Tech firms pledge to pay for AI data centre power costs. But will they?

Market structure: Hyperscalers (GOOGL, GOOG, MSFT, AMZN, META) are the immediate winners — the pledge reduces long‑run marginal energy risk for AI compute and preserves growth runway, while regulated utilities can win contracted revenues and ratebase additions from interconnection/upgrades. Residential rate relief is uncertain because enforcement sits with state PUCs and grid operators; expect project-level outcomes over 6–24 months as interconnection queues and PPAs are negotiated.

Risk assessment: Key tail risks are enforcement failure (pledge = PR), rapid natural‑gas price shocks from geopolitics (Henry Hub > $6/mmBtu would likely push US power prices +20–40% in 1–3 months) and local bans/permit delays that strand tech capex. Shorter horizon (days–weeks) sees headline volatility; medium (3–12 months) sees margin and cash‑flow impacts from on‑site generation capex; long (1–3 years) is grid reshaping and possible favorable regulatory rate design for large customers.

Trade implications: Favor selective exposure to hyperscalers while hedging policy and commodity risk: size modest longs in GOOGL/MSFT/AMZN/META (1–2% each) funded from defensive cash, and buy 3–6 month 10–20% OTM call spreads to cap cost. Add a conservative 1% long in regulated utilities ETF (XLU) to capture ratebase upside, and a 1% 3–6 month natural‑gas call spread ($5→$7) as a geopolitical hedge.

Contrarian angles: Consensus assumes pledge converts to built capacity; history (telecom/renewables) shows many corporate promises become REC purchases or negotiated below‑market rates — meaning utilities may not get the anticipated capex and tech FCF could compress 1–4% if firms fund generation. Watch state PUC filings and company CAPEX guidance in the next 30–90 days as the decisive signal; if filings are absent, downside for utility/tech sentiment is underpriced.