What Lenders Need To Underwrite in AI Data Centers

The AI boom is creating a major financing opportunity for banks—but data centers come with risks that extend well beyond traditional commercial real estate underwriting. 

The growth case is substantial. Revenue in the data processing and hosting industry rose 13.9% year over year in the fourth quarter of 2025, reaching nearly $114 billion. Industry sales are forecast to grow 9% annually through 2030, while AI-focused facilities alone are projected to account for $5.2 trillion of the $6.7 trillion in global data center capital spending expected by then. 

That expansion can generate demand for construction and commercial real estate loans, equipment financing, working capital, and treasury management services. Recurring contracts, multiyear service agreements, and relatively high barriers to entry can support predictable cash flow. The industry also generated median EBITDA margins of 16.2% and after-tax net profit margins of 6.3% in 2024–25. 

In an analysis for ProSight, Amy Short, director of research at Vertical IQ, identifies five areas that lenders should examine closely when evaluating AI-related data center projects: 

Start with power and water. According to U.S. Department of Energy figures, data centers consumed about 4.7% of U.S. electricity in 2024, and that share could reach 11.8% by 2030. Grid constraints, water availability, utility pricing, and permitting delays can all affect whether a project remains viable. Underwriters should confirm that power and water access is secured, utility commitments are credible, and energy-management plans are realistic. 

Account for regulatory and community pressure. Short noted that lawmakers in 14 states were pursuing measures that could limit or delay data center construction. That concern has since produced a major state-level action: New York imposed the nation’s first statewide moratorium on new hyperscale data centers, pausing state environmental permits for up to one year while regulators develop new standards. For lenders, the development reinforces the need to consider permitting requirements, environmental concerns, grid reliability, water use, and local opposition when assessing project timelines and costs. 

Build reinvestment into the credit view. AI infrastructure can become outdated much faster than conventional real estate. Servers, networking equipment, software, and computing architecture may require frequent upgrades. Lenders should test whether capital expenditure forecasts account for replacement cycles and whether management has a credible strategy for keeping the facility competitive. 

Examine where the revenue comes from. Large technology companies and hyperscale cloud providers can anchor a project, but they can also create significant concentration risk. Contract duration, renewal provisions, customer credit quality, and exposure to a small number of clients deserve close scrutiny. Long-term visibility is strongest when revenue is diversified across customers and end markets. 

Evaluate the operator, not only the asset. Successful data centers require technical expertise, cybersecurity capabilities, reliable systems, and access to specialized talent. Management’s experience operating complex infrastructure—and its ability to recruit and retain engineers, systems analysts, developers, and cybersecurity professionals—can be as important as the physical collateral. 

The takeaway: As Short explains, AI data centers may offer lenders access to a fast-growing sector with recurring revenue and strong demand. But sound underwriting must account for resource access, technology replacement, customer concentration, operating expertise, cybersecurity, and regulatory exposure. In this market, collateral value and debt service coverage tell only part of the story.