"You Can't Buy a Policy That Big" Is a Concrete Problem
The premium figure reads easily as an industry-growth story, but what the report actually identifies is a structural gap: the asset value of a single hyperscale campus now exceeds the limits the reinsurance market can supply at a sensible price. This is not an abstract worry. Broker figures reported earlier put it plainly: a large data-centre project can reach $20 billion in construction cost before equipment is installed, while policy limits for mega data centres across Marsh's book run roughly $1.5 billion to $3.5 billion, with only three known placements above $5 billion. That is also why Swiss Re names catastrophe bonds and sidecars — the upper layers traditional capacity cannot absorb have to be filled by capital-markets instruments.
Site Selection Has Stacked the Risk Into the Same Weather
The second point is concentration. Data centres need large parcels of land and cheap renewable power, and those two requirements push new campuses toward the same regions — so assets pile up geographically while sharing grids and backbone networks. The catastrophe scenarios Swiss Re lists in this report are earthquake, windstorm and flood. Swiss Re Institute had previously given a sharper US breakdown using its CatNet catastrophe assessment tool: about 40% of US data-centre capacity sits in areas rated significant-to-very-high for tornado days, and more than a quarter in areas with substantial hail exposure. The consequence brokers report is deductibles — catastrophe deductibles in tornado-exposed regions typically run 2% to 5% of total insured value, which on a $2 billion facility means $40 million to $100 million retained before coverage responds. On the record, Gianfranco Lot, Swiss Re's Chief Underwriting Officer for P&C Reinsurance, framed it as the digital economy becoming a real economy: AI needs data centers, power grids and increasingly complex infrastructure, and all of it needs insurance.
What It Means If You Buy AI Services
On its face this is insurance-industry news. For teams consuming cloud and APIs it lands as two things. The first is a real boundary on business continuity. Your model serving runs inside a handful of enormous campuses, and those campuses are more correlated on natural catastrophe than they used to be — the probability of one weather event hitting several sites at once can no longer be treated as zero. In an SLA conversation, the question worth more than an availability percentage is where the failover path goes during a regional outage, to which region, and how long it takes. The second is long-run cost. Deductibles and coverage ceilings eventually land in operating costs, and they shape where new campuses go and when they can be built. None of that shows up in today's inference pricing, but it will show up in capacity planning past 2027.
via: Swiss Re press release, Artemis, Business Insurance on data-centre catastrophe exposure