How difficult is it becoming to secure insurance for data centers?

Is it the case that the scale of AI data centers projects make them now more difficult to insure?

A press report recently suggested that Meta only has a fraction of its data center exposure fully insured? 

Across this past year, insurance companies have become more vocal about the vulnerability of AI data centers as investment capital sums have soared, and as McKinsey has projected, around US$7 Trillion needs to be invested by 2030.

AI Project values

Average project values have surged from around US$150 million to US $3 billion, and hyperscalers alone are likely to end this year with around US$710 billion in capital expenditures. 

With very large campus projects, capex runs into tens of billions of dollars.  

The industry are suggesting that there is insufficient insurance capacity in the market to insure these projects to their full value. No single insurer can absorb the risk of a US$30 billion data center project, so large projects need many insurers layered together, and even then full value often can't be covered. 

Elevated Risk Profiles

The integration of lithium-ion battery backup units into server racks has created an ignition source that didn't previously exist in data processing rooms, and liquid cooling systems adopted for high-heat GPUs present another growing exposure. An example occurred in 2025 where a tornado at a data center site caused extensive loss in one major insurance firm.

The US in particular has seen a shift to interior locations with power advantages but sited in areas with significant natural hazard exposure, including hail- and tornado-prone regions. 

Outages and the rise of parametric insurance products

An increasing constraint is not physical damage to buildings but protection against revenue losses from outages and the risk of disproportionate losses that lenders rely on to ensure debt repayment. 

Calculating the financial impact of an outage on an owner-operated facility (as opposed to a leased one with clear tenant contracts) can be considerably more difficult and claims can take one to two years to adjust, which is part of why parametric (trigger-based, fast-payout) products are gaining interest. 

Contracts are paramount

Policy wording, valuation, and how the indemnity period is defined matter enormously, and gaps are often not discovered until a claim arrives, by which point the wording is fixed.

Cyber Risk is pushing towards third-party capital

Insurance coverage remains limited for cyber risk specifically, because of the difficulty in modelling policies. Risk is being pushed towards third-party capital outside the traditional insurance industry. 

Premiums are escalating 

Global premiums tied to data centers are expected to more than double to US$24.2 billion by 2030 from US$10.6 billion today. The escalation reflects strained capacity as much as insurer appetite. Smaller/mid-size data centers can generally still find reasonably conventional coverage; the real difficulty is concentrated at the hyperscale end, where developers increasingly have to prove a project is "insurable and bankable" before it can even get financed, and where multiple insurers, reinsurance layers, and sometimes parametric or alternative-capital structures have to be stitched together to get anywhere close to full coverage. 

+  +  + 

Insurance is becoming top of mind for data center developers and Platform Markets Group will be reflecting these concerns in its forthcoming event series. 

The next opportunity to meet the investors behind the infrastructure will be in Antibes at the annual Platform Global conference – 8-10 September. Book today as tickets are almost sold out: www.platform-markets.com/global