Data‑center fights slow AI rollouts
AI projects are hitting the ground-level politics of water, power and rates, not just boardroom budgets — local communities in Virginia and Florida have pushed back or delayed large AI data‑centre proposals over environmental and utility concerns. That local resistance matters because less than 10% of U.S. data‑centre capacity is currently “AI‑ready,” and academics warn AI growth could add hundreds of terawatt‑hours to global power use without clearer measurement and permitting frameworks. (whsv.com, wpbf.com, datacenterknowledge.com, brookings.edu)
In Strasburg, Virginia, residents are arguing over an 87-acre project called Project Tallmadge before a shovel even hits the ground, with questions centered on water use, electricity demand, and what the town gets in return. In Palm Beach County, Florida, a vote on another artificial intelligence data center called Project Tango was pushed from April 23 to July 15 at the developer’s request after months of local backlash. (whsv.com, wpbf.com, wusf.org) A data center is a warehouse full of computers, and an artificial intelligence data center is a version packed with denser chips that pull more power and throw off more heat. More heat means more cooling equipment, and more cooling equipment usually means more water systems, more substations, and bigger utility bills somewhere in the chain. (datacenterknowledge.com, brookings.edu) That is why these fights are showing up at county meetings instead of just in company earnings calls. The bottleneck is no longer only money for servers; it is permits for land, hookups for power, and proof that nearby residents will not absorb the cost in water stress or higher rates. (brookings.edu, dbrs.morningstar.com) In Strasburg, the proposed site sits in the town’s industrial park near Interstate 81 at Exit 296, and the town’s own project page says the plan covers roughly 87 acres. Local reporting says residents have been pressing officials online for details about environmental effects and power demand as the proposal moves through public scrutiny. (strasburgva.com, whsv.com) In Palm Beach County, the controversy has been simmering longer. WUSF reported that county commissioners voted 7-0 in December to delay the application to the April 23, 2026 zoning meeting for more impact study, and the county has now shifted that hearing again to July 15 after a request from the project’s agent. (wusf.org, cbs12.com, wpbf.com) Project Tango became more combustible because it was first approved in 2016 as a regular data center and later came back with a request for 64 additional acres tied to an artificial intelligence buildout. That turned an old land-use approval into a fresh political fight over scale, traffic, noise, energy demand, and whether the original deal still fits the neighborhood around it. (wusf.org, wptv.com) These local fights matter because the United States cannot simply repurpose most existing server buildings for artificial intelligence work. Data Center Knowledge, citing real estate firm Jones Lang LaSalle, reported on April 10 that less than 10 percent of United States data center capacity is currently ready for production artificial intelligence. (datacenterknowledge.com) “Artificial intelligence ready” mostly means a building can handle much heavier electrical loads per rack and the cooling systems needed to keep advanced chips from overheating. Jones Lang LaSalle has said those upgrades can require structural changes for floor loading, power delivery, and heat removal, which is why many older facilities are not easy conversions. (datacenterknowledge.com, jll.com) The energy math behind the push is getting large enough that academics are warning regulators to stop guessing. Brookings said artificial intelligence computing consumed about 460 terawatt-hours of electricity globally in 2025, about the annual electricity use of Japan, and projected that demand could rise to roughly 945 terawatt-hours by 2030. (brookings.edu) A terawatt-hour is one billion kilowatt-hours, which is the unit on a home power bill, so the gap between 460 and 945 terawatt-hours is not abstract. It means every new cluster of chips can trigger very physical questions about transmission lines, substations, backup generation, water access, and who pays when utilities build ahead of demand. (brookings.edu, deloitte.com) That is why the delays in one Virginia town and one Florida county are not small-town side stories. When most of the country’s server space is not built for artificial intelligence, every contested permit, postponed hearing, and disputed water or power study can slow the timetable for the next wave of artificial intelligence rollouts. (whsv.com, wpbf.com, datacenterknowledge.com, brookings.edu)
