• Marsh (NYSE:MRSH) is lobbying US regulators to reduce capital requirements on bank loans protected by credit risk insurance, as AI data-center lending surges.
  • Barclays (NYSE:BCS) is seeking insurance on over $10 billion in data-center debt, while HSBC (NYSE:HSBC) is exploring similar deals.
  • Global data-center spending is projected to reach $6.7 trillion by 2030, and a Fed decision could come by year-end.

The Regulatory Push

Marsh, the global insurance brokerage and risk advisory firm, is urging US regulators to reduce capital requirements on bank loans backed by credit risk insurance, according to Federal Reserve records. The push comes as lenders scramble to manage their exposure to a surge in AI data-center financing.

At the heart of Marsh’s proposal is a seemingly technical fix: allowing investment-grade insurance subsidiaries to qualify as eligible guarantors even if their parent, rather than the subsidiary itself, issues the investment-grade debt. Guy Carpenter, Marsh’s reinsurance arm, met with Fed staff on June 30 to discuss the issue, arguing that current ambiguity obstructs US banks’ access to insurance-based credit-risk transfer.

This is narrower than an across-the-board capital reduction for data-center loans. Whether protection produces capital relief would still depend on the guarantor, contract, applicable rules and supervisory treatment. The proposal retains restrictions addressing correlated—or “wrong-way”—risk and predominantly credit-protection-focused monoline insurers.

The practical problem is corporate structure: an insurance operating subsidiary may be investment grade but does not issue its own unsecured debt; its parent does. Marsh’s proposal would explicitly allow an investment-grade subsidiary to qualify through an appropriate parent’s investment-grade debt issuance. It argues that clarification would improve US competitiveness, particularly for smaller banks without large capital-markets operations. That is an industry advocacy position, not a regulatory finding.

Banks Line Up

The lobbying effort comes as major banks are actively seeking insurance on their AI-infrastructure loans. Barclays has purchased insurance on AI-infrastructure loans after earlier reporting described exploratory discussions covering more than $10 billion of such debt, potentially transferring more than half the credit risk on some facilities. That does not establish that the entire $10 billion-plus portfolio has been insured.

HSBC is exploring insurance on project-finance loans, including data centers, characterizing the initiative as normal portfolio management and its data-center exposure as small. Separately, it has worked on a significant risk transfer involving more than $1 billion of data-center-linked loans. Public reporting does not confirm a completed insurance transaction.

JPMorgan (NYSE:JPM) considered credit insurance but decided against using it, according to November 11 reporting. Some insurers were already near or at their data-center sector limits in a May 2026 Willis (NASDAQ:WTW) survey.

The stakes are high: global data-center spending is projected to reach $6.7 trillion by 2030, according to McKinsey, a cumulative investment projection that is fueling demand for risk-transfer solutions. A Fed decision could come by year-end, according to the supplied headline, though the official meeting record does not promise that timetable or an approval.

A Question of Capacity

While banks’ efforts to insure AI-infrastructure loans have advanced from exploratory discussions toward actual use, insurer capacity is already becoming a constraint. The project-finance credit-insurance market has expanded from a handful of providers a decade ago to more than 70, yet current demand is testing that broader capacity.

Premiums can absorb 50%–70% of the bank’s earnings on the insured portion of a loan—not 50%–70% of the loan principal. Banks may accept that expense to preserve client relationships and obtain room for additional lending.

The central issue is whether insurance genuinely diversifies lending risk—or mainly moves growing AI exposure from banks to insurers. For Marsh, clearer US capital recognition would plausibly expand the addressable market for arranging these transactions. That is a business implication—not announced revenue guidance—and the benefit cannot yet be reliably quantified.

Marsh, formerly Marsh & McLennan, reported approximately $27 billion in annual revenue and more than 95,000 employees across 130 countries. For the third quarter ended September 30, 2026, revenue rose 8% to about $6.9 billion, with adjusted earnings per share up 10% to $2.03. The company has not separately quantified the contribution from AI-loan insurance in reviewed disclosures.

A spokesperson for Marsh declined to comment beyond the Fed meeting record. Barclays and HSBC did not respond to requests for comment by press time.

The Broader Debate

Internationally, this is primarily a regulatory-competitiveness and cross-border risk-allocation issue: different rules influence which banks can lend economically and which insurance markets absorb the exposure. European banks can obtain regulatory-capital benefits from qualifying insurance arrangements, while US banks generally lack equivalent treatment for the arrangements discussed in current reporting.

Two benefits must therefore be distinguished: internal exposure relief, where insurance may reduce a bank’s retained borrower or sector exposure and help it stay within internal lending limits; and regulatory-capital relief, where supervisors recognize qualifying protection when calculating required capital. The first does not automatically deliver the second.

The relevant US rulemaking involves the Federal Reserve, FDIC and OCC. While the headline suggests a possible year-end Fed decision, the official meeting record does not promise that timetable or an approval.

The economic stakes extend beyond technology. The IMF says investment is becoming more capital-intensive and reliant on external financing. The Bank of England warns that long-term debt may finance facilities whose economic usefulness deteriorates faster than expected as chip and cooling requirements evolve, reducing collateral values and cash available for debt service. Data-center borrowers also face many shared risks—power costs, financing conditions, regulation and AI demand—so losses may arrive together rather than as isolated defaults.

For now, all eyes are on the Fed. Without clearer capital treatment, banks may continue to seek insurance for internal risk management, but the cost of capital relief could remain a hurdle.