Aon's expanded data center program covers a tenth of one $50 billion campus

A campus at the top of that range is worth about five times what the whole world's data center insurance pool collects in a year: $50 billion of replacement value against $10 billion to $11 billion of annual premium ^[1][3][1]. Construction alone can exceed $20 billion before the first GPU is installed ^[2]. Aon's program capacity went from $3.5 billion to $5 billion inside 2026, a rise of 43% ^[7][3].

Of the roughly $200 billion Swiss Re describes, $111 billion comes from the renewable energy projects built to power the sites and $91 billion from the data centers, so generation accounts for 55% of the headline figure ^[5][6][4]. The $91 billion is cumulative. Spread across 2026 through 2030 it averages $18.2 billion a year ^[5]. A straight line from today's $10.5 billion midpoint to the $25 billion midpoint projected for 2030 sums to about $88.75 billion over those same five years, within about 3% of Swiss Re's number ^[3][4][7][8]. On that reading the estimate is the entire pool.

Where the exposure sits is the harder part. About 40% of US data center capacity is in significant tornado zones and more than 25% in high-hail areas ^[8][9]. Verisk launched a US Data Center Exposure Database on September 4, 2026, covering more than 2,500 facilities, to improve catastrophe modelling of those assets ^[10].

Cat bonds price best on well-modelled perils such as named hurricane and earthquake ^[14]. The losses that define data center risk are fire, water damage from cooling system failures, and business interruption that can dwarf the physical damage, and modelling those for bond structures is still in progress ^[15]. Cryptobriefing does not give a rate ^[9].

Without a rate, the repricing case has to be tested on the peril schedule of the first deals, which Cryptobriefing says are anticipated within 12 to 18 months ^[12]. In my view the model for the non-catastrophe perils binds harder here than investor appetite: a bond triggered only by wind and quake leaves fire, cooling failure and downtime with insurers and reinsurers whose capacity is already not scaling with the buildout ^[13][15]. Two other outcomes are live. Investors may accept a modelled-loss trigger for fire and equipment failure at a wide enough spread, in which case much of the $20 billion to $30 billion pool of 2030 sits with capital markets ^[4]. Or the campuses do not burn, losses stay attritional, and nobody ever tests the $50 billion replacement figure ^[1].