These 2 Companies Are Quietly Powering the AI Boom. No, It's Not Nvidia or Palantir.

AI data-center expansion is creating an electricity and grid-access bottleneck, positioning power suppliers and on-site generation providers as potential beneficiaries. Constellation Energy operates roughly 55 GW of generation capacity and recently signed 920 MW of long-term clean-power purchase agreements with 15- to 20-year average terms. Bloom Energy reported Q2 2026 revenue of $1.1 billion, up 166% year over year, and said major U.S. hyperscalers and more than a dozen AI labs, neoclouds, and data-center operators have validated its fuel-cell systems.

Analysis

The investable scarcity is not aggregate generation but deliverable, firm capacity at constrained load pockets. CEG’s earnings upside depends on its uncontracted nuclear output and the price/repricing cadence of its contracted book; long-duration PPAs improve cash-flow visibility but can cap near-term participation if wholesale power rises faster than contracted escalators. The larger second-order beneficiaries may be GE Vernova (GEV), Eaton (ETN), Vertiv (VRT), and Quanta Services (PWR), which monetize interconnection, substation, backup, and power-quality spend regardless of which generator wins.

BE is a higher-beta solution to queue delays, but validation is not equivalent to binding orders, installed economics, or attractive project-level returns. Fuel-cell deployments expose customers to natural-gas basis risk, permitting/emissions constraints, and potentially expensive financing; therefore, the relevant proof point is backlog conversion, gross-margin progression, and disclosed contracted capacity rather than top-line growth alone. A rapid decline in AI capex, accelerated utility interconnection approvals, or lower regional capacity prices would compress the scarcity premium within 1-3 months.

Consensus is likely over-crediting generation owners while underestimating the transmission and electrical-equipment bottleneck. Over 6-18 months, hyperscalers can diversify power procurement geographically and sign PPAs, but they cannot easily bypass transformer, switchgear, and transmission lead times; this favors ETN/PWR/GEV more durably than a pure merchant-power thesis. For CEG, monitor forward power curves and incremental contracted volumes; for BE, require order visibility and margin evidence before treating AI demand as recurring earnings.

Key Decisions for Investors

• Prefer a 6-12 month basket long ETN, PWR, and GEV over a concentrated CEG position: these names capture capex required by both grid-connected and on-site-power solutions. Reassess if utility/data-center capex guidance weakens or order backlog growth decelerates materially.
• Maintain CEG as a tactical 3-6 month long only on evidence of favorable power-price repricing or incremental high-value PPAs; hedge with a partial short in XLU if the objective is scarcity exposure rather than broad regulated-utility beta. Falsify on declining forward power curves or contracting volume that fails to support earnings guidance.
• Do not chase BE solely on AI qualification claims. Set an alert for disclosed hyperscaler purchase commitments, backlog, financed deployments, and sustained gross-margin expansion; absent those data, BE is a high-volatility watch item rather than a core long.
• For a higher-risk 3-6 month relative-value expression, long ETN/PWR versus short BE can monetize the more certain grid-capex cycle while hedging broad AI-power enthusiasm; exit if BE reports binding, large-scale data-center orders with margins that demonstrate attractive unit economics.