Where AI data centers are reducing power bills : The Indicator from Planet Money

Wholesale electricity prices near data centers have risen 267% over the last five years, sparking local backlash against new AI data centers. Experts contend data centers may be a scapegoat for an aging U.S. grid, raising questions about who should bear costs and whether the grid can handle accelerated load growth from AI infrastructure.

Analysis

Local political pushback against hyperscale power draws is already compressing the supply of shovel-ready interconnection capacity; expect clusters of delayed projects and rising land/interconnection bid premiums within 6–24 months as utilities tighten queue rules. That reordering will create a geography shift in where compute gets built — states with faster permitting but weaker grids will see outsized short-term wholesale-price volatility and higher capacity charges.

The dominant commercial response will be economic: hyperscalers will accelerate onsite generation + storage and more aggressive PPA contracting to avoid volatile wholesale exposure, creating a multi-year demand tailwind for grid-scale batteries, inverter/plant integrators, and renewables developers. Contracts moving from merchant pricing to fixed long-term PPAs also shift credit risk away from local utilities and toward developers and corporate buyers, concentrating counterparty risk on large tech balance sheets.

On the supply side, expect 12–36 month lead times to create pricing power for transmission and substation OEMs (transformers, switchgear, HV cable) and for turnkey EPCs; that means order books and margin expansion before broad utility rate-base recovery is recognized. Offsetting these capex-driven winners is a credible technological risk: a 10–20% improvement in compute energy efficiency per generation (chips/packaging/thermal) would materially shave long-run grid demand growth and cap a multi-year capex cycle.

Policy is the wild card: accelerated federal transmission funding or FERC queue reform could materially shorten upgrade timelines (18–60 months), compressing premiums on transmission OEMs and easing data-center siting constraints. Conversely, aggressive local moratoria and punitive demand tariffs enacted over the next 3–12 months would strand data-center growth in high-cost regions and reprice nearby REITs and municipal utility credit spreads.

Key Decisions for Investors

• Long AES (AES) — 12–24 months: buy shares or 18-month calls. Rationale: primary beneficiary of grid-scale storage and integrated renewable+storage PPAs as hyperscalers build behind-the-meter solutions. Risk/Reward: ~30–40% upside if storage adoption accelerates; 15–20% downside if merchant power prices collapse or project economics stall.
• Long ABB (ABB) — 12–36 months: buy shares or buy-dated LEAPS. Rationale: transformer/switchgear supplier with pricing power given 12–36 month lead times and order backlog from transmission/substation upgrades. Risk/Reward: asymmetric 2:1 upside/downside if transmission build-out materializes; downside from prolonged permitting freezes.
• Pair trade — Short Digital Realty (DLR) / Long American Electric Power (AEP) — 6–12 months: short DLR (growth-exposed REIT reliant on new builds) and long AEP (regulated transmission footprint). Rationale: moratoria and tighter queue access compress data-center growth but increase regulated utility rate-base investments. Risk/Reward: target 20–30% net pair return; cut losses at 10% on either leg if sector sentiment reverses.
• Convex view on semis — Long NVDA Jan-2028 calls (select strike): 18–30 months. Rationale: even with potential siting slowdowns, secular AI demand persists and tech-side efficiency gains are a longer-cycle offset — NVDA captures the majority of upside. Risk/Reward: 3:1 asymmetric payoff if AI deployments continue, but significant volatility if hyperscaler capex is deferred in the near term.