Concentration, Correlation, and Why Capacity Runs Out

The single most important concept in insuring large physical infrastructure is one that rarely gets said plainly: insurance works because losses are independent, and it breaks when they are not. Understand that sentence and the rest of the market becomes legible.

The original logic of insurance is the pooling of unrelated risks. A thousand houses in a thousand towns will not all burn down in the same year; the unlucky few are paid for by the premiums of the many, and the arithmetic is stable because the fires have nothing to do with one another. This is the law of large numbers doing quiet, load-bearing work. The whole apparatus rests on one assumption — that your loss and my loss are uncorrelated. When that assumption holds, insurance is close to magic. When it fails, insurance becomes something the industry fears: a correlated catastrophe, where thousands of policies pay out at once and the pool is drained faster than premiums can refill it.

Data centers strain the assumption in two directions at once. The first is physical concentration. Put half a billion dollars of equipment under one roof and a single fire, flood, or electrical fault becomes a very large claim — a severity problem. But underwriters can price severity if they can bound it. The harder problem is the second one: correlation. The industry has been clustering these facilities in a handful of favored regions — cheap power, fiber, permitting, tax treatment — and clustering is the enemy of independence. If ten campuses sit in the same river valley, a single hundred-year flood is not ten independent claims; it is one event that hits all ten. If a regional grid failure or a wildfire season sweeps a metro area, the losses arrive together. The pool that was supposed to spread risk instead concentrates it.

This is why capacity — the total amount of cover the market is collectively willing to provide for a given risk or region — is the variable that actually governs whether a project gets built. Any single insurer limits how much it will expose to one location or one peril, because no prudent carrier wants a single event to threaten its solvency. Above that line, risk is passed upward to reinsurers — the insurers of insurers, names like Munich Re and Swiss Re — and syndicated across the specialist market at Lloyd's of London, where large and unusual risks have been shared among competing underwriters for centuries. A big data center is rarely insured by one carrier writing one policy. It is a tower of layered participants, each taking a slice, assembled by brokers who thread the placement together. When people say a risk 'cannot get covered,' they almost never mean no policy exists. They mean the tower cannot be built to the height the lenders require at a price the project can bear.

Here is where the discipline connects to the central thesis of this domain. Power and water are siting constraints because a facility physically cannot run without them. Insurability is becoming the next constraint in that sequence, and it behaves differently. Power and water are engineering problems with, in principle, engineering answers. Insurability is a market judgment about correlated risk, and it can withdraw from a whole region at once — not because any single site got worse, but because the aggregate got too concentrated for the market's appetite. A perfectly engineered facility can find its terms deteriorating because of what its neighbors built. That is a genuinely new kind of siting constraint, and it does not respond to a better sprinkler system.

The practical takeaways are structural and durable. First, aggregation is the underwriter's true fear — not any single building, but the sum of correlated exposures in one place. Second, capacity is finite and regional; it is allocated, and it can be rationed. Third, the reinsurance layer sitting above your primary carrier ultimately determines how much cover the whole system can offer, which means a data center's insurability depends on global reinsurance conditions it cannot see and cannot influence. And fourth, diversification of siting — the very thing that cheap-power economics discourages — is what keeps a portfolio insurable. The market rewards spreading out and punishes clustering, which puts insurance in direct tension with the location logic driving the buildout. That tension is not a footnote. It is the reason this subject earns a place beside power and water in every serious siting decision.