"At this scale, insurance either isn't available in the market, or it's just prohibitively expensive" [At this scale]
"If a data centre has a power outage for 45 minutes, it will wipe out half a year of revenue" [45 minutes]
The article documents that mega AI data centres (1+ gigawatt, $30B+ cost) cannot obtain full replacement insurance coverage, forcing lenders like KKR and Blackstone to walk away from deals. This signals a fundamental structural problem: the velocity and scale of AI infrastructure investment has outrun the insurance industry's ability to price and underwrite tail risks. This creates a bifurcated market where only hyperscalers with balance-sheet capacity can fund projects, while newer lenders face unacceptable uninsured exposure.
The article documents that power and water supply interruptions—not physical damage—pose the largest financial threat to AI data centres, yet these risks were uninsurable three years ago and remain difficult to price. Parametrix's recent entry into this market shows insurers are still learning to model these risks. The fact that a 45-minute outage can eliminate half a year's revenue reveals that AI data centre economics are extremely fragile to operational disruption, and that insurance markets have not yet developed adequate tools to transfer this risk, leaving lenders exposed to tail events they cannot hedge.