Data centres – how can we move beyond the hype to long term insurability?

On a recent trip to Texas I was lucky enough to explore a data centre under construction. The scale and speed which they are constructed is truly impressive. As the AI race drives significant construction investment spike into market, forecasts have suggested that global data centre spend could reach $7 trillion by 2030.  

The team at Swiss Re Institute recently looked into the numbers behind what this growing sector could mean for the re/insurance sector. There is a clear opportunity, and a potentially substantial market, as hyperscalers want to protect assets which are often valued in the tens of billions of dollars.  

However, while navigating this push to build it's vital that attention is also being paid to the complexities that come with building and operating these high-value facilities. Like many emerging risks before, it is one that can only be economically insured if the risks are properly understood and managed from the outset, especially interdependent exposures and accumulation risk.

One thing became immediately obvious on my trip was that building the facility is no longer the biggest challenge. Powering it is. 

To support AI, modern data centres use advanced GPUs that require 10-15 times more power and produce up to 10 times more heat than CPUs. Indeed, while traditional servers typically generated 5-10Kw per rack, today’s AI-optimised configurations now regularly exceed 100kW. 

This has naturally required a re-engineering of data centre infrastructure. Many facilities now have their own power sources such as gas turbines or renewables on site, with grid capacity often being insufficient to meet demands.  

Fire is also a major risk area, with many data centres reliant on a substantial number of lithium-ion batteries that are prone to terminal runaway issues.

Cooling systems

Traditional air-cooling has been replaced by water-based cooling systems, with the latter reliant on complex networks of high-pressure pipes. Liquid cooling changes several assumptions that have traditionally underpinned data centre underwriting. 

This presents the obvious risk of leakages, where several claims have already emerged. However, modern cooling systems also add risk to data centre constructions. Water chillers are often located externally and are therefore often left exposed to adverse weather and potential perils such as hail, windstorms and freeze events. Pipework also extends deeper into white space, water volumes increase, equipment values become more concentrated and structural loading changes significantly. These developments influence both loss frequency and potential loss severity. 

Power availability, land costs and tax incentives all naturally play a role in determining where and how data centres are built, yet long-term resilience planning often receives less attention than it deserves.  

An underappreciation of potential risks can lead to several potential points of failure and exposures that can have costly consequences – and these are already materialising. The insurance industry has already seen losses linked to hail, fire and water damages running into tens of millions of dollars. And it would be unrealistic to assume these events represent the upper limits.  

While there is yet to be a market-defining loss event, this should not be mistaken for an absence of risk.

In the case of the insurance industry, this preparedness is particularly important when considering the interconnected nature of data centre coverages and exposures. Policy interaction remains an underappreciated issue. 

Construction, operational property, specialist equipment like GPUs, battery systems, renewable generation and cyber cover may all be placed separately. The interaction between those policies during a complex loss is not always straightforward. 

Questions around causation, policy triggers, subrogation and allocation of business interruption losses deserve greater industry attention than they currently receive. 

There’s also a clear phasing issue between engineering and operational policies which may result in accumulation risks, while highly connected buildings such as these are naturally susceptible to cyber-attacks that could lead to material damages.  

Business interruption exposures become particularly interesting where physical damage occurs outside the insured premises. Damage to substations, transformers, transmission infrastructure or other insured critical suppliers may interrupt operations at a data centre even though the facility itself has suffered no physical damage. Understanding those dependencies, together with the scope of contingent business interruption cover, becomes increasingly important as AI infrastructure grows. 

Large power transformers illustrate how business interruption severity may evolve. A transformer loss following a covered physical event can result in replacement times measured in months or even years. In those circumstances, the interruption loss may significantly exceed the property damage itself. The engineering challenge therefore extends beyond preventing failures to understanding recovery capability and replacement strategies.

The point is that there are a multitude of real, highly complex and interconnected risks behind data centres. So, while the opportunity for insurers in this market is undoubtedly significant, with data centres representing a large, growing and strategically important market, it is vital that participants focus on creating an economically sustainable ecosystem insurable over time.  

Discussions across the data centre value chain should focus on ensuring that data centre insurance and reinsurance become economically sustainable over the long term. And that starts with risk understanding. 

Insurers, reinsurers and brokers can consider building a better understanding of policy interaction, operational dependencies, risks dependencies, cyber-physical events and business interruption. But our role extends beyond that.

At Swiss Re, we aim to support the evolving data centre market. Our goal is to bring engineering expertise, underwriting discipline, risk management insights and claims experience to the table to help insurers and their clients navigate this complex risk landscape. However,

We all have a role to play in examining how facilities are designed, constructed and operated. Site visits, engineering engagements, discussions about risk and advocating for resilience should all be considered as fundamental to long-term data centre insurability. If we’re able to influence constructions, through engineering engagement and risk dialogue, to ensure they are more focused on resilience, then we’ll help to ensure that avoidable vulnerabilities are addressed, and operational risks mitigated.  

The most successful partnerships in this space will be built on transparency, where we can work together with our clients to assess risks, influence resilience measures, develop robust policy wordings and create sustainable solutions that work for all parties. 

The long-term viability of data centre insurability and re-insurability relies on this approach. Ultimately, long-term partnerships matter more than short-term growth or headline premiums.