US insurance regulators are looking at the credit risk behind AI data centre financing, and that matters because insurers have become a key buyer of the private debt keeping the buildout moving.
The AI infrastructure boom is no longer just a story about chips, power and land. It is becoming a story about who ultimately carries the credit risk when billions of dollars are raised to build data centres whose economics still depend on demand arriving years into the future. That question has now landed with US insurance rulemakers.
The National Association of Insurance Commissioners is examining how insurers are exposed to data centre projects, particularly through private credit and privately rated securities. As the Financial Times reported, the NAIC's Securities Valuation Office is reviewing investments across asset types and sectors, including data centres, as part of its work for state insurance regulators. That sounds technical. It is also important. The capital rules applied to an insurer can change materially depending on how risky a bond is judged to be.
The pressure point is private ratings. Many early-stage data centre projects are seeking investment-grade treatment so they can be bought by large institutional investors, including insurers and pension funds, even before the facilities are finished. In theory, the rating reflects the strength of tenants, lease terms, construction history and expected cash flows. In practice, regulators are asking whether that package is being assessed with enough caution in a market where AI demand, power availability and construction costs are all moving quickly.
The NAIC has had more room to challenge those ratings since January 2026. If a private rating diverges sharply from the regulator's own analysis, the NAIC can override it for capital treatment purposes. That matters because lower-quality investments require insurers to hold more capital. A data centre bond that looks cheap and efficient under one rating can become much less attractive if regulators decide the risk has been understated.
This is where the AI funding machine starts to look more fragile than the headlines suggest. Big technology companies and AI infrastructure providers have been raising extraordinary sums through bond markets, private credit, leases and special-purpose vehicles. US companies issued more than $200 billion of AI-related bonds in 2025, according to earlier Financial Times reporting, while a recent Axios analysis of S&P Global Market Intelligence data found that Alphabet, Amazon, Meta, Microsoft and Oracle had already raised $255.34 billion through equity and debt in the first half of 2026. The capital is flowing, but it is flowing through increasingly complex channels.
Insurance money has become one of those channels because insurers need long-term assets to match long-term liabilities. Data centre debt can look attractive in that context. The borrowers are often tied to cloud giants with strong credit profiles, and the projects are backed by long leases or contractual commitments. But the risk is not the same as buying plain corporate bonds from Microsoft or Amazon. A project can be delayed, power interconnections can slip, construction budgets can rise, and the economics can change if AI workloads do not generate the revenue investors expect.
That is why the NAIC's review is not just housekeeping. About 20 percent of US life insurers' fixed-income portfolios are now linked to private and illiquid bonds, according to Moody's Ratings data cited in recent reporting. Private credit can be perfectly legitimate, but it is harder for outsiders to price, harder to sell in stress and more dependent on assumptions that may not be tested until conditions turn. If too much exposure sits in one fast-growing sector, the question becomes whether insurers are being paid enough for the concentration risk.
Treasury has also been circling the issue. Treasury Secretary Scott Bessent met with state insurance commissioners in May to discuss regulatory responses to life insurers' rising exposure to private assets, with the department emphasizing regulation that supports innovation while managing risk. That language is careful, but the direction is clear. Washington does not want to stop AI infrastructure financing, yet it also does not want a private-credit boom to migrate quietly onto insurer balance sheets without adequate capital behind it.
The market implication is straightforward. If regulators force tougher capital treatment for some data centre debt, insurers may buy less of it or demand better terms. That would not end the AI buildout, but it could make financing more expensive and slow the projects that depend on cheap private capital. The strongest hyperscalers will still have options. More speculative projects, especially those relying on private ratings and optimistic lease assumptions, may find the next round harder.
For investors and operators, the signal is worth watching. The AI infrastructure trade has been treated as a one-way race to build as much capacity as possible. Insurance regulators are asking a different question: who is holding the risk if the numbers stop working. That answer could shape how quickly the next wave of data centres gets funded.
Also read: Infineon opens its €5 billion Dresden megafab on July 2 as Europe's power chip ambitions meet their real test • OpenAI offers equity to Washington to preempt a far costlier forced takeover • EngineAI takes its humanoid robot factory story to Hong Kong's public markets
