‘Irresponsible’: backlash as Utah approves datacenter twice the size of Manhattan

Utah approved the Stratos AI datacenter, a 40,000-acre project across three sites that would require about 9GW of power and significant water use, triggering major public backlash. Critics say it could raise the state's planet-heating pollution by roughly 50%, worsen Great Salt Lake stress, and increase local temperatures by 2F to 12F; the governor now says the project must not harm the lake or raise power bills. A referendum effort is underway, and developers have withdrawn an initial water-diversion filing while planning a new application.

Analysis

This is less an isolated Utah zoning fight than an early signal that AI infrastructure is colliding with physical scarcity constraints. The first-order losers are not just the project sponsors; it is the adjacent power stack—utilities, gas infrastructure, cooling, water services, and eventually ratepayers—because mega-loads force capital spending upfront while monetization remains uncertain over a multiyear buildout. The market is underpricing the likelihood that permitting risk, not compute demand, becomes the binding constraint on datacenter growth in inland western states.

The second-order impact is a policy wedge: once local backlash becomes tied to power bills, drought, and ecosystem damage, governors and regulators will force phased approvals, bespoke water contracts, and possibly curtailment rights. That favors incumbents with regulated generation, grid interconnection optionality, and site diversification, while hurting pure-play datacenter developers and hyperscale-adjacent land banks that rely on frictionless scale. The more this becomes a referendum on “AI vs. local resources,” the more every new project inherits a higher cost of capital and a longer permitting timeline.

The contrarian miss is that the bearish ESG headline may not translate into immediate project cancellation; instead, it can accelerate a redesign toward smaller phased assets, on-site generation, and higher-value workloads. That means the near-term trade is not to short AI broadly, but to fade the “infinite buildout” narrative in names exposed to power scarcity and approval bottlenecks. Over 3-12 months, the key catalyst is whether local referendum mechanics and state-level conditions force a precedent that other jurisdictions copy.

Key Decisions for Investors

• Short a basket of datacenter-enablement developers / land plays most exposed to speculative giga-campus expansion; cover on any state-level approval with strict phased limits. Risk/reward: 2-3x downside if permitting precedents spread, but sharp squeeze risk on a single headline approval.
• Long regulated utilities with large renewables/nuclear backlogs and transmission upside; they can monetize the buildout while controlling the rate case process. Prefer 6-12 month horizon as grid interconnection bottlenecks become more visible.
• Pair trade: long utility infrastructure beneficiaries (NEE, DUK, NEE if seeking transmission exposure; or grid equipment via ETN/PWR) vs short unregulated data-center REIT exposure (EQIX/AMT only if valuation already embeds peak growth). Target is multiple compression in the losers as capex intensity rises.
• Buy out-of-the-money puts on high-beta AI infrastructure enablers into any new local approval fight. The catalyst window is 1-3 months; this is a volatility trade more than a directionally clean fundamental short.
• Accumulate LNG/gas midstream only on pullbacks if you believe onsite gas generation will be normalized; otherwise avoid outright long exposure because political backlash can force curtailed economics and stranded-capital risk over 12-24 months.