Insuring New Data Centers Against Disaster Risk

 The Wall Street Journal has published an intriguing piece about the rising demand by AI data centers for insurance.

In this Substack, I want to quickly cover a lot of ground.

• Given that data centers are new, Insurers can’t rely on past loss ratio data to calculate their risk exposure.

• Data centers are place based, multi-billion dollar investments. If a $10 billion dollar facility faces a 1/10000 chance of being knocked out by a tornado, then the risk neutral price of a policy is more than $1 million per year.

• Now, where did that 1/10000 probability come from? How does the Insurer estimate this? If a tornado literally struck the data center, how much damage would take place?

• The data center is brand new. If it has been located in a tornado alley, why wasn’t it built to withstand a serious tornado? Does the data center receive a serious insurance discount if it has taken these self-protection steps?

• Why would a major AI lab place a data center in a tornado alley? What role do government subsidies play in attracting data centers to risky places? What role does land use regulation in safer areas play in deflecting data centers to riskier places?

• In a risky location, what steps can a data center take to fortify the Building so that a tornado causes less damage?

Grok offers some insights;

To reduce physical damage from a serious tornado, data centers should use hardened construction. Build with thick reinforced concrete walls—often 12 inches or more—and steel frames designed to ASCE 7-22 tornado load standards for Risk Category IV essential facilities. These resist winds of 150–250+ mph depending on location and building size. Secure roofs against uplift with robust anchoring and high-wind-rated systems. Minimize openings; use impact-resistant or blast-rated doors and few or no windows to block flying debris. Enclose generators, HVAC, and electrical systems in interior vaults or protected spaces. Follow FEMA P-361 and ICC-500 principles for critical areas. Site layout should limit nearby debris sources. Regular inspections maintain integrity. These measures cannot ensure survival of a rare direct EF5 hit but substantially reduce damage from most tornadoes, protecting the facility’s structure.

I spoke to Chat GPT about how an AI lab could protect itself from a major data center disaster shock. Note the spatial diversification point here. Don’t keep all of your eggs in one basket!

Suppose that a leading AI Lab experiences a shock to one of its data centers that knocks it out permanently. Data Center equipment depreciates quickly over time. The shock allows the Center to be rebuilt with better equipment.

The AI Lab can raise the price of issuing prompts and collect revenue from the rest of their non-destroyed inventory of data centers. They can rent space from Musk who has surplus capacity. Pricing can be used to efficiently allocate AI capacity if data centers are hit with place based disasters.

This is AI adaptation!

From the insurer’s perspective, I am puzzled about how they will calculate their expected losses. Unlike people, cars or homes, there isn’t a large sample of AI data centers to run A/B tests on concerning the demand for such insurance and the losses that the insurers bear when bad things happen.

I also think that the major AI Labs can self insure. Does Meta and Google need to buy insurance?

The reason this story interests me is the link to natural disaster adaptation. If new capital can take a punch, then data centers (while very valuable in terms of installed capacity and the value of this capacity) are not at great risk.

When the Palisades fires were burning in January 2025, Pepperdine University ordered students to relocate to the safe buildings.

Each owner of a data center will have to decide how much to invest in risk mitigation. If government subsidies such data centers by offering FEMA recovery bailouts, then this will create a moral hazard effect and the for profit firms will invest less in their own defense.

The private insurers can be the “adult in the room” by encouraging these firms to invest in risk mitigation by offering them insurance premium discounts for taking these actions.