Kevin O'Leary's 40,000-Acre Utah AI Data Center: Inside the Collapse

Kevin O'Leary set out to build what would have been the largest data center on earth: a 40,000-acre AI campus in Utah drawing nine gigawatts of power, more than double the average electricity consumption of the entire state. The project did not come together, and The Verge spent months reconstructing why. Features writer Josh Dzieza walked through the findings in a conversation published this week.

The scale is the first thing to understand. Nine gigawatts is roughly the output of nine large nuclear reactors running at full tilt, and no single customer has ever asked a U.S. grid for that much capacity. For comparison, the largest existing hyperscale campuses operate in the hundreds of megawatts. O'Leary's pitch bundled land, power and AI compute into one package at a size the electricity system has never been asked to serve.

According to the investigation, the deal ran into the mundane machinery that governs big infrastructure: land assembly across thousands of acres, water rights, transmission access and the patchwork of local and state approvals that a project of that footprint requires. Promotional momentum and renderings moved faster than permits and interconnection queues. Dzieza's reporting traces how the gap between the announced vision and the paperwork on the ground widened over time.

The story lands in a moment when data center proposals are colliding with real constraints. Grid operators across the country have reported multi-year backlogs for new large-load connections, and several utilities have revised forecasts upward as AI training and inference demand climbs. Communities in states like Virginia, Georgia and Ohio have pushed back on water use, rate increases and noise. Utah's project ran into the same friction, just at an unusually public scale.

O'Leary, known best as a television investor, is not a hyperscaler, and the venture depended on assembling partners, capital and offtake agreements that never fully materialized. The Verge's account suggests the project functioned partly as a marketing exercise, with announcements outpacing signed commitments. That pattern is becoming familiar as developers court attention and investors before the engineering is settled.

What to watch next is whether any developer can actually deliver gigawatt-class single sites, or whether the industry settles into a collection of smaller, distributed builds stitched together by new transmission. Regulators in several states are already weighing how to allocate grid costs when one customer's demand dwarfs everyone else's. Utah's outcome may end up as the case study both sides cite.