Your 2023 AI Cost Model Is Now Wrong in 42 States

On July 18, 2026, opponents of rapid data center expansion staged 142 coordinated protests across 42 states. It was the first nationally organized campaign against the AI infrastructure boom — and it barely registered as a tech industry story. The protests were local, loud, and largely about things like water bills, property values, and substations. But for CIOs planning AI deployments, those local complaints have become a strategic risk factor. The math has shifted underneath them. As of mid-July, 10 states have active data center moratoriums, with eight more considering similar legislation. Twenty-three states have approved large-load tariffs that force data centers to pay full infrastructure costs. And at least nine states have introduced bills to repeal tax incentives entirely. Add it up, and the power cost assumptions baked into 2023 business cases are wrong in close to half the country. This is no longer a permitting issue. It's a balance sheet issue.

New York was the first state to impose a statewide data center moratorium. On July 14, 2026, Governor Kathy Hochul signed an executive order blocking permits for new hyperscale facilities consuming 50 megawatts or more for one year. The pause gives regulators time to develop rules around energy, water, and environmental impact — and Hochul has pledged to pursue legislation repealing sales tax exemptions for large data centers. The order followed a Gallup poll finding that 7 in 10 Americans would oppose a new data center in their area. "As data center development threatens to hike up utility bills, deplete our natural resources, and create uncertainty for New Yorkers, it's my responsibility to take action and lead," Hochul said. But the epicenter of the backlash is Virginia, home to the densest concentration of data centers on earth — "Data Center Alley" in Loudoun and Prince William counties. Virginia has produced 30 distinct laws from 38 data center-related bills introduced in 2026, with 15 bills signed into nine laws as of mid-June. The state has gone from the industry's most welcoming host to its most aggressive regulator, enacting a first-of-its-kind electricity consumption tax of $0.011 per kilowatt-hour, effective July 1 through July 2028. The tax carries a $600 million annual cap, with excess refunded to operators. It's a modest sum for a 100-megawatt facility, but politically, it's a watershed. Virginia also missed out on more than $1.6 billion in tax revenues in fiscal 2025 due to data center exemptions — a number that lawmakers are now citing openly. The message: the era of subsidized AI infrastructure is over.

The Regulatory Wave, State by State

The moratoriums and incentive pauses are not isolated events. A wave of legislative action has swept through statehouses in 2026: | State | Action | Details | |---|---|---| | Arizona | 3-year incentive moratorium | HB 4168 froze new data center tax incentives effective July 1, 2026 | | Illinois | Administrative pause | Gov. Pritzker paused new incentive agreements beginning July 1, 2026 | | Ohio | Administrative pause | Gov. DeWine suspended consideration of new tax exemption requests | | Maryland | Study requirement | SB 116 mandates a comprehensive environmental, energy, and economic impact study | | Massachusetts | Incentive halt | Gov. Healey paused data center sales tax exemption applications | | Georgia | Multiple local moratoriums | Paulding County, Albany, Chamblee, DeKalb County, and Augusta have all enacted pauses | | Oklahoma | Proposed moratorium | SB 1488 would pause development until 2029 to study water supply and property value impacts | | Montgomery County, MD | 18-month moratorium | Unanimous vote on permits for data centers ≥25MW | | Prince George's County, MD | 2-year moratorium | On hyperscale data center development | | Washington County, MD | 12-month moratorium | Extended from proposed 6 months | | South Dakota | Local authority preservation | SB 135 preserves local government authority to regulate, limit, or prohibit development | Ohio offers a cautionary tale. The state projected tax revenue losses of $136 million in 2025 and $142 million in 2026 from data center tax breaks. Actual losses ballooned to $554 million in 2024 and approached $1.6 billion in 2025. With numbers like that, the political calculus shifts quickly. The opposition spans partisan lines, a fact that makes it hard for tech companies to predict outcomes. Mother Jones summed it up this way: "America's data center backlash refuses to stick cleanly to partisan boundaries, with opposition to data center buildout nearly as high among Republicans as it is among Democrats."

Why the Backlash Happened: Energy, Water, and Transparency

The proximate cause of the pushback is visible in the numbers. Hyperscaler capital expenditure on AI infrastructure is projected to exceed $690 billion in 2026 alone, while U.S. utilities have committed to $1.4 trillion in grid infrastructure spending through 2030. Global data center capex is expected to surpass $1 trillion for 2026. The top four U.S. cloud providers — Amazon, Google, Meta, and Microsoft — increased data center capex by 78 percent. Microsoft alone projected $190 billion in capital expenditures for 2026. Supply, however, is not keeping pace. Wood Mackenzie estimates that only about a third of planned data center capacity is under active development; the rest has stalled, largely due to power constraints. Bernstein estimates that construction costs per megawatt have risen approximately 20-25 percent since 2023, driven by electrical equipment, steel, and specialized labor. A 1-gigawatt AI data center now costs between $35 billion and $50 billion to build. Water is the more emotionally charged issue. Large data centers can consume up to 5 million gallons of water per day — equivalent to a town of 10,000 to 50,000 residents. Total U.S. data center water demand could climb to 73 billion gallons annually by 2028, up from about 17 billion gallons in 2023. Senator Ron Wyden (D-OR) pushed back against big tech executives, writing: "Large data centers can consume up to 5 million gallons per day, or the equivalent to the water use of a town of 10,000 to 50,000 people." A lawsuit filed in Virginia on July 15, 2026, by Dr. Nathan Wangusi, a former AWS water sustainability program manager, alleges that Amazon Web Services published false and misleading statements about water use in its Northern Virginia data centers. Wangusi obtained billing and consumption records via FOIA requests that allegedly "materially differ" from AWS's public claims. The complaint specificially challenges AWS's assertion of a 42% year-over-year reduction in Northern Virginia water use, claiming the FOIA records from Prince William Water Authority show decreases of only 0.8% and 32.1% — and no 12-month window producing 42%. The case is ongoing and unverified in court, but it has pushed issues of resource transparency to the forefront of the data center debate.

The Cost Wave Hitting CIOs

Arif Gasilov, a partner at sustainability advisory firm Gasilov Group, puts the situation bluntly: "What this means for CIOs is that power cost assumptions built in 2023 are wrong in close to half the country." His advice: "A CIO planning an AI deployment that depends on colocation or cloud capacity in any of these states should be asking their provider what the rate structure looks like under the new tariffs and recalculating economics." The large-load tariff wave is the most direct mechanism. In 2025 alone, state regulators approved 29 large-load tariffs. The logic is straightforward: if a substation or power line "would not be built but for a data center, the developer covers the cost, even where the wider grid benefits." This is a fundamental shift from the previous model, where infrastructure costs were socialized across all ratepayers. A database tracking emerging tariffs now lists 77 approved and proposed tariffs across 60 utilities. Add the incentive reversal. Illinois, Arizona, and Ohio have paused incentives; Massachusetts halted data center sales tax exemption applications; New Jersey froze its program; North Carolina is phasing exemptions out. Ohio's projected $142 million in 2026 tax revenue loss turning into an actual $1.6 billion in 2025 has a way of concentrating minds. Greg LeRoy, executive director of Good Jobs First, put it more directly: "Data centers are the last thing states should be subsidizing in 2026." Speaking of Ohio — the states' experience shows that pauses may not be temporary at all. The incentive culture that the data center industry relied on has been broken at the state level, and reversing that shift will be difficult for several years.

Cancellations and Capital Flight

The backlash has already claimed major projects. On July 2, 2026, Blackstone's QTS canceled its $5 billion+ Digital Gateway project in Prince William County, Virginia, withdrawing all filings. The project, an 850-hectare campus that would have been the world's largest data center park, was undone by "entitlements, politics, and infrastructure stress," the company said. QTS reaffirmed commitment to Virginia but redirected capital to areas with "lower community resistance." Three days later, Prince William County supervisors unanimously rejected the Dulles South Innovation Center, a proposed 43-million-square-foot data center campus. In Montgomery County, Maryland, Atmosphere Data Centers' 300-megawatt campus on a former coal plant now faces an 18-month moratorium. Community-driven delays have accelerated dramatically. Data Center Watch reports that grassroots organizations pushed at least 75 data center projects into delay or cancellation in Q1 2026 alone, involving more than $130 billion in investment. Morgan Stanley warns that $156 billion in projects were cancelled or delayed in 2025. And the pipeline is now severely impaired: worldwide, 30-50% of planned data center capacity may not go into service on time, and only about one third of planned capacity is under active development, per Wood Mackenzie. Some of that capital is going overseas. Anthropic is reportedly in discussions to lock down at least 1.4 gigawatts of data center capacity in Australia, with a total investment expected to reach $15 billion. Microsoft is building its first Finnish data center with operations scheduled for late 2026 and a second facility in Lahti, targeting 128MW at full scale. TikTok reportedly is spending $25 billion on Southeast Asia data centers. Malaysia, with construction costs roughly 15% lower than Saudi Arabia, keeps drawing international tech capital. The U.S. is not losing all new builds — but the friction has increased. For every successful project, there are now community meetings, environmental studies, and legal challenges that didn't exist five years ago.

What This Means for Infrastructure Strategy

The practical implications for CIOs break down into four areas. 1. Cost models need rebuilding. Existing colocation contracts and cloud commitments should be audited for tariff exposure. Large-load tariffs can add 20-40% to power costs. For a large AI training cluster, this can swing a business case by tens of millions of dollars annually. Tax incentives that were assumed in a multi-year TCO may disappear at renewal. 2. Geographic concentration is a liability. The QTS withdrawal shows that even large, approved projects can be abandoned. CIOs who rely heavily on Northern Virginia should diversify into regions with more favorable regulatory environments. That might mean emerging U.S. markets like Texas (which has its own grid constraints), the Midwest, or international locations. It also means rethinking edge architecture to sidestep hyperscale moratorium thresholds — distributed smaller facilities can avoid many of the restrictions aimed at 50MW+ or 100MW+ campuses. 3. Lead times will get longer. Bernstein's estimate of 20-25% construction cost inflation and extended timelines means capacity plans need to extend well into 2028. Uptime Institute reports that AI is compressing once-separate risks—costs, capacity forecasting, power, and talent—forcing phased builds and pushing workloads toward third-party sites. The days of quickly spinning up new capacity in core markets are on hold. 4. Contract leverage has shifted. GPU rental prices are climbing steeply — B200 hourly rates are projected to rise from $2.63 to $5.10 by October 2026, a 94% jump. H100 one-year leases have climbed nearly 40% in five months. The compute supply squeeze is making power-aware architecture more important than ever. This is reshaping how CIOs design their AI workloads: with an eye on energy efficiency, not just raw performance. The shift to "power and cooling first" is not just a technical adjustment but a governance one. Energy strategy is now a required component of the digital transformation roadmap.

The Federal Wildcard

Local action is leading, but the federal picture is more complicated. The House Energy and Commerce Committee has taken the position that a federal moratorium is "not on the table," and the committee unanimously advanced the Electric Consumer Protection Act in July 2026, which would require large data centers above 100MW to cover the full grid upgrade costs they trigger. But on the other end, Representative Alexandria Ocasio-Cortez introduced the AI Data Center Moratorium Act in June 2026, proposing an immediate federal pause until "strong national safeguards" are enacted. A Senate draft from Senator Markey, dated July 13, 2026, would require data centers to obtain a federal certificate of public interest before permitting or construction, meeting minimum standards for energy, environment, and economic benefit. And FERC has ordered NERC to develop reliability standards specifically for data centers and other large computing loads by December 31, 2026. The sum of these proposals suggests federal action is more likely to be regulatory cost-shifting than an outright ban. But the direction of travel is clear: data center construction is increasingly treated as a public utility matter for which the industry must bear externalities.

The Community Factor

The protests on July 18 were not just noise. In Paulding County, Georgia, commissioners approved a moratorium as the room "erupted with applause from the public." One resident thanked the commissioners: "I thank you very much for humbling yourselves and making the statement that you did and listening to us today, listening to us for the last three months." Hacker News discussions reflect a more nuanced view. Some commenters note the risk of simply pushing investment to other states — a game of regulatory arbitrage. Others point to reports of Amazon employees facing termination threats after publicly backing limits on new data center construction in Seattle. "Skepticism about corporate tolerance for internal dissent" was a recurring theme — what does it say about the industry when employees who want limits are put at risk? Chuck Girt, CTO at FiberLight, offers a less alarmist take: "I don't think the rate of data center construction changes the direction AI is headed, but it could influence how organizations deploy and access AI at scale. Most enterprises aren't going to build this infrastructure themselves; they're going to rely on cloud and data center environments to provide the compute AI requires." Yet the industry consensus is reflected in Standard Chartered's research note: "Data center pushback in the US signals early limits to the AI capex boom." PitchBook analysts similarly note that "community pushback against data center construction has accelerated in 2026 as new moratoriums have been introduced, with the majority of the change happening at the local level."

The Real Bottleneck Is No Longer GPUs

For the past two years, CIOs focused on securing GPU supply. That was the bottleneck. But in 2026, with construction costs up 20-25% per megawatt, PPA prices spiking 40-120% for wind and solar in markets like Texas, and grid interconnection queues stretching five or more years, the bottleneck has shifted. The constraint is physical infrastructure and its cost structure, not chip supply alone. Oakland, California voted to ban data centers entirely. Maine saw the first proposed statewide moratorium in spring 2026. Even in states where moratoriums haven't passed, the uncertainty itself is costly. Developers are hedging, re-timing, and re-placing projects. Utilities are issuing new tariffs faster than their customers can model them. The climate angle adds another layer of complexity. Microsoft's 2025 fiscal year carbon emissions rose 25% year-over-year to 20 million tonnes CO2e, driven primarily by AI data center construction. Microsoft maintains its 2030 carbon-negative ambition with an internal carbon tax, but the trajectory is clearly moving away from near-term decarbonization. HSBC has already pushed back its net-zero target for its own operations to 2050. If hyperscalers can't reconcile AI growth with carbon targets, the pressure for regulation will only increase.

Recalibrating for the New Reality

The 42-state pushback represents a fundamental realignment in how data centers are perceived and governed. Not every state will enact a moratorium. Not every incentive is gone. But the assumptions of predictable power pricing, generous tax policy, and community acceptance in major hubs need to be reevaluated. CIOs should run scenario planning on their AI infrastructure portfolios: one scenario where large-load tariffs and lost incentives raise costs 20-40%; one where new capacity is delayed by 24 months in their preferred regions; one where distributed edge or international capacity becomes the primary deployment path. As you do this, keep in mind that the current wave of state legislation is not coordinated — and it is not consistent. The regulatory map is shifting quarter to quarter. The deeper question is whether the AI buildout can find equilibrium with the communities that host it. The data center industry has never had to ask this question before, because it has never faced broad, cross-partisan resistance. The fact that in 2026 it does should be treated not as a public relations problem, but as a fundamental input to infrastructure planning. The $690 billion in AI infrastructure spending projected for 2026 will encounter headwinds that didn't exist 12 months ago. The CIOs who navigate this period will be those who diversify their footprints, recalibrate their cost models, and design for flexibility rather than scale alone. Waiting for the regulatory storm to pass is a reasonable instinct — but the storm isn't going to pass. It's the new climate.

As one analyst put it: "A 46% failure rate on 2025-2026 planned capacity is priced in by most vendors, but not by most buyers." The buyers are now catching up. For any CIO whose AI roadmap assumes stable power pricing and available capacity in the usual places, the assumptions themselves have become the biggest risk factor.