The level of investment into data centers globally may soon make it necessary for insurers involved to begin selling securitization products to spread the risk among a broader pool of investors, according to Zurich Insurance Group AG.
“I do anticipate with the number of these projects being built at those values, that those types of discussions are going to become more and more necessary,” Kelly Kinzer, Zurich’s global head of construction and surety told Bloomberg, even though the market is “not there today.”
Lenders across private and public markets have been lining up to invest into the build out of global infrastructure for cloud and artificial intelligence services as they bet on demand from consumers and industry. Tech behemoths including Oracle, Meta and Alphabet have helped propel global bond issuance to more than $6.57 trillion in 2025.
Five years ago the average value of a data center project within Zurich’s portfolio was $150 million, whereas now it is $3 billion, Kinzer said.
The expansion of private credit into the financing of infrastructure projects such as data centers has changed the amounts at play as well as the delivery dynamics of such projects, the insurer said in its “Future of Construction” report published this week.
“Compared with traditional bank lending, they typically impose sharper performance thresholds, shorter decision horizons, and less capacity for operational deviation,” according to the report.
The increased involvement of private credit has also led to the elimination of loss limits, which would cap the total amount at risk for insurers.
“We have seen more of a push towards the expectation that full limits are purchased on these projects, which has put the entire industry in a very challenging position,” Kinzer said. “There simply is not enough insurance capacity in the marketplace today.”
Figures cited by XDI indicate that the global AI build-out will cost $7 trillion over the next four years, with the four largest hyperscalers having already spent more than $400 billion on infrastructure last year alone.
Those massive investments are ushering in a new era of risk. According to S&P Global, total insurable values for a single data center are expected to be as much as $30 billion per location, compared to $10 billion for some of the world’s biggest bridges.
Against that backdrop, investors are looking to capitalize on the insurance gap that’s likely to emerge. Insurance-linked securities are instruments where the return is tied to specific events materializing. The most common type is catastrophe bonds, where investors in those notes earn a premium unless a specific weather or other event actually occurs.
Insurable risks for data centers include damages from weather, comparable to other property and casualty insurances, but the required coverage is often higher than for other assets due to the soaring cost of GPU computing chips. The opportunity to offer risk coverage around data centers comes as insurers seek new streams of revenue to help offset a slowdown in premium growth across property and casualty insurance.
Cumulative global insurance premiums associated with data centers are estimated to reach $134 billion between 2026 and 2030, according to a report by Artemis, citing a broker at Aon Plc.
Alternative investors such as Euler ILS Partners are teaming up with insurance companies to underwrite specialist policies, Bloomberg reported earlier. Insurer Aon has said it was also approached by ILS investors asking how they can participate in this market.
Insurance-linked securities help to distribute the risk associated with betting on data centers at a time when the ultimate use case for artificial intelligence is not yet clear and expectations for realized productivity gains have yet to be proved. While investors into insurance-linked securities buy risks associated with damage to the asset, they don’t bear the risk that the data center proves unprofitable.
Photograph: Data processing in Zurich; photo credit: Adrian Bretscher/Getty Images
Related:
Copyright 2026 Bloomberg.
Topics
Data Driven
Interested in Data Driven?
Get automatic alerts for this topic.
