Everything About AI
NextFin News - Voters across 13 states have delivered the clearest warning yet to the artificial-intelligence buildout: they do not want data centers in their communities, and they will use the ballot box to stop them. In the 2026 midterm elections, data-center measures appeared on at least 50 local ballots; of the 47 decided as of November 4, voters backed new restrictions or rejected pro-development measures in 46 cases. The lone exception was Harwood, North Dakota. The revolt reached the top of the ticket, helping Democrats unseat a Texas governor and an Ohio senator who had aligned with the industry, and it spilled into the market: power producers tied to AI data centers lost more than 13 percent in a single session while hyperscalers initially rose. The question now is whether a scattered set of local votes becomes a structural brake on the most capital-intensive infrastructure boom in a generation — or whether the industry simply reroutes around the opposition.
The Vote: A Near-Unanimous Rebuke
The numbers are lopsided in a way that rarely happens in American politics. Election trackers recorded at least 50 local ballot measures related to data centers across 13 states — Arkansas, California, Florida, Illinois, Kansas, Maine, Michigan, Nevada, New Hampshire, New Jersey, North Dakota, Ohio, and Wisconsin. Forty-one of the 50, or 82 percent, landed on the November 3 general election ballot. As of November 4, with 47 decided, a majority favored greater limits on data-center development in 46 of them, either by approving restrictions and requirements or by rejecting measures that would have facilitated development.
Redbud Consulting, which has been tracking the initiatives, framed the results by what the votes actually did rather than how the questions were worded: of 39 nationwide referendums fully counted, 38 enacted new restrictions on potential new data centers. The margins were not close. In Alhambra, California, Measure DB to prohibit data centers in the city passed with 82 percent. In Ottawa, Kansas, voters approved a ban on data centers using more than 25 megawatts by 72 percent to 28 percent. Emporia, Kansas, approved its "Prohibit High Impact Data Centers Initiative" 60 percent to 40 percent. Even in places where the measure failed, the politics moved against the industry: in Harwood, North Dakota, voters rejected additional review requirements for industrial developments such as data centers — the only loss for the restriction side.
The backlash was not confined to ballot language. In Ohio, 21 local measures were on the ballot, and Democrat Sherrod Brown defeated incumbent Republican Senator Jon Husted after casting Husted as "the face of data centers." Brown took 50 percent of the vote with 98 percent reporting, reclaiming a seat he lost in 2024 by about 3.5 points — in a state Donald Trump carried by 11 points two years earlier. In Texas, Governor-elect Gina Hinojosa unseated three-term Republican Greg Abbott, blaming the incumbent for a rapid buildout that put the tech industry's interests ahead of local communities. Texas is on track to become the world's largest data-center market by 2030. Democrats who campaigned on pausing or regulating new projects also won governorships in Georgia and Ohio.
The intensity behind the votes is measurable. A post-election voter poll found that 56 percent of voters said data centers were a crucial or major factor in their congressional vote, with Democrats ranking the issue higher than Republicans. An Election Day exit poll put the same figure at 56 percent for AI and data centers as a "crucial or major factor" in the House vote. And a University of Massachusetts Amherst poll earlier in the year found that only 11 percent of Americans would support the construction of an AI data center in their community. Opposition to data centers is not a niche environmental complaint; it is a majority position with teeth.
Why Voters Turned: Power, Water, Noise, and a Broken Bargain
The objections are concrete, and they map directly onto the physics of the AI boom. Residents cite the drain on electricity and water, the noise from cooling systems, and the limited economic benefit that flows back to their towns. These are not abstract concerns. A typical 100-megawatt AI data center consumes 1.5 million to 3 million cubic meters of water per year for evaporative cooling. A 500-megawatt facility running at 90 percent utilization burns through roughly 3.9 terawatt-hours annually — comparable to the yearly electricity use of about 360,000 U.S. homes. A single NVIDIA GB200 NVL72 rack draws 120 to 140 kilowatts, versus 10 to 15 kilowatts for a traditional enterprise rack; the facilities built for the old generation cannot physically host the new one without a full infrastructure redesign.
On the grid side, the scale is just beginning to register. The International Energy Agency projected that global data-center energy consumption could jump from 460 terawatt-hours in 2022 to more than 1,000 TWh in 2026. The Electric Power Research Institute said in May that U.S. data centers' electricity consumption could double to more than 9 percent of total U.S. generation by 2030. The Belfer Center at Harvard estimates data-center energy usage could reach 325 to 580 TWh by 2028, or 6.7 percent to 12 percent of national consumption. Local officials are not reading these reports in the abstract — they are signing interconnection agreements and watching their own substations fill up.
The political economy of the deal has also soured. For years, communities were sold data centers as high-tech job creators with a light footprint. In practice, a hyperscale AI facility employs relatively few permanent workers, draws enormous amounts of power that could otherwise serve homes and factories, and locks municipalities into payment-in-lieu-of-taxes agreements that limit revenue. When the trade-off became visible — higher electricity bills, strained water supplies, noise, and few local jobs — the political coalition behind the projects collapsed. That is why the issue cut across the usual partisan lines: local measures passed in Republican and Democratic areas alike, even as the voter surveys showed Democrats ranking the issue higher than Republicans. When a Republican governor loses a red state by running as the data-center candidate, the industry's natural political coalition has fractured.
The money did not save the industry. Silicon Valley spent record sums trying to shape the 2026 midterms, yet an analysis of Federal Election Commission records found that 14 of the 20 largest donor groups backed Republicans almost exclusively, with crypto, AI, and online-gambling interests pouring in more than $377 million. Elon Musk, who spent hundreds of millions of dollars this cycle, largely came up empty. His post-election assessment on his social platform was sardonic: "Data centers in space looking pretty appealing right now." When a billionaire can buy advertising but cannot buy local consent, the constraint has moved beyond campaign finance.
The Market Reaction: A Two-Day Selloff With Two Different Engines
Wall Street's first read of the election was precise about who was exposed. On Wednesday, November 4 — the first full trading day after the vote — the major averages rose: the Nasdaq 100 closed up 0.91 percent, the S&P 500 gained 1.58 percent, and the Dow rose 3.03 percent, helped by a drop in oil prices. Under that calm surface, the election trade was violent. Vistra fell 13.8 percent, NRG Energy 13.5 percent, Talen Energy 8.5 percent, and Constellation Energy 6.7 percent. The Philadelphia Stock Exchange Semiconductor Index fell 3.1 percent. But the data-center landlords Equinix and Digital Realty rose 3.9 percent and 2.6 percent, and Microsoft and Amazon each gained more than 6 percent.
That first-day split is the cleanest signal of what the market actually priced. Investors punished the power producers whose growth narrative was most tightly bound to AI data-center demand — especially those with Texas exposure — while sparing the REITs that own the facilities and the hyperscalers that rent them. The distinction matters: the vote does not threaten the demand for compute; it threatens the permitting, the interconnection queue, and the local political license to build.
Thursday brought a second engine. The Nasdaq 100 fell more than 6 percent before lunch on the East Coast, and by 9:52 a.m. Central Time the index was down 6.25 percent with the chip index down 11.6 percent. That move had a second driver: a weak outlook from Qualcomm, a debate over Arm Holdings' results, and Micron's pause of a Taiwan plant. By Thursday, the election fears and the earnings fears had merged, and almost everything fell together, REITs included. The honest reading is that Wednesday isolated the election effect; Thursday compounded it with valuation concerns.
Analysts are framing the risk as "slower and costlier," not "stopped." KeyBanc downgraded Vistra two days before its November 6 earnings, citing its Texas exposure. UBS put about 30 percent of 2028 data-center capacity at risk and wrote that "data center pauses will slow the pace of buildout into 2028." Wells Fargo estimates Texas accounts for about 27 percent of U.S. data-center gigawatts. MRB Partners said the midterm results "will add to headwinds for big tech and the data center buildout." None of these assessments describes a halt. They describe a boom that now has to negotiate with every county board along the route.
The Mechanism: Why This Is Structural, Not Cyclical
The central question for investors is whether the backlash is cyclical — a temporary wave of NIMBY sentiment that will recede as projects deliver jobs and tax revenue — or structural, a durable change in the political cost of building AI infrastructure. The evidence points to structural, for three reasons.
First, the constraint is physical and visible before the benefit ever arrives. A cyclical backlash typically follows a disappointment: jobs that do not materialize, tax revenue that falls short. Here, voters are reacting to inputs — megawatts, water, noise — that are knowable at the permitting stage and cannot be hidden. Once a community has seen a 500-megawatt facility draw power equivalent to a small city, that knowledge does not revert. The opposition is based on a cost that is immediate and a benefit that is distant and diffuse.
Second, the opposition is cross-partisan and local, which makes it hard to nationalize away. National politicians can be flipped by campaign spending and party messaging; zoning boards and city councils cannot be easily nationalized, and the issue cuts against the pro-development Republican orthodoxy that the industry relied on. When a Republican governor loses a red state by running as the data-center candidate, the industry's natural political coalition has fractured. Local land-use politics is the least scalable form of politics there is — which is exactly why it is so hard for a national industry to solve.
Third, the projects themselves have changed. The old data-center model — a warehouse of servers drawing steady, moderate power — has been replaced by AI campuses whose power density per rack has increased nearly tenfold. That means each new project is a bigger shock to the local grid than the last one, so each one triggers more opposition. The industry is not climbing a learning curve that reduces friction; it is climbing a curve that increases the friction per project even as it needs more projects.
This is why the second-order effect is more important than the first. The first-order effect is obvious: some projects will be delayed or blocked. The second-order effect is that the cost of capital for data-center development rises, because political risk is now a line item in every pro forma. A developer who once underwrote a project on construction risk and power-price risk must now underwrite voter risk — the risk that a newly elected county board, a ballot initiative, or a state attorney general will intervene after the money is committed. That re-pricing does not stop demand; it concentrates the industry toward players who can absorb delay and litigation, and it pushes development toward jurisdictions with fewer democratic veto points. The boom does not end; it becomes more expensive, more concentrated, and more politically fragile.
"The revolt against data centers was one of the top issues of the 2026 election," said Representative Greg Casar, the Texas Democrat who chairs the Congressional Progressive Caucus. "And in a few years we will look back on it as just the early tremor of a much bigger AI earthquake upending American politics."
Casar also called for AI company CEOs to testify under oath, signaling that the local revolt has a federal legislative shadow. "The House Democratic majority must make it a top priority to protect Americans' jobs, freedom, and lives from AI billionaires," he said. Whether or not that agenda advances, the threat of federal scrutiny adds another layer of uncertainty to capital planning.
The Counter-Thesis: Fragmented, Local, and Legally Vulnerable
The strongest case against the structural reading is that the backlash is fragmented, local, and legally vulnerable — a collection of municipal votes, not a coordinated national policy. Local ordinances can be preempted by state legislatures. Ballot measures can be tied up in court. And demand for AI compute is not going away: the hyperscalers have multi-year capacity plans, and the power purchase agreements are already signed. From this angle, the election is a speed bump, not a wall. The industry has survived local opposition before — fracking, wind farms, pipelines — and it has learned to route around it. The buildout slows, reroutes to friendlier states, and continues.
There is real evidence for this view. Wednesday's market action spared the REITs and hyperscalers, which is what a rerouting scenario looks like: the assets that own and use the capacity are fine; only the marginal power producers in the hot spots get hit. UBS's 30 percent at-risk figure implies 70 percent is not at risk. And the measures are overwhelmingly local — city and county ordinances — which states can and do override when they want to.
But the rerouting argument underestimates two things. First, the industry is running out of friendly states. Texas, the single largest and most politically hospitable data-center market, just elected a governor who ran against the buildout. Virginia, the other great hub, has faced its own local revolts. When the two biggest markets both turn hostile, "reroute" starts to mean "go somewhere with worse power, worse fiber, and worse economics" — which is another way of saying the cost of the boom rises structurally.
Second, the legal vulnerability cuts both ways. Yes, local measures can be challenged. But litigation takes years, and data-center projects are financed on the assumption of certainty. A project stuck in court for three years is a project whose returns have been destroyed by discounting alone, even if the company eventually wins. Delay is a form of defeat in infrastructure finance.
The falsifying signal for the structural view is specific: if, over the next two election cycles, states with large data-center pipelines pass preemptive legislation that strips localities of the power to block or delay facilities — and if data-center gigawatts under construction return to their pre-election trajectory — then the backlash was cyclical and local after all. Watch state legislatures in Texas, Georgia, Ohio, and Arizona in 2027. If they reassert state control over siting, the structural thesis is wrong. If they do not, the industry has entered a new regime.
What Comes Next: Three Horizons
Short term (next 3–6 months): Expect volatility in the power producers most exposed to AI data-center demand, especially those with Texas and Virginia exposure. The November 5 intraday move showed how quickly the trade can reverse when earnings and politics collide. Developers will pause ribbon-cuttings and re-evaluate projects in jurisdictions where new officials have taken office. The December votes in Clarksville, Arkansas — referendums on up to $50 billion in Hatchbo bonds and $5 billion in SF ARK1 bonds — and Mandan, North Dakota will test whether the November pattern holds.
Medium term (2027–2028): The buildout slows rather than stops. UBS's estimate that 30 percent of 2028 capacity is at risk is the right order of magnitude for a base case. Projects in friendly jurisdictions proceed; projects in contested ones face delays, redesigns, or relocation. The cost of capital for data-center development rises as political risk gets priced in. Hyperscalers with balance sheets and in-house development teams gain share relative to speculative cloud builders, which is why the market punished CoreWeave-style names harder than Microsoft and Amazon.
Long term (2029 and beyond): The structural question resolves one of two ways. If states preempt local control, the boom resumes its pre-election trajectory and the 2026 revolt becomes a case study in local powerlessness. If they do not, the AI buildout becomes a permanently negotiated enterprise, with higher costs, slower deployment, and a geographic footprint shaped as much by county boards as by power prices. Either way, the era of building first and asking questions later is over.
The investors who get this right will not be asking whether AI demand is real — it is. They will be asking who has the political license to deliver it. The 2026 election answered the first question and made the second one the only one that matters.
Explore more exclusive insights at nextfin.ai.
