Skip to content

Invest1 publisher3 min readPublished Updated

One hyperscale campus can carry insurable value worth 45% of every cat bond outstanding

The case for putting AI data-center risk into catastrophe bonds rests on one investor's ratio between a single campus and the whole market. Spreads in that market have been softening while he waits for a first deal.

The Investor · Invest desk

Illustration accompanying One hyperscale campus can carry insurable value worth 45% of every cat bond outstanding

What happened

  • Powell puts the insurable value of a single hyperscale campus at $20bn to $30bn, against roughly $66bn outstanding across every catastrophe bond in force.
  • Catastrophe bond issuance has reached $18.9bn so far in 2026 with first-time buyers coming in, and Artemis.bm's Steve Evans said reinsurance and cat bond spreads have softened.
  • Hanni Ali of Radix ILS said lenders to data centers may also use the catastrophe bond market to offload sabotage, war and cyberattack risk.

Compiled by The InvestorSomething wrong?How this is made

Why it matters

  • contradiction Powell's capacity argument and Evans's softening spreads point in opposite directions: a market whose pricing is falling back toward historical averages has capital available to deploy, so the binding constraint is a model for the peril.
  • constraint The exposures that most often take a data center offline are the ones Powell calls hardest to price, so a first deal covers wind and quake and leaves fire, water damage and business interruption sitting with the traditional market.
  • exposure A data center tranche would put insurance-linked securities investors, who can lose all of their principal once a trigger is met, behind a single site holding $20bn or more of value.
  • precedent Whichever deal prices first sets the template for how AI infrastructure enters this market, as a named-peril property trade sponsored by insurers or as cover bought by the banks financing the buildout.

The ratio depends on which end of the range you take. Twenty billion dollars of campus value against roughly $66bn outstanding is 30 per cent; $30bn against the same base is 45 per cent [4][5][6]. Powell's "roughly a third" sits at the bottom of his own range [7]. Both versions compare two different quantities: the full insurable value of one site and the notional limit outstanding across every catastrophe bond in force [4][5]. Cat bonds take a layer, and the entry point Powell describes is a property catastrophe tranche covering hurricane and earthquake, perils the insurance-linked securities market already models [8][21]. A first deal would cede a fraction of $20bn.

The comparison still matters. "One campus can carry insured value equal to roughly a third of every catastrophe bond in existence. You cannot solve that with the traditional market alone. The arithmetic doesn't work, and that's why this ultimately ends up in the capital markets," Powell told CNBC by email [7]. The report does not include a figure for how much property cover hyperscale operators buy today, so the capacity claim rests on that one ratio.

The market he is pointing at is getting cheaper. Issuance has reached $18.9bn so far in 2026 with first-time buyers coming in [11], which means one campus at the low end of Powell's range is worth more insurable value than the entire market has placed this year [19]. Steve Evans of Artemis.bm said "the well-capitalised nature of the global reinsurance and insurance-linked securities market means that pricing of reinsurance and cat bond spreads have softened, making buyer conditions even more favourable" [12], and he saw "no signs of investor interest waning" even as spreads move back toward historical averages [13].

So the scarce input is a model for the peril. "The challenge is that some of the biggest data center exposures, including fire, water damage, power outages and business interruption, are harder for the cat bond market to price today," Powell said [9]. Geography is moving too, with more building in Texas and Arizona pushing exposure away from coastal hurricane and toward tornado and hail [17]. An investor in the resulting tranche can lose some or all of the principal once a covered trigger is met [15].

"As those risks become better modeled and structures become more standardized, I would expect the first dedicated data center cat bond deal within the next 12 to 18 months," Powell said [10]. I would expect that first deal to be small and single-peril, a wind-and-hail tranche that no operator's risk manager would call a solution. Two other paths are live. Hanni Ali of Radix ILS points to lenders using the market for sabotage, war and cyberattack cover [14], and a bank-sponsored structure could reach investors before a property one does. Or reinsurers keep writing the risk privately through quota shares, sidecars and new facilities while spreads soften, and nothing reaches the bond market at all [3]. The test of that last one is a data center layer placed inside a sponsor's existing property catastrophe programme at the next renewal.

What to watch

  • The terms of the first dedicated data center cat bond to price: peril covered, attachment point and size.
  • Whether cat bond spreads keep softening toward historical averages, since a hardening market changes who wants to cede risk and at what price.
  • Whether a lender-sponsored structure covering sabotage, war or cyberattack reaches investors before a property catastrophe tranche does.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories