Palm Beach rejected a proposed data center project, signaling heightened regulatory risk for AI infrastructure siting in high-density regions.
Palm Beach County's rejection of a proposed 600 MW AI data center signals a critical vulnerability in Wall Street's forecasts for AI electricity demand: announced capacity may not translate into actual grid load if local communities block projects.
On Wednesday, commissioners voted 5–1 to reject Project Tango, an AI-focused campus, despite months of redesigns, additional technical studies, and a recommendation for approval from county planning staff. The outcome has prompted Jefferies to recalibrate its assumptions. The investment bank estimates that Florida Power & Light's capital plan through 2032 includes roughly 6 GW of prospective data center load, representing about $12 billion in capital investment, or 12% of the utility's planned spending. For Jefferies, the Palm Beach vote exposes a deeper uncertainty: local approval, not grid capacity, may be the binding constraint on AI infrastructure deployment.
"Local permitting risk is now load risk. Community acceptance is now forecast risk," said Neil Osnato, founder of Persistence Analytics Group. He argued that utilities and investors should distinguish between announced, permitted, financed, construction-stage, and operating AI loads rather than treating all proposals as equally likely to materialize.
Project Tango underwent an unusually thorough review process. The developer revised the proposal, reduced the data center portion, expanded warehouse space, met with residents and officials, and submitted updated studies covering noise, traffic, and concurrency. Planning staff ultimately recommended approval, finding the application compliant with land development regulations if paired with conditions including noise mitigation, operational limits, and utility service requirements.
Commissioners nonetheless cited unresolved concerns about land-use compatibility, low-frequency noise, water consumption, traffic, and the limited operating history of hyperscale AI campuses. The company had addressed objections from the Palm Beach County School Board with an independent assessment from specialists in acoustics, occupational medicine, and environmental health, which concluded the project would not create adverse health effects for nearby students or residents. The facility was designed as a 600 MW campus supporting AI and cloud workloads that would rely on battery energy storage rather than routine diesel-generator operation. These assurances did not sway the commission.
Whether Palm Beach proves an outlier or an early indicator remains unclear. Communities in Northern Virginia, metro Phoenix, and parts of Georgia have similarly challenged large data center proposals over noise, water use, and land compatibility, though many projects have moved forward after redesigns or negotiated conditions.
The denial without prejudice allows the developer to reapply with revisions, but it introduces material delay. That lag is prompting some AI developers to reconsider deployment timelines. Michael Webber, professor of energy resources at the University of Texas at Austin, noted that large AI projects facing extended utility interconnection timelines are increasingly turning to on-site generation as a bridge. "On-site generation is a 'bridge to power' service," Webber said. "It provides them power to bridge the time delay between when the data center turns on and a utility connection can be made." He expects most such generators will eventually revert to backup power and peak-demand operation once permanent utility service becomes available.
For utilities and investors, the larger question now is how to reflect proposed AI demand in long-term load forecasts before projects secure local approval. As Osnato summarized: "Submitted load is not realizable load. Forecast demand is not durable demand."