The Shadow Grid: Big Tech’s Quiet Takeover of America’s Power

Operation Epic Fury (Feb 28, 2026) struck >1,000 Iranian targets and killed Supreme Leader Ali Khamenei, with Anthropic’s Claude (via the Pentagon’s Maven system and Palantir) used operationally — a catalyst reframing AI compute as national security infrastructure. Energy implications are acute: U.S. data-center power demand doubled from 2018–24 and could triple by 2028, ~680 planned U.S. data centers would need the energy equivalent of ~186 large nuclear plants, and PJM capacity auctions cleared near $329–$333/MW‑day while Brent approached ~$120/bbl. Portfolio action: overweight infrastructure and on-site generation/value chain suppliers (e.g., GEV, ETN, HUBB, PWR, VST, CEG, CCJ, UEC, BWXT, XYL, WTS, MWA, PANW, FTNT) and underweight utilities with heavy data-center exposure (e.g., D, DUK, AEP) given the risk of hyperscalers going behind-the-meter.

Analysis

The military use of AI has recast hyperscaler power as a de facto national-security asset, shifting the value driver from pure LCOE to resilient capacity and sovereign control of supply chains. Expect hyperscalers to favor behind‑the‑meter generation, long-duration storage, and dedicated transmission where 100–500 MW increments of resilient capacity can be contracted or built faster than regulatory markets can adapt; that changes the profitability calculus for equipment OEMs, EPCs, and fuel suppliers for the next 3–7 years.

Second‑order winners are capital‑intensive industrial suppliers (turbines, switching, modular reactors), specialist utilities pivoting to bespoke commercial contracts, and OT/cybersecurity vendors that protect bespoke grids from nation‑state actors. Stranded‑asset risk concentrates in regulated utilities that built transmission for data‑center load forecasts — when >10–15% of a utility’s projected incremental load is defeasible, credit spreads and allowed ROE assumptions should be re‑priced within 12–36 months.

Key tail risks: a regulatory backlash that forces hyperscalers to remain on public grids, rapid normalization of Middle East logistics that eases fuel price pressure, or a major cyber incident that raises capital costs for private grids. Near‑term catalysts to monitor are multi‑GW PPAs, high‑profile SMR licensing milestones, Pentagon procurement contracts tying cloud providers to classified compute, and large AI IPOs that unlock hyperscaler capex commitments; any of these can compress timelines from years to quarters.

Key Decisions for Investors

• Long Quanta Services (PWR) 2–4% portfolio weight: entry on pullback to 5–10% below 30‑day VWAP. Thesis: EPC execution and grid buildout demand supports 25–40% upside over 12–24 months. Risk: project delays and commodity inflation; stop at 12%.
• Long GE Vernova (GEV) 3% weight for diversified gas turbine + grid equipment exposure: target 30% upside in 12–24 months as captive generation orders accelerate. Hedge execution risk with 6–9 month covered calls if share move is rapid.
• Pair trade — Short Dominion Energy (D) 1–2% vs Long Constellation Energy (CEG) 1–2%: time horizon 6–18 months. Rationale: D is exposed to stranded transmission asset risk; CEG benefits from long PPAs and creditable generation. Risk/reward ~2:1; close if state regulators mandate captive‑customer protections.
• Long BWX Technologies (BWXT) + Uranium Energy (UEC) split 60/40: BWXT for defense/SMR components, UEC as uranium‑price lever. Timeframe 12–36 months; target combined return 40% if SMR and PPA tailwinds materialize. Volatility high; use 15% stop.
• Long Palo Alto Networks (PANW) 2% as insurance on OT/cyber demand: buy on weakness or into earnings, target 20–30% in 9–12 months as defense/enterprise OT spend re‑rates. Key risk: gross margin compression if competition intensifies.