In short
There is no single off the shelf insurance policy that covers every risk for an operating AI data center. Instead, an owner or operator layers several types of coverage. These include commercial property, business interruption, equipment breakdown, cyber, technology errors and omissions, commercial general liability, and directors and officers insurance. The global market for AI data center insurance was valued at $2.26 billion in 2025 and is projected to reach $4.84 billion by 2030. Research and Markets The values at stake are measured in billions. A single site can cost over $20 billion to build, and after adding GPUs and other technology the value can double. Traditional insurance capacity cannot cover the full value at competitive rates. Meanwhile, insurers are beginning to write AI specific exclusions into standard policies, and a few brokers and carriers are launching new tailored products. This article explains what coverage is available, where the gaps are, and how the insurance market is adapting for owners, operators, and lenders.
What insurance policies cover an operating AI data center?
When an AI data center finishes construction and begins serving customers, the risk profile changes. Construction policies end, and operational insurance takes over. Because there is no single ready made policy for an AI data center, the owner must buy several separate policies, each covering different harms. Law firm analysis
Commercial property insurance
Commercial property insurance pays to repair or replace the physical assets of the AI data center when they are damaged by a covered event such as a fire, storm, or explosion. Policyholders should ensure their property policies cover the building, servers, GPUs, cooling equipment, power infrastructure, and business personal property. Law firm analysis
Standard commercial property policies often exclude or limit coverage for water intrusion, flood, faulty construction, wear and tear, loss of electronic data, and failure of utility services. An owner must check that the policy defines covered property to include every structure on campus and all equipment, and that sublimits for certain items do not leave a hole. Law firm analysis
Business interruption and contingent business interruption
Business interruption (BI) coverage is usually part of a commercial property policy. It replaces lost revenue and pays for extra operating costs when the AI data center must suspend operations because of a covered physical loss, for example a fire in the server hall. The coverage is triggered by physical damage, not by a network outage alone. Law firm analysis
Contingent business interruption (CBI) extends that protection to a loss at a separate property that the AI data center depends on, such as a key power plant or a critical equipment supplier that suffers a fire. Without CBI, a shutdown at the utility or a supplier could leave the AI data center with no income and no coverage. Law firm analysis
Equipment breakdown insurance
Standard property insurance typically does not cover internal mechanical or electrical failure. Equipment breakdown insurance fills that gap. It covers sudden and accidental breakdown of boilers, compressors, HVAC systems, electrical panels, and computer equipment. Major carriers include Hartford Steam Boiler and FM Global. The global equipment breakdown insurance market was valued at $11.8 billion in 2025. DataIntelo
Cyber insurance
Cyber insurance covers losses from data breaches, ransomware attacks, privacy violations, and third party liability. Policies often exclude losses caused by the insured’s failure to follow its own cybersecurity procedures, war or terrorism, and social engineering. Social engineering coverage can usually be bought back. Law firm analysis
In 2024, U.S. cyber insurers wrote $9.14 billion in direct premium, a 7 percent decrease from 2023. Reported claims increased nearly 40 percent to roughly 50,000. NAIC 2025 Cybersecurity Insurance Report
A growing number of cyber insurers are now adding language that defines AI or excludes certain AI related losses. As of mid 2025, at least three cyber insurers had introduced specific AI definitions or terms in their policies, according to the Betterley Report. Law firm analysis
Technology errors and omissions insurance
Technology E&O insurance covers claims that the insured’s technology services or products caused a loss, including claims from mistakes in AI systems. These policies are not written on standard forms and are highly negotiable. Many are combined with cyber coverage. A typical exclusion bars coverage for power, utility, or telecommunications failures that occur outside the AI data center’s direct control. Law firm analysis
Commercial general liability insurance
Commercial general liability (CGL) insurance covers bodily injury, property damage, and personal and advertising injury caused by the AI data center’s operations. Pollution exclusions in standard CGL policies may block coverage for diesel emissions from backup generators, PFAS from cooling systems, and noise or vibration complaints from neighbors. In Loudoun County, Virginia, residents have complained about construction noise and vibration, and these could turn into liability claims. Bloomberg, cited by Law firm analysis An operator can buy back some pollution coverage through an endorsement.
Directors and officers insurance
Directors and officers (D&O) insurance protects executives and board members from claims that they failed to properly oversee AI risks. Insurers now examine a company’s AI governance during underwriting. Some carriers have added absolute AI exclusions to D&O policies, as explained below. Law firm analysis
| Coverage | What it covers | Key exclusions to watch | Notes |
|---|---|---|---|
| Commercial property | Physical damage to buildings, servers, GPUs, cooling, power infrastructure | Water, flood, faulty design, utility failure, electronic data | Check that whole campus is included and sublimits are adequate |
| Business interruption | Lost income and extra expenses after a covered physical loss | No coverage without physical damage trigger | Add contingent BI for utility and supplier dependencies |
| Equipment breakdown | Internal mechanical or electrical failure of HVAC, generators, electrical panels, servers | Gradual deterioration, wear and tear | Carried by Hartford Steam Boiler, FM Global |
| Cyber insurance | Data breaches, ransomware, privacy liability | Failure to follow security procedures, war, terrorism (social engineering buy back available) | AI definitions and exclusions emerging |
| Technology E&O | Errors in technology services, including AI systems | Power or telecom failures outside the facility’s control | Highly negotiable, often combined with cyber |
| Commercial general liability | Bodily injury, property damage, personal injury | Pollution (diesel, PFAS, noise, vibration) unless endorsed | Important for neighbor claims |
| Directors and officers | Executive liability for oversight failures, including AI governance | AI exclusions appearing in some markets | Insurers reviewing AI governance |
What are the biggest physical risks for an AI data center?
The technology inside an AI data center is expensive, but the biggest threat to operations is often ordinary physical damage. A 15 year study by FM Global of traditional AI data centers found that fire caused only 10.9 percent of loss events but accounted for 42.3 percent of total loss costs. Water from sprinklers or cooling systems caused nearly 24 percent of total loss costs. Power supply failures caused 45 percent of all AI data center outages, according to the Uptime Institute.
AI servers amplify these risks. They can draw more than 100 kilowatts of power per rack, compared with 5 to 15 kilowatts for traditional servers. This higher density increases fire and cooling loads. Many AI data centers are also installing lithium ion battery backup units (BBUs) directly in server racks, creating an ignition source that did not previously exist inside the equipment room. FM Global’s 2026 data sheet revision raised the recommended fire resistance rating for walls from one hour to two hours in response.
Two recent incidents illustrate the exposure. In September 2025, a lithium ion BBU failed during maintenance at a South Korean government AI data center, causing an explosion and fire that knocked out 647 services including emergency response and tax systems. Swiss Re sigma 07/2026 In September 2024, a lithium ion battery fire at a Digital Realty AI data center in Singapore burned for more than 36 hours and heavily disrupted Alibaba Cloud services. Swiss Re sigma 07/2026
On site power generation also introduces new risks. Roughly 30 percent of planned US AI data center capacity could have power generation on site. Battery energy storage systems being integrated into AI data centers also bring fire, explosion, and toxic gas hazards. Swiss Re sigma 07/2026
Natural catastrophes are another large exposure. Over one quarter of U.S. AI data center capacity may be in areas that see three or more large hail days per year, and over 40 percent could sit in zones with significant to very high tornado activity. Swiss Re sigma 07/2026 Swiss Re estimates that a single AI data center location could face close to $10 billion in natural catastrophe losses. Reuters
How are insurers handling AI as a risk?
Most AI related risks are currently covered silently under traditional policies. A commercial general liability policy that covers bodily injury from a product may also respond if an AI system causes harm, because the policy does not mention AI at all. This silence is similar to how early cyber risks were handled before dedicated cyber insurance existed. But insurers are now deciding what to cover and what to exclude. WTW
In 2025, the Insurance Services Office (ISO) published three optional endorsements for commercial general liability policies that exclude generative AI. They are
- CG 40 47, which excludes all bodily injury, property damage, and personal and advertising injury coverage for any loss arising out of generative artificial intelligence.
- CG 40 48, which removes coverage only for personal and advertising injury.
- CG 35 08, which removes coverage for bodily injury or property damage under products and completed operations.
These endorsements are optional. Carriers can choose whether to attach them. There is no public data on how many have done so as of mid 2026.
Several individual carriers have introduced their own AI exclusions. The Berkley Absolute AI Exclusion, used in D&O, E&O, and fiduciary liability policies, excludes any claim based upon, arising out of, or attributable to any use, deployment, or development of AI. It defines AI broadly to include any machine based system that infers how to generate outputs such as predictions, content, recommendations, or decisions. Law firm analysis
The Hamilton Select AI Exclusion excludes any claim, wrongful act, damages, or defense costs based upon, arising out of, or in any way involving any actual or alleged use of generative artificial intelligence by the insured. Law firm analysis
Not all carriers move toward broad exclusions. Philadelphia Indemnity has an endorsement that carves out content created using generative AI in performance of the insured’s services while keeping coverage for other offenses. Law firm analysis
At least three cyber insurers now include explicit AI definitions or terms, but most are still in a wait and see mode, according to the June 2025 Betterley Report. Law firm analysis
What new insurance products address AI data center risks?
In response to the scale of the market, several brokers and carriers have launched tailored products.
Aon Data Center Lifecycle Insurance Program. Launched in July 2025, this program provides up to $1.5 billion in total coverage for both construction and operational risks. It includes $400 million for construction cyber physical damage and uses advanced risk engineering and cyber impact modeling. Research and Markets
Zurich Data Center Project Guard. Available from January 1, 2026, Zurich’s product is primarily a builders risk policy that tacks on operational coverage. After construction is complete, the policy can provide up to 12 months of property and business interruption coverage, parametric weather protection during the first year, climate control system failure coverage, and impounded water coverage. Zurich has insured more than 250 AI data center projects across 20 plus states with a total project value over $350 billion. Zurich North America
Willis Digital Infrastructure Protector. Launched by WTW’s Willis business in April 2026, Digital Infrastructure Protector is described as an end to end lifecycle insurance solution for data center owners and operators. The solution provides access to more than US $3 billion in capacity. Reinsurance News
IMA Financial Group $4 billion placement. In mid 2026, IMA Financial Group closed a $4 billion property insurance placement for a publicly traded AI and high performance computing AI data center company. This is one of the largest single property placements in the digital infrastructure sector. LinkedIn / AIM Media House
Other offerings. The CRC Group, through Insurisk, offers a product that provides up to $500 million in combined property (including builders risk) and casualty capacity for high density AI data centers. CRC Group Armilla Insurance Services introduced an AI liability insurance product in April 2025 underwritten by Lloyd’s of London syndicates including Chaucer Group. It covers losses from AI hallucinations, degrading model performance, and mechanical or algorithmic failures. Law firm analysis Google Cloud partnered with Beazley, Chubb, and Munich Re to offer a tailored cyber insurance solution with affirmative AI coverage for its customers. Law firm analysis Munich Re has offered performance guarantee coverage for AI through its aiSure product since 2018. Law firm analysis
What capacity constraints and market challenges exist?
The insurance industry is struggling to provide enough coverage for the largest AI data center campuses. Swiss Re reports that the re/insurance market can only support a fraction of the full construction cost limits at competitive rates for traditional risk policies. Reuters
Global insurance premiums tied to AI data centers are expected to rise to $24.2 billion by 2030, up from $10.6 billion. Swiss Re sigma 07/2026 The narrower AI data center insurance market alone is forecast to grow from $2.64 billion in 2026 to $4.84 billion by 2030, a compound annual growth rate of 16.3 percent. Research and Markets
A key challenge is accumulation risk. Large AI data centers are often insured through separate programs for buildings, equipment, and power plants. A single physical event, such as a fire or tornado, could trigger claims under multiple programs at the same location, making it difficult for insurers to track their total exposure. Swiss Re’s Head of Engineering and Nuclear, Jimmy Keime, warned that a single AI data center location could generate close to $10 billion in natural catastrophe losses. Reuters, Swiss Re sigma 07/2026
Rate increases reflect these pressures. In the Northern Virginia AI data center corridor, which holds the largest concentration of AI data centers in the world, property and business interruption rates have risen 15 to 25 percent over the past two years, according to one broker report. Hotaling Insurance
To expand capacity, some insurers are looking to the insurance linked securities (ILS) market to bring in third party capital. Guy Carpenter’s CEO, Dean Klisura, indicated that insurers are considering ILS to support AI data center capacity needs. LinkedIn / Steve Evans
What should owners, operators, and lenders do now?
The AI data center insurance market is evolving quickly. Policyholders and their advisors should take several practical steps.
- Review all policies for AI exclusions. Check whether your CGL, D&O, E&O, and cyber policies contain the new ISO endorsements or carrier specific AI exclusions. If coverage is silent, that may be favorable today, but be prepared for exclusions at renewal.
- Negotiate narrow exclusions. If a carrier proposes an absolute AI exclusion, push for a narrower carve out, like the Philadelphia Indemnity approach, that only removes coverage for AI generated content while preserving protection for other AI related risks.
- Ensure contingent business interruption covers critical dependencies. Map out which third party failures would shut you down (the local utility, a key cooling equipment supplier, a cloud provider) and confirm that your CBI coverage responds.
- Value property accurately. An AI data center’s value can double after installing GPUs and other technology. You need a regular appraisal and policy limits that match replacement cost. A $20 billion build may need a $40 billion property limit after equip.
- Work with risk engineers to meet the latest standards. FM Global Data Sheet 5-32 now recommends two hour fire rated walls in equipment rooms with lithium ion batteries. Adopting these standards can help with underwriting and reduce premiums.
- For lenders, require adequate coverage and monitor the AI exclusion landscape. A loan agreement should mandate that the borrower maintain all necessary insurance with limits reviewed annually. Ask whether an AI exclusion could reduce the coverage that protects the collateral.
Key takeaways
- There is no single AI data center policy. You must layer property, BI, equipment breakdown, cyber, E&O, CGL, and D&O coverage, and tailor each.
- AI data centers face concentrated physical risks from fire (42% of loss costs), water (24%), and power outages (45% of outages). New battery and on site generation technology adds to the hazard.
- Most AI risks are currently covered silently, but insurers are rapidly introducing exclusions. The ISO’s 2025 generative AI endorsements and the Berkley absolute exclusion are the most significant.
- New tailored products from Aon, Zurich, Willis, and others are beginning to fill gaps, but capacity remains limited. The biggest sites cannot be fully insured at competitive rates.
- Accumulation risk from single site events is a major underwriting challenge. A $20 billion campus could cause $10 billion in natural catastrophe losses.
- Owners and lenders should actively review their policies, negotiate narrow AI exclusions, secure contingent BI for critical suppliers, and adopt the latest fire protection standards.
Frequently asked questions
Q:Is there a single insurance policy that covers everything for an AI data center?
A:No. You must buy multiple policies and coordinate them. No standard packaged policy covers all the necessary risks.
Q:What does business interruption coverage do for an AI data center?
A:If a covered physical loss (like a fire) forces the center to stop operations, BI coverage replaces the lost income and pays for extra expenses during the shutdown. Contingent BI extends that to losses at suppliers or utilities the AI data center relies on.
Q:Does my cyber insurance cover AI related failures?
A:Most current cyber policies do not specifically mention AI. Some losses caused by AI, such as a hack that uses an AI tool, may be covered. But a growing number of insurers are adding AI definitions or exclusions. You must review the policy language.
Q:What is the ISO generative AI exclusion and does it apply to my policy?
A:In 2025, ISO released three optional endorsements that insurers can attach to commercial general liability policies to remove coverage for injuries or damages arising out of generative artificial intelligence. They are optional, so whether your policy includes one depends on the carrier’s decision. Ask your broker.
Q:How much does AI data center insurance cost?
A:Pricing is proprietary, but the market is hardening. In Northern Virginia, property and BI rates have increased 15 to 25 percent in two years. Insurance premiums tied to data centers are estimated at about $10.6 billion now and are projected to reach $24.2 billion by 2030, according to Swiss Re. The narrower AI data center insurance market specifically is about $2.64 billion and is growing roughly 16 percent annually, according to Research and Markets.
Q:How do I make sure my property insurance covers the full value of servers and GPUs?
A:The property policy must list all covered property categories and set limits high enough for replacement cost. After installing expensive GPUs, the campus value can double. You should get a professional appraisal and update the policy schedule.
Q:My AI data center’s generators and cooling systems might disturb neighbors. Does my insurance cover that?
A:Standard CGL policies often contain pollution exclusions that could bar coverage for noise, vibration, diesel exhaust, or cooling system chemicals. You can buy an endorsement to bring back some of that coverage. Loudoun County, Virginia, has already seen resident complaints over these issues.
Q:What new insurance products exist for operational AI data centers?
A:Aon’s Data Center Lifecycle Program offers up to $1.5 billion. Zurich’s Data Center Project Guard adds up to 12 months of post construction operational coverage. WTW’s Willis launched Digital Infrastructure Protector. The CRC Group provides up to $500 million in combined capacity. For AI liability, Armilla and Google Cloud partnerships offer targeted coverage.
Q:Can a $20 billion AI data center get fully insured?
A:Not entirely, or not at competitive rates. The re/insurance market can only handle a fraction of the full value. Large owners often self insure part of the risk or use multiple programs that cover different pieces.
Q:How can a lender protect its interest in the insurance?
A:A lender should require in the loan documents that the borrower maintain all necessary coverages, name the lender as an additional insured on property policies, and periodically review the policy terms, especially any new AI exclusions that could reduce the insurance’s value as collateral protection.
Subscribe to The Compute Law Brief
The Compute Law Brief is a free weekday newsletter on the law of AI infrastructure across tax, real estate, construction, power, and deals. The big US build markets and federal law. Three minutes a morning. No paywall, and no email gate to read the blog. Subscribe if you want it in your inbox.
Junde Liu, JD, LL.M. (Taxation) candidate at UF Law. Originally published on Compute Law Blog. This article is general information and does not constitute legal advice. Reading it does not create an attorney client relationship. The reader should not act on the basis of any content here without first consulting a licensed attorney in the relevant state. Last reviewed for accuracy May 23, 2026.