Why the Dominant Insurers Are Engineers

The structure of the data-center insurance market is unusual, and the reason is buried in a statistical problem. Ordinary property insurance is priced actuarially: an insurer holding thousands of broadly similar buildings, with decades of claims across them, can estimate the expected loss of any one from the behaviour of the population. The law of large numbers does the work. Large data centers defeat this. There are comparatively few of them, they are recent, and each is individually distinct in design, protection, and location. The population that actuarial pricing needs does not exist. So the insurers that dominate this space price differently — from the physics of the specific building rather than the statistics of a category. This is engineering-led underwriting, and it reshapes the entire commercial relationship.

FM Global is the archetype and the clearest illustration of the model. It is built around a research campus where fire and natural-hazard behaviour are tested physically, and its commercial doctrine is that the great majority of large losses are preventable — that a properly engineered and maintained facility should rarely suffer a serious fire. The proposition it sells is not principally the transfer of risk; it is the prevention of loss, with indemnity as the residual. Its engineers inspect the insured facility, measure it against detailed loss-prevention standards, and issue recommendations that the insured is expected to act on. The underwriting decision, the premium, and the risk-improvement plan are one continuous conversation rather than separate transactions. An operator that implements the recommendations is rewarded; one that lets recommendations lapse finds the terms hardening or the cover withdrawn. The insurer behaves less like a counterparty to a bet and more like an engineering partner whose interests are aligned with the facility never failing.

This model has consequences the practitioner should internalise. First, the inspection is not a formality but the core of the product, and the recommendations that follow are commercially binding in practice even when not strictly contractual — ignoring them is the surest way to erode both price and coverage. Second, the standards are prescriptive and physical: the spacing and rating of fire barriers, the specification of detection and suppression, the treatment of battery rooms, the discipline around hot work and maintenance. Meeting them is what makes a facility a 'highly protected risk', the category these insurers are built to write, and the gap between a highly protected risk and an ordinary one is visible directly in the premium and the capacity offered. Third, because the model depends on the insurer's own engineering judgement of the specific site, it does not transplant cleanly onto risks the insurer cannot inspect — which is exactly why contingent exposures at third-party sites, treated earlier, sit awkwardly in this framework.

The engineering-led model also explains a broader market dynamic now bearing on the whole sector. Because these insurers price from physical risk rather than historical averages, they respond quickly and visibly to changes in physical risk. Rising natural-catastrophe losses, the novel fire behaviour of large lithium-ion installations, the extreme value concentration of the newest facilities — these feed directly into engineering assessments and therefore into pricing, capacity, and appetite. When the engineers conclude that a class of hazard is worsening, the market moves without waiting for the claims to accumulate, because the model was never waiting for claims in the first place. That responsiveness is why insurance has become a leading rather than lagging signal in data-center development, and it sets up the course's final argument: that the insurer's engineering verdict on a location, delivered before anything is built, increasingly determines whether it should be built there at all. The practitioner's job is to engage that verdict early — to bring the engineers in at design, treat their standards as siting and construction inputs, and build a facility that the model wants to insure rather than one it has to be argued into.