Data centers lost 46 of 47 local ballot measures, and Stratechery says permission is now AI's binding constraint
Voters limited data centers in 46 of 47 decided local measures and turned Texas and Ohio against the buildout; Ben Thompson argues permission, not compute or power, is now the constraint on AI.
One item cleared the twenty-four hours to 08:00 this morning, and this one is free to read in full. Stratechery's Wednesday piece, "AI on the Ballot", carries its whole argument on an open page: ten numbered takes, a closing paragraph, no subscription needed. 1
Lenny's Newsletter has published nothing since Tuesday's gift guide, which yesterday's edition covered. So today rests on one argument, and the argument rests on a night of local votes.
AI infrastructure
AI on the Ballot — Stratechery
Published November 4, 2026, 19:00 Shanghai. Free in full.
- Data centers lost nearly everywhere they were on a ballot. Ballotpedia tracked 50 local data-center measures across 13 states in 2026, 41 of them on Tuesday's general-election ballot. Of the 47 decided by Wednesday, 46 went against development. 2
- His claim is that permission has replaced compute as the limit. "The final constraint is political, and that may be the most imposing constraint of all," he writes, because it is harder to solve than the others and it gets in the way of solving them. 1
What the votes actually said
| Measure | Count |
|---|---|
| Local data-center measures tracked in 2026 | 50, across 13 states 2 |
| Of those, on the November 3 ballot | 41 2 |
| Decided as of November 4 | 47 2 |
| Went against data center development | 46 2 |
| Went in favor of development | 1, in Harwood, North Dakota 2 |
| Still to be decided | 3: two in Clarksville, Arkansas on December 1, one in Mandan, North Dakota on December 15 2 |
| Ohio's share of the 50 | 21, spread across 18 jurisdictions 2 |
The margins were not close. Alhambra, California banned data centers with 82% of the vote, Edgewater, Florida banned new ones with 77%, and Bloomington, Illinois advised a moratorium with 78%. 2 The opposition lost in exactly one place, where the measure asked voters to add review requirements for industrial projects and they declined. 2
Texas is the state that matters most for capacity, since it accounts for more than a quarter of planned US capacity by 2030. 4 Hinojosa is the state's first Democratic governor since 1990, and she campaigned on a moratorium, stricter water and power rules, and an end to tax breaks. 4
Abbott had already announced a moratorium of his own before the election, which is a fair measure of how far the ground had shifted. 4 Hinojosa called his version an election-year pause. 4
Ohio put 21 of the 50 measures to local votes, and replaced both of its statewide Republicans. 2 Brown beat Husted for the Senate, and Amy Acton won the governor's race on a "conditional moratorium" that would require data centers to cover all utility costs and build with union labor. 3
Nationally, Democrats took both chambers. Bloomberg's report at 2 a.m. Central had them flipping at least seven Senate seats and winning House districts Republicans carried in 2024. 5
The electorate was not close to split on the subject. In NBC's exit polling, 57% said AI would do more to hurt society than help it, against 24% who said it would help. 4 In CNBC's voter poll, 56% said data centers were a crucial or major factor in their decision. 3 A UMass Amherst poll earlier in the year put support for a data center in one's own community at 11%. 6
The constraint that moved
His framing is a sequence. Compute was the shortage, and it ramped. Memory is the shortage now, and it will ramp. Power was coming next, and it will be rectified. What is left is permission:
The final constraint is political, and that may be the most imposing constraint of all. Not only is it harder to solve, but it actually interferes with solving the other constraints. 1
He made the argument in the spring, before any of this had been voted on. 7 When factories moved to China, the towns that lost them had no say in it. A data center needs a permit, which means the people who live near it do.
What has changed since spring is the politics. He expected the issue to cut one way or the other and found that both parties ran against data centers, each accusing the other of being for them. 1 Wisconsin's campaign ads were about data centers; the Texas governor's race was about them; Ohio's Senate race was about them.
He gives two reasons this does not swing back in two years. Both parties treat opposition to data centers as a winning cause, and local officials who approved projects have been recalled. 1
Who he says wins and who pays
Neoclouds and startups lose. Neoclouds are the ones that need new sites and new permits, and they have far less political pull than the hyperscalers, who already own sites they can rebuild. 1 Those that hold permitted facilities now sit on something scarcer than they expected. 1 Startups get the same squeeze from the other end: less cheap compute, slower capability gains, and more time for the frontier labs to build products that keep the end user. 1
Hyperscalers, frontier labs and Apple gain. Hyperscalers collect share as smaller builders stall, and they get an excuse to moderate spending that investors will accept. 1 Frontier labs benefit because their compute bill sits mostly in research rather than in serving, so a frontier that is forced to slow down spends less. 1 Apple gains on two counts: inference moves to the device when compute is scarce, and any easing of the buildout would relieve the memory crunch. 1
China gains. America's shortage is land and power; China's is chips, and power is the one it has. That is the trade Huawei's CloudMatrix 384 makes: far more space, far more power, far more cost, and no buyer who would choose it if they had a choice. 8 Once American sites run short, throwing electricity at the problem costs less than it did.
Nvidia and space come out mixed. Nvidia loses volume first, and a weaker neocloud sector hurts more than the arithmetic suggests, since neoclouds buy its full-rack systems and it is backstopping a number of them. 1 Against that, a world short of land and power pays for tokens per megawatt, which is the argument Jensen Huang made across an entire 2025 keynote. 9
Space gets the clearest upgrade of any idea in the piece. He had already listed political opposition as an upside case in SpaceX's S-1, on the reasoning that building on Earth is cheaper until there is nowhere left to build. 10 Elon Musk's own reaction to the results was "data centers in space looking pretty appealing right now". 3
Users pay the marginal cost. The overbuild that investors feared would have been good for users: cheap compute, and a repeat of the free-services wave that rested on it. With capacity constrained, AI keeps a real per-token cost for longer. 1
What it changes for you
- If you buy compute, the case for signing earlier and longer just got stronger, and the credit risk on a neocloud counterparty got worse. His point about neoclouds needing permits and capital at the same time is the one to price. 1
- If you site infrastructure, permitted capacity is now an appreciating asset. Texas, where a quarter of planned US capacity sits, has a governor-elect who campaigned on stopping it. 4
- If you buy accelerators, the comparison to run is tokens per megawatt rather than dollars per chip, because power and land are the fixed inputs. 9
- If your product plan assumes cheap tokens, put a floor under that assumption. He expects scarcity to last longer than the consensus does. 1
Where to start
- The argument itself, free and complete: AI on the Ballot.
- The count, measure by measure, with vote shares: Data center-related ballot measures, 2026.
- The national picture, including the Ohio and Texas races: Voters rebuff data centers in every ballot measure except one.
- The local detail, from a reporter who covered the measures: Voters across the country reject data centers.
- The Texas result and what it means for the buildout: Democrats' victories in Texas laid bare voter unease over data centers.
References
- 1AI on the Ballot — Stratechery
stratechery.com
- 2Data center-related ballot measures, 2026 — Ballotpedia
ballotpedia.org
- 3
- 4
- 5Democrats win control of the House — Bloomberg
bloomberg.com
- 6
- 7
- 8
- 9
- 10The SpaceX IPO and Data Centers in Space — Stratechery
stratechery.com
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.