On 13 October we covered Texas halting data-centre permits and Holyrood voting to park decisions on schemes over 50MW. Both were governments acting on behalf of voters. Three weeks later the voters acted for themselves. On 3 November, the US midterms put data centres on the ballot in roughly 40 localities across 13 states, and the result was about as one-sided as local democracy gets.
Local opposition to data-centre expansion is translating into electoral and policy setbacks across the US, as voters increasingly push back against new developments.
Public opposition is gaining ground: Of 33 measures restricting data-centre development, all had passed, with six more leading and just one defeated.
A rare setback: The only failed measure was a Harwood, North Dakota proposal requiring independent reviews and public hearings before approving large industrial projects. It was rejected by 75% of 430 voters.
A wider pattern: A separate tracker covering 46 measures across 10 states found all 33 restrictive measures passing, while 11 of 13 pro-data-centre measures lost.
Local resistance is widespread: Ballotpedia counted 50 local data-centre measures in 2026, with 41 decided on 3 November.
The political stakes are rising: Texas governor-elect Gina Hinojosa has pledged to halt all new data-centre approvals until the state legislature introduces protective measures.
This is important for anyone whose pipeline touches the sector, which in the US is now a large share of everything. ConstructConnect's report of 29 September put year-to-date data-centre starts at $98.5bn across 153 projects, 24% of all non-residential building starts, with an average project cost of $1.18bn.
The votes will not stop that money. Tom's Hardware notes that of 300-plus local bans enacted so far, only three developments totalling 2.3GW have been delayed. What the votes change is the terms on which the money gets spent, and that is a project manager's and quantity surveyor's problem.
What voters actually said yes to
The instruments were anything but uniform. Alhambra, California and Edgewater, Florida, voted on outright bans. Bloomington, Illinois, passed an advisory measure for a moratorium on large-scale facilities; Peru, Wells, and Wiscasset in Maine passed moratoria of their own.
El Dorado, Kansas banned data centres and battery storage together. Janesville, Wisconsin, now requires voter approval for any development above $450m on the former GM/JATCO site.
Five Illinois townships rejected advisory questions asking them to permit data centres at all. Piqua, Ohio, recalled its mayor and two council members; Yukon, Oklahoma, recalled its mayor.
There were exceptions. Wilmington, Ohio approved rezoning rural residential land for a data centre, and Granville, Ohio rejected a charter ban. Reason's summary is fair: almost nowhere did a pro-data-centre measure pass or an anti-data-centre measure fail, and in most places it was not close.
The practical reading is that permitting risk is now site-specific in a way it was not a year ago. A programme-level assumption that a county will approve because the neighbouring county did is no longer safe. Each site carries its own ballot history, its own recall risk and, increasingly, its own voter-approval trigger.
Meta says it will be harder, and describes the model it thinks works
Dina Powell McCormick, Meta's president, spoke at the WSJ Tech conference the day after the vote.
I think it's going to be harder. It has been hard, I should say, but I also think that it's becoming clear what models work, and the states want those models that work.
For a delivery team, that is the owner of five planned gigawatt-scale campuses telling you the approval path has lengthened and that the acceptable deal structure is narrowing.
She then described the structure. Successful Meta projects, in her account, involve Meta paying for its own electricity, directing tax surpluses to teachers and first responders, keeping the labour force local and securing bipartisan support. She expects future projects to face more review and to be weighed against whether constituents want them. Read that as a cost plan: local-labour clauses, ratepayer protections and community contributions are moving from goodwill to precondition, and they need a line in the estimate before the bid goes in.
OpenAI's compact: NDAs out, assurances and curtailment in
Two days after the election, OpenAI published a National Community Compact for campuses it will own and operate, alongside a policy blueprint it wants legislators to adopt.
The Neuron's reading of the documents is the clearest I have seen. The compact states: "We will not use non-disclosure agreements (NDAs) with public officials," with carve-outs for security, trade secrets and competitively sensitive information.
The Neuron notes OpenAI says it already voided NDAs in Effingham County, Georgia, in July, so the pledge predates the vote even if the election gives it urgency.
The commercial terms are part of what a QS should study. If these spread, expect to see them in employers' requirements and utility agreements:
The developer pays for the electricity its campus consumes and for the incremental grid infrastructure needed to serve it, rather than socialising those costs across other customers.
Financial assurances are posted before major utility spending begins so that other ratepayers are protected if the project is delayed, reduced or cancelled.
Public reporting covers water and electricity use, construction impacts and independent noise studies before operations begin and annually thereafter, with water disclosures separating initial cooling-system fills from ongoing use.
Curtailment is written into the power contract. TNW reports that under OpenAI's 25-year Grid Protectors deal at the 3.2GW Effingham campus, Georgia Power can cut up to 1,000MW at busy times, and OpenAI pays the full cost of the lines and service the campus needs.
Two caveats. Georgia Power expects deals with large users to save customers about $950m a year from 2029, but TNW notes OpenAI admits flexibility cannot replace new plants and lines, and no other sites have been named.
The Neuron is right that the blueprint is a proposal, not law, and the compact does not automatically bind partner-operated facilities that supply OpenAI with compute. A curtailment clause also has a construction consequence: on-site generation or batteries become scope, and a 1,000MW swing in available power is a programme and commissioning risk, not a footnote.
Texas, and what Holyrood can take from it
Texas is the sharpest case because the state had already moved before the vote. Reuters, in a report carried by The Star, recounts that Greg Abbott halted new grid connections on 3 August and ordered an audit, with grid operator ERCOT tracking more than 400GW of data-centre connection requests at the time. Hinojosa's pledge goes further.
It's a brand-new day in Texas for a ton of what was probably considered settled policy -- not just for data centers, but energy as a whole.
He expects some projects to be delayed or to leave the state, and others to accelerate to energise before new rules land. If you are on a Texas scheme, that acceleration pressure is coming your way.
Brittany Kaiser, chief executive of developer Alpha Compute, told Reuters that developers without signed contracts and financing are more likely to look elsewhere, including Europe. That is the read-across for UK readers. Holyrood's pause on schemes over 50 MW is the same instinct as Harwood's failed review amendment and Janesville's $450m trigger: give the community a formal say before the fence goes up.
The US evidence is that once the say is granted, it is used and used against. The lesson for a UK scheme is to assume the pause becomes a condition and to arrive with the ratepayer protections, local-labour terms and curtailment flexibility already costed rather than conceding them under pressure in a committee room.
Takeaway
The 3 November results did not end the US data-centre boom. ConstructConnect's $98.5bn of starts is already in the ground, and only a handful of gigawatts have been delayed by bans so far. What ended is the version of the pipeline where a developer negotiated behind an NDA, had the utility socialise grid costs and expected planning to follow.
The replacement is visible in Meta's and OpenAI's own words: pay for your power, post assurances, report publicly, accept curtailment and keep labour local. For a PM or QS, those are not reputational niceties. They are scope, risk and cost, and they belong in the bid before the vote, not after it.
If you want this kind of reading of the data-centre and grid story every week, with the numbers checked and the hype filtered out, the Project Flux newsletter is at projectflux.ai.
Links and Stuff
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
