FSB Warns AI Private Credit Boom Risks Sizeable Investor Losses

More than a third of private credit deals signed in 2025 were AI bets. That share sat at 17% as recently as five years ago. The Financial Stability Board’s May 6 warning on private credit vulnerabilities put the figure on paper Wednesday, and the global watchdog said the rapid pivot leaves a roughly $1.5 trillion lending market exposed to a single industry’s electricity supply, valuation moves, and demand curve.

The FSB monitors central banks and regulators in 24 countries. Its verdict on Wednesday was blunt. A sharp correction in AI asset prices, the body said, could trigger “sizeable” credit losses for the funds that wrote those checks.

What started as cheap structured financing for hyperscalers and chip companies has, in five years, become the largest single concentration in non-bank corporate lending. And the regulators just told banks to look hard at how much of it sits on their own books.

AI Now Drives One In Three Private Credit Deals

The headline number in the FSB’s report is the deal-share jump. AI lending grew from 17% of private credit transactions over the prior five-year window to over 33% in 2025 alone. Healthcare and broader services round out the top three borrower sectors.

That kind of concentration is what regulators flag first. The FSB wrote that the focus on specific sectors “may leave private credit funds exposed to idiosyncratic risks” and increase exposure to “region or industry-specific shocks.” Translation: when AI sneezes, a third of the book catches it.

Private credit funds raise money from pension funds, insurers, and increasingly retail investors, then lend directly to companies outside the regulated banking perimeter. The borrowers tend to be lower-rated and more leveraged. The FSB’s 6 May 2026 report on vulnerabilities in private credit notes that documentation and reporting standards are weaker than in the syndicated loan market, which makes the next stress test harder to model.

The growth has been ferocious. The Bank for International Settlements pegged the global private credit pool at $1.5 trillion to $2 trillion in 2024 figures, with assets under management nearly doubling since 2020. The BIS bulletin on financing the AI boom from cash flows to debt documents how hyperscalers have shifted from internal cash funding to leveraged structures as compute demand outpaced retained earnings.

Why Power Grids Just Became A Credit Risk

Tucked inside the FSB report is a line that should reset how analysts price datacentre debt. Repayment, the watchdog warned, depends on electricity arriving on time.

Any “significant shortfall in the supply of electricity, a critical factor in the construction and operation of datacentres, could lead to delays or cancellations of projects,” the report said. That is a financial-stability body listing kilowatt hours next to interest coverage as a determinant of whether a loan gets paid back.

The numbers behind the warning sketch out the squeeze:

• 33%+ of 2025 private credit deals tied to AI infrastructure, per the FSB.
• 17% was the AI deal share averaged across 2020 to 2024.
• $1.5T to $2T total private credit market size in 2024 figures.
• 24 countries whose central banks and regulators feed the FSB’s surveillance data.

How A Datacentre Glut Could Crush Returns

Electricity is one tail risk. Demand collapse is the other. The FSB warned that AI company valuations could take a hit if the wave of datacentre construction outpaces actual workload demand, leaving operators with empty racks and lenders holding the bag.

That scenario has analyst support. The thesis: model-training demand is lumpy, inference demand has become commoditised, and the marginal datacentre going up in 2026 may not find a tenant willing to pay 2024-level rents. If revenue per rack falls 20%, debt service coverage on highly levered project finance falls faster.

Hyperscalers themselves are leaning harder on outside capital. Record venture funding into AI startups in early 2026 is one part of that story; private credit financing the picks-and-shovels layer is the other. Both depend on the same demand assumptions holding.

The FSB’s framing is unusually direct for a multilateral. The body normally uses softer prose. Listing electricity supply, datacentre construction timelines, and AI demand as three distinct triggers for a credit shock is a marker that staff economists modelled this scenario in detail.

Investors have started watching the same variables. Private credit fund quarterly letters reviewed in coverage of the report increasingly disclose the share of the portfolio tied to AI compute build-outs, and several large managers have introduced sector caps to limit the overhang.

Tricolor And First Brands Showed The Exit Door

The watchdog did not need to imagine what private credit losses look like. It pointed to two recent corporate failures already on the books.

Both First Brands and Tricolor failed due to company-specific frauds rather than broader macroeconomic challenges or poor lending practices indicative of systemic issues, the analysts at Cambridge Associates wrote in a November 2025 client note assessing the collapses.

That is the optimistic read. The pessimistic read, which the FSB embraces, is that the same opacity that allowed alleged double-pledging at Tricolor exists across the private credit book. Banks including JP Morgan and Barclays took losses on Tricolor warehouse facilities, while UBS and Jefferies disclosed material exposure to the First Brands collapse. The Cleary Gottlieb restructuring analysis of the Tricolor failure documents how the subprime auto lender allegedly submitted manipulated loan files for years before the bankruptcy filing in September 2025.

Banks Are Tangled In The Web

The FSB’s most consequential argument is about the regulated banking system. Banks are no longer at arm’s length from private credit. They are funding it, partnering on it, and lending to companies that also borrow from it.

The watchdog flagged three channels of exposure that fed into the Tricolor and First Brands losses:

The integration is what the FSB called “the intricate web of exposures in corporate credit.” That phrasing matters. A regulator using the word web is signaling that contagion paths are not always visible from a single bank’s risk dashboard.

US lawmakers are already pressing the same point. A November 2025 Senate Banking Committee letter to federal banking agencies on credit risk asked supervisors to detail their view of private credit exposure across the largest US banks following the two collapses.

Funds Cap Withdrawals As Confidence Wobbles

The pressure is showing up in fund flows. Several large managers, including KKR, Apollo, BlackRock and Blue Owl, capped retail investor redemptions in their semi-liquid private credit vehicles in recent weeks as withdrawal requests piled up.

Those caps are written into fund documents. They exist precisely so a manager does not have to fire-sale loans to meet redemption requests. Investors get the cash back over months, not days, and the fund avoids realizing losses on illiquid positions.

The optics are still bad. Cambridge Associates and other allocators have noted that retail investors arriving in private credit during the 2023 to 2025 boom may not have priced in the lockup terms they signed. The first widespread test of those terms is happening now.

Frequently Asked Questions

Is My Pension Fund Exposed To AI Datacentre Loans?

Possibly yes, depending on the plan’s allocation to private credit. Most large public pension funds have allocated 5% to 12% of assets to private credit since 2020. Check your plan’s annual allocation report, usually posted on the fund’s investor-relations page, for the share tagged “private credit” or “direct lending,” and ask the trustee whether sector concentration limits exist.

How Do I Tell If A Private Credit Fund Is AI-Heavy?

Read the latest quarterly letter. Most large managers, including the four named in the redemption-cap stories, now disclose sector breakdowns. Look for line items labelled “technology,” “infrastructure,” or “data infrastructure.” If those three combined exceed 25% of the book, you are sitting in an AI-tilted vehicle. Ask the manager directly for the AI-specific share if the disclosure is not granular.

Can I Pull My Money Out Of A Capped Fund Early?

No, not without paying penalties or accepting a discount on a secondary sale. Once a manager activates redemption gates, the contractual terms allow them to limit monthly outflows to a percentage of net asset value, typically 2% to 5%. The secondary market for fund interests exists but pricing in May 2026 has tightened, with discounts of 8% to 15% reported on AI-heavy vehicles.

Will The FSB Report Force New Rules?

Not directly. The FSB issues recommendations, not rules. Member jurisdictions, including the US Federal Reserve, the Bank of England, and the European Central Bank, decide whether to translate the warning into supervisory action. Watch the next round of bank stress tests in 2026 for whether private credit exposure becomes a separate scenario, which would be the clearest sign regulators agree with the FSB’s reading.

The Wednesday report is not a forecast that AI lending will blow up. It is a checklist of the variables that would have to break for it to. Electricity supply, valuation moves, demand curves, and bank counterparty links are now sitting on the same page in the same regulator file. Investors who treated private credit as the calm corner of corporate finance no longer have that luxury.

Disclaimer: This article reports on a Financial Stability Board publication and related market commentary. It is for informational purposes only and does not constitute investment advice. Private credit funds carry liquidity, credit, and concentration risks, and individual exposure varies by fund structure and jurisdiction. Readers considering allocations to or redemptions from private credit vehicles should consult a licensed financial adviser. Figures, deal shares, and fund manager actions cited are accurate as of publication on May 6, 2026, and are subject to change.