Deep Dive Sep 19, 2026

Data-center opposition has reached the US midterms, and it lands on governors, not Congress. Most candidates want data centers to bring and pay for their own power, by MultiState's reading. Our call: for developers who generate power on site, that demand is a cost, not a barrier

Key takeaways

Opposition to US data centers did reach the election this year. Where it lands is the governor's executive order, not federal law. By the reading of MultiState, a firm that tracks state-level policy for companies, most candidates in the governors' races want data centers to carry the cost of power, and an outright building ban is the minority position. That is MultiState's qualitative reading. No candidate-by-candidate count has been published.

On September 11 this publication filed an internal call and did not run it, because it rested on a single sample. The call said that for the past year US data-center resistance had lived at the level of local permits and governors' executive orders. That is a level where a developer can negotiate case by case, where a new administration can revoke, and where community-benefit payments and tax agreements do the work. On August 20 a Republican candidate for the US Senate in Michigan backed a one-year moratorium on new data centers, and the call read that as the first time the resistance had taken the form of a federal campaign pledge. From there the call inferred that once that form spread, both of the developers' standing tools would fail at once. One tool is case-by-case negotiation with local government. The other is building generation on the data-center site, which the industry calls behind-the-meter and which mostly means gas turbines, so that the project does not wait in line for the grid operator's interconnection approval. This piece takes that call apart and re-tests it. The re-test reaches three conclusions, set out in the next three paragraphs.

The correction comes first. Neither "first" nor "federal" holds. The candidate, Mike Rogers, opened by saying he does not support a federal ban. What he asked for was a one-year pause decided at the state level, in Michigan. A federal moratorium bill had existed since March 25, when Senator Sanders introduced one. By July 21 a news outlet had counted candidates for governor in at least twelve states supporting a pause. What happened for the first time on August 20 is narrower. In the coverage reviewed, a Republican Senate nominee in a swing state reversed position to support a pause, days after the National Republican Senatorial Committee had circulated an internal warning. The NRSC is the party body that runs Senate campaigns for Republican candidates.

The election delivers the resistance to the governor's office, and the governor's instrument for keeping the promise is still the executive order. Data-center land use and permits are issued by state and local government. A US senator votes in Congress and has no power to approve or forbid a specific campus, so a Senate candidate's pledge cannot be kept from that seat. A candidate for governor can keep it, and the most specific pledge found promises an executive order on the first day in office. The original call said the resistance had changed channels. The reading here is that the channel has not changed: policy still lands as a governor's executive order, and the next governor can revoke it. What changes is who gets to sign, which the election now decides, and on a known calendar: votes counted on November 3, inauguration in January. This piece says an official 'holds the pen' when, once elected, that official can sign or veto a ban.

Does this close off on-site generation as a way around the grid queue? In most cases no, and the campaign is pushing projects toward it. By MultiState's reading, the majority position among candidates is that data centers should bring power and pay the full electricity cost, meaning self-supplied generation or the full cost of grid expansion. The developers who can pay that cost are the ones already building generation on site. This is a directional call. This piece holds no cost comparison between on-site generation and paying the full cost of grid expansion. The only thing on-site generation cannot get around is a blanket ban on construction, and in the campaign the blanket ban is the minority position.

If only this part gets read: when the next candidate comes out for a data-center ban, two questions sort it. Whether the office the candidate is running for can sign that ban once won. Whether the demand is a ban, or a requirement that data centers bring power and pay for it. A "no" to the first means a campaign position in most cases. A "bring power" answer to the second is not bad news for developers who generate on site.

The site has no evolution map for this topic yet, so this piece carries no chart.

One: dates first. What the election already held before August 20

The original call rested on one sample. On August 20 a political reporter at the Detroit News reported that Mike Rogers, the Republican candidate for the US Senate, supported a one-year moratorium on new data centers. The Detroit News original is paywalled. The text read for this piece is the same article as syndicated on Yahoo, checked line by line against the same-day report by E&E News, a trade outlet that covers energy and environment policy. Rogers' first sentence to the Detroit News was that he does not support a federal ban, and that Michigan needs stronger guardrails.

He then listed four reasons: community say, electricity bills, water, and favors to insiders. Only after that did he propose a one-year pause on new data centers, to last until Michigan sets up an approval process that puts local communities first. E&E News put it more bluntly. What Rogers supports, in that outlet's account, is a one-year Michigan pause approved by state decision-makers, not a national ban.

Here is Rogers's statement placed on the timeline:

Date	Event	Level
March 25	Senator Sanders and Representative Ocasio-Cortez announce an AI data-center moratorium bill. Sanders says a moratorium is needed at the federal level	Federal bill
June 24	The House version is formally introduced, with nine cosponsors, all Democrats	Federal bill
July 14	New York's governor reportedly signs a pause	Governor's executive order
July 21	Newsweek reports that candidates for governor in at least twelve states support an AI data-center pause	Governors' races
August 3	Texas's governor reportedly orders an audit of every data center in the state's interconnection queue	Governor's executive order
August 8	Michigan's Democratic Senate candidate El-Sayed says no further data center should be approved until federal guardrails exist. Ohio's Republican candidate for governor, Ramaswamy, publishes a pledge demanding legislation, with a statewide pause until it passes	Senate and governors' races
August 18	Pennsylvania's governor reportedly signs an executive order requiring local community consent. The same week, the NRSC circulates an internal memo calling data centers a sleeper issue for the whole midterm	Governor's executive order; party machine
August 20	Rogers reverses course and backs a Michigan pause	Senate race; state-level policy

Against this table, "the first federal campaign pledge" in the original call needs three corrections. The federal position had existed as a bill since March. Pause pledges had appeared in governors' races in a dozen states by July. And Rogers said in plain terms that he wanted state policy and rejected a federal ban. Even inside his own race he was not first: his Democratic opponent, El-Sayed, had voiced a conditional pause twelve days earlier.

So what was actually new on August 20? The answer is narrower than the original call, and still informative. In the coverage reviewed, Rogers is the first Republican Senate nominee in a swing state, meaning a state either party could win, to reverse toward supporting a pause. The timing follows the memo from his own party's Senate campaign committee. The memo warned that unless voters' perception of data centers was corrected quickly, the movement against them would spread well beyond Ohio. An investigative outlet also reported that Rogers changed position days after that outlet disclosed AI-related holdings in his portfolio. Only the summary of that report has been reviewed. Taken together, these say that the party machine has started to manage data centers as a campaign risk. That is a different thing from federal policy moving toward a ban.

Two: a Senate seat cannot sign a data-center ban. Whether a pledge can be kept depends on the office

The second pillar of the original call was that a campaign pledge is harder to revoke than an administrative action, because a pledge runs for a term and under public scrutiny. When that sentence was filed, no evidence backed it. This piece went looking. The result: the sentence holds for some candidates and fails for others, and the dividing line is the office.

Political science has comparative research on pledge fulfillment. A 2017 cross-national study in the American Journal of Political Science reports in its abstract that in the United States, on average, more than sixty percent of the pledges made by the president's party are at least partly fulfilled. The same abstract says the side without executive power fulfills markedly fewer. The authors are Thomson, Royed, Naurin and colleagues. The study looks at governing parties at the national level and counts "at least partly fulfilled", a low bar. Only the abstract has been reviewed. Applying it to candidates for governor and senator is this piece's analogy, not the study's conclusion. Our read, by that analogy: whether a pledge becomes policy depends mainly on whether the person who made it has the power to act once elected.

Applied to this case:

Candidates for the US Senate: Michigan's data-center permits are issued by state and local authorities, and a senator holds no such pen. Rogers handed the matter to state decision-makers in his own answer. What he can do is take a position in the campaign and vote in the Senate if elected. Structurally, his pledge cannot be kept from the office he is running for.
The federal bill: the text of the House version says that construction or expansion of an AI data center may not begin or continue until Congress passes a set of laws. That set includes federal review and approval of AI products before they reach the market. Enforcement goes to the Secretary of Energy, who may tie compliance to future permits. The condition for lifting the pause is that Congress passes a set of AI laws, pre-market review among them, not that siting or power problems get solved. So the bill is about AI, and data centers are the lever. Its nine cosponsors are all Democrats. As of September 19, govinfo and Sanders's press releases show no committee vote. Some pages on the congressional website could not be loaded at the time. According to syndicated reporting, House Speaker Johnson called a federal ban a dangerous idea a few days before September 19. Miles Brundage, formerly head of policy research at OpenAI, wrote on X on August 23: "I think a national datacenter moratorium is a bad idea... I also don't think it'll happen".
Candidates for governor: these are the people with a pen. Among the dozen or so candidates for governor that Newsweek listed, Oklahoma's Democratic candidate Cyndi Munson was the most specific, pledging to sign an executive order imposing a pause on her first day in office. Maine is clearer still. The sitting governor vetoed a pause bill the legislature passed this year, and both parties' candidates to succeed her support a pause; the Democratic candidate, Pingree, has said publicly that she would sign. In Maine, what the election decides is exactly whether that veto gets reversed.

So "a pledge is harder to revoke" narrows to this: the sentence means something only when the person who said it holds the pen after the election, and the people who hold it are mainly the candidates for governor.

Three: a governor's fastest tool is still an executive order. The election decides who signs it, and from which day

Here a turn appears that the original call never handled. The fastest instrument a governor has for keeping the promise is the executive order. That is exactly the layer the call described as negotiable case by case and revocable at the next change of administration.

SemiAnalysis, a research firm covering semiconductors and data centers, put it plainly in a September 15 survey of moratoriums. New York's and Texas's governors "bypassed their legislatures to impose pauses they can ultimately control and revoke". Durable statewide limits remain much harder: "13 bills died in 2026, with almost none reaching a floor vote".

So the real change comes down to two concrete things.

First, the election decides who signs the order, and the signing date is known in advance. For the past year the governors' pauses a developer faced could land on any day: New York on July 14, Texas on August 3, Pennsylvania on August 18, none on a fixed date. Now, at least in states where a pro-pause candidate can win, the risk concentrates on a known calendar: votes counted on November 3, inauguration in January, then the 2027 legislative sessions. New Hampshire's sitting governor has reportedly said a multi-year pause will go into the budget, taking effect in July 2027 at the earliest. Only the headline and summary of that report have been seen.

Second, once the pen changes hands, revoking gets more expensive, but only for the person holding it. A governor elected on a first-day pause pledge who then withdraws the order faces the voters and allies who delivered the win. That is a far higher political price than for a governor who signed on short notice. On that point the original call was right. But the next governor can still revoke. Local pauses are revoked a different way, through developer lawsuits. The counter-examples found for this piece show them revoked faster. Hill County, Texas, passed a one-year pause in May by a 3-2 vote and lifted it on June 12, after a developer filed in federal court, replacing it with a disclosure checklist. Imperial County, California, reportedly had its pause stayed by a court order. The election slows revocation, but it does not turn an executive order into law. Becoming law takes the state legislature. SemiAnalysis counts thirteen such bills dead this year, and the one Maine's legislature passed was vetoed by the governor.

Four: the campaign mainstream is "bring power", and that is the door on-site generation passes most easily

The sharpest sentence in the original call was about the developers' way around interconnection approval. Generating on site skips the grid queue, the call said, but it cannot get past a ban aimed at the data center rather than at the power source. The logic was sound. The error was assuming that the campaign would produce that kind of ban. By MultiState's count, the campaign is mostly producing something else.

MultiState surveyed the governors' races on September 1. Candidates in thirty-six states are debating data-center policy, and MultiState's conclusion is that most candidates support data-center development but want stronger regulation, above all a requirement that facilities pay for their own energy infrastructure. That is MultiState's qualitative reading. No candidate-by-candidate count has been published. Thirty-six is the number of states where candidates are debating data-center policy; the earlier "at least twelve states with a candidate supporting a pause" is a different count, and the two cannot be subtracted from each other. Some named examples:

Texas Governor Abbott's guardrails: a data center may not take water the community needs, may not take power the grid needs, and must bring electricity prices down.
Pennsylvania Governor Shapiro's grid standard requires developers to build, bring or buy the full generating capacity their own load needs, and to pay the full cost of it.
Florida's Republican candidate for governor, Donalds, reportedly wants data centers barred from public power and water, and has not called for a pause.

Abbott's guardrails also cover water. On-site generation only solves the power half; the limit on cooling water still applies.

What these three share is that they fold the opposition into a cost rather than a ban. The developers who can pay that cost are the ones generating on site. This is a directional call; this piece holds no cost comparison between on-site generation and paying the full cost of grid expansion. SemiAnalysis read the Texas pause the same way. The interconnection queue stalls, "but is offset by an acceleration of behind-the-meter demand. For projects that never needed the grid, it means nothing at all, making it a net positive for BtM developers and on-site generation suppliers". The same firm forecast in June that from 2028 on-site generation would supply more than half of new US data-center capacity.

Two things on-site generation cannot get around:

A blanket ban that does not distinguish by power source. Rogers spoke of "new data centers" without separating out those with their own generation. The federal bill defines a data center by capacity, anything above twenty megawatts that meets a high-density or liquid-cooling condition, and likewise carves out no exemption for on-site power. Munson's line that Oklahoma's water and power belong to Oklahomans also points at a blanket pause.
The local politics of on-site generation. The on-site cases seen for this piece are mostly gas turbines. That trades a grid dispute for a dispute over air pollution and noise. One reported case: Musk's second supercomputer in Mississippi installed dozens of gas turbines without the required permits. One more escape hatch belongs on the record. SemiAnalysis documented a project that got out from under Brownsville's city pause by having its land taken outside the city limits. Once the land leaves the city's jurisdiction, the city's pause no longer reaches it.

So "both tools fail at once" has to be rewritten. Case-by-case negotiation really is getting harder, because the counterparty now includes candidates who need a public position. On-site generation becomes more valuable, not less, under the rules most campaigns are producing, and fails only in the minority of states with a blanket ban. The resistance has pushed construction toward places a ban cannot reach. It has also turned a grid dispute into a local fight over air pollution.

Five: the ballot payoff is uneven, and the industry has already moved money into the campaign

Public opinion points one way. Whether putting a pause in a platform wins votes is less clear. The evidence runs both ways, and most of it sits at headline or summary depth, not enough to settle a net direction. Developers should not use polls to forecast policy. The new thing in this section is that the industry is now spending directly on the election.

The first step in the original call's mechanism was that where public opinion exists, candidates price it into their positions and write it into platforms. The opinion half is clear. Gallup's March survey found seventy-one percent of Americans opposed to an AI data center being built near their home. A separate May poll with a sample of more than four thousand reportedly found seventy-five percent opposed; the polling organization is not named in the report seen.

Meanwhile, the industry has not stopped at local negotiation. TechCrunch reported on August 31 on Build American AI, a group funded by the venture firm a16z and OpenAI co-founder Brockman. The group plans to spend millions of dollars on advertising in selected states to sell voters on the benefits of data centers. Developers in fact have a third tool, and are already using it: spending money to move voters directly. The original call saw two tools failing and missed the third coming onto the field. The shape of the contest changes from a one-to-one negotiation between developer and local government, with community payments attached, into an open campaign with an opponent and an advertising budget. On that direction the call was right; what it underestimated was how fast the industry would turn.

But on "a platform pledge brings votes", the evidence is mixed:

Losses: South Carolina Representative Nancy Mace ran for governor on a state-level pause and lost the primary (E&E News). Wisconsin candidate for governor Francesca Hong ran on a pause and, by several outlets' headlines, lost the Democratic primary by less than a percentage point.
Wins: in Michigan's August 4 primary, by one climate outlet's count, at least fourteen candidates opposed to data centers won. Only the summary of that item has been seen.
The party machine's reading: the NRSC memo says that if the Ohio Senate race is lost over data centers, the industry will be blamed, and both parties will take note.

Six: so what to count

The original call proposed a leading indicator: how many states have a candidate who put data centers in the platform. That count has already hit its ceiling as a signal. Governors' races in thirty-six states are debating it, and at least twelve states have a candidate supporting a pause; pushing the count higher says nothing about who can actually make a pause happen. To carry information it has to split into two columns:

Column	The question	Why
Holds the pen	Can the office the candidate is running for sign or veto this ban (a governor can, a senator cannot)	Decides whether the pledge becomes policy
What is demanded	A blanket ban, or "bring power and pay the full cost"	Decides whether on-site generation still works as a way around

A candidate who fits both columns is a candidate for governor who has pledged to sign a blanket pause on taking office. That is the new risk to developers.

Pause pledges found for this piece (as of September 19; not a complete list)

State	Candidate	Pledge	Exemption for on-site generation stated?
Oklahoma	Cyndi Munson (Democrat)	Executive order imposing a pause on day one	None in the coverage reviewed
Ohio	Ramaswamy (Republican)	Demand legislation, with a statewide pause until it passes	None in the coverage reviewed
Maine	Both parties' candidates to succeed the governor (Democrat Pingree says she would sign)	Support a pause; the sitting governor vetoed the legislature's pause bill	None in the coverage reviewed

Munson comes closest to fitting both columns. The other states can be checked one by one after the November 3 count.

Our call

Data-center opposition reached the election this year, but it lands on the governor's office, not on Congress. The federal moratorium bill was introduced in March, candidates for governor in at least twelve states supported a pause by July, and on August 20 Mike Rogers, Michigan's Republican Senate candidate, said he does not support a federal ban. What the election ultimately delivers is still a revocable executive order. What changes is who signs it and on which day, with votes counted on November 3 and inauguration in January. By MultiState's reading, most candidates want data centers to supply power and pay the full cost. That is a qualitative reading, not a count. For developers generating on site, that demand is a cost, not a barrier. The only thing they cannot get around is a blanket ban, and blanket bans are the minority position. So the thing to watch is not how many states mention data centers. It is how many candidates for governor who pledged to sign a blanket pause win in November, and how many states write "bring power and pay the full cost" into rules. Developers' timelines move from the permit-by-permit schedule to the election calendar.

What this changes, and for whom

Data-center developers and site-selection teams: for new sites in the first half of 2027, the reference list becomes the governors elected in November who pledged to sign a pause on taking office. In states outside that list the risk is mostly a requirement that data centers pay for their own power, a cost question that an on-site generation plan clears. In states on the list the question is no longer how much more it costs, but whether the project can be built at all.
Generation equipment suppliers and gas-power investors: the campaign's mainstream position pushes demand toward on-site generation. Our read: as long as blanket bans stay in a minority of states, this election cycle is a positive for on-site generation demand. The reverse risk to watch is air-quality permitting for on-site gas turbines, which is where the next local dispute lands. If any state starts rejecting on-site gas projects in bulk on air permits, that positive reading comes down.
Lenders and infrastructure funds: Our read: local pauses do little to projects already approved, and short pauses often expire before a later-stage project needs its approval. In states with a change of governor, the item that enters loan terms is the new governor's first executive order, an event with a known date; one form is making that order's coverage of the project a condition for drawdown. In states where the sitting governor is re-elected or not up for election, a pause can still land on any day.
Policy and public-affairs teams: the room for one-to-one community-benefit negotiations with local government is shrinking, and the counterparty becomes candidates who need a public position. Industry groups are already buying campaign advertising. No evidence found for this piece shows that advertising works better than a concrete bring-power-and-pay-in-full commitment.

What would prove this wrong

1. A state passes a statewide ban by legislation, with no exemption for on-site generation. Then two sentences in this piece fail together: "what the election delivers is still a revocable executive order" and "a cost for on-site developers, not a barrier". The observation window, set by this publication, runs to June 30, 2027, covering the first legislative session after the new governors take office.

2. Any federal data-center moratorium bill reaches a committee vote. Then section two's reading that no committee-vote record was found, and the inference that a federal ban is not close, are withdrawn. Same window.

3. Every candidate for governor who pledged a pause loses in November, and no new governor signs a pause in the first half of 2027. Then the election route added no new risk for developers. This piece and the original call were both wrong in the same direction, overestimating the risk. Same window.

4. A governor elected on a pause pledge withdraws that pause within the first year in office. Then even "revocation gets more expensive for the person holding the pen" fails.

5. On-site generation projects in any state get blocked in bulk because of a campaign pledge rather than air permits. Then section four's "on-site generation becomes more valuable under the rules most campaigns are producing" has to be rewritten.

Written from the same research and judgments as the Traditional Chinese edition. Sources are linked; we distinguish original documents from reporting and mark what we could not verify.

Sources cited in other language editions

These sources are linked in the body of this piece's other language editions but not in this one.