Inside our months-long investigation into Kevin O’Leary’s Utah data center debacle

Kevin O’Leary’s Stratos/Wonder Valley proposal for a 40,000-acre, 9GW Utah AI data-center campus—more than double Utah’s average statewide power use—has been effectively paused after intense local opposition, political losses, and litigation over county approval. The project’s scale, potential gas-powered generation, water concerns near the shrinking Great Salt Lake, and opaque permitting process triggered a bipartisan backlash; two county commissioners and Utah Senate President J. Stuart Adams lost their seats amid the controversy. The case highlights rising execution, permitting, and political risk for the AI infrastructure buildout, even as hyperscalers and private-credit providers continue to finance powered-land and data-center development.

Analysis

The relevant read-through is not a demand shock for AI compute but a rising cost-of-time for capacity. Community opposition, litigation and electoral sensitivity convert permitting from a largely local execution item into a portfolio-level bottleneck; projects with contracted power, interconnection rights and operating campuses should command a scarcity premium versus speculative "powered land." This favors Digital Realty (DLR), Equinix (EQIX), Vertiv (VRT), Eaton (ETN), GE Vernova (GEV), Constellation Energy (CEG) and Vistra (VST) more than greenfield developers whose value rests on announced—not deliverable—gigawatts.

For hyperscalers, the near-term effect is capex inefficiency rather than lost AI revenue: delays force capacity to be sourced in higher-cost markets, extend temporary leases, and may require paying for generation or transmission before utilization. META has the highest narrative and multiple sensitivity because its AI spend has the least directly observable external revenue offset; MSFT is relatively insulated by enterprise demand and its broader geographic footprint. NVDA is protected in the next 1-3 quarters by supply-constrained accelerator demand, but a 6-18 month accumulation of delayed campuses would push out cluster deployments and raise cancellation/deferment risk at the margin.

Consensus is likely underpricing the political-feedback loop: each poorly handled project increases scrutiny of tax abatements, water rights, air permits and utility cost allocation elsewhere. The important indicator is not announced data-center power capacity but the conversion rate from announced capacity to signed utility service, turbine delivery, and a creditworthy tenant lease. A broad AI infrastructure short is premature while hyperscaler balance sheets remain strong; the more actionable expression is long scarce, already-operational infrastructure against developers and capex-sensitive buyers.

Key Decisions for Investors

• Initiate a 6-12 month long DLR / short META pair at roughly equal dollar exposure. DLR monetizes scarce commissioned capacity and enterprise demand, while META bears disproportionate greenfield permitting and capex-duration risk; target 15-20% relative return, with thesis invalidated if META materially raises AI monetization guidance or DLR lease spreads soften for two consecutive quarters.
• Overweight VRT and ETN on 3-9 month pullbacks rather than chase headline-driven strength. Delayed builds do not eliminate demand for power distribution, cooling and retrofit work, and more difficult site development can increase equipment content per delivered MW; reduce exposure if hyperscaler aggregate capex guidance falls more than 10% or backlog conversion slows.
• Own CEG or VST selectively versus broad utilities for a 6-18 month horizon, favoring contracted/dispatchable power scarcity over merchant greenfield development. The risk is regulatory intervention shifting grid-upgrade costs to large-load customers or a sustained decline in power-forward curves; use a 12-15% downside stop from entry.
• Avoid adding incremental long exposure to META solely on AI-capex optimism until management discloses a clearer split between operational capacity, contracted power and merely planned capacity. Monitor quarterly capex guidance, lease commitments and depreciation growth; a capex increase without a corresponding revenue or utilization disclosure is a negative quality signal.
• Set an alert for state-level changes to data-center tax incentives, large-load interconnection rules, or water/air permitting requirements in Texas, Virginia, Arizona and Utah. A cluster of restrictive actions would be a catalyst to rotate from hyperscaler capex beneficiaries toward existing data-center REITs and regulated/dispatchable power owners.