XLU's AI Power Story Crumbles as Texas Freezes Data-Center Demand

Texas ERCOT's Batch Zero review has paused new data-center interconnections, while AEP Ohio's pipeline fell by more than half after application fees were introduced, challenging the load-growth assumptions supporting utility AI-capex plans. Exelon reduced its data-center pipeline to 36GW from 43GW, with only 11GW deemed high probability and 4GW supported by signed agreements and $1B in collateral. XLU has returned roughly 2% YTD versus 13% for SPY and faces valuation pressure as its AI-demand thesis is audited, while the 4.77% 10-year Treasury yield and lower-cost VPU offer alternatives.

Analysis

The key valuation question is no longer aggregate load requests but conversion from request to collateralized, commission-approved load. Utilities that have embedded speculative demand in transmission timing face asymmetric downside: capex can be deferred, but a rate-case disallowance or lower authorized ROE would impair both EPS growth and the premium multiple assigned to the AI-power narrative. AEP is most exposed because its capital-plan credibility is now directly linked to a queue being tested for economic commitment; DUK and SO merit monitoring for similar concentration, even if their demand mix is presently more diversified.

This is favorable for companies monetizing power through executed contracts rather than through forecast rate-base additions. CEG's earnings sensitivity is tied to delivered energy, capacity values, and contracted nuclear output—not simply a utility commission accepting a future construction budget. EXC is a quieter relative beneficiary: disciplined queue filtering reduces headline gigawatts but can improve capital efficiency, regulatory defensibility, and the probability that its authorized spend earns on schedule. Over the next 1-3 months, each utility's next disclosure of deposits, signed service agreements, and revised load conversion rates matters more than headline pipeline size.

Consensus may overstate the read-through from Texas to the national power buildout. A tighter interconnection process can delay load realization but also shifts bargaining power toward incumbents with existing generation, transmission access, and creditworthy contracted customers. The bearish case becomes structural only if hyperscalers cancel rather than relocate projects, or if regulators require utilities to socialize major dedicated-network costs; absent that, the likely outcome is slower, better-funded load growth. Higher long-end yields remain the more immediate broad-sector multiple risk, making a diversified utilities ETF a poor expression of a selective AI-power thesis.

Key Decisions for Investors

• Initiate a 3-6 month pair trade: long CEG / short AEP in equal dollar amounts. The trade isolates contracted merchant-power monetization from speculative rate-base growth; target 15-20% relative outperformance, with a stop if AEP demonstrates a material increase in fully collateralized load or CEG loses contracted-price support.
• Prefer EXC over AEP for regulated-utility exposure over 6-12 months. Add only after management quantifies high-probability load conversion and confirms capital-plan funding; thesis is falsified by a capex-plan reduction, adverse rate-case treatment, or meaningful deterioration in customer collateral quality.
• Avoid adding broad XLU exposure ahead of next earnings until constituent disclosures distinguish signed load from nonbinding requests. For passive utility exposure, use lower-cost VPU only as a rates-sensitive defensive allocation, not as an AI infrastructure position.
• Set an event-driven alert around 10-year Treasury yields: a sustained move above 5.0% would likely pressure utility valuation multiples regardless of data-center demand quality. Conversely, a 50bp decline in long rates paired with validated load commitments would favor NEE and SO more than AEP because their diversified growth platforms can re-rate without relying solely on one interconnection queue.