Power Buildout and Utility Exposure to AI Capital Spending Slowdown
The physical layer of the artificial intelligence infrastructure buildout—encompassing power generation, electrical equipment, and cooling systems—is experiencing a critical structural divergence. While the long-term spending narrative remains highly bullish, physical constraints, supply chain bottlenecks, and complex project execution are delaying revenue recognition for physical equipment providers. This has introduced significant financial volatility, as demonstrated by Vertiv Holdings Co.'s (VRT) Q2 2026 revenue miss and subsequent legal backlash, alongside escalating debt levels at major power providers like Vistra Corp. (VST) and Constellation Energy (CEG).
Physical Bottlenecks and the Vertiv Q2 2026 Case Study
Vertiv, a pure-play provider of critical digital infrastructure and liquid-cooling thermal chains, reported its second-quarter 2026 financial results on July 29, 2026. While the company posted strong bottom-line results, its top-line performance revealed the operational friction of building massive AI data centers:
"On July 29, 2026, Vertiv reported its financial results for the second quarter of 2026. Among other items, Vertiv reported net sales of $3.27 billion, missing the consensus estimate of $3.38 billion. Vertiv said that its revenue reflected 'minor timing shifts,' primarily due to temporary supply chain congestion and multi-phased project execution as deployments become larger and more complex."
This revenue miss triggered a sharp 10% single-day decline in Vertiv's stock price, which has continued to slide, down 9.8% over the past month and 18.1% over the past three months to $261.95. Despite the Q2 revenue delay, Vertiv’s management actually raised its full-year 2026 guidance, forecasting net sales of $14.0 billion (up 37% YoY) and adjusted EPS of $6.65 to $6.75, signaling that demand remains robust but is being throttled by physical delivery constraints.
The top-line miss and stock drop immediately sparked a wave of shareholder class-action lawsuits and securities fraud investigations by Pomerantz LLP, Schall, Brown & Schwartz LLP (SBS), and other firms, alleging that management made misleading disclosures about its supply chain capacity and project timelines. This legal backlash highlights the extreme market sensitivity and financial volatility surrounding any delay in the AI spending-to-revenue conversion chain.
Escalating Debt and Leverage in the Power Buildout
The independent power producers (IPPs) and utility companies funding the massive electrical grid expansions required for AI data centers are taking on substantial debt, leaving them highly exposed to any potential capex slowdown.1
Vistra Corp. (VST)
Vistra's balance sheet remains highly leveraged as it finances its capital-intensive power buildout. As of June 30, 2026, the company’s total debt stands at $19.89 billion (approaching the $22 billion threshold being tracked in Power Buildout and Utility Exposure to AI Capital Spending Slowdown) against just $435 million in cash. Vistra's trailing twelve-month (TTM) revenue contracted 5.5% YoY to $19.21 billion, and its stock has plunged 16.6% over the past month to $136.21, reflecting a significant cooling in the hyper-bullish AI utilities trade.
Constellation Energy Corp. (CEG)
Constellation Energy, which is actively positioning its nuclear fleet to power AI data centers, has also accumulated a massive debt load. As of June 30, 2026, Constellation's total debt stands at $24.70 billion against $1.08 billion in cash. Although the company raised its 2026 adjusted operating earnings guidance to $11.50–$12.50 per share following a Q2 earnings beat of $2.55 per share, its stock has declined 5.3% over the past three months to $272.88, indicating that public markets are starting to price in the heavy financing costs of the power buildout.
Eaton Corporation PLC (ETN)
Eaton, a global leader in electrical power management, reported solid Q2 2026 revenue of $8.53 billion (up 21.4% YoY) and an EPS beat of $3.15. However, Eaton is also carrying a substantial debt load of $21.33 billion as of June 30, 2026, to fund its capacity expansions.
The Re-Leasing and Stranded Asset Risk of Custom AI Facilities
The physical infrastructure of AI data centers carries unique residual value risks that traditional data center funding models fail to account for. As detailed by legal and financial analysts:
"AI-focused data centres serving neocloud counterparties (specialist AI cloud operators such as CoreWeave, Lambda Labs, Crusoe, Nebius, Applied Digital and Fluidstack) exhibit risk profiles that diverge from traditional infrastructure... AI-optimised facilities are purpose-built with specialised cooling systems, high-density power configurations and custom layouts designed for specific GPU rack topologies. Liquid cooling alone often accounts for 10% of total construction costs on average. This customisation creates material re-leasing risk: converting an AI facility to general-purpose use is expensive and complex... In a downside scenario, these facilities may function as stranded assets with limited alternative use and depressed liquidation value."
If the neocloud providers operating within these custom facilities default or fail to renew their short-term (3-to-5-year) customer contracts, the physical landlords face a narrow universe of replacement tenants. The extremely high customization of liquid-cooling and high-density power configurations makes converting these facilities back to general-purpose enterprise data centers economically prohibitive. This creates a severe second-order exposure for the banks, private credit funds, and insurers who have financed these physical developments.
-
An instance of Debt-financed physical infrastructure cannot survive a pause in hyperscaler spending. — Utility companies and power providers are amassing billions in debt to fund the grid expansions needed for AI, exposing themselves to severe distress if tech capex slows down. ↩︎