Verisk’s Newbold highlights data centre exposures and AI underwriting risks

The expansion of data centres to support artificial intelligence (AI) is creating large insurance exposures that underwriters struggle to assess, according to Rob Newbold, president of Verisk Extreme Event Solutions.

Verisk has developed a data centre exposure database to help clients understand the locations, construction and contents of these facilities, combining that information with its catastrophe models.

Newbold caught up with GR at Verisk’s conference hosted in London, reflecting on growing re/insurance sector concerns about the scale of data centre exposures.

Newbold said the US alone has roughly 2,500 data centres, with Virginia represending one particular concentration.

Finding sites with limited catastrophe exposure is becoming harder as suitable locations fill up, or local residents complain about data centres in their midst.

Clients are seeking better information about facilities that can represent individual risks.

“As an industry, we have to underwrite these specialty risks, and we don’t know enough about where they are, meaning we don’t know enough about what we’re underwriting,” he said.

Newbold suggested a single data centre could generate a $20bn loss, a staggering figure, underlining the importance of understanding what insurers are taking onto their books.

Location remains critical, with coastal areas, fault lines and the wildland–urban interface presenting different catastrophe exposures.

AI needs human judgement

Alongside the physical exposures created by AI infrastructure, Newbold sees risks in relying too heavily on the technology for underwriting decisions.

He warned against handing workflows to AI agents without human oversight, because repeated questions can produce different answers.

“The robot can’t decide,” Newbold said.

Verisk is using AI to process atmospheric datasets, support model development and reduce inefficient data handling within catastrophe modelling workflows.

Newbold also sees opportunities to identify repeated exposure to the same property across different reinsurance treaties, helping insurers understand portfolio concentrations.

AI is accelerating software development, but employees still need to check that its output is logical. Its role, he said, is to augment people’s capabilities.

Catastrophe risk beyond the cycle

Verisk’s average annual global modelled property catastrophe loss estimate stands at $171bn.

Newbold stressed that this represents a mathematical average across its models, rather than a forecast for any individual year.

“What it means is the industry needs to be capitalised to respond to 170 plus billion dollars in any given year,” he said.

Market pricing does not determine the Verisk’s scientific priorities, Newbold emphasised.

“We don’t change our view based on whether the rates are low or high,” he added.

Climate impacts may prompt faster updates from the catastrophe risk modeller, but changing rates do not alter its view.

More frequent updates must also be balanced against insurers’ need for stability, he suggested.