Zurich Insurance Group disclosed on August 6, 2026 that it has bought up to $1 billion of quota-share reinsurance dedicated solely to its data-center construction portfolio, alongside a first-time breakout of a distinct US peak-peril catastrophe tower carrying a $150 million Turicum Re cat bond above a $650 million retention.

Both moves say the same thing: this book has outgrown a general property program.

Key Takeaways

  • The quota share covers one asset class. Zurich has underwritten at least 500 data-center projects and leads roughly 70% of them, a footprint built as construction premiums grew 18% globally on what the company attributes to "growing AI demand."
  • A proportional treaty was the only structure available. A catastrophe excess-of-loss attachment needs a credible loss distribution, and hyperscale AI campuses are a 2024-onward phenomenon.
  • A hyperscale campus can cost more than $20 billion to build before computing equipment, and roughly double that once GPUs, cooling and networking go in.
  • Delay-in-startup exposure can exceed $1 billion on a single campus, against sublimits measured in tens of millions on conventional construction projects.
  • The peak-peril tower isolates named storm and earthquake, the two US perils the ILS market prices most efficiently, funded partly by Zurich's first cat bond since 2012.

What Zurich Disclosed

The group reported net income of $3.5 billion, up 14% year over year, on a property and casualty combined ratio of 92.7% and gross written premiums of $29.9 billion, up 7%. Construction premiums grew 18% globally, which Zurich tied directly to data-center buildout across the United States and internationally.

Two reinsurance structures arrived alongside those numbers. One is up to $1 billion of quota-share cover dedicated to data-center construction. The other is a US catastrophe program disclosed for the first time as two towers rather than one all-perils placement: a general US property tower, and a peak-peril tower isolating named storm and earthquake above a $650 million retention (Reinsurance News, August 2026).

Layer / InstrumentAmountDetail
Retention$650 millionUnchanged from January 1, 2026 renewal
Turicum Re 2026-1 cat bond$150 millionAttaches at $650M, exhausts at $850M; 9.22% initial attachment probability; 7.88% initial expected loss; three-year term to April 2029
North America earthquake swap$225 millionUp from $215 million at January 1, 2026
Europe all-perils retention$489 millionDown from $505 million at January 1, 2026
Europe regional cat treaties$449 millionDown from $463 million at January 1, 2026

Why a Quota Share and Not an Excess Layer

An excess-of-loss treaty needs an attachment point, and an attachment point needs a credible loss distribution. This class does not have one.

Hyperscale AI campuses are a 2024-onward phenomenon, and the portfolio's total insured value is still compounding as projects break ground. Setting an attachment against a book whose premium base grew 18% in a half year risks the point going stale inside the treaty year, either too low, ceding premium the primary should keep, or too high, leaving Zurich exposed on a portfolio still finding its loss experience.

A quota share shares a fixed percentage from inception, so the reinsurer's exposure scales with the book. It also relieves underwriting volatility and the capital strain of rapid growth in one instrument, because the reinsurer takes a pro-rata share of reserves as well as losses.

The exposure itself explains the caution. A hyperscale campus can cost more than $20 billion before any computing equipment is installed, roughly doubling once GPUs, cooling and networking hardware go in (Swiss Re Institute, July 2026). That is one insured location, one builders-risk program, and lenders demanding limits sized to full replacement cost even where the probable maximum loss from any single peril is far lower.

Delay-in-startup coverage widens the gap further. A large AI campus can generate delay-in-startup exposure exceeding $1 billion on its own, because the revenue and contractual penalties tied to a hyperscaler's compute-leasing commitments dwarf the physical structure (Risk & Insurance, 2026). Conventional builders-risk forms carry sublimits in the tens of millions. Add a chip supply chain concentrated in a handful of foundries, and a fire or delayed delivery at one campus can trigger contingent business-interruption claims across several insureds at once, through a channel with nothing to do with weather.

Leading 70% of 500 projects cuts both ways here. On a lead-placed risk, Zurich sets wording, negotiates claims and sets the case reserves that become the industry's eventual loss triangles, which is a real pricing advantage over any following market. It also puts the practical burden of adjusting builders-risk and delay-in-startup losses on one claims organization across a portfolio whose insured values compound faster than any other commercial property class. Ceding proportionally spreads the capital consequence of a bad year while Zurich keeps the claims control that generates the data.

Matching Instrument to Peril, and What the Match Leaves Out

Splitting the peak-peril tower out is a capital-allocation decision rather than a disclosure preference. Named storm and earthquake carry decades of modelling and a deep base of dedicated capital willing to take the tail, so they are where the capital markets price most efficiently against traditional reinsurance.

The funding reflects that. The $150 million Turicum Re Ltd. Series 2026-1 notes carry an indemnity trigger on a per-occurrence basis, were structured by GC Securities with AIR Worldwide as modelling agent, and priced at a final 15.75% risk interest spread after upsizing from a $125 million target and clearing roughly 7% below the midpoint of initial guidance (Artemis deal directory, 2026). "Insurance-linked securities are an established and strategic source of reinsurance capacity that can provide additional flexibility and cost efficiency," Head of Group Reinsurance Paolo Mantero said of the return after fourteen years away.

The limit of that logic is that the data-center accumulation does not run mainly through named storm and earthquake. Swiss Re's CatNet modelling against US Department of Energy data found more than 25% of US data-center capacity sits where three or more large-hail days occur annually, and roughly 40% in zones rated significant to very-high for tornado risk. Neither peril has an ILS market of comparable depth.

Severity concentrates somewhere else again. An FM Global fifteen-year study cited in the same Swiss Re report found fire accounts for about 11% of loss events at industrial and commercial facilities but more than 42% of loss costs, which argues for weighting fire-protection engineering credits against catastrophe-model output rather than reading the two as one number.

And the exposure has an operational side the towers do not reach at all, because insurers increasingly run underwriting, claims and pricing systems on the same hyperscale infrastructure they now insure. Uptime Institute found 57% of respondents' most recent major outage cost more than $100,000 and one in five exceeded $1 million, with roughly one in ten reporting serious or severe impact, even as per-site outage frequency continued a five-year decline. Fewer events on a larger, more interconnected base of value is the same shape the catastrophe perils have taken, arriving through a channel no reinsurance tower is structured against.

Further Reading

Sources