Swiss Re's $200 Billion Sits on Specific Ground

At the Monte Carlo Rendez-Vous this week, Swiss Re Institute put a number on something the market has been circling for a year: the build-out of AI data centers and renewable energy could be a $200 billion premium opportunity between 2026 and 2030. The scale behind that number is hard to picture. Global data-center capex is expected to pass $1 trillion; the five largest US hyperscalers alone are on course for close to $800 billion of AI capital spending in 2026; global energy investment is heading for $3.4 trillion, with $2.2 trillion of it in renewables and related sectors. Artemis reported the findings alongside two lines that matter for the ILS market. As the insurance towers behind these assets are assembled, "reinsurance and alternative capital sit above and behind the subscription tower." And Swiss Re "see[s] a future role for catastrophe bonds and sidecars to provide additional capacity for upper layers of property programmes exposed to large natural catastrophe scenarios" — with brokers and reinsurers already at the prospective stage on cat bond and sidecar structures for data center risk.

That is an invitation. We think it is also a warning, and the warning is about geography.

The capex super-cycle is a construction program, not a spreadsheet

The phrase "capex super-cycle" makes the build-out sound abstract. It is not. It is steel, concrete, transformers, chillers and diesel, going up on specific parcels of land, most of them chosen for the price of power and the availability of fiber rather than for the hazard profile of the ground. A hyperscale campus is a billion-dollar-plus structure with a replacement schedule measured in years, a contents value (the compute) that can exceed the building, and a business-interruption exposure that starts accruing the second the utility feed drops.

Insurers know how to write that risk one site at a time. What is new is the scale, the concentration, and the speed. The data center half of Swiss Re's $200 billion is towers stacked on a few hundred campuses that did not exist five years ago, in clusters that did not exist ten years ago. The renewables half is spread wider, but it is not spread evenly: a utility-scale solar farm is square miles of glass laid flat in the places with the most sun, and the places with the most sun in the United States overlap the hail belt to an uncomfortable degree. Wind is on the Plains, under the same convective storms. Neither asset class chose its ground for the hazard.

Where the buildings actually are

Follow the fiber and the substations and the map is short. Northern Virginia's Loudoun and Prince William counties carry more data center capacity than most countries. Phoenix and its western suburbs. Dallas–Fort Worth. Columbus and central Ohio. Atlanta. Hillsboro, Oregon and the Columbia River corridor. The Bay Area and Santa Clara. Chicago's western suburbs. Then the second wave chasing cheap power: Iowa, Nebraska, Texas panhandle wind, the Tennessee Valley.

Read that list as a hazard modeler and each cluster has a peril attached to it. Northern Virginia sits under the remnant-rain track of Atlantic hurricanes; the region's floods are fluvial, sudden and increasingly frequent. Dallas is in the middle of the hail and tornado belt, and a data center roof carrying a few hundred rooftop units is a large, flat, expensive target for two-inch hail. Phoenix's problem is heat and the water that the cooling plant needs, which is not a peril a property policy names but is a very real driver of outage. The Columbia River and the Bay Area are seismic: a Cascadia or Hayward event does not need to collapse a building to take a campus offline for a quarter. Central Ohio and Iowa are convective storm country. The Southeast is hurricane country, full stop.

A homeowners book spreads a peril across a million roofs. A data center book concentrates it on a few dozen; a solar book concentrates it on a few hundred fields that all face the same sky. That is the opposite of what a cat bond investor is used to holding, and it changes what the capital sitting "above and behind" the tower needs to know.

What the alternative capital will need

Three things follow from the concentration.

Site-level hazard, not zone-level. A rating territory or a county-level modeled loss is the wrong resolution for a portfolio of fifty campuses. The question an ILS investor or a sidecar sponsor will ask is: for this parcel, in this floodplain, at this distance from this fault, with this convective storm history, what is the hazard? That is a location question, and it has to be answered from the ground up, with the sources named.

Correlation across a handful of points. A convective outbreak that crosses Dallas and Oklahoma City on the same afternoon, a remnant tropical system that stalls over Virginia, a Cascadia rupture: each hits several campuses in one event. The tail of a data center book is driven by a small number of correlated sites, which is precisely the shape of exposure that stochastic event sets exist to characterize. A per-site expected loss added up fifty times will understate it.

Parametric structures that match the loss. Swiss Re is pointing cat bonds and sidecars at the upper layers of these programs, which is where the large-scenario tail lives. Data center loss is dominated by interruption, and interruption starts on a clock. That is a natural fit for parametric triggers — wind speed at a station, ground motion at a sensor, flood gauge height on the adjacent river — where the basis risk is explicit and the payout is fast. The industry-loss triggers that dominate today's cat bond market are a poor match: a data center campus can be down for a month in an event that barely registers in an industry index.

What we can and cannot say about it

Calchis scores hazard, not price. For a US location we return a hazard score for each peril the model covers — earthquake, flood, hurricane, wildfire, hail, tornado and winter storm — with the components, the sources and the parameters behind each number, and since this week the response also says which perils are not scored at that location and why. A hail score in Ohio is a decision with a stated reason, not an absence. Modeled dollar losses are available for hurricane only; for the other perils we return the hazard and refuse to invent a loss figure. And a location outside the modeled coverage is refused rather than scored against the nearest state's parameters, which matters for a build-out that is global even where our coverage is not.

None of that is a data center product or a renewables product. What it is, is the kind of site-level, source-cited, honestly-scoped input that a tower built on a few hundred campuses and a few hundred fields will need underneath it. The $200 billion is real. So is the ground it sits on.

Source: Swiss Re Institute, presented at the Monte Carlo Rendez-Vous de Septembre, as reported by Artemis, 5 September 2026. Figures are Swiss Re's; the geography and the hazard reading are ours.