South Korean insurers are developing new approaches to AI data center risks as the country prepares for a major expansion of artificial intelligence infrastructure, putting underwriting standards, available capacity and placement requirements under greater scrutiny.
Samsung Fire & Marine Insurance is preparing AI data center risk assessment guidelines with external experts, with release expected in the second half of 2026, according to the industry report. Hyundai Marine & Fire Insurance and Hanwha General Insurance are also reviewing their approaches to the market.
The response comes as South Korea plans to develop 18.4 GW of AI data center capacity by 2035, with total investment expected to exceed KRW 1,000 trillion. An initial 8.4 GW, representing about KRW 550 trillion of investment, is planned by 2029.
The broader Korean insurance industry is also examining the risks. In July, the Korea Fire Protection Association held a data center risk management and underwriting seminar attended by representatives from 11 non-life insurers, including Samsung Fire & Marine Insurance, Hyundai Marine & Fire Insurance and Hanwha General Insurance. The event focused on risk assessment and claims issues associated with data centers.
For brokers, the development of clearer underwriting approaches could affect the information, engineering assessments and risk controls required to secure coverage as more AI data center projects enter the South Korean market. Rising project values, overlapping property and technology exposures, and limited loss experience could also make it harder to secure sufficient limits and satisfy lender requirements.
Those challenges are already emerging internationally.
Swiss Re Institute estimates global insurance premiums tied to data centers will rise from $10.6 billion to $24.2 billion by 2030. Construction costs for a single site can exceed $20 billion, with insured values increasing further once GPUs and other technology are installed. At that scale, Swiss Re said traditional insurance and reinsurance capacity may be insufficient to provide all the limits sought by financing institutions at competitive rates.
That makes insurance capacity more than a risk-transfer issue. Lenders can require limits reflecting the full cost of construction, meaning the availability and structure of coverage can become part of the financing equation for major projects and a key consideration for brokers assembling placements.
International brokers have meanwhile been increasing dedicated capacity as project values grow.
Aon raised the capacity of its Data Center Lifecycle Insurance Program to $5 billion in July, following an increase to $3.5 billion earlier in the year. The program provides up to $5 billion for construction all risks, delay in start-up, property damage and business interruption, alongside separate capacity for cyber and technology errors and omissions, project cargo, liability and terrorism risks.
Marsh has also expanded its Nimbus data center facility, which now provides limits of up to $2.7 billion, including delay in start-up and business interruption coverage, for major construction projects across the U.S., UK, Canada, Europe, Australia and New Zealand.
The capacity issue is compounded by changes in the underlying risk.
AI data centers operate at substantially higher power densities than traditional facilities and increasingly rely on liquid cooling, battery storage and complex power infrastructure. Swiss Re cited industry data showing that fire accounted for 42.3% of traditional data center loss costs despite representing 10.9% of loss events, while liquid-related incidents represented nearly 24% of loss costs. Power supply was responsible for 45% of data center outages, according to Uptime Institute data cited by Swiss Re.
Those exposures can also cut across several insurance programs. Large data centers may insure buildings, equipment and power infrastructure separately, potentially making aggregate exposure harder for carriers to identify. Shared power, cooling and fire-protection systems can also allow one event to generate multiple property and business interruption claims across tenants and policies.
For brokers, that increases the importance of understanding not only individual policy limits but also how property, construction, cyber, equipment and business interruption exposures interact across an entire site. It could also make insurer appetite, aggregation limits and the sequencing of different covers more important when building large placements.
Underwriting remains difficult because relatively few next-generation AI facilities have accumulated meaningful operating and claims histories. Swiss Re said limited empirical loss experience means insurers will increasingly depend on specialized technical assessments and tighter management of accumulated exposures when deploying capacity.
For brokers, limited claims data could translate into greater variation in underwriting approaches, pricing and required risk information between carriers, increasing the importance of early engagement with markets when structuring coverage. In practice, that means bringing in a specialist technical risk engineer or assessor at the submission stage, before financing terms are finalised, rather than treating engineering input as a step that follows once a placement structure is already set.
The Korea Insurance Research Institute has also identified a widening gap between the insurance values of large data center projects and the capacity available from insurers and reinsurers at competitive rates.
Kim Jin-eok of the Korea Insurance Research Institute said in the report, "Global insurance companies are rolling out dedicated programs that cover all stages from construction to operation under a single contract," and noted, "As large-scale AI data center investment plans have been announced domestically, it is necessary to review the current fire insurance industry classification standards for data centers and the underwriting criteria for dedicated insurance."
