How Data Centers Are Testing Insurance Capital, Underwriting, and Claims
Analysis
By James W. Moore
Key Takeaways
- AI data centers are not creating entirely new property risks. They are concentrating familiar risks inside projects of unfamiliar size, complexity, and value.
- The insurance market is responding with larger facilities, integrated coverage, and engineering-led capacity. The question is whether that response can keep pace with construction and concentration.
- Capital gets a project insured. Underwriting determines how the pieces fit together. Claims reveal whether they actually do.
Put two insurance CEOs at a table and tell them a single data center may need billions of dollars in property coverage.
Neither one starts by asking whether the building might catch fire. Of course it might. That part is familiar.
The first question is more likely to be: How much of this do we want?
The second: Who else is taking a piece?
Then the conversation moves to the terms, the engineering, the handoff from construction to operations, and what happens if something goes wrong before that handoff is complete.
That is what makes Marsh’s new Stratus facility interesting. It can provide up to $10 billion in property capacity for operational data centers and other digital infrastructure, drawing on 30 traditional and alternative-capital providers.
Ten billion dollars sounds like an answer. It is really evidence of the size of the question.
AI may feel like software, but it runs on some of the largest and most valuable physical infrastructure projects being built today. Those facilities require power, cooling, specialized equipment, connectivity, and almost uninterrupted operation. Some data halls may already be running while construction continues elsewhere on the same campus.
Insurance has to keep up with all of it.
Who Takes How Much?
There is no single way to assemble enough capacity for these projects.
Marsh’s answer is an exchange that spreads the risk among many providers. Aon’s Data Center Lifecycle Insurance Program takes a similar coordinated approach. In July, Aon expanded the program to as much as $5 billion in first-party construction and property capacity, backed by a panel of insurers. Cyber, cargo, liability, and terrorism sit alongside it with their own limits.
FM offers a different model. Through FM Intellium, the mutual can provide up to $5 billion from one insurer, supported by the engineering and loss-prevention work that has always been central to FM’s approach.
The identical headline numbers describe two very different strategies. Aon is organizing a panel. FM is putting substantial capacity from one balance sheet behind risks it has helped engineer.
Anyone who has assembled a large property program knows the headline limit is the easy part to explain. The harder work is deciding which balance sheet takes which layer, on what terms, and whether those commitments will still satisfy the project’s next lender.
Neither model suggests that data centers are uninsurable. Both suggest that insuring them requires more thought about where the risk ultimately sits.
Moody’s describes credible insurance as a gatekeeper for investment in data centers. That makes the insurance program part of the capital strategy. Private credit may accept more uncertainty during construction and early operations, although it will price that uncertainty into rates, fees, and covenants. Banks and long-term infrastructure investors generally expect clearer coverage and a more established operating record.
The building may not change as those investors rotate through. The definition of adequate insurance can.
So the capital discussion is not simply about whether the market can advertise a large number. It is about how much limit is actually available, how many carriers share it, where the concentration lands, and whether the resulting program is good enough for the next source of financing.
Capacity gets the project to the table. It does not finish the job.
What Exactly Are We Insuring?
A large data center is not one property risk with a few endorsements attached.
Equipment has to be transported and stored. The campus has to be built. Power and cooling have to be installed and tested. Individual data halls may be commissioned at different times. One hall may be serving customers while contractors are still working in another, even though both depend on shared power, cooling, or ventilation systems.
That can pull cargo, builders risk, delay in startup, equipment breakdown, operational property, business interruption, cyber, liability, professional liability, and surety into the same project.
None of those coverages is unusual on its own. Large accounts have always required several policies. The trouble is that a data center can move from one phase to another building by building, or even hall by hall, while the insurance contracts still expect cleaner dividing lines.
The challenge is not finding one perfect policy. It is making several reasonable policies work together.
The construction-to-operations handoff is where that becomes difficult. Practical completion, ownership, acceptance, and operational control do not always occur on the same day. Delay-in-startup coverage eventually has to give way to business interruption. A building that looks like one project may contain both active operations and unfinished construction.
If the handoff is not clear before the loss, it will not become clearer afterward.
Insurers and brokers are already adjusting. Aon built its program around risks that were traditionally handled separately. Zurich’s Data Center Project Guard can extend operational property coverage for up to 12 months after construction, helping bridge phased handovers. It also addresses climate-control failure, a particularly important issue when humidity and condensation can damage sensitive equipment without producing the kind of dramatic scene most people associate with a major property loss.
This is what market adaptation looks like in practice. It is not a grand reinvention of insurance. It is carriers and brokers finding the seams and trying to close them before a claim does it for them.
Then Something Goes Wrong
The causes are familiar. The concentration of consequences is not.
Allianz Commercial reviewed 221 data-center claims with a total reported value of approximately €677 million, including the shares of other insurers. Water damage was the most frequent cause. Fire represented 59% of claims value. By coverage line, business interruption was the leading driver of severity.
Those findings are not contradictory. Fire was the most expensive cause of loss. Business interruption was the coverage line through which much of the financial consequence emerged.
One Allianz example involved a hot-work fire late in construction. Smoke and soot traveled through the air-circulation system into large parts of the building. The reported claim was between $50 million and $100 million.
There is nothing exotic about welding sparks starting a fire. What changes the result is where it happens, what the ventilation system carries with it, how much sensitive equipment is exposed, and how long replacement or restoration takes.
The same problem appears with water, condensation, hail, power disturbances, and equipment breakdown. A component may be technically repairable but no longer covered by the manufacturer’s warranty. A replacement chiller, transformer, or other specialized system may take months to obtain. Temporary power, interim hardware, relocation, forensic accounting, and several groups of technical experts may all be needed at once.
Meanwhile, the meter is running. Construction is delayed. Revenue is lost. Service commitments may be missed. Lenders, tenants, contractors, owners, and insurers may all have different interests in the outcome.
This is where the program gets tested.
Zurich offers a useful example of the feedback loop. Its claims experience with climate-control failure, condensation, and white rust exposed friction among owners, contractors, policy language, and equipment warranties. Zurich responded by adding coverage intended to address that problem.
Claims informed underwriting. Underwriting changed the product. The revised product may influence how much capital the carrier is willing to deploy on the next project.
That is how the system is supposed to work.
Can Insurance Keep Up?
The data-center market does not look like an insurance market that has stopped functioning.
Marsh and Aon are assembling more capacity. FM is putting substantial engineering-led capacity behind selected risks. Zurich is addressing the handoff between construction and operations. Allianz is turning claims experience into better loss prevention and underwriting.
The market is adapting.
The harder question is whether it can adapt as quickly as project values, construction schedules, and geographic concentration are growing.
Capital determines whether the risk can be insured. Underwriting determines how. Claims reveal whether the structure worked. Then the lessons from claims flow back into underwriting and eventually into the amount and price of capital available for the next project.
That loop has always existed. AI infrastructure is asking it to move much faster and carry much more weight.
Marsh’s $10 billion facility is not evidence that data centers are uninsurable. It is also not proof that the problem has been solved. It is evidence that the insurance market is reorganizing around physical infrastructure that no longer fits comfortably inside conventional programs.
The real test will not be whether insurers remembered that a data center can burn, leak, or lose power. They know how to insure those events.
It will be whether a loss lands between two policies, two project phases, or two layers of capital, and whether anyone owns the whole problem before a containable event becomes a prolonged financial loss.
The loss may be physical. The insurance failure begins at the seams.
Sources
- Marsh launches Stratus, a $10 billion property insurance exchange
- Aon expands Data Center Lifecycle Insurance Program to $5 billion
- FM announces significant increase in capacity for FM Intellium clients
- Moody’s: The Big Deal About Data Centers
- Allianz Commercial: The Data Center Construction Boom, Risks and Claims Trends
- Zurich: Lessons From Data Center Claims
- Zurich Data Center Project Guard
