The artificial intelligence boom is creating a problem that has little to do with chips, GPUs or algorithms. As AI data centres become larger and more power intensive, they are also becoming increasingly difficult for insurers to cover.
Howden’s latest report, Insuring the Data Centre Supercycle, describes data centres as one of the fastest growing commercial risk classes and argues that the rapid expansion of AI infrastructure is creating a fundamentally different insurance environment.
The scale is striking. Global data centre capacity is projected to increase by around 100 GW between 2025 and 2030, effectively doubling over five years. Howden estimates this could create around $1.2 trillion of real estate value.
But the biggest AI campuses are becoming difficult propositions even for the insurance market.
Howden says mega campuses can have $20 billion-$50 billion of total insurable value at a single site. The concentration of so much value in one location creates an accumulation problem for insurers because a single incident can generate enormous losses.
In other words, the AI infrastructure boom is creating assets whose financial value can be larger than the amount of risk that insurers are comfortable carrying at a single location.
The report says that despite the expansion of underwriting capacity, insurers can economically cover only a portion of the total insured property value of the largest AI-focused data centres.
This is where reinsurance and alternative forms of risk transfer become increasingly important.
The problem is not just physical assets
A data centre is fundamentally different from a conventional commercial building. Its value comes from continuous computing activity. A physical event can damage the building, electrical systems or cooling equipment, but an outage can also interrupt the operations of hundreds of customers.
Howden describes data centre risk as “heterogeneous, interconnected and centred on downtime.”
That creates multiple layers of exposure at the same time. A single event can potentially result in: property damage, business interruption, cyber losses, liability claims, customer claims, equipment failure, service-level penalties
The report therefore highlighted that data centre insurance cannot simply be treated as conventional property insurance.
Insurance is becoming part of the financing equation
There is another important implication. As projects become larger and require enormous upfront capital, insurance can influence whether financing can be deployed. Howden says structured credit insurance and related solutions are increasingly supporting project finance by allowing banks, institutional investors and infrastructure funds to deploy more capital while reducing retained credit risk. That means insurance is moving from being a back-end protection mechanism to becoming part of the infrastructure financing ecosystem.
The opportunity could be substantial. Howden estimates that the six major data-centre risk segments could represent around $5 billion of annual US operational premiums by 2030, equivalent to around 60% of annual operational premiums in the US.
The AI data-centre boom, therefore, is creating another industry alongside the visible ecosystem of chips, servers, power, cooling and construction. It is creating an increasingly sophisticated risk-transfer industry.
