Why reinsurers may be underestimating their data centre exposure

Inconsistent exposure data across different business lines prevents reinsurers from aggregating their true total line on a single site. 

Key points:
AI drives bigger data centre builds
Single sites span multiple classes
Higher values demand tighter lines

Reinsurers could unknowingly accumulate significant exposure to a single multibillion-dollar data centre through different cedants, treaties and business lines, creating concentrations that may not be visible in conventional exposure data.

Several insurers may each cover part of the property risk, while a reinsurer could ultimately carry exposures spanning energy, cyber, casualty and political risk tied to the same site, according to Luke Wedmore and Antonin de Benoist.

“The value of these new data centres is so high that any single insurer is having to put out their maximum line on any data centre,” Wedmore, senior research analyst at global insurer and reinsurer MS Amlin, told Monte Carlo Today. “If we reinsure multiple of these underlying insurers, we’re suddenly exposed to this aggregation risk as a reinsurer.”

The challenge is becoming more pressing as AI drives investment in larger, more complex data centres and pushes development into areas exposed to natural catastrophe and other risks.

Citing the March 2026 Iranian drone strikes on AWS facilities in the UAE and Bahrain, de Benoist, chief operating officer and co-founder of exposure management technology company Allphins, said the resulting physical damage, power disruption and cloud outages affecting banking and payment systems illustrated how a single event could cut across several insurance classes.

“In that instance, you have multiple touch points across property, political risk, energy, casualty and cyber,” he said. “The common denominator to it is the location of the asset, but location is actually not the core current attribute. Property, Political Violence give you an asset. Casualty, Political Risks and Cyber give you an insured name. 

Multiple tenants add another layer of complexity. The same facility may be insured by different insurers and ultimately reach one reinsurer through several cedants and treaties, de Benoist said.

“These new data centres that are required for AI are much larger than the previous data centres that were being built,” Wedmore said. “Because they’re larger, they need more space, more energy, more cooling facilities.”

Hidden aggregation

“We know where the data centres are, but we just don’t necessarily know where the data centres are in our exposure data.”

In May, MS Amlin said it had examined more than 670 US data centre projects under construction or planned. It found 51% of the projects, representing $670 billion of investment, were in states at high risk of tornadoes, large hail and damaging winds.

But while that research showed where data centres faced storm risk, it did not tell MS Amlin how much exposure it had accumulated to individual sites through its own reinsurance portfolio.

Its reinsurance underwriters therefore asked Wedmore’s team to identify each data centre in the portfolio and aggregate its exposure to each site across cedants and classes.

“We know where the data centres are, but we just don’t necessarily know where the data centres are in our exposure data,” Wedmore said.

Public lists identified the sites, but MS Amlin could not calculate its exposure across entire campuses and associated infrastructure.

“We wanted to make sure we captured all of that infrastructure in any aggregation analysis that we were doing,” Wedmore said.

That gap led the company to Allphins, which combined public data with MS Amlin’s proprietary list of newer sites missing from open-source databases. The portfolio could then be matched against those locations and exposures aggregated across reinsurance classes.

“We were able to run our portfolio against it and make decisions off the back of that,” Wedmore said. “This visibility allows us to deploy capacity responsibly to support the sector’s growth while maintaining underwriting discipline.”

He said the cross-class view had become important to underwriters, with the capability now being developed further.

Infrastructure drives risk

Identifying accumulation, however, is only the first step. Knowing the total exposure to a campus does not reveal what might cause a loss or how that risk should be priced.

Allphins has footprints for about 2,000 sites, but those boundaries do not yet capture all the operational characteristics that can determine risk.

“What is missing today is really the type of power source, the backup generation, the fuel type, the type of cooling system, the interdependencies with other assets nearby,” de Benoist said.

“The next phase, which is happening now, is really to build up this database,” he added. “We are aiming for a lot more attributes and information.”

For Wedmore, that information will become increasingly important as the values at stake rise.

“These are such high-value assets that they require diligent underwriting, careful management of your line size, and careful selection of risks,” he said.

No loss benchmark

The challenge for reinsurers is that they are having to understand these concentrations without the benefit of a major data centre loss to demonstrate how exposures might interact.

“It’s an industry that is evolving very fast, and where we haven’t seen any large market event to learn from, and where losses are a bit hidden,” de Benoist said.

“Facilities are built by very large companies which have their own captives, and where the retention is very high, so you might not see all the attrition on the claim side.”

Antonin de Benoist is chief operating officer and co-founder of Allphins. He can be reached at: antonin.debenoist@allphins.com

Luke Wedmore is senior research analyst at MS Amlin. His contact details are luke.wedmore@msamlin.com

For more news from Monte Carlo Today, click here.

Did you get value from this story? Sign up to our free daily newsletters and get stories like this sent straight to your inbox.