The AI race is on and it feels like it’s accelerating. We recently did a piece about the five ways we think AI will impact cat bonds, but one topic bears a deeper dive: the reality that data centers make AI output possible.
Right now, reinsurers are trying to find ways to price and cover data center risk. It’s a big undertaking, considering data centers earmarked to handle AI processing loads are expected to require over $5 trillion (according to McKinsey). To further break it down, there are two correlating events happening that impact the $5T figure: first, the amount of data centers being built is increasing. Second, the amount of computing power these facilities need to accommodate is also increasing. In fact, the amount of megawatts per data center has nearly doubled from 2019 to 2025. Large infrastructure projects - like bridges and tunnels - require $5-10 billion in coverage limits, and AI hyperscale campuses are expected to incur nearly three times that for construction alone. This means that there is a lot of money on the line should something devastate one of these facilities.
But it’s more complicated than just how much protection data centers need, it’s also about how the market goes about building that financial emergency lever. Data centers are credit-driven versus insurance-driven, so it’s the lender financing the build that is ultimately demanding protection. This is a structure that’s different than what we traditionally see in cat bonds, where the insurer is managing retention.
The number of perils that can affect a data center is far greater and more difficult to predict than singular incidents that most cat bonds track to. Now, a cat bond will specify an event, such as a hurricane, covering where it could hit and the level of damage it causes, or parametric triggers it would have to reach to activate the cat bond. When we’re thinking about a physical structure, we have to qualify everything that could harm it, and to what degree. Yes, a storm could destroy the building, but what about power outages, an equipment failure, or a fire that breaks out on-site? These are all unnatural acts that could just as easily happen to a data center, and there’s no precedent for how we predict and model any of these events.
Expanding on the above point, there are other losses to consider - such as revenue. If a data center goes down, the downstream capital impact could be significant, especially considering that they are producing nearly three times more revenue than eight years ago. In addition, data centers are susceptible to other factors, such as energy pricing volatility, and can be at the mercy of government policy or private company timelines. Imagine if the underlying power grid that a data center relies on needs an upgrade?
Another consideration is where these data centers are located. Cat bonds have traditionally focused on areas most likely to experience a large-scale peril, such as Florida and California. However, data center construction is mostly concentrated in the central US. In fact, in the Midwest alone, there are 419 planned data centers in addition to the 655 existing ones, which represents a 64% increase.
So, where does this leave cat bonds? As of today, no data center cat bond exists, but it’s likely that they’re coming. As insurance premiums mount, traditional insurance markets will need to look for other solutions, like capital markets, to alleviate the pressure. We’re already seeing hints of this becoming a reality. In January of this year, Aon’s reinsurance team created its first-ever data center treaty aligned to $5 billion of capital through the insurance value chain through a single insurer. It’s a space to watch, and we’re keeping a sharp eye on how the market will develop.
Sources:
ILS Brookmont
McKinsey
S&P Global
Pew Research
Reinsurance