The scale of individual projects taking shape amid this capex super-cycle is immense: a modern AI data centre campus can approach USD 50 billion in total replacement value, with new investment often clustered in a small number of strategic locations, connected through infrastructure and specialised suppliers. 

Shared electricity grids, telecommunications infrastructure, cloud services, and specialist suppliers can create correlated losses. At the same time, interconnected facilities may be financed, owned, operated and occupied by different parties, creating overlapping property, business interruption, cyber and liability programmes.

The reality for insurers is that risk is both accumulating and concentrating. Losses can become more correlated and severe, with business interruption as a driver. Single events affect larger pools of value, while specialised equipment, supply chain bottlenecks and extended replacement timelines can all boost the cost of recovery.

Hyperscale data centres illustrate the challenge: facilities costing billions of dollars can cluster in hazard-prone regions where they share critical infrastructure, networks and suppliers. A single disruption can therefore affect multiple insureds, sectors and lines of business simultaneously.

A Swiss Re analysis concludes that Texas and Virginia combined represent over 40% of current and planned US data centre capacity. While there are economies-of-scale advantages to locating facilities in specific geographies, this concentration also means many AI data centre projects are being planned or built in regions vulnerable to natural catastrophes such as severe convective storms and flooding.

Concern about this issue appears to be shared broadly: a Swiss Re analysis shows that the top 25% of Fortune 100 companies now cite asset concentration, clustering around enabling resources, and digital infrastructure dependencies as key risks in their public disclosures far more frequently than in 2018.

Critical component suppliers of the current capex super-cycle also pose concentration challenges. For instance, about 88% of Taiwan's semiconductor fabrication plants, a major source of advanced chips necessary for the expansion of AI, are situated in extreme or very extreme earthquake risk zones. Should disruptions materialise, the impact could be substantial: delays, downtime, and significant financial losses. 

In insurance, risks should ideally be distributed uniformly, with loss events that happen frequently, are limited in size and occur independently from one another. However, some risks emerging with the capex super-cycle don't share these attributes. 

This means insurers must adapt. Underwriting hard-to-insure risks requires innovative approaches. Larger, more complex risks may require specific technical underwriting capabilities, as well as more risk distribution via co-insurance, syndication and reinsurance.