One data centre campus can sit in a reinsurer’s book many times over, through different cedants and under different names. The clash question is which of those lines respond to the same event, and how many treaties that one event reaches. I am building the analytics that read the campus once.
Giuseppe PancucciDublin, Ireland
One tornado, one campus built in phases.Illustration
The transformer that sets the recovery time.Illustration
Swiss Re Institute estimates that around 40% of US data centre capacity, operating, under construction and planned, could sit in significant-to-very-high tornado-day zones. One path can cross several halls.Source: Swiss Re Institute, sigma insights 07/2026, 27 March 2026.
One event, read by five wordings. Physical damage triggers construction and property; cyber and delay in start-up depend on the policy text; liability needs a third-party claim.
Each treaty counts the event its own way: an hours clause in the catastrophe cover, an event limit in the per risk, a cession limit in the quota share. Each pays above its own retention.
Business interruption pays only up to its indemnity period. Recovery runs at the pace of the slowest component, and campuses waiting on the same maker share the queue.
Illustrative. Asset ownership, indemnity periods, wordings and treaty terms, including each treaty’s definition of event, vary by campus. Tornado exposure: Swiss Re Institute, sigma insights 07/2026, 27 March 2026.
Read on one campus, and on what it depends on
5lines
One site, many treatiesConstruction, property, power, cyber and liability, each written under its own name and ceded in its own treaty. The model sets them side by side on the campus, and shows which of them one event can reach.Accumulation across lines12 v 36
Months: indemnity against replacementA 12-month business interruption indemnity period on two rated US campuses, against 36 months, and up to 60, to deliver a large power transformer. Beyond 18 to 24 months of restoration, the tenant can terminate the affected leases.S&P Global Ratings; US Department of Energy80%+
One component, every campusMore than 80 per cent of large power transformers are imported, and there is a single US producer of the grain-oriented electrical steel at their core: a dependency shared by campuses that sit in different portfolios.US Department of Energy, 2024US data. The campus measures come from public documents first and from cedant data as it arrives; the transformer figure is national. None is a total of market exposure.
The model
One campus, one record.
What it gives you
Three answers for every campus in a portfolio.
For each data centre campus a cedant writes, the model sets out three things that today sit in different files, under different names.
Accumulation by campus
Every line exposed to the same site, construction, property, power, cyber and liability, and every treaty it is ceded into, set side by side on one campus, with what one event can reach. The figure a cedant needs to decide whether to buy dedicated protection, and a reinsurer needs to price it.
The indemnity gap
For each campus, the time it would take to rebuild and refit, set against the indemnity periods its covers provide: business interruption, contingent business interruption, delay in start-up.
The technology it depends on
The components that set the recovery time, who makes them, how long they take to replace, and which other campuses in the same portfolio depend on the same ones.
Built first from public documents, rating presales, prospectuses, grid and permit registers, then from the cedant's own submissions and bordereaux, matched to the same campus.
The model provides analytics. It does not underwrite or place risk. First phase completes at the end of 2026. Presentation to the market planned for 2027.
Four kinds of delay, where they strike, and who carries each
Permitting
Before the switch-on
An air permit refused, a grid connection that waits. No damage, no delivery, no rent.
Carried by the tenant in one contract, the developer in the next.
Supply
Before delivery
A transformer with a lead time counted in years, a rack generation that ships late.
Carried by the contractor through liquidated damages, up to their cap, then by a reserve, the developer or the tenant, as the contract says.
Loss
After the event
A fire, a flood, a destroyed transformer, during construction or in operation. The delay is measured in months of lost rent.
Carried by delay-in-start-up cover in construction and business interruption in operation, each up to its indemnity period; beyond it, by the owner.
Financing
When value falls short
At refinancing or at lease end, the technology inside is worth less than the debt on it.
Carried by the guarantor of residual value, up to its cap, then by the equity, then by the lenders.
The blind spot
One campus, many books.
The same campus is financed by lenders, insured by underwriters and relied on by the firms that run on it, each under its own name. The links between them are contractual: lenders do not fund a campus without cover, and that cover is split across lines, layers and carriers, sometimes in separate programmes for buildings, equipment and power.
“For data center owners and hyperscalers, insurance is not just balance-sheet protection: it is a critical prerequisite for capital to flow at all.”
Portfolio data rarely arrives with the granularity needed to trace one site across lines, cedants and treaties. An accumulation that is obvious on a map is invisible in the data.
Why no institution adds this up today: three books, no sum.Read the articles→About
Built one, merged one.
I led technology and operations at a reinsurer regulated by the Central Bank of Ireland: I built the function from the ground up as Chief Technology Officer and sat on its Audit and Risk Committee, then, as Chief Operating Officer, led the operational delivery of its cross-border merger into its parent group.
Before that, technology programmes in Italian insurance and banking. Today I work on the governance of data and AI in regulated finance, and on the model described on this page.
On the reinsurance side, the critical point is usually the data that arrives from cedants: its quality, its granularity, and the level at which it is aggregated. The model is built for that gap.
Contact
Compare notes.
If you write, place or reinsure data centre risk, I would like to test the model on one portfolio, with insured names removed but locations kept, under a confidentiality agreement, and compare what it finds with what you see today.