Marsh Wants the Fed to Count Insurance as Capital Relief on AI Data Center Loans

By Hector Herrera | November 13, 2026 | Finance

Insurance giant Marsh & McLennan is pressing US federal regulators to change how banks account for AI data center loans when those loans are covered by credit risk insurance — a shift that would let lenders hold less capital against one of the fastest-growing loan categories in the world. According to Bloomberg, at least three of the world's largest banks have already explored or purchased this type of credit insurance coverage. The problem: current US rules don't recognize it as capital relief, leaving banks carrying full regulatory capital against loans they've partially hedged.

This is a significant policy ask at a critical moment. As AI infrastructure lending scales toward the trillions, the gap between how banks manage risk in practice and how regulators account for it is becoming a material constraint on credit availability for data center construction.

The Capital Rules Problem

Under current US bank capital rules derived from the Basel III framework, banks must hold capital reserves — a buffer against potential loan losses — calibrated to the risk weight of their exposures. When banks buy credit risk insurance to hedge a loan, that insurance is supposed to reduce the effective risk weight, allowing the bank to hold less capital against that position.

The complication is that US regulators have not formally recognized insurance products as eligible credit risk mitigants for AI data center loans in the same way they recognize other forms of credit enhancement, such as guarantees or collateral. Under the current treatment, banks that pay to insure AI loans still carry the same capital burden as if the insurance didn't exist.

Marsh's argument to the Federal Reserve is straightforward: if a bank has transferred meaningful credit risk to an insurer, regulators should recognize that transfer. Forcing banks to hold capital as if unhedged creates a perverse incentive — it penalizes risk management rather than rewarding it.

Why AI Data Center Lending Is Different

The scale of AI infrastructure buildout has created a loan category with unusual risk characteristics:

• Concentration: A handful of hyperscalers — Microsoft, Amazon, Google, Meta, and Oracle — are the primary borrowers or anchor tenants in most AI data center transactions. This creates lender concentration risk even when individual loans look sound.
• Asset specificity: AI data centers are purpose-built for specific compute architectures. If a borrower defaults, the facility is harder to repurpose than a general commercial real estate asset.
• Demand uncertainty: AI compute demand projections are based on extrapolations from a relatively short adoption curve. If demand growth slows or shifts to new hardware architectures, the asset values underpinning these loans could deteriorate faster than traditional real estate.
• Speed: Banks are being asked to close large construction loans quickly to capture the AI buildout cycle. That compression reduces due diligence time.

Credit risk insurance addresses concentration and default risk. It doesn't eliminate asset specificity or demand uncertainty — those remain on the bank's book regardless. But for lenders whose concern is credit default rather than asset liquidation value, the insurance coverage is real risk transfer.

The Three-Bank Signal

Bloomberg's reporting that three major global banks have already explored or purchased credit risk insurance for AI loans — without disclosing which banks — is a significant data point. It means the product is real, the market is functioning, and lenders are already managing risk in ways that regulators haven't caught up to.

Marsh's lobbying push likely reflects frustration from bank clients who have bought coverage and are still carrying the full capital cost. If the Fed and other US regulators updated their treatment — aligning US rules with how some European regulators handle insured credit risk — it would retroactively validate what those three banks have already done and open the market more broadly.

What Changes If the Fed Agrees

If US regulators update capital treatment to recognize AI loan insurance:

• Banks can extend more AI credit without raising capital. This directly increases the supply of construction lending for data centers.
• Insurance pricing becomes more competitive. A larger addressable market with regulatory tailwind would attract more insurers, lowering premiums over time.
• The capital relief creates a new risk transfer market. Insurers absorbing AI credit risk would likely reinsure portions of their exposure, distributing risk more broadly across global capital markets — the same structure that developed in commercial real estate and project finance.

The downside scenario is that capital relief encourages overlending into AI infrastructure at a moment when the market may already be overbuilding in some regions. Regulators pushing back on Marsh's proposal are likely citing exactly that concern — that loosening capital requirements mid-cycle in a hot sector is precisely when such loosening is most dangerous.

What to Watch

Watch for the Federal Reserve's response — formally or through guidance updates — to Marsh's advocacy. The Basel III endgame rulemaking that US regulators have been finalizing through 2025-2026 is the most likely vehicle for any change in credit risk mitigation treatment. Banks with existing AI loan insurance coverage are watching closely: capital relief applied retroactively would materially improve their return on equity calculations for deals already on their books.