NextFin News - The artificial-intelligence build-out is running into a wall that no amount of capital can buy its way out of: voters. Maria Goodpaster, the McKinsey & Company partner who led the firm's research on the roughly $7 trillion data-center construction cycle, said in a televised interview this week that the political "backlash is very real" - and the numbers behind her warning suggest the industry has underestimated how fast local opposition can harden into law.
During the first three months of 2026 alone, 75 data-center projects worth about $130 billion were blocked or delayed by local opposition, according to a tally of project-level opposition compiled by market reporters covering the sector. Separately, more than 300 data-center-related bills were filed across over 30 states within the first six weeks of 2026 - more legislation in a month and a half than in all of 2025, when roughly 200 bills appeared across 40 states. This is not a handful of disgruntled towns. It is a coordinated, bipartisan political response that is rewriting the economics of the AI infrastructure race.
The central question this piece answers: is the data-center backlash a cyclical public-relations flare-up that will fade once companies improve their community outreach, or is it a structural regime shift that will permanently raise the cost and slow the pace of the AI build-out? The evidence - a polling reversal of nearly 50 percentage points in four months, moratoriums in New York, South Dakota and Oklahoma, tax-credit repeals in at least eight states, and both parties weaponizing the issue in the 2026 midterms - points decisively toward structural. That judgment matters because it changes who wins and who loses in the $7 trillion build-out, and it changes the timeline on which AI compute actually reaches the market.
Layer 1: The Backlash Has Numbers, Not Just Noise
Goodpaster's warning carries weight because it comes from the supply side of the build-out, not from an activist camp. As lead author of McKinsey's March 2026 report on the $7 trillion data-center build-out, she has mapped how industrial suppliers can capture a share of the spending. Her point was not that the build-out is ending, but that the path to capturing it has narrowed - and that the companies assuming a smooth, incentive-driven rollout are misreading the political terrain.
The political data bear her out. A panel discussion hosted by Boston Review in August 2026 cited polling showing how quickly opinion moved: in January 2026, when asked whether they would support or oppose a data center in their area, 43 percent of Americans said support and 42 percent said oppose. By May 2026, the same question produced 71 percent opposed and 21 percent in favor - a shift of roughly 50 percentage points in four months. Nate Silver's analysis of search behavior found Google queries for "data center" running about 3.5 times higher in 2026 than in 2025, and nearly 10 times higher than in 2024. Public attention is not abstract; it is searching for the thing it is angry about.
That attention is converting into legislation at a pace the industry has not seen. MultiState, a policy-tracking firm, counted more than 300 state data-center bills across 30-plus states within the first six weeks of 2026. The content of those bills marks a sharp break from the past. In earlier years, states competed to attract data centers with tax breaks and state-funded infrastructure. Now the same legislatures are rolling those incentives back, imposing energy-consumption taxes, and pausing construction outright.
The moratoriums are the clearest signal. New York introduced legislation to halt all data-center construction for up to three years while regulators study rate impacts. South Dakota proposed a one-year moratorium on hyperscale expansion. Oklahoma placed a moratorium on any facility with an electrical load above 100 megawatts until November 2029. These are not environmental fringe proposals - they are mainstream legislative vehicles in states that previously welcomed the industry.
The tax-incentive reversal is equally broad. A tally of state legislation found that lawmakers in at least 28 states introduced bills to roll back data-center tax incentives in 2026. The Center on Budget and Policy Priorities, cited in trade coverage, counted eight states that enacted rollback legislation this year with another 17 considering it. Virginia - home to more than 600 data centers, the most of any state, and long called the data-center capital of the world - imposed a consumption tax on data-center energy usage under new Democratic control. Broadband Breakfast noted that 38 states still offer data-center tax incentives, but the political direction is now one-way: toward restriction, not inducement.
The bipartisan nature of the backlash is what makes it structurally durable rather than electorally cyclical. In Wisconsin, Democratic primary winner David Crowley framed his difference with his opponent in a single line: "I disagree with her stance on data centers." Republican Tom Tiffany responded with visible frustration: "You don't disagree about defund the police? You don't disagree about this? Why is it data centers?" The answer is that data centers touch a coalition of grievances - rising electricity bills, water use, noise, land consumption, and a sense that distant tech companies are imposing costs on communities that see few of the long-term jobs - that neither party wants to be on the wrong side of heading into the midterms.
Layer 2: Why This Is Structural, Not a Public-Relations Problem
The Transmission Mechanism: From Town Hall to Cost of Capital
The industry's first instinct has been to treat this as a communications problem. OpenAI and Meta have reportedly sought outside help to combat the public-relations fallout, and trade coverage in August described a tech "charm offensive" aimed at data-center opponents. That response misunderstands the mechanism. The backlash does not travel through media coverage; it travels through local permitting, state legislatures, and utility commissions - and each of those channels converts opposition into hard cost.
The chain runs like this. Local opposition stops or delays a specific project. Those delays then give state legislators a visible constituency issue, which produces bills that raise compliance costs or withdraw subsidies. Utility commissions, facing pressure over residential rate increases, impose special rate classes or demand that large users fund grid upgrades. The result is not that projects never get built; it is that each project takes longer, requires more capital, and clears a higher regulatory bar. In infrastructure finance, time and uncertainty are priced directly into the cost of capital. A build-out that was modeled on 24-month permitting and subsidized power now faces 36-to-48-month timelines and market-rate or above-market-rate electricity. That spread - measured in hundreds of basis points of weighted-average cost of capital across a $7 trillion program - is the real cost of the backlash.
Pennsylvania's executive order, signed by Governor Josh Shapiro on August 18, 2026, shows the mechanism in one policy. Data centers must source power from new generating resources within the same local PJM Interconnection zone, and the required share of firm clean energy ramps from 10 percent at the start of the year to 14.5 percent three years later and 32 percent by January 1, 2035. The order's stated logic - "growth should pay for growth" - is a template other states are already studying. It transforms the data center from a welcomed taxpayer into a ratepayer of last resort that must build its own grid capacity.
Cyclical Versus Structural: The Evidence Floor
A cyclical backlash would show three features: a short-term trigger, a history of mean reversion, and a fix that restores the prior equilibrium. A structural shift shows a permanent change in the rules of the game. The data-center backlash meets the structural test on three counts.
First, the driver is a change in political rules, not a temporary sentiment spike. Eight states have already enacted tax-subsidy rollbacks; more than 300 bills are in play across 30 states; three states have active moratorium proposals. Once a tax credit is repealed or a moratorium is codified, it does not revert when sentiment cools - it requires a new legislative majority to undo.
Second, the opposition is bipartisan and multi-level, which makes it resistant to a single election cycle. The issue is live in red states such as Oklahoma, South Dakota and Texas - where Governor Greg Abbott told critics that data centers "basically dug their own grave" and "got the backlash they deserve" - and in blue states such as New York, Pennsylvania and Illinois, where Governor JB Pritzker called for a two-year pause on tax incentives and said data centers must "pay their fair share." An issue that produces the same policy direction under both parties is not a partisan pendulum; it is a regime shift.
Third, the underlying grievance is material, not perceptual. Data centers are land- and power-intensive by design. A single hyperscale campus can occupy hundreds of acres - one facility in Port Washington, Wisconsin spans 700 acres - and AI-optimized racks draw 30 to over 100 kilowatts each, compared with 5 to 15 kilowatts for traditional racks. When a community sees its substation upgraded for a facility that promises dozens of permanent jobs, the cost-benefit arithmetic that convinced officials five years ago no longer convinces voters. That arithmetic has changed permanently; it will not mean-revert.
There is, however, a cyclical leg layered on top of the structural one, and it matters for timing. The public-relations response - the charm offensive, the community-benefits agreements, the better siting disclosure - can reduce the number of projects blocked in any given quarter. But PR can only smooth the path; it cannot restore the subsidy regime or repeal the moratoriums. The structural shift sets a higher floor for cost and timeline; the cyclical PR effort determines how much above that floor any single project lands.
The Second-Order Effect: The $7 Trillion Gets Slower, Costlier, and More Concentrated
The first-order effect of the backlash is obvious: some projects get delayed. The second-order effect is what the market has not fully priced. McKinsey's $7 trillion figure - of which roughly $5.2 trillion is tied to AI workloads - is a spending estimate, not a delivery guarantee. When the cost of capital rises and permitting lengthens, the same $7 trillion buys fewer gigawatts of delivered capacity over the same period. Global data-center capacity is projected to nearly triple between 2025 and 2030, from about 82 gigawatts to roughly 200 gigawatts. The backlash puts the upper end of that range at risk.
The concentration effect follows. If only projects with the deepest balance sheets and the most patient capital can clear the new regulatory bar, the build-out concentrates further among the hyperscalers and a handful of large colocation operators. Smaller developers, independent power producers without pre-negotiated grid access, and industrial suppliers that were counting on a broad, competitive field of customers face a narrower window. McKinsey's own framing - that the window for industrial suppliers to secure leading positions "is narrowing" - cuts both ways: it is an opportunity for the prepared and an exit door for the late.
The power-and-cooling slice of the spend becomes the bottleneck that matters most. McKinsey estimates that $1.3 trillion - about 25 percent of total AI infrastructure investment - flows to power, cooling and related infrastructure. When states require on-site or same-zone generation, that share rises. Companies positioned in transformers, switchgear, backup generation, liquid cooling, and behind-the-meter power are the relative beneficiaries; generic construction and commoditized hardware suppliers face margin compression as projects re-bid under tougher terms.
The third-order implication reaches AI itself. A slower, costlier data-center rollout means compute supply grows more slowly than demand models assumed. That supports pricing power for cloud providers with existing capacity, but it also means the AI applications betting on exponentially cheaper compute face a supply constraint that no software breakthrough can fix. The companies that already locked in power and permits before the political turn - roughly the cohort that broke ground through 2025 - hold an option value that late entrants will pay a premium to access.
The Counter-Thesis: The China Imperative and the Manageable NIMBY
The strongest case against the structural view is the national-security argument. The AI race with China creates a federal imperative that could override local objections - through preemption, through fast-track permitting for projects deemed critical, or through direct subsidies that offset state-level costs. Industry advocates argue that data centers are essential infrastructure for the AI era, bringing high-paying jobs and tax revenue to communities, and that the opposition is a concentrated, well-organized minority rather than a durable majority.
There is evidence for the softer version of this view. A Cato Institute commentary noted that data centers used just 0.3 percent of the U.S. public water supply in 2023 - far below the public's perception of their water intensity. And the Boston Review panel noted that the backlash episode also surfaces real local benefits: major tax revenue and high-paying jobs that can last seven to ten years or longer. If voters weigh those benefits against the costs, opposition could soften.
But the counter-thesis fails on two grounds. First, federal preemption is itself a political act that requires congressional majorities or aggressive executive action, and data centers have become a midterm flashpoint precisely because neither party wants to be seen overriding local communities on behalf of large tech companies. Second, even if the national-security argument eventually wins at the federal level, the lag is measured in years, and the projects delayed in the interim do not come back. The China imperative may change the terminal outcome for some projects; it does not change the cost and timeline of getting there.
The falsifying signal is specific and observable. If the project block-and-delay rate falls back below roughly 20 projects per quarter - from the current 75 per quarter pace - and fewer than 10 states file rollback or moratorium bills across the next two legislative sessions, then the structural thesis weakens and the backlash looks more like a cyclical NIMBY flare-up that PR and better siting can manage. Until those numbers move, the burden of proof sits with the industry.
Layer 3: What Comes Next - Scenarios by Time Horizon
Short term (6-12 months): volatility and deal-by-deal friction. Expect continued project delays, more state-level bills, and a wave of community-benefits agreements as developers try to buy local peace. Data-center REITs and power-exposed developers will trade on headline risk - each moratorium vote and each delayed interconnection becomes a catalyst. The hyperscalers with the deepest balance sheets absorb the friction; smaller players stall or sell.
Medium term (1-3 years): a bifurcated market. Projects that secured power and permits before the political turn deliver strong returns as compute scarcity supports pricing. Late-cycle projects re-bid at higher cost of capital, with some cancelled outright. The industrial-supplier landscape consolidates around vendors with hyperscaler qualification pathways and the ability to deliver under the new regulatory bar. Colocation vacancy stays tight - McKinsey's own exhibit work has shown available data-center space scarce at least through 2027 - and that scarcity deepens if new supply keeps slipping.
Long term (3-5 years): federal resolution or permanent local veto. The base case is a messy hybrid: federal fast-tracking for projects designated as critical AI infrastructure, layered on top of a permanently higher state and local compliance floor. The upside case is a national-compromise framework that trades federal preemption for enforceable community benefits and ratepayer protections - restoring some permitting predictability. The downside case is a patchwork of state vetoes that fragments the U.S. grid build-out and hands the AI infrastructure advantage to jurisdictions with fewer democratic constraints.
For investors and operators, the watch list is concrete: the quarterly count of blocked or delayed projects (currently 75, worth about $130 billion in the first quarter of 2026); the number of states enacting rollback or moratorium legislation (eight enacted so far in 2026, with 17 more considering); and the spread between modeled and actual interconnection timelines for hyperscale projects. Each of these is a direct read on whether the backlash is accelerating or plateauing.
The central judgment: Goodpaster is right that the backlash is very real, but the deeper point is that it is very durable. The AI build-out will still happen - the strategic and commercial incentives are too large - but it will happen on a slower, costlier, more politically negotiated path than the $7 trillion models assumed. The winners will not be the companies that build the best servers; they will be the ones that secured power, permits, and political cover before the voters noticed what was being built in their backyard.
"They basically dug their own grave for the problem that's been caused for them, and that's why they got the backlash they deserve," Texas Governor Greg Abbott said this month, arguing that the industry's approach to communities earned the opposition it now faces.
Whether or not one agrees with his framing, it captures the strategic error at the heart of the build-out: the industry spent a decade optimizing for speed and scale, and only now is it being forced to optimize for consent. In infrastructure, consent is not a line item - it is the schedule.
Explore more exclusive insights at nextfin.ai.
