AI Data Centers Are Too Big to Insure. Companies Are Doing it Themselves.

Building a modern AI data center can exceed $20 billion, a figure that Swiss Re Group has said can double once the graphics processing units (GPUs), tenants, and servers are in place. Post-construction, the infrastructure still requires specialized cooling systems, robust security infrastructure and high-voltage power grid connections to become (and remain) operational.

Due to the need for vast land and renewable energy access, developers are clustering these massive facilities in regions that have historically been highly exposed to natural disasters. This leaves multi-billion-dollar data centers vulnerable to windstorms, floods and droughts — climate events capable of destroying hardware, throttling processing speeds or triggering emergency shutdowns.

Internal tech-driven risks further complicate underwriting for AI data centers. Water damage caused by cooling system leaks, cyber breaches targeting expanded operational technology and ignition hazards from lithium-ion batteries are all potential risks within the data centers. Together, these exposures drive the risk index of an AI facility far above that of a traditional commercial building, making its risk pricing tough on insurers.

It is unsurprising, then, that global insurance premiums for data centers are projected to climb to $24.2 billion by 2030, up from $10.6 billion at the moment.

"As data center development and construction spreads, the required insurance coverage is evolving, as it is currently beyond what the traditional property/casualty industry has previously experienced," said David Blades, associate director of Industry Research and Analytics at AMBest.

As a result, Michael Serricchio, regional leader at Marsh Captive Solutions, expects "explosive growth" in the use of captives — insurance subsidiaries that companies rely on to keep underwriting in-house.

The global captive market, currently valued at $258.2 billion, is projected to reach $332.7 billion by 2031. Meanwhile, AM Best-rated U.S. captives generated an estimated $8.2 billion in savings for their parent companies over a five-year period, the rating agency reported in July.

But even with captives, AI data center risks remain unpriceable due to the absence of relevant historical data. Alex Musy, CEO at Huscarl, which is building an autonomous AI actuary for corporations and their captives, offers one answer: AI-powered actuarial modeling.

Musy told International Business Times via email that the model "it enables more precise sizing of required insurance limits – in many cases, they can be decreased significantly with actuarial confidence. It also enables risk modeling based on proxies or synthetic datasets to price emerging risks with limited historical data."

The shift towards self-insurance isn't limited to data center operators, though. While captives are typically associated with large enterprises, the self-insurance model is rapidly moving downmarket into the broader tech sector.

Jonathan Selby, Tech & Media Practice Lead at The Baldwin Group, Innovation Practice, said that "coverage gaps, rather than premium pricing alone, are driving mid-market tech firms toward alternative risk solutions like captives."

"As standard insurers apply broad AI and IP exclusions to traditional E&O and Cyber lines, high-growth companies with non-standard risk profiles are left underinsured," he added.

For these businesses, "Captives offer a compelling way to turn insurance costs into balance-sheet assets," he added.

The balance sheet benefit seems to be driving record adoption across the industry. Global broker Marsh recorded 118 new captives formed in 2025 — a jump from 92 the previous year. This growth occurred despite a 4% drop in commercial insurance pricing, proving a structural shift in risk management rather than a temporary response to high premiums.

Venture capital investors are increasingly backing the self-insurance momentum, pouring capital into specialized risk management startups. Recent deals highlight the growing market opportunity, from a $45 million Series B round raised by captive manager Luzern Risk in September 2026, to Huscarl's $5.6 million seed round from FRST and Y Combinator for its autonomous AI actuary platform.

As traditional P&C insurers hesitate over massive, unpredictable exposures, major hyperscalers may start to bypass them entirely. Moving forward, the question of insuring data centers may no longer be about which carriers can cover their risk but whether commercial insurers can compete with the self-insured ecosystems tech giants are building for themselves.