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

AI data-center expansion is creating a potential electricity-supply and grid-connection bottleneck, benefiting both centralized generation owner Constellation Energy and on-site power-system provider Bloom Energy. Constellation operates roughly 55GW of capacity and recently signed 920MW of clean-power purchase agreements with 15-20 year average terms. Bloom reported Q2 2026 revenue of $1.1B, up 166% year over year, and said its systems have been validated by major U.S. hyperscalers and more than a dozen AI labs, neoclouds, and data-center operators.

Analysis

The investable bottleneck is not aggregate U.S. generation but deliverable, firm power at constrained load pockets. That favors equipment that bypasses interconnection queues—BE, gas turbines (GEV), switchgear (ETN) and data-center electrical infrastructure (VRT)—before it fully accrues to incumbent generators. CEG's upside is more conditional: long-dated contracted generation provides visibility and credit quality, but also means new demand may first raise renewal values and forward prices rather than create an immediate volume-driven earnings step-up.

BE is the higher-beta expression, but its valuation should be anchored to deployable backlog, service economics, fuel pass-through, and customer financing—not revenue growth alone. On-site fuel cells solve time-to-power, yet expose customers to natural-gas price and pipeline-availability risk; a sustained Henry Hub spike or local gas constraint can impair the total-cost-of-ownership advantage versus grid power. The likely 1-3 month catalyst is disclosed hyperscaler orders, backlog conversion and gross-margin progression; the 6-18 month risk is that utility interconnection timelines improve or large customers choose temporary turbines and then migrate to grid supply.

Consensus may be overpaying for the broad 'AI power' label while underweighting the distinction between power availability and power affordability. CEG is a lower-volatility scarcity/contract-renewal vehicle, whereas BE must prove that AI demand translates into repeatable, financed deployments rather than pilot validation. A meaningful compression in forward power prices, nuclear outage issues, or PPA pricing below market expectations would challenge CEG; BE is falsified by order growth without operating-margin and operating-cash-flow conversion.

Key Decisions for Investors

• Maintain a core long CEG on a 6-18 month horizon, but size as a defensive power-scarcity exposure rather than a near-term AI revenue trade. Add only after evidence that contracted volumes are being repriced above prior portfolio economics; reduce if forward power curves weaken materially or nuclear availability/guidance deteriorates.
• Use BE only as a tactical 1-3 month long following independently disclosed hyperscaler purchase commitments, funded backlog, and gross-margin guidance. Target at least 2:1 upside/downside from entry; exit if bookings accelerate without positive operating-cash-flow trajectory or if customer financing expands materially.
• Prefer a barbell pair of long CEG / short BE for investors seeking lower AI-power beta until BE discloses contract economics and deployment cadence. The pair captures dependable firm-power repricing while hedging speculative multiple risk; cover the short if BE reports multiple large, non-cancelable hyperscaler orders with margin-accretive service attachment.
• Watch GEV, ETN and VRT as second-order beneficiaries of constrained interconnection capacity. These names monetize generation and electrical-delivery capex regardless of whether the end customer selects nuclear PPAs, fuel cells, or utility-scale generation; initiate only after data-center capex guidance confirms that power-delivery spending is not being deferred.