Concentration by Construction: The Accumulation Risk Hiding in the AI Data Center Boom

he build-out of data centers, motivated by the scaling needs of AI, has become a rapidly growing exposure in property and casualty insurance. S&P Global Ratings called it a “meaningful growth opportunity” in an April report,¹ projecting that rising demand for data center insurance coverage could generate $10 billion in new premiums in 2026, and the Swiss Re Institute’s March Sigma report estimated that global premiums linked to data centers could more than double to $24 billion by 2030.² The opportunity is clear, but the hard question is whether insurers can see how much of that exposure may move together.

That question starts inside the facility. A modern data center relies on shared power, cooling, and fire-suppression systems. Swiss Re, citing FM Global data, noted that fire accounted for only 10.9% of data center loss events but 42.3% of loss costs over a 15-year period. Liquid cooling, adopted to manage the heat from dense AI chips, adds another path for loss if water escapes near sensitive equipment, and a power or cooling failure can leave the building intact but idle the economic value inside it.

The insurance placement can obscure that accumulation. As the Sigma report notes, a single campus may reach carriers through separate programs for buildings, equipment, and on-site power plants. Each piece may look manageable when reviewed separately, but one event can trigger claims across several insurance programs at once. Phased projects add complexity when part of a campus has been handed over for operation while construction continues elsewhere.

Data centers also cluster geographically. Large facilities need reliable power and fiber connectivity, sufficient land, and a viable cooling strategy, narrowing the pool of suitable locations. Swiss Re estimates that more than one quarter of U.S. data center capacity may be in areas with at least three large hail days annually, while about 40% may sit in significant-to-very-high tornado day zones. A single tornado path can damage multiple structures on one campus — and in some cases multiple campuses — through wind, debris, or wind-driven water intrusion, creating losses that exceed what a single-location maximum probable loss estimate may capture.

A less visible form of concentration may sit on the asset side of the balance sheet. In its March Quarterly Review, the Bank for International Settlements observed that off-balance-sheet financing structures channeling private credit into AI-related data center infrastructure “strengthen links between hyperscalers⁴ and non-bank investors such as private credit vehicles and insurers.”³ For an insurer that both writes the property risk and invests in credit tied to the same build-out, risk could become correlated across underwriting and invested assets.

The data center boom challenges the diversification assumptions embedded in pricing, catastrophe aggregation, and capital models. Loss estimation that treats buildings on campus or at nearby sites as independent may understate the tail risk. Tracking accumulation across lines and across separately submitted programs may be necessary. With limited loss history for next-generation AI facilities, actuaries will need engineering judgment, dependency mapping, and scenario testing to identify exposures that look separate in normal conditions but converge in the tail.