In mid-July 2026, anti-data-center protests broke out in roughly 125 cities across the United States in a single weekend, the first coordinated national mobilization of its kind. Days earlier, New York had become the first state in the country to impose a moratorium on new AI data centers. Neither of those things would have been imaginable two years ago, when data centers were still a niche infrastructure story. Now they're one of the defining local-politics fights of the AI boom, and the reason is simple: the pitch communities were sold and the experience they're actually having don't match.
The Simple Version
Data centers are the physical buildings that make AI possible: warehouses full of servers that need enormous amounts of electricity and, often, water for cooling. Tech companies are building them at a record pace to keep up with AI demand, and they typically pitch host communities on billions of dollars in investment, construction jobs, and new tax revenue.
Residents are increasingly pushing back. A Reuters/Ipsos poll from June 2026 found only 33% of Americans think the current pace of data center construction is a good thing, and just 14% would be comfortable having one built near them. That’s not a fringe reaction: 77% of Americans, across party lines, worry data centers will make their electricity bills more expensive. Local opposition has already blocked or delayed $130 billion worth of projects in 2026 alone, matching all of 2025 in less than seven months.
Why This Is Happening Now
The backlash isn’t really about data centers as buildings. It’s about what residents believe those buildings will cost them, layered onto anxiety about a technology that feels abstract and hard to hold accountable.
Electricity is the biggest flashpoint. A single large data center can draw as much power as a mid-sized city, and if utilities spread the cost of new power plants and transmission lines across all ratepayers instead of billing the data center directly, everyone else’s bill rises to subsidize a private company’s AI buildout, exactly what the 77% figure above reflects people fear is already happening. States are responding with “large load” tariffs: a special rate class that bills data centers directly for the infrastructure built to serve them, the same logic as a restaurant paying for its own dedicated water line rather than raising rent for the whole building. Nevada, Pennsylvania, and Kentucky are all writing or debating rules like this.
Water is the second flashpoint. Many hyperscale facilities use evaporative cooling that consumes large volumes of water, and reporting has found new data centers disproportionately sited in drought-prone regions. In one widely cited case, a facility reportedly drew 30 million gallons unnoticed until residents complained about low water pressure. Most states, including California, don’t require site-level water-use disclosure, so nobody can say with confidence how big the aggregate problem really is.
Then there’s the jobs-and-taxes mismatch: data centers are capital-intensive, not labor-intensive, employing hundreds of construction workers for a year or two before settling into a permanent staff of a few dozen, often alongside multi-year property-tax abatements that have triggered fights in Missouri, Kentucky, and elsewhere over who pays for the schools and roads a facility depends on. Communities are increasingly countering with community benefit agreements: negotiated, legally enforceable contracts that lock in local hiring, wages, and infrastructure commitments up front, turning a developer’s “hundreds of good jobs” pitch into something residents can actually hold the company to.
Underneath all of that is something less tangible: data centers have become the place where broader anxiety about AI’s effect on jobs and the economy gets a physical address. Brookings has described this as AI’s most visible, place-based proxy fight: you can’t organize a protest against “AI,” but you can show up at a zoning hearing. That’s attracted real political money on both sides, including a pro-AI super PAC that launched with roughly $140 million from AI investors, facing off against a smaller, employee-funded group pushing for AI guardrails.
Are the Water and Electricity Concerns Overblown?
Genuinely mixed. The individual incidents, like the unnoticed 30-million-gallon draw and the drought-region siting patterns, are real, but most states don’t require the site-level disclosure that would confirm how big the problem is in aggregate, so it can’t be confirmed as a crisis or ruled out either. Newer closed-loop and air-cooling systems use far less water than older evaporative designs, so severity varies a lot by facility.
Electricity worry is more forward-looking: the 77% figure is sentiment, not documented bill increases, gathered in a high-salience election year. But regulators aren’t treating it as manufactured: Nevada, Pennsylvania, and Kentucky are actively writing large-load tariff rules to prevent cost-shifting, and the market agrees: Constellation Energy (CEG) and NRG Energy (NRG) are both down double digits in 2026 despite a bullish AI-power narrative, suggesting investors are pricing in real regulatory risk, not hype.
Is the Buildout Ahead of Real Demand?
There’s a deeper question underneath the backlash: is this construction actually necessary yet, or is the industry building ahead of demand that hasn’t fully arrived?
The case for real demand is strong: TSMC cited “strong, multi-year structural demand” while expanding its Arizona investment to $265 billion, Meta and Anthropic are reportedly negotiating a $10 billion compute-leasing deal, SpaceX is in talks to supply the Pentagon with AI computing capacity, and Micron crossed a $1 trillion valuation in May on real memory-chip shortages.
But the pace may be outrunning it. UBS expects hyperscaler capex to rise 76% in 2026 before decelerating sharply to 25% in 2027 and 6% in 2028. Financing is tightening too: bond cover ratios for hyperscaler debt fell from nearly 5-to-1 in February to below 2-to-1 by July, and in June the Bank for International Settlements warned disappointing returns “could trigger a sudden pullback in financing and turn the capex boom into a protracted bust.” The market is already repricing around this: Apple (AAPL, $333.74) overtook Nvidia (NVDA, $202.81) as the world’s most valuable company in mid-July, as investors rotate toward companies less exposed to capex intensity.
None of this means AI demand is fake. It means the sharper question isn’t “is this a bubble,” but “did construction get ahead of demand by a year or two.” A slowdown from opposition or tighter financing wouldn’t mean demand dried up, just that the buildout pace is recalibrating toward it.
Is the U.S. Really Losing Ground to China?
A common argument against restricting data centers is that delays could hand AI infrastructure leadership to other countries, but naming China specifically overstates it. China’s buildout can move faster than America’s, with far less community veto power, but U.S. export controls keep the most advanced Nvidia chips largely out of the country (state media calls this an “AI Iron Curtain”), so China can often build faster without being able to fill those buildings with equivalent compute. What’s closing quickly is the model gap, not the hardware gap: Moonshot’s Kimi K3, unveiled at Shanghai’s World AI Conference in mid-July, the same week Xi Jinping pitched a rival 29-country “World AI Cooperation Organisation” against the 35-country, U.S.-led alternative.
The more credible competitive threat is the Gulf states: the UAE and Saudi Arabia can legally secure advanced Nvidia chips, offer cheap subsidized land and power, and face no local zoning fights, a more plausible destination for redirected hyperscaler growth than China. And domestically, the backlash still doesn’t look like a demand-killer: UBS expects hyperscaler capex up 76% in 2026, and Vertiv and Equinix are having record years, with most blocked projects relocating to friendlier states rather than leaving the country.
Why the Market Cares
Not every company tied to data centers is exposed to this fight the same way, and that split is visible in 2026 stock performance.
Infrastructure and colocation companies profit from whichever data centers get built, wherever that ends up being: a slowdown in one state just shifts where the revenue comes from, it doesn’t eliminate it. That’s likely part of why Vertiv (VRT, $289.56, +78.73% YTD) and Equinix (EQIX, $1,020.00, +33.13% YTD) have had standout years, with Digital Realty Trust (DLR, $173.88, +12.39% YTD) also higher, if more modestly.
Regional power utilities have the opposite exposure: their growth story depends specifically on new data centers getting built in their service territory, which puts them directly in the path of moratoria, ratepayer-protection fights, and local political risk. Constellation Energy (CEG, $252.39, -28.56% YTD) and NRG Energy (NRG, $129.11, -18.92% YTD) are both down sharply this year despite bullish narratives around AI power demand, a sign that the market is already discounting this risk rather than ignoring it.
The hyperscalers sit in between. Microsoft (MSFT, $393.82, -18.57% YTD), Amazon (AMZN, $247.23, +7.11% YTD), Alphabet (GOOGL, $346.77, +10.79% YTD), and Meta (META, $646.01, -2.13% YTD) keep announcing large projects (Meta’s roughly $50 billion Louisiana campus among them), but their 2026 stock performance has been mixed to negative, reflecting broader volatility in the AI trade as much as data-center-specific headline risk.
The Bottom Line
For consumers, the practical stakes are electricity bills, water access, and whether a nearby project delivers the jobs it promised: outcomes that hinge on whether states adopt large-load tariffs and water-disclosure rules before communities discover the downside on their own. For companies, the playbook is now clear: pair project announcements with enforceable community benefit agreements, invest in water-efficient cooling, and expect a state-by-state patchwork of tariff rules rather than one national standard. Developers that skip this and lean only on the size of their investment figure are the ones landing on Data Center Watch’s blocked-or-delayed list.
The backlash isn’t the whole story, though. Opposition isn’t universal (some rural areas still see data centers as a rare chance to diversify a shrinking tax base), and blocking too many projects carries a real national-competitiveness risk, though the sharper version of that threat points to chip-accessible Gulf states rather than China, whose faster buildout is offset by U.S. chip export controls. The tariffs and community benefit agreements meant to fix this are also new and largely untested, and a slowdown in one jurisdiction can simply push the same water and power draw to a more permissive one rather than resolving it.
Data centers can be a genuine economic opportunity for the towns that host them, but only if those towns negotiate for it rather than approving whatever gets pitched. The states writing enforceable rules now are trying to capture the investment and tax base without absorbing the higher bills and broken promises that triggered 2026’s backlash in the first place. For investors, the same split applies: companies that profit no matter where a project lands look structurally safer than the regional utilities and hyperscalers now carrying real political and regulatory risk on their books.