The AI-driven construction of hyperscale data centers is producing individual sites worth $30 billion to $45 billion or more in insurable value, but the conventional property insurance market’s effective limit for a single risk generally tops out at $5 billion to $10 billion. That gap, documented in a new paper by LGT Capital Partners titled Too Big to Insure? Hyperscale Data Centers and the Role of Insurance-Linked Strategies, is not just an underwriting challenge. It has started to constrain how these projects are financed.
LGT, a Liechtenstein-based alternative asset manager with an insurance-linked securities (ILS) division, cites market estimates that one-third to half of data center value is currently uninsured, sitting on developers’ own balance sheets. That framing aligns with S&P Global Ratings’ April 2026 assessment, which found that total insurable values for a single hyperscale campus have reached $20 billion to $30 billion, while full coverage at that scale remains largely unavailable. S&P noted the gap could function as a capital-structure constraint for lenders who fund data center projects.
Moody’s insurance analysts noted in March 2026 that investors including Blackstone were reported to have passed on data center debt opportunities because of insufficient insurance coverage. Insurance is typically mandatory before banks and private capital markets will fund construction.
The capacity problem has two distinct layers. The first is structural. A single hyperscale campus can require more coverage than any one insurer can underwrite, requiring collaborative structures where multiple carriers share the risk.
The second is technical. The contents of data centers, including servers, graphics processing units (GPUs), and chips, depreciate rapidly, become obsolete quickly, and are sensitive to smoke, water, and vibration. Supply chain constraints mean replacement lead times are long.
Taken together, those characteristics make business interruption and replacement cost appraisals difficult. Many insurers either exclude contents coverage or apply sublimits rather than attempting full-value cover.
LGT also highlights natural catastrophe accumulation as an underwriting concern. Data centers are clustered in a small number of geographic hubs, including Northern Virginia, Dallas-Fort Worth, Silicon Valley, Phoenix, Chicago, and the Pacific Northwest. Many of those hubs sit in counties with elevated catastrophe risk.
Earthquakes and severe convective storms account for the largest share of expected annual losses across current and projected portfolios. Texas and Virginia alone account for more than 40% of current and planned US data center capacity, according to Swiss Re Institute’s sigma 3/2026, Time to Build: Expanding the Frontier of Insurability for the Capex Super-Cycle. More than a quarter of that capacity sits in areas with three or more large-hail days per year, and around 40% falls within zones with significant tornado exposure, based on Swiss Re’s modelling of US Department of Energy capacity data.
Brokers and carriers have been building capacity structures specifically for this segment. Marsh launched Stratus in August 2026. The property insurance exchange offers up to $10 billion in single-placement capacity for the operational phase of digital infrastructure projects, drawing on 30 traditional and alternative capital providers. Stratus extends Marsh’s construction-phase Nimbus facility, which offers limits up to $2.7 billion.
Aon, meanwhile, expanded its Data Center Lifecycle Insurance Program to $5 billion in July 2026, its third capacity increase in just over a year.
LGT’s paper adds that ILS capital is positioned to complement these traditional structures, particularly for low-frequency, high-severity natural catastrophe tail risks. Where PML-based underwriting leaves a portion of the risk uninsured, ILS structures such as catastrophe bonds and collateralised reinsurance can absorb layers that conventional carriers will not write.
LGT estimates the cumulative global insurance premium opportunity for data centers at between $87.7 billion and $141.5 billion from 2026 to 2030, based on McKinsey and Dell’Oro capex forecasts. These are LGT’s own projections, not independent market data, and should be read accordingly.
The scale of the pipeline makes the capacity question urgent. Capital spending by the five largest hyperscalers is forecast to exceed $600 billion in 2026, a 36% annual increase, with roughly three-quarters of that tied to physical AI infrastructure.