Kevin O'Leary details massive Utah AI data center to rival China's tech dominance

Kevin O'Leary unveiled plans for a Utah data center spanning roughly 40,000 acres, with an initial power target of about 3 gigawatts to support AI infrastructure. The project is designed to use a nearby natural gas pipeline for independent power generation, reducing strain on local grids and potentially supplying excess electricity back to the grid. The initiative is positioned as a U.S. effort to expand AI computing capacity and compete with China, with potential interest from hyperscalers and government partners.

Analysis

This is less a single real-estate announcement than a signal that AI capex is migrating toward vertically integrated power generation. If even a fraction of these projects get built, the bottleneck shifts from GPU availability to gas turbines, transmission equipment, switchgear, and water/cooling systems — a more industrial, less software-centric spend stack. That favors the picks-and-shovels names with order backlogs already stretched, while weakening the relative scarcity premium in pure-play data center REITs if self-powered campuses become the template.

The second-order effect is on regional gas demand and localized power pricing, not just national AI sentiment. A dedicated multi-gigawatt load would tighten nearby pipeline utilization and improve takeaway economics for gas producers with direct basin exposure, but it also raises the risk of permitting backlash, methane scrutiny, and interconnection delays that can push monetization out by 12-24 months. In other words, the market may be pricing the headline as immediate AI infrastructure demand when the more investable edge is in the industrial supply chain that can monetize before the campus itself does.

Contrarianly, the market should be careful not to extrapolate this into a clean win for hyperscalers. Self-generated power reduces grid friction but increases execution complexity, which usually means capex overruns, phased buildouts, and lower near-term returns on capital than investors assume. If this becomes politically symbolic rather than economically disciplined, the winners are not the operators but the EPCs, turbine suppliers, transformers, and gas infrastructure owners selling into a multi-year backlog cycle.

Key Decisions for Investors

• Long VRT / ETN on a 6-12 month horizon as the highest-conviction ‘AI power stack’ expression; risk/reward is attractive if data center buildouts keep shifting toward self-generation, with backlog conversion and pricing power supporting estimate revisions.
• Long KMI or WMB versus short IYR over 3-6 months: own gas-pipeline names that benefit from incremental baseload demand and hedge the risk that data center REIT enthusiasm gets capped by self-powered projects.
• Buy calls on GEV or long GEV/short SMCI for 6-12 months: the market is likely underappreciating how much AI capex leaks into turbines, power equipment, and maintenance rather than GPUs alone.
• If the theme broadens, use pullbacks to accumulate DD or CNR-style industrial input names; the risk/reward is skewed because order visibility improves before revenue does, creating a 2-3 quarter lagged upside setup.
• Avoid chasing pure hyperscaler exposure until permitting and power financing are clearer; if this narrative stalls, the downside is primarily in speculative AI infrastructure multiples rather than in the utility/industrial beneficiaries.