The $1.5 trillion AI credit bet is starting to look crowded

The world's largest publicly listed hedge fund is now openly warning the people who allocate to it: the AI private-credit complex has the shape of a bubble, and the unwind, when it comes, will not be polite. Man Group's chief investment officer for private credit told the Financial Times on 17 June 2026 that the $1.5 trillion market for loans funding data centres, chip designers and AI-adjacent infrastructure is vulnerable to a "violent" correction if revenue assumptions for the underlying companies fail to keep pace with the leverage stacked on top of them.

The warning lands at a fragile moment. AI has been the single biggest driver of credit origination in private markets for the better part of two years, pulling in pension funds, sovereign wealth funds and yield-hungry insurers who were told, repeatedly, that the technology's infrastructure build-out would generate utility-like cashflows for decades. The premise is not absurd. The pricing is another matter. Spreads on AI-linked private credit have compressed sharply even as the underlying borrowers' commercial trajectories remain largely unproven, and the structural protections that private credit traditionally offered — first-lien security, covenant packages, equity cushions — have been quietly relaxed to push volume out the door.

The trade, and who is on the other side

Man Group is not a disinterested observer. The London-listed manager runs roughly $200 billion and has been a major participant in the very market it is now flagging. That is the point. A manager with skin in the game warning that the game is dangerous is a more credible signal than a sell-side strategist calling for a top after the fact. It also means Man is positioning for the move it predicts — buying protection, tightening underwriting on new originations, and quietly telling large LPs that the entry point on AI credit in mid-2026 will not be available for long.

The buyers on the other side are easy to name. US pension funds — CalPERS, CalSTRS, the New York State Common Retirement Fund — have all increased private-credit allocations, partly because public-market fixed income no longer offers the yield their actuarial assumptions demand. Middle Eastern and Asian sovereign wealth funds have done the same, both for return and for the strategic relationship that comes with co-investing alongside US GPs. Insurance balance sheets, freed from some of the post-2008 capital constraints by lighter NAIC designations on private credit, have been the marginal buyer of the more senior, lower-yielding tranches. None of these counterparties are momentum traders. They are sticky, long-duration capital — which is precisely why a violent correction is the scenario being warned about: a sudden mark-to-market reset in a market with no mark-to-market.

What the lenders actually own

The composition of the $1.5 trillion matters more than the headline number. Roughly half sits in senior secured loans to hyper-scale cloud and AI-compute operators — names whose revenue is tied to a small group of foundation-model customers and whose capacity build-out is financed on the assumption that AI workloads continue to grow at 60-100% a year for at least the rest of the decade. The other half is a more heterogeneous pool: GPU-financing facilities, neocloud operators leasing capacity to the same handful of customers, fibre and power-infrastructure projects, and a fast-growing tail of "AI-adjacent" borrowers — software roll-ups, vertical-AI applications, sales-tooling startups — whose revenues are real but whose path to debt service is harder to underwrite without optimistic assumptions about customer retention.

It is the second bucket that worries people like Man. A senior secured loan to a hyperscaler, even one levered to AI capex, is collateralised by receivables from investment-grade counterparties and by assets with a deep secondary market. A unitranche loan to a 2024-vintage AI application startup, sized at eight times forward EBITDA, is collateralised by assumptions. The 2024 and 2025 vintages have not yet had a full credit cycle. The first real test of underwriting will be the 2027-28 refinancing wave, when the interest-only periods on a large cohort of AI-linked loans expire and amortisation begins.

The structural frame: leverage, liquidity and the second wave of the cycle

Private credit spent the last decade being sold as a structural improvement on leveraged finance: better covenant protection, closer lender-borrower relationships, faster execution. The pitch was plausible when the asset class was $500 billion and dominated by direct lenders underwriting mid-market industrial companies with twenty years of operating history. At $1.5 trillion, and with a meaningful share of new origination flowing to companies less than three years old, the asset class has begun to exhibit the same symptoms that pre-date every credit correction in modern memory — compressed spreads, looser documentation, and a growing reliance on payment-in-kind features that let borrowers pay interest in additional principal rather than cash. PIK toggles are a useful tool in a downturn. PIK as a baseline feature in a booming market is a sign that underwriters are anticipating the cash will not always be there.

There is also a second-order exposure that rarely shows up in the marketing decks. The same institutions that are buying AI private credit are also exposed to public-market equity in the AI complex. A 10-15% drawdown in the leading AI equities would not, on its own, trigger a private-credit correction. A 30% drawdown, accompanied by a freeze in the IPO market and a contraction in strategic acquisitions, would. AI startups are the marginal exit route for venture and growth equity; a closed exit window is a closed refinancing window for the credit that sits beneath them. The correlation between private AI credit and public AI equity is rising, which is the opposite of the diversification story that brought most of this capital in.

What a 'violent' correction would actually look like

Man Group's choice of word is deliberate. A correction in private credit is mechanically different from a correction in public credit. Public bonds can be marked down intraday, but the cash coupon continues, the issuer keeps operating, and the holder can usually wait for maturity. Private loans are typically held to maturity by design, but the NAV is marked at the discretion of the GP, often on a quarterly basis with significant lags. A 2024-vintage AI loan that the GP marked at par in March 2026 will, in a serious downturn, be marked at 60-70 cents. That is the moment a "violent" correction stops being an abstraction and becomes a redemption-queue event.

The mechanism is familiar from the 2022-23 open-end private-fund liquidity episode. Investors who signed up for private credit with quarterly liquidity and a 12-month notice period discover, in a downturn, that the notice period is a queue, not a right. Forced sellers — pensions rebalancing to meet liability-driven funding targets, insurers meeting capital requirements, fund-of-funds facing their own LP redemptions — sell what they can, which is the public-market sleeve of their AI exposure, accelerating the equity drawdown that is causing the private-credit problem in the first place. The circuit is not hypothetical. It is the same loop that ran in 2008 and 2009, with a different cast of characters and a different set of underlying assets.

What remains uncertain

The sources do not specify how much of the $1.5 trillion Man Group considers mispriced, nor the size of the mark-down it expects in a stress scenario. The firm has not publicly disclosed its own net positioning. The 14-point US-Iran deal reported on the same day by Reuters — which would, if implemented, free up regional capital flows and possibly redirect some Middle Eastern sovereign allocations — is one of several external variables that could either accelerate or dampen the rotation. So is the trajectory of long-end US Treasury yields, which determines whether public-market fixed income once again becomes a viable substitute for the private-credit allocation. Most importantly, the operating data on the underlying AI borrowers will only become legible over the next 18-24 months. Until then, every mark is a model output, and every model output in a market with no clearing price is, by definition, contestable.

The honest read of the warning is this: Man Group is telling its LPs that the next twelve months of returns on AI private credit will look attractive in a way that disguises the build-up of risk, and that the appropriate response is to reduce gross exposure, lengthen manager lock-ups where possible, and prepare for a liquidity event that the structure of the asset class is poorly equipped to absorb. That is not a call for a crash. It is a call for a market that has stopped being a market to be reminded, gently but firmly, what one looks like.

This publication reads the Man Group warning as a positioning signal from a participant, not as a forecast. The data points that would confirm or refute it — refinancing defaults in 2027-28, NAV marks in the next two quarterly cycles, public-equity drawdowns in the AI complex — are not yet in hand. The sources for this piece are the inputs the wire read; the conclusions are Monexus's own.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://x.com/unusual_whales/status/2065259029706752000
• https://x.com/reuters/status/2067295044537548800