Who Insures the AI Boom?

A data centre can have expensive servers, long-term customers and substantial financing. Yet its ability to earn money still depends on something remarkably basic: keeping the equipment running.

Electricity must reach the building. Cooling systems must control the heat. Critical components must work together, including when something fails.

When that chain breaks, the financial consequences can extend beyond a repair bill. The operator may lose income, customers may experience disruption and lenders may still expect repayment.

That raises a question behind the excitement about artificial intelligence: who pays when the infrastructure supporting it goes wrong?

Insurance can absorb part of the loss. Understanding which part matters to anyone building, financing or investing in a data centre.

A growing business for insurers

Swiss Re Institute forecasts that AI data centres could generate approximately US$91 billion in cumulative insurance premiums globally between 2026 and 2030, covering construction and operation.

That figure needs careful reading. It is a forecast of premiums across five years, rather than an annual total, the value of the buildings or the amount insurers expect to pay in claims. It also covers the global market, rather than Asia alone.

The opportunity comes with difficult risks. A single location can concentrate valuable equipment, sophisticated cooling systems and several customers whose businesses depend on continuous service.

For an insurer, collecting more premiums is only one side of the equation. Claims, expenses and the cost of transferring risk to reinsurers also affect whether the business is profitable.

Who actually provides the insurance?

The company developing or operating a data centre typically works with an insurance broker to arrange suitable protection. Insurers accept specified risks in exchange for premiums. Reinsurers, in turn, can take on agreed portions of insurers’ exposure.

Large programmes may involve several insurers, each accepting a share of the risk or covering a different layer of potential losses.

A real example shows how this market is developing. In July 2026, Aon announced that its Data Center Lifecycle Insurance Program had expanded to up to US$5 billion in capacity for specified construction and operational covers, supported by a panel of insurers.

Aon’s role in arranging that programme should not be confused with a single insurer promising to pay US$5 billion for every project. Available capacity remains subject to underwriting, policy terms and the protection purchased.

The ownership structure matters too. A landlord might own the building, an operator might run the facility and a customer might own the servers inside it. Their financial interests and insurance arrangements can differ.

A policy protecting the building does not automatically protect every business using it.

Different problems need different protection

“Data centre insurance” is a convenient phrase, but it can describe a collection of policies and extensions.

The practical starting point is to ask what happened, whose property or income was affected, and which policy could respond.

Construction delays are not all treated alike

Imagine a facility scheduled to open in December. A fire damages essential electrical equipment during construction, and replacement pushes opening into the following year.

Construction insurance may address the physical damage. Delay in start-up cover may address specified financial losses from the resulting delay, provided the insured trigger and other conditions are met.

Now imagine the same project opens late because an approval takes longer than expected or a customer changes its plans.

Those situations should not be assumed to qualify under the same insurance. Aon’s guidance on delay in start-up risk in Asia-Pacific explains that the cover is tied to physical damage under the relevant first-party insurance arrangement, with wording tailored to the project.

For a developer, the distinction has a direct financial consequence: buying insurance does not remove the need to fund delays that fall outside it.

An outage does not automatically mean a payout

Once a facility is operating, the distinction between damage and disruption becomes especially important.

A fire that damages insured equipment and stops operations presents a different coverage question from a software error that interrupts service without damaging anything.

Under many traditional property arrangements, business interruption protection depends on insured physical damage. Cover for failures at an external electricity supplier, or for interruption without physical damage, requires particular attention to the policy and any extensions.

Cyber insurance may address certain system disruptions, but it should not be treated as universal protection against every technical failure.

Customer compensation also needs separate review. A service agreement may require credits or payments when performance falls below an agreed standard. That obligation does not automatically become an insured loss under the operator’s property policy.

Specialist solutions do exist. Munich Re, for example, offers tailored liquidated damages cover for data centres, including certain operational service-level obligations. Whether it fits a project depends on the terms agreed.

For operators and lenders, the useful exercise is to compare customer promises with insurance protection before an incident occurs.

One incident can reach several balance sheets

Data centres connect businesses that may otherwise appear unrelated.

A problem at one facility can affect the operator, equipment owners and customers. A failure in shared infrastructure can disrupt several facilities at once.

For insurers and reinsurers, this creates a concentration problem: policies that look separate may respond to the same event.

Swiss Re highlights how shared power, water and transport infrastructure can expose multiple insured businesses to a common loss. An insurer therefore needs to understand the connections between sites, rather than simply count the number of buildings it covers.

The same thinking helps investors. Owning interests in several facilities does not necessarily provide strong diversification if all depend on the same critical electricity connection or other shared resource.

The question is whether the underlying dependencies are different enough to prevent one failure from spreading through the portfolio.

Good engineering affects the insurance conversation

Insurance starts with an assessment of what could go wrong and how large the resulting loss could become.

Allianz’s 2026 analysis of data centre claims identified fire as the leading contributor to loss severity and water damage as the most frequent cause within the claims examined. Those findings describe that analysis; they should not be read as a ranking of every data centre outage worldwide.

For a project in Asia, the practical review should reflect its particular location and design. Flood exposure, electricity arrangements, cooling equipment, fire protection and access to replacement components all deserve attention where relevant.

Backup equipment also needs scrutiny. A second power source offers limited reassurance if the same event can disable both supplies. Spare components help only if they are suitable and can be installed quickly enough.

Developers should be able to explain how systems are tested, how maintenance is managed and what happens when a critical component fails.

These details help insurers assess the risk. They also help owners understand how much disruption the business may need to absorb itself.

What should investors and lenders ask?

An insurance certificate confirms less than a full understanding of the programme.

Before relying on insurance in a financing or investment decision, useful questions include:

• Which company owns each major asset, and whose financial interests are insured?
• What events trigger protection for construction delays and operating interruptions?
• How are external utility failures and technology providers treated?
• What deductibles, waiting periods, exclusions and limits could leave the business exposed?
• Does the time allowed for interruption losses reflect realistic replacement and recovery periods?
• Which customer obligations have separate protection, and which remain with the operator?
• How much cash could the business need while repairs and claims are being resolved?

These questions connect the insurance documents to the project’s cash flow.

They also keep insurance in its proper place within the investment case. Property and interruption policies should not be assumed to protect against weaker customer demand, poor pricing decisions or equipment becoming commercially outdated.

A well-insured facility still needs a viable business model.

The AI boom is creating demand for infrastructure that must operate reliably under demanding conditions. Insurance helps distribute some of the financial consequences when things go wrong. The strongest projects can explain both what their policies protect and how they will fund the risks that remain.