which is itself the point but the recurrence of the same package makes the resemblance worth noticing.

So the mechanism was not translation. Nobody restated the public’s objection in the commission’s language, and the commission did not hear a muffled version of the county meeting. A utility protecting its own balance sheet drafted a rule whose incidental effect was to protect households. A consumer counsel and some industrial users improved it at the margins, which is the closest thing to representation in the story and is real, though the Ohio Consumers’ Counsel is a statutory office, not a movement. A commission approved it on grounds internal to ratemaking. Then it spread, because a design that has survived a contested proceeding somewhere is the cheapest thing for a commission elsewhere to adopt, and because every utility faces the identical problem.

Be precise about what the chronology does and does not establish. It does not show that no public concern touched the Ohio proceeding; Ohio ratepayers were already unhappy, and the Consumers’ Counsel was in the room. What it shows is that the design preceded the national political wave by two years, and that the wave is now producing legislation which codifies rules the utilities had already written. The politics is downstream of the design.

This is a better outcome than most public fights get, and it should be said plainly: the protection is real, it is spreading, and households are better off for it. But it is not a victory in the sense the word usually carries. It is an alignment. For as long as the utility’s interest in avoiding stranded assets points the same direction as the household’s interest in not paying for them, the household is protected by a rule it did not ask for, drafted by a party that does not represent it, on a rationale that has nothing to do with fairness.

Alignments end. The one that produced this depends on a specific set of facts: uncertain load forecasts, expensive transmission, and a real risk that announced capacity never arrives. Change any of those and the incentive inverts. A utility confident the load is coming has every reason to build first and allocate later. The exemptions are already visible in the original, AEP Ohio’s tariff grandfathered existing data centers and expansions below the threshold and every exemption is a place where the alignment has already stopped holding.

Someone Tried to Move the Room

Which brings us to the part of this year that received almost no coverage at all.

In October 2025 the Secretary of Energy invoked Section 403 of the Department of Energy Organization Act, a rarely used authority, to direct the Federal Energy Regulatory Commission to open a rulemaking on the interconnection of large loads to the interstate transmission system, generally those above twenty megawatts, and to take final action by April 30, 2026. [15] Load interconnection had been a state and local matter for as long as there has been a division of authority over the grid.

Read as a venue question rather than an engineering one, that is an attempt to redraw the boundary around a decision. Be precise about what it was not: the proposal explicitly disclaimed reach over retail sales, local distribution, siting, behind-the-meter arrangements and loads under twenty megawatts. [15]

The state commissioners did not treat the disclaimer as dispositive. On November 11, 2025 their association, NARUC, passed a resolution urging the Commission to preserve state retail authority. Ten days later it filed comments arguing that FERC had never asserted jurisdiction over end-user load interconnections, that doing so falls outside the Federal Power Act’s boundaries, and that states set rates across customer classes. The filing also said something that will matter at the end of this essay. The venue for a retail customer affected by the service, NARUC wrote, is a state commission, and those customers get recourse by participating in state proceedings and through electing or influencing the appointment of state regulators. [16] Roughly a hundred and fifty parties filed initial comments. [17]

On June 18, 2026, the Commission did not decline federal action. It declined the single nationwide rule the directive had sketched. Instead it issued six orders to show cause under Section 206 of the Federal Power Act, one to each jurisdictional grid operator and its transmission owners, each preliminarily finding the region’s tariff unjust and unreasonable for want of clear provisions on large load integration, and each giving sixty days to justify the status quo or propose revisions. [18]

The federal venue expanded. The boundary was restated. The Commission took up transmission service to large loads, the study processes behind it, and the network upgrade costs that enter wholesale rates, and it said the orders were not intended to intrude on state authority, framing its action as reaching Commission-jurisdictional transmission service and transmission cost shifting while leaving retail customer protection to state regulators. [19]

Be careful how much that settles. These are preliminary findings opening proceedings, not a final rule and not an adjudication of the jurisdictional line. The accurate description is narrower than “the room held” and more interesting: the first federal attempt to redraw the boundary ended by formally restating the boundary it had put under pressure. That restatement is now on the record, in an order, available to be cited by the next state commission that needs it. It is also only that.

Notice how it was produced. By an association of state regulators filing comments in a federal docket, against a cabinet secretary, on a record most of the affected population will never know existed. Members of the public could have filed; the Commission maintains an office whose purpose is to help them do it. There was no rally, no referendum, no public event around which a constituency could form. The defense was conducted by the professionals already in the room, which was fortunate, because nobody else was coming.

The venue was contested within eleven months of producing the result, at the level of jurisdiction rather than merits, and it survived on a legal boundary and the people paid to notice it. Next time the drafting may be better, or the Commission differently composed.

$200 Million in the Wrong Room

It would be tidy to say the money is aimed at this. It is not.

The two largest AI super PAC networks have raised more than two hundred million dollars between them this cycle. Leading the Future, funded by OpenAI president Greg Brockman, Andreessen Horowitz and others, has raised about $140 million. Public First Action, launched in February with a $20 million donation from Anthropic that its spokesman says is restricted to public education rather than political spending, had raised $80 million by the end of June. Together the two put at least $44 million into forty House and Senate candidates in the first half of the year. [20] [21] Almost none of the advertising, on either side, mentions artificial intelligence at all. [22]

Nor do they want the same statute. Leading the Future argues for a national standard against a state patchwork; Public First Action was created to oppose federal efforts to freeze state progress on AI oversight. [21] The disagreement is real, not staged.

And AI governance is a different question from the one that decided your exposure to the buildout. The December 2025 executive order that launched the preemption campaign excludes from its proposed preemption, in terms, AI compute and data center infrastructure, other than generally applicable permitting reforms alongside child safety and state procurement. [23] The White House framework built on it, released March 20, 2026, preserves state police powers, and on ratepayers runs the other way entirely: it advises Congress to ensure that residential ratepayers do not face higher electricity costs as a result of new AI data center construction and operation. [24] The rate class survived the year not because the preemption campaign spared it. The campaign was never pointed at it, and on this question is nominally on the same side.

So the mismatch worth naming is not one of money. Campaign spending and docket intervention are not substitutable goods, and no quantity of the first buys a better outcome in the second. The mismatch is one of attention. Two hundred million dollars is contesting the governance of a technology in the one venue where a constituency can be assembled and rewarded. The question of who pays for the physical plant that technology runs on was divided between federal and state authority in a docket where no constituency existed at all.

The Win With No Owner

The problem with an unowned victory is not that it is unfair. It is that nobody will fight for it.

A movement can defend what it built. It cannot defend what it did not build and does not know it has. Test that operationally rather than rhetorically. No recognizable electoral constituency has formed around large load cost allocation. Few candidates run on it and no incumbent has lost a seat for weakening it. And no mass-membership organization exists to make one answer for it, consumer counsels and regulators’ associations do this work, but they are statutory offices and professional bodies, not a base that can be mobilized or disappointed. Three absences, each checkable.

The visible fight, if it goes well, makes this worse rather than better. A candidate wins in November on data center anger. Georgia’s commissioners are already gone. The sensation is that something was accomplished, and something was: seats changed hands over a real grievance, which is how the system is supposed to work. But the part of the grievance that got answered was answered elsewhere, by other people, for other reasons, and it can be unwound the same way. A public that believes it has already won is the least likely public to notice.

Four things follow. They are shaped like the venue rather than like a movement, because a movement is what the venue cannot produce.

Put the retail boundary in statute. The line held this June on a commission’s characterization of its own intent, in preliminary orders, which is the weakest durable form a legal boundary takes. It survives the current commission’s composition and the current drafting. A statutory reservation of retail cost allocation to the states is a specific ask, addressed to a specific committee, and it is the only version of this year’s outcome that does not have to be re-won.

Fund a standing intervenor whose mandate is the boundary. Consumer advocates already appear in these dockets and did real work in Ohio. Their mandate is rates. Nobody’s mandate is the jurisdictional line as such, which is why its defense in June depended on an association of regulators having the institutional reflex to file. A reflex is not a plan.

Send the anger to the appointment. This is the one place where volume genuinely migrates, and it is not my suggestion. It is NARUC’s, on the record in the federal docket: retail customers get recourse through state proceedings and through electing or influencing the appointment of state regulators. [16] Roughly ten states elect their commissioners. The rest appoint them, usually by a governor, usually with almost no public attention on the choice. Anger cannot file a brief. It can decide who sits on the commission, and Georgia has already demonstrated that it will.

Watch three specific things, since no public can monitor a docket. A special contract regime, in which individual projects negotiate around the tariff class rather than joining it, which turns a rule into a default. Erosion of thresholds and grandfathering, which was in the original Ohio design and is where every subsequent exemption will be argued from. And a federal definition of large load broad enough that retail cost allocation gets settled at the wholesale level before any state gets to argue it, which is the drafting error the next attempt will not make.

The party memo saw the shape of this correctly, which is why it is worth reading twice. It is frightened about Ohio. It is not frightened about anything upstream of Ohio, and it does not spend a sentence pretending to be.

You are winning a fight you do not know you are having, and you are winning it because someone else’s interest currently points your way. It does not appear on any ballot, and it was defended this year by an association of the people who do this for a living. Nobody defends a victory that was never theirs.

Notes

[1] Alex Isenstadt, “Exclusive: GOP warns AI companies that data centers are politically radioactive,” Axios, August 19, 2026 (Maria Curi contributing). The memo, headlined “Ohio Data Center Risk,” was obtained by Axios and posted in full by the reporter. Quoted phrases are the memo’s own language as reported. The comparison of data centers to spent nuclear waste is attributed in the same piece to internal party and industry polling described to Axios.

[2] NBC News, “New York to impose the country’s first statewide moratorium on data centers,” July 14, 2026; CNBC, “New York becomes first U.S. state to impose AI data center ban,” July 14, 2026, reporting that fourteen state legislatures had introduced bills restricting new data center construction and that none had been signed into law.

[3] Maine LD 307, passed by both chambers and vetoed by Governor Janet Mills in April 2026. The bill would have paused projects requiring at least 20 MW until late 2027. The veto cited the effect on a redevelopment project at the former International Paper mill in Jay.

[4] New York Executive Order No. 62 (July 14, 2026), imposing a one-year moratorium on discretionary state environmental permits for data centers of 50 MW or more and directing the Department of Public Service to examine large load interconnection in Case 26-E-0045. The legislature’s Responsible Data Center Development Act, containing a 20 MW moratorium, had not been signed as of the order.

[5] Seattle one-year moratorium on projects of 20 MW or more, adopted June 9, 2026; Monterey Park, California ballot measure prohibiting data centers citywide, approved by 88 percent of voters June 2, 2026; Texas pause on new approvals pending Public Utility Commission of Texas and ERCOT audits, announced August 2026.

[6] Georgia Public Service Commission special elections, November 4, 2025. Alicia Johnson (D) defeated incumbent Tim Echols (R) in District 2 and Peter Hubbard (D) defeated incumbent Fitz Johnson (R) in District 3, each with about 59 percent. Reported by the Atlanta Journal-Constitution, WABE and Axios Atlanta. The six Georgia Power rate increases since 2023 and the roughly $500 annual household increase are from AJC reporting.

[7] Fox News poll conducted July 17–20, 2026 by Beacon Research (D) and Shaw & Company Research (R); 1,003 registered voters; margin of error ±3 points. Asked whether respondents would favor or oppose the building of a data center in their area to support AI technology: 30 percent favor, 70 percent oppose. Nearly eight in ten preferred slower construction to rapid buildout.

[8] Gallup, “Americans Oppose AI Data Centers in Their Area,” May 2026, reporting a March survey in which 71 percent opposed construction in their area and 48 percent were strongly opposed.

[9] POLITICO poll conducted by Public First, July 2026: 41 percent would oppose a data center built within three miles of their home, up from 28 percent in January.

[10] Edison Electric Institute, “Large Load Projects and Tariffs,” updated July 2026: as of July 2026, 24 states have approved at least one large load tariff and another 6 have pending large load tariffs.

[11] Smart Electric Power Alliance, Database of Emerging Large Load Tariffs (DELTa), March 31, 2026 update: 51 approved and 26 proposed tariffs and service rules across 36 states and 60 utilities. Virginia GS-5 terms (25 MW threshold, 14-year contracts, $1.5 million per megawatt collateral) as reported in contemporaneous trade analysis of the Virginia State Corporation Commission’s approval.

[12] New Jersey A796/S731, signed July 2026. The enacted text directs the Board of Public Utilities to identify the defining characteristics of a “large data center” and “large data center customer” for each electric public utility, including a minimum megawatt size designation “which shall not be greater than 50 megawatts,” and to aggregate the peak monthly demand of data centers under common ownership or control, on the same or contiguous sites, or sharing substantial physical, operational or interconnection infrastructure, treating them as a single large data center. Utilities apply the rules within 180 days of the board’s order, to new and existing facilities at or above the threshold. The statute requires financial guarantees that a large data center customer will pay for at least 85 percent of the service it requests for not less than 10 years, and upfront deposits toward new transmission.

[13] Virginia HB 30, the 2026–2028 biennial budget, passed June 22 and signed by Governor Abigail Spanberger on June 30, 2026. It imposes a data center electricity consumption tax of $0.011 per kilowatt-hour effective July 1, 2026 through June 30, 2028, capped at $600 million annually with pro rata refunds above the cap, and preserves the existing data center retail sales and use tax exemption under Va. Code § 58.1-609.3(18).

[14] Public Utilities Commission of Ohio, Case No. 24-0508-EL-ATA. AEP Ohio applied on May 14, 2024. On July 9, 2025 the commission adopted a stipulation joined by AEP Ohio, commission staff, the Ohio Consumers’ Counsel, the Ohio Energy Group, the Ohio Manufacturers’ Association Energy Group and Industrial Energy Users-Ohio, requiring qualifying customers above 25 MW to pay for a minimum of 85 percent of subscribed load for up to 12 years with creditworthiness, collateral and exit-fee provisions, and directing AEP Ohio to file the tariffs and lift its connection moratorium. The moratorium had been imposed unilaterally in March 2023. Existing data centers and expansions below the threshold were grandfathered.

[15] Interconnection of Large Loads to the Interstate Transmission System, Advance Notice of Proposed Rulemaking, FERC Docket No. RM26-4-000, transmitted October 23, 2025 with the letter of Secretary of Energy Chris Wright under Section 403 of the Department of Energy Organization Act, 42 U.S.C. § 7173, requesting final action by April 30, 2026. The ANOPR addressed loads generally above 20 MW and disclaimed federal authority over retail sales, local distribution, siting, behind-the-meter and intrastate arrangements, and loads under 20 MW.

[16] National Association of Regulatory Utility Commissioners, Resolution Urging the Federal Energy Regulatory Commission to Preserve and Affirm State Retail Regulatory Jurisdiction in Its Large Load Interconnection Proceeding (passed by the Committee on Electricity Nov. 10, 2025; adopted by the NARUC Board of Directors Nov. 11, 2025). See also Initial Comments of the National Association of Regulatory Utility Commissioners, FERC Docket No. RM26-4-000, at 3–6 (filed Nov. 21, 2025), Accession No. 20251121-5132. The comments state that FERC has never asserted jurisdiction over end-user load interconnections and that doing so would exceed the boundaries imposed by the Federal Power Act. They further explain that “the venue for a retail end-use customer who is directly affected by the services provided by the electric supplier is a state commission,” with recourse available “by directly participating in state proceedings and through the process of electing or influencing the appointment of state regulators.”

[17] More than 150 initial comments were filed in Docket No. RM26-4-000, as reported in contemporaneous trade coverage.

[18] Six orders to show cause under Section 206 of the Federal Power Act, all issued June 18, 2026 by unanimous vote: PJM Interconnection, L.L.C., 195 FERC ¶ 61,211 (2026) (Docket No. EL26-67); Midcontinent Indep. Sys. Operator, Inc., 195 FERC ¶ 61,212 (2026) (Docket No. EL26-70); Sw. Power Pool, Inc., 195 FERC ¶ 61,213 (2026) (Docket No. EL26-68); Cal. Indep. Sys. Operator Corp., 195 FERC ¶ 61,214 (2026) (Docket No. EL26-71); ISO New England Inc., 195 FERC ¶ 61,215 (2026) (Docket No. EL26-72); N.Y. Indep. Sys. Operator, Inc., 195 FERC ¶ 61,216 (2026) (Docket No. EL26-69). Each makes preliminary findings that the tariff appears unjust and unreasonable with respect to large load integration and sets a 60-day response deadline. The intervention deadline was July 9, 2026; replies to the show cause responses were due September 16, 2026. See also Interconnection of Large Loads to the Interstate Transmission Sys., 195 FERC ¶ 61,045 (2026) (order regarding intent to act).

[19] The Commission stated that the orders were not intended to intrude on state authority, framing its action as directed to Commission-jurisdictional transmission service and transmission cost shifting while leaving retail customer protections to state regulators, and indicating that the findings were not intended to preempt state large load tariffs. See PJM Interconnection, L.L.C., 195 FERC ¶ 61,211 (2026).

[20] CNBC, “What AI companies want for the millions they’re spending on elections,” July 9, 2026: Leading the Future raised $125 million by the end of 2025 and spent more than $24 million on primaries through the end of June; Public First Action raised $80 million through the end of June and spent $20 million, including a $20 million donation from Anthropic that a PAC spokesman describes as restricted to educating the public on AI policy rather than political purposes; the two PACs together put at least $44 million into 40 House and Senate candidates through the end of June. Reuters reporting in August 2026 put Leading the Future’s total raised at about $140 million, with funders including Greg Brockman, Andreessen Horowitz, Joe Lonsdale, Ron Conway and Perplexity.

[21] NPR, “Groups tied to OpenAI and Anthropic are spending big on the midterms,” June 22, 2026, describing Leading the Future’s opposition to stricter regulation and Public First Action’s stated purpose of opposing federal efforts to freeze state progress. Anthropic’s $20 million contribution was announced in February 2026.

[22] Ben Kamisar, “Ads funded by AI industry are flooding the 2026 election. They’re about everything except AI.,” NBC News, February 27, 2026.

[23] Executive Order 14365, “Ensuring a National Policy Framework for Artificial Intelligence,” December 11, 2025. The order directs that the recommended federal framework not propose preemption of state laws concerning child safety protections, AI compute and data center infrastructure other than generally applicable permitting reforms, state government procurement and use of AI, and other topics to be determined.

[24] The White House, National Policy Framework for Artificial Intelligence, released March 20, 2026 pursuant to Executive Order 14365. It calls on Congress to preempt state AI laws that impose undue burdens while preserving state police powers, and advises Congress to ensure that residential ratepayers do not experience increased electricity costs as a result of new AI data center construction and operation.