Swiss Re Sizes a $200B Data-Centre Opportunity Its Own Underwriter Says the Industry Cannot Write

Swiss Re Institute's sigma 3/2026, published September 5, estimates that AI data centres and renewable energy infrastructure could generate around $200 billion in commercial insurance premium between 2026 and 2030 (Swiss Re, September 5, 2026). Four days later Gianfranco Lot, chief underwriting officer for P&C reinsurance at the same company, told the Monte Carlo Rendez-Vous that "the insurance industry is not able to cope with the demand" (Artemis, September 9, 2026). Both statements are correct, and the gap between them is being filled by the buyers.

The five largest US hyperscalers are expected to spend nearly $800 billion on AI capital expenditure in 2026, inside a global data-centre capex estimate above $1 trillion. A single campus can cost up to $50 billion to replace (sigma 3/2026 via Insurance Business, September 7, 2026). No property tower in the market reaches that figure, so the premium Swiss Re is counting is largely premium the hyperscalers will pay themselves.

Key Takeaways

$200 billion is Swiss Re Institute's five-year premium estimate for AI data centres and renewables, published at Monte Carlo alongside a warning that these assets "cluster in the same locations and increasingly depend on shared infrastructure and networks" (Artemis, September 5, 2026).
80% to 85% of large peak risks are reinsured, according to Lot, who added that "there's no insurance possible without reinsurance of these risks" and that the sidecar and cat bond structures proposed for the gap have not yet materialised (Artemis, September 9, 2026).
$20 billion to $30 billion is S&P's estimate of total insurable value on a single hyperscale campus, against carrier per-risk limits in the low single-digit billions and required limits of $5 billion to $10 billion (Artemis, July 10, 2026; Insurance Business, September 7, 2026).
$10 billion is the largest single placement on offer, Marsh's Stratus exchange launched August 26 with 30 traditional and alternative capital providers, next to its $2.7 billion Nimbus construction facility (Artemis, August 26, 2026).
$11.5 billion of the $79 billion written by the 1,900 captives Marsh manages is reinsured, and Marsh's captive leader expects "explosive growth in captives to take on portfolio risks for data centers" (Insurance Journal, September 14, 2026).

Two Swiss Re Statements Four Days Apart

The sigma report frames the buildout as the largest commercial P&C opportunity in decades. Global energy investment reaches $3.4 trillion in 2026, $2.2 trillion of it in renewables, nuclear, grids and storage, on top of the hyperscaler capex. Swiss Re projects annual AI data-centre spending of $1.6 trillion by 2031 and $7.6 trillion cumulatively from 2026 (Artemis, September 9, 2026). Its prescription is engineering-led underwriting, better accumulation modelling and disciplined accumulation management, and a spreading of exposure "across insurers, reinsurers and capital markets" (Insurance Business, September 7, 2026).

The report names cat bonds and sidecars as capacity for the upper layers exposed to earthquake, windstorm or flood, and describes brokers and reinsurers as "in prospective stages of developing" those structures (Artemis, September 5, 2026).

Lot's Monte Carlo remarks put numbers on the dependency. Because 80% to 85% of large peak risks are ceded, the primary market's line size is set by what reinsurers will take, and reinsurers are already at the limit of what they will hold per campus. Alternative capital "needs to be effective as well, in terms of the economics" for every party in the chain, he said. That is the condition under which the sidecars and infrastructure cat bonds would appear, and Swiss Re is "actively thinking" about them with existing partners (Artemis, September 9, 2026).

Jérôme Haegeli, the group chief economist, called for "a different risk architecture" and "syndicated and layered capacity" (Artemis, September 9, 2026).

S&P had already measured the architecture problem in July. Total insurable value on a hyperscale campus runs $20 billion to $30 billion, construction alone $10 billion to $30 billion, while the largest carriers offer per-risk capacity only "in the low single-digit billions". The agency expected the shortfall to "drive greater use of self-insurance through captive insurers, and potentially, alternative capital such as insurance-linked securities" (Artemis, July 10, 2026). Two months on, the captive half of that sentence has arrived and the ILS half remains prospective.

Where a $30 Billion Campus Goes When the Tower Stops at $10 Billion

The arithmetic of a layered tower explains the sequence. At per-carrier lines of one to three billion dollars, a $25 billion programme needs dozens of participations, each carrier tracking its aggregate across every other data-centre programme it has joined, as AIG's chief executive described in August. Marsh's Stratus bundles that syndication into one placement of up to $10 billion across 30 providers for operational assets, sitting beside the $2.7 billion Nimbus construction facility (Artemis, August 26, 2026). Zurich bought up to $1 billion of quota share dedicated to its data-centre construction book (Zurich, August 2026). Stack every one of those and a $30 billion campus is still half uninsured.

The owners have answered with balance sheet. Marsh manages about 1,900 captives writing $79 billion of premium, of which $11.5 billion is reinsured. Its US and Canada captive leader Michael Serricchio expects explosive growth in captives taking data-centre portfolio risk, and globally more than 6,000 captives now write about $240 billion, up nearly 20% in two years (Insurance Journal, September 14, 2026). Meta structured a special guarantee for the bondholders of its Hyperion campus through a Blue Owl special purpose vehicle rated A+ by S&P, which is a credit instrument doing an insurer's job.

GPUs inside the buildings have no obvious coverage product, and business interruption policies carry 12 to 24 hour waiting periods that a compute outage can exhaust before cover attaches (Insurance Journal, September 14, 2026).

The actuarial consequence sits in what a captive retains. A captive that keeps the working layer and buys only the peak-peril excess hands the market the thinnest premium and the fattest tail. The everyday equipment, water and outage losses that would build a damage curve stay inside the hyperscaler. The earthquake and windstorm limit goes to reinsurers who, by Lot's own account, cannot cope with the demand. Verisk mapped 2,500 US data centres this month and acknowledged the vulnerability curve is still missing; a captive structure keeps the data that would draw it out of the industry's hands.

The Accumulation the Captive Keeps and the Market Inherits

The layer the market does write is the one where diversification is thinnest. Texas and Virginia hold more than 40% of US data-centre capacity. Over 25% of capacity sits in hail zones and 40% in areas with three or more tornado days a year, while 88% of Taiwan's semiconductor fabrication sits in extreme or very extreme seismic zones (Insurance Business, September 7, 2026). Campuses are built in clusters that share power, cooling and network, so a peak-peril excess layer over a portfolio of them is a bet on one hail corridor rather than a spread of independent risks.

Maren Josefs at S&P said ILS "can play a role in those big project line sizes" but that "investors don't like surprises" (Artemis, September 7, 2026), and a concentrated, unmodelled peril is the definition of one.

The cat bond market that sigma nominates for the upper layers closed the half at $65.6 billion outstanding on record $18 billion issuance (Gallagher Re, September 1, 2026). That is two campuses' replacement value spread across every sponsor and peril in the world. Its 2026 issuance ran 78% indemnity with aggregate cover shrinking to 36% of outstanding risk capital (Artemis data, September 2026). The market is moving toward named-sponsor, per-occurrence structures at the moment data centres would need it to price a clustered, secondary-peril, multi-site exposure it has never held.

The $200 billion is a demand figure. The supply figures published the same fortnight are the $10 billion Stratus can place, the $2.7 billion Nimbus can construct, the $1 billion Zurich ceded, and the $11.5 billion of reinsurance behind $79 billion of captive premium. The difference between the first number and the sum of the rest is what the hyperscalers hold themselves, on campuses whose accumulation nobody outside them can yet model. Lot's rule that no insurance is possible without reinsurance runs the other way too: the reinsurer that sized the opportunity will see only the layer the owners chose to cede.

Swiss Re Sizes a $200B Data-Centre Opportunity Its Own Underwriter Says the Industry Cannot Write

Key Takeaways

Two Swiss Re Statements Four Days Apart

Where a $30 Billion Campus Goes When the Tower Stops at $10 Billion

The Accumulation the Captive Keeps and the Market Inherits

Further Reading

Sources