# Moody's warns that the data center boom is creating fiscal risks governments cannot ignore > Source: > Published: 2026-06-25 16:43:32+00:00 *As AI infrastructure demand runs hot, a credit warning from Moody's and legislative pushback in at least 14 states are shaping which regions will actually get built out and which will be left behind.* The math on the AI infrastructure buildout has always been staggering. The five largest hyperscalers are set to spend $710 billion in capital expenditures in 2026 alone, and data centers are where most of that lands. What the projections rarely factored in is what happens when the towns, utilities, and water systems on the receiving end say they can't absorb it. Moody's Ratings now has an answer, and it's blunt: rapid data center growth is creating meaningful credit risks for state and municipal governments, driven by electricity and water demands that strain the very infrastructure these projects depend on. The credit warning matters because it formalizes a trend that local officials have been raising for two years. Data centers are extraordinarily power-hungry and, depending on cooling method, can consume roughly 2 million liters of water per day per 100 megawatts of capacity, according to the International Energy Agency. When a hyperscale facility arrives in a mid-size county, it doesn't come with its own substation or reservoir. It arrives expecting a grid and a water system that were not sized for it, and the costs of upgrading them fall somewhere between the developer and the public in ways that local governments are increasingly unwilling to absorb quietly. At least 14 states have considered or enacted moratoriums in 2026, according to the National Conference of State Legislatures. New York's legislature passed the Responsible Data Center Development Act in the final hours of its 2026 session, clearing the Senate 44-16 and the Assembly 102-39. The bill, championed by State Senator Kristen Gonzalez, would impose a one-year statewide freeze on new permits for hyperscale facilities with peak loads above 20 megawatts and now awaits Governor Hochul's signature. Vermont's S.205 proposes a moratorium running through July 2030. Oklahoma's SB 1488 freezes facilities above 100 megawatts until November 2029 while the state public utilities commission studies grid and water impacts. Local governments haven't waited for their legislatures. Denver imposed a one-year pause in May. Minneapolis paused facilities above 350,000 square feet in the same month. Baltimore City's pause went to the mayor in early June. In Michigan, more than 25 communities had active moratoriums or zoning rewrites underway as of spring 2026, with the Ypsilanti Community Utilities Authority going further and simply cutting off new water and sewer connections to data centers entirely for twelve months. An OpenAI-Oracle data center project in Saline, Michigan was voted down by residents before construction began anyway, according to Fortune's reporting in May. The deeper problem isn't public sentiment. It's time. Between May 2024 and June 2025, at least 36 US data center projects were delayed or blocked, disrupting an estimated $162 billion in investment, according to data cited by Ropes & Gray. PJM Interconnection, which manages the grid for much of the mid-Atlantic and Midwest, is reporting interconnection wait times averaging roughly eight years for new energy projects. Nearly half of US data center projects stall at the zoning and permitting stage. Demand isn't the constraint. Getting a shovel in the ground is. ## Where capital goes next Moody's noted directly that resistance in some states will push developers toward jurisdictions that are ready to move. That's not a prediction so much as a description of what's already happening. When a project gets blocked in a community with grid constraints and an active moratorium, the developer doesn't abandon the project. They call Texas, or Georgia, or look at what Ohio and Indiana are offering. The states that have kept permitting open, built transmission capacity, and offered tax incentives without attaching environmental review hurdles are receiving disproportionate inbound interest from hyperscalers and co-location developers alike. The leverage question for AI infrastructure investors and startups is real. A site that arrives with power, permitting, and community approval already secured is not just faster to build. It commands better financing terms, as lenders and investors increasingly price permitting risk into cost of capital. Developers are now advancing permitting well before any tenant commitment, treating an approved site as the asset itself. In that environment, states that streamline approvals aren't just being developer-friendly, they're compressing the timeline between hyperscaler demand signal and actual rack capacity online. The grid cost negotiation is also evolving. In Minnesota, Google signed a deal with Xcel Energy to fully fund 1,900 megawatts of clean energy additions, covering wind, solar, batteries, and grid infrastructure upgrades. In Louisiana, Meta committed to funding seven natural gas plants, more than 200 miles of transmission lines, and associated battery systems as part of its arrangement with Entergy. Both deals reflect a policy shift: more than 30 states have proposed or enacted additional tariffs or cost-sharing frameworks requiring large load customers to pay for the infrastructure their demand creates. That's a reasonable outcome for ratepayers. It's also a higher upfront cost for developers, and in tighter permitting environments, it's one more reason to choose the path of least resistance. Frankly, the Moody's warning lands at the exact moment hyperscalers can least afford to ignore it. A $710 billion capital commitment is only as valuable as the sites it can actually reach. If permitting bottlenecks throttle supply while demand keeps accelerating, the shortage won't show up in earnings calls immediately. It'll show up twelve to eighteen months from now, when the rack capacity that was supposed to be online isn't, and the AI workloads waiting for it have nowhere to go. **Also read:** [Andreessen Horowitz bets $15 million that network automation is the next AI infrastructure bottleneck](https://startupfortune.com/andreessen-horowitz-bets-15-million-that-network-automation-is-the-next-ai-infrastructure-bottleneck/) • [DeepSeek is doubling its headcount and the real question is whether its famous efficiency survives the growth](https://startupfortune.com/deepseek-is-doubling-its-headcount-and-the-real-question-is-whether-its-famous-efficiency-survives-the-growth/) • [Adobe is buying Topaz Labs because building on-device AI from scratch would take too long](https://startupfortune.com/adobe-is-buying-topaz-labs-because-building-on-device-ai-from-scratch-would-take-too-long/)