(Re)in Summary
- Data centres remain mainly property risks, but on-site generation and battery storage are extending underwriting into power, energy and engineering lines.
- Facilities taking three years to build can be outdated on completion, pushing developers towards powered shells fitted out to tenants’ specifications.
- Rapid equipment obsolescence makes new-for-old replacement after a loss difficult, as successors to discontinued models can cost several times more.
- Clustering around power, fibre and cooling means one earthquake or other catastrophe could damage several high-value facilities at once.
- Accurate valuations and asset records reduce underinsurance and claims disputes, while recovery planning belongs in risk management from the start.
Data centres in Asia-Pacific remain predominantly property risks, but on-site power generation and battery storage are pulling underwriters into energy-type exposures, speakers at the PARIMA Singapore conference said on Thursday (15 October).
Data centre risks now stretch beyond traditional property lines into power generation, energy and engineering, said Joe Wee, head of construction underwriting at Sompo.
Risks for data centres remain centred around property, but other aspects of data centre risk complicate cover, Wee said. “When we incorporate all this to data centre construction, suddenly the risk becomes a complex one.”
Ewan Smith, head of business development at real estate services firm Cushman & Wakefield’s data centre group, said data centres are becoming harder to build and insure as AI changes their design, adding that many operators are likely underinsured.
As AI pushes rack densities from 3-5 kW to 100-150 kW, buildings will have to be designed differently, and denser technology raises risk. “They are technically complex, and therefore they’re expensive,” Smith said.
Hyperscalers are also increasingly shaping how third-party facilities are designed, creating engineering risks, Smith said.
“If you’re a hyperscaler, and you’re moving into a data centre that perhaps doesn’t really fit what you want in terms of the engineering requirements, there’s risk,” said Smith. “If you’re a builder and you’re building a facility in the hope that some hyperscaler might like it, there’s risk.”
Obsolete before completion
The pace of technological change means facilities risk being out of date before they open, Smith said.
“A lot of the facilities that are being built now take three years to build,” Smith said. “By the time they’re finished, commissioned, they’re out of date. You’ve got to be really careful what you build.”
Developers are moving to build powered shells and cores before engaging hyperscalers and AI operators, consulting them on how they can fit data centres out, Smith added. “They don’t want to build on spec, because by the time it’s finished, it’s a dinosaur, and no good to anybody.”
Obsolescence, replacing equipment after a loss, will also raise issues. “How many people are drilling down into their policies, and looking at the replacement [part] and what they get replaced with?” Smith said. “Not only that, but is the [part] going to be available in five years or 10 years’ time? Everything’s moving at such a pace that you can’t do a new for old… and the next one’s three times the price, because it’s three times the capacity and it’s three times as fast.”
The uncertainty will persist, Smith added. “The industry is moving very, very quickly. Construction techniques are changing. Usage of the facility once it’s completed is changing. Cooling is changing, power generation, everything is changing.”
Developers are proposing data centres in increasingly unconventional sites, from abandoned shopping malls and decommissioned ships to undersea locations, raising questions about how underwriters assess untested designs.
These unconventional proposals will have to go through pilot testing, Wee said, and insurers will only come in once they are assessed as acceptable for risk transfer.
“We [underwriters] don’t like anything that’s first of its kind in the world,” said Wee. “So this ought to happen in the lab and all the research scientists and whatnot should be involved and data sets collected and then analysed.”
Major loss possibly inevitable
Grid constraints are pushing data centre providers around the world towards alternative power, said Derek Yong, senior risk engineer at HDI Global. “Due to the grid constraints, we do see providers who are turning to on-site power generation,” he said.
Risk engineers now have to assess data centres for both property and power risks. Battery energy storage systems also carry thermal runaway fire risk, and burning lithium-ion batteries often release toxic gas, which complicates firefighting, Yong added.
Operators short of space are packing battery banks too closely together, raising the risk of fire spreading, Smith said. “Spacing between the banks of batteries would greatly reduce the risk of the fire spreading should one battery catch fire,” said Smith. “But you try telling that to a guy that’s built a data centre where he’s only got half the space to put his lithium-ion.”
As data centres group around power availability, fibre routes and cooling, one catastrophe can hit several facilities at once, posing aggregation risks. “When you have a bunch of data centres located in one [place], a single severity event, for example an earthquake, would smash that entire complex”, said Wee.
The largest data centres are now potentially worth more than what any one insurer can absorb in a single placement, as hyperscaler AI campuses carry total insurable values of up to US$50bn, according to a new Howden report released earlier in October—making the consequences to aggregate risk potentially “catastrophic”, said Yong.
“We can imagine the impact if we can see multiple data centres in the same area,” he added. “We do try to minimise or mitigate the consequence of any natural catastrophe from happening, but I would say, essentially, it’s still a very big risk.”
A major loss may be inevitable, Smith warned. “We haven’t had a big fire, we haven’t had a big catastrophe, but it’s going to happen, and a lot of people will pay.”
Mitigate early
Mitigation needs to start well before a policy is bought, the panel said.
Insurance is “one of the many tools” available to risk managers, alongside business continuity plans, Wee said. “If it’s a technological issue, you really need to have your engineers heavily involved.”
Preventing losses is always cheaper than recovering from a loss, added Smith. Accurate valuations and asset records reduce disputes and underinsurance, he said, and recovery planning should be conducted “as part of risk management, not after recovery is needed”.
While risk managers may not be able to prevent an event from happening, they can help select data centre providers and review their risk mitigation measures, said Yong. “You have your insurance panel, insurers have risk engineers… so if there are any queries or questions, do not hesitate to ask them.”
