‘Clearly, people hate data centers’: Sam Altman admits Americans hate what he’s wrought as his own data centers head quits

OpenAI signaled mounting delays risk: CEO Sam Altman said Americans are “pretty negative on AI” and “people hate data centers,” amid 70% public opposition (48% strongly). The article cites 120+ state moratorium proposals in 38 states, Texas ERCOT facing ~474 GW of new grid-connection requests, and Pennsylvania removing AI data centers from fast-track permitting—potentially pressuring OpenAI’s planned $50B+ compute spend and Stargate ramp-up (nearly 7 GW planned capacity, with added >1 GW Michigan, 3.2 GW Georgia, and ~8 GW in southern Ohio). Overall, regulatory/community pushback is likely to slow data-center capacity builds for hyperscalers and related power infrastructure.

Analysis

This is less a demand shock than a timing shock: the market is underpricing how quickly local permitting hostility can turn AI capex into a longer-duration cash flow stream. That hits the highest-multiple infrastructure names first because their valuation assumes fast conversion from announced megawatts to shipped product, installed equipment, or leased capacity; a 2-4 quarter slip can compress forward multiples even if the end demand case is intact.

The second-order winner is not necessarily "power" broadly, but incumbents with already-locked interconnects, rate-base protection, or dispatchable generation that can monetize scarcity while new campuses get delayed. That favors names with existing footprint and permitting moats, while punishing the ecosystem built around rapid campus replication: electrical gear, cooling, construction services, and data-center REITs with exposure to new-build absorption. The more regulators add affordability/community screens, the more the marginal project becomes a bespoke political asset rather than a scalable rollout.

Contrarian view: consensus may be too bearish on total AI electricity demand and too bullish on the pace of greenfield buildout. If hyperscalers respond by concentrating spend into fewer, larger, politically safer sites or by securing behind-the-meter generation, the bottleneck shifts from "how much compute" to "who controls the scarce infrastructure," which is structurally better for incumbents than for project developers. Falsifiers are fast permitting wins, federal preemption, or next-quarter guidance from major AI/infra beneficiaries showing no slowdown in order intake or backlog conversion.

Key Decisions for Investors

• Short VRT on any 3-5% bounce; 1-3 month catalyst is backlog timing risk and multiple compression if data-center permits/interconnects slip. Cover if management confirms unchanged 2026 order conversion or if bookings reaccelerate.
• Pair trade: long CEG / short VRT for a 3-6 month horizon. Thesis is that approved generation and scarcity pricing are cleaner beneficiaries than equipment vendors exposed to delayed campus starts. Risk is a rapid policy fix that restores buildout velocity.
• Buy XLU selectively on pullbacks as a lower-beta hedge against regulatory friction; utilities with already-approved rate-base investment can still monetize grid upgrades even if speculative data-center projects stall. Falsify if state regulators cap cost recovery or data-center load growth rolls over materially.
• Avoid chasing data-center REITs and electrical contractors until we see actual permitting recovery; if the market bids them on headline AI enthusiasm alone, use that strength to fade.