US data centres storm risk is now a systemic insurance gap, not a niche underwriting concern. 51% of planned US data centre projects — worth 670 billion USD — are located in states at high risk of severe convective storms including tornadoes, large hail, and damaging winds, according to a major exposure analysis by MS Amlin. As hyperscaler capex floods interior US states that standard commercial property products were never designed to serve at this scale, a convergence of concentrated values, inadequate models, and genuine capacity ceilings is emerging as one of the most consequential underinsurance problems in commercial P&C.
High-Storm-Risk States: The Data Centre Concentration Architecture
MS Amlin’s analysis mapped 320 of the 670 planned US data centre facilities into high-SCS-risk states — a hazard category that encompasses tornadoes, large hail, and derecho-class damaging winds. The broader catastrophe footprint is even larger: 56% of all 670 planned projects, representing 800 billion USD in total investment, sit in states with any major natural catastrophe exposure. The relative scale of the problem becomes clear when set against the current baseline: existing data centres in high-SCS-risk states are valued at approximately 20 billion USD — the planned AI infrastructure in the same zones could be 40 times that value.
The geographic driver is AI compute demand. Northern Virginia — the traditional data centre hub — is saturated; hyperscalers are building across Texas, Tennessee, Wisconsin, and Ohio for power, land, and water availability. most new data centre capacity under construction in 2026 is already outside Virginia, distributed across states with elevated SCS exposure. As MS Amlin’s Chief Underwriting Officer Martin Burke observed, “hundreds of billions of dollars of new digital infrastructure are being directed towards regions at higher risk of potentially destructive severe convective storms” — and when assets of this scale cluster in hazard-prone regions, the potential loss severity from a single storm event can rise very quickly.
How Hyperscale Values Broke the Standard Property Policy
The problem is not simply that data centres are in storm-prone states — it is that the insurance products deployed there were calibrated for a different era. Average insured data centre project values have risen from 150 million USD to 3 billion USD over the past five years, according to Zurich, which reports that severe weather has been the leading cause of loss in its builders-risk portfolio for three consecutive years. At the extreme end of the hyperscale spectrum, individual campuses now carry total insurable values of tens of billions for construction alone — a range at which no single carrier can provide full coverage.
Zurich’s Kelly Kinzer, who leads the insurer’s data centre practice, has been direct: “there simply isn’t sufficient insurance capacity in the market to insure these projects to their full value.” S&P Global Ratings estimates that data centre insurance demand could generate 10 billion USD in new premiums in 2026 alone — reflecting the gap between what exists and what the sector needs. Against the 2 trillion USD in total insurable value represented by approximately 11,000 data centres currently operating globally, even that premium estimate signals a profound under-penetration of the sector relative to its risk exposure.
The accumulation risk amplifies the capacity problem. MS Amlin has built a proprietary aggregation monitoring database for data centre exposures precisely because insurers can unknowingly accumulate exposure to the same facility across multiple policies — property, construction, business interruption, and cyber covers can all point to the same physical asset without a system to flag the concentration. Specialist capacity structures for concentrated infrastructure buildouts — as developed for nuclear new-build — are now the logical model for the data centre sector.
SCS as a Primary Peril: Three Consecutive Years at Record Levels
The severity of the exposure trajectory becomes clear in the context of recent loss history. According to Swiss Re sigma, 2025 global severe convective storm insured losses reached 51 billion USD — the third consecutive year above that threshold — while total global natural catastrophe insured losses were 107 billion USD, of which secondary perils including SCS, wildfire, and flood accounted for a record 92%. SCS losses have grown at approximately 7% annually in real terms since 2008, driven by a combination of increased storm frequency, exposure growth, and the very geographic expansion of insurable assets that the data centre buildout now exemplifies.
For reinsurers, the concentration risk in interior US states adds a new aggregation scenario to already elevated SCS budgets: a derecho or hail swarm that crosses West Texas or central Ohio can now simultaneously hit multiple co-located hyperscale campuses representing tens of billions in values within a twenty-mile corridor. Climate analytics tools for commercial property pricing — such as those developed for extreme heat — face an analogous modelling gap for data centre SCS accumulation.
