Private Credit: The Two-Trillion-Dollar Portfolio No One Can Price

Opening — two true statements about the same fund

In the second quarter of 2026, shareholders of Blue Owl Technology Income Corp.¹ asked to take out 38.1% of the fund's shares. They were paid 5% of net asset value — the fund's own estimate of what its loans are worth, divided across the shares outstanding — pro rata, which worked out to roughly thirteen cents of every dollar requested. In the same letter, the manager reported that loans on non-accrual — the standard measure of a lender's bad debts — stood at 0.2% of fair value, with no new non-accruals in the prior quarter. Sixty-four per cent of the portfolio was lent to software companies, the sector then being repriced hardest by artificial intelligence. Both statements were true. They describe two different worlds.

The durable profit in private credit does not sit in the loans; it sits in who cannot leave and in who sets the number the outside world sees — and the two are the same question, because permanent capital is only permanent for as long as its owner believes the mark.

These loans do not trade, so nobody quotes a price for them. Each quarter the manager that made the loan produces a valuation instead — the mark — and that number becomes the fund's reported net asset value, the base its fees are charged on, and the price at which an investor is allowed to leave. Nothing in that arrangement is improper. It does mean the number the outside world sees is produced by the party whose fees depend on it.

The consensus reading of this industry has two halves and both are incomplete. The bullish half says the listed managers have built annuity businesses: fee streams on locked capital, growing at twenty per cent, deserving a multiple somewhere between an asset manager and a toll road. The bearish half says two trillion dollars of unmarked corporate loans, financed partly by banks and partly by retail money that can ask for its cash back, is a run waiting for a trigger. The bullish half is right about roughly half the capital and wrong about the rest, because the word permanent is doing four different jobs across the eight managers covered here and only one of them is contractual. The bearish half has the mechanism upside down: the fragility does not run from defaults to redemptions, it runs from disbelief in the marks to redemptions to forced sales to defaults. In 2026 the second sequence began without the first.

This machine comes apart into layers — origination, fund management, funding, leverage, valuation, workout — and at each one the question is the same: where does the money actually stay after competitors, a bad vintage, a regulator and the banks have all had their turn? Growth is abundant here. Rent is not. Rent, in the sense used throughout this study, is profit that survives competition — the money a business keeps because a rival cannot simply arrive with capital and take it. The rent has concentrated in two places: insurance liabilities that cannot be redeemed by the people whose money it is, and origination breadth wide enough to move when direct lending stops paying. Everything else — spread, scale in direct lending, the wealth channel, the fee rate itself — is being competed away in front of us, and the 2026 stress is accelerating it.

Evidence cutoff: 30 August 2026. Company financials are as of the quarter ended 30 June 2026 unless stated. This is an assessment of industry economics and competitive position. It is not investment advice, a recommendation, or a solicitation.

1. 1Reported credit quality and investor behaviour have decoupled, and the investors are ahead. Among the six largest non-traded BDCs — business development companies, funds that lend to mid-sized firms and, in the non-traded version, are sold to individuals rather than listed on an exchange — most disclose non-accruals in the 0.2%–0.4% of fair value band, fair value being what the fund itself says the loans are worth today, but BCRED is 1.1%; quarterly redemption requests at those same vehicles ran 7.9% to 40.4% of shares. The marks, not the loans, are the fault line. High confidence
2. 2Apollo owns the only genuinely contractual permanence at scale. $537bn of its $621bn perpetual capital (the industry's term for money that carries no fixed end date and no scheduled return to its owner) — 87% — is annuity liabilities at Athene and Athora, money whose owners hold a policy, not a redemption right. Blue Owl's $225bn of "permanent capital" includes the BDCs that just capped withdrawals. High confidence
3. 3The retail investor's share of the illiquidity premium is roughly half what the marketing shows. The illiquidity premium is the extra return an investor is supposed to earn in exchange for holding something they cannot sell. On Blue Owl's own index comparison, its flagship Class I beat leveraged loans by 325bp — basis points, hundredths of a percentage point — since inception; the load-bearing Class S share the retail client actually buys beat them by 161bp. High confidence
4. 4Incentive income has already broken; fee-related earnings have not. Incentive income is the manager's cut of investment profits, paid only in years the returns clear a threshold; fee-related earnings are the flat annual charge on the assets themselves, paid whatever happens. Blackstone Secured Lending's income-based incentive fee fell 95% year over year, to $1.7m from $34.7m; Blue Owl's GAAP performance revenues fell 90%. Management fees at both rose. That divergence is the whole case for valuing the two streams differently. High confidence
5. 5The banking system never left the business; it moved up the capital structure. Asset managers and funds are now Bank of America's largest single industry concentration at $244.9bn committed, JPMorgan's third at $176.8bn (up 15.7% in six months, equal to 58% of CET1, the core equity capital a bank is required to hold against its risks), and inside Wells Fargo's largest C&I category. The next cycle's bank losses arrive dressed as facility and counterparty exposure. Moderate confidence
6. 6Default statistics are the wrong dashboard and will stay wrong for a year. Lincoln's covenant default rate fell to 3.1% in Q1 2026 while lender takeovers ran $39.4bn across 2025–26 against $13.6bn in the prior three years combined. Income quality moves first; defaults confirm afterwards. High confidence
7. 7Origination, not capital, is the binding constraint. Origination is the work of finding borrowers and writing the loans — the supply of deals, as distinct from the supply of money. Gross spreads compressed roughly 75–100bp into 2025 while Ares alone accumulated a record $170bn of dry powder, money committed by investors and not yet lent; they widened back to 554bp in June 2026 only because the retail channel stopped supplying capital. The second act — asset-based finance and AI infrastructure — is where the surviving rent goes. Moderate confidence

1. The loan — what a borrower actually signs

A private credit loan in its dominant form is a unitranche: a single senior secured facility that collapses what used to be a first-lienThe claim that gets paid first out of a borrower's assets if things go wrong. A first-lien lender stands at the front of the queue, a second-lien lender behind them, and the owners' equity last of all. term loan and a second-lien or mezzanineDebt that sits between the senior loans and the equity — repaid after the senior lenders, before the owners, and priced higher to compensate for the position. piece into one blended instrument, priced at one blended spread, held by one lender or a small club. The borrower is usually a company owned by a private equity sponsorThe private equity firm that owns the borrower, having bought it with a mix of its investors' equity and borrowed money. In this market the sponsor, not the company, is the lender's real counterparty: it runs the financing auction, it negotiates the amendments, and it decides whether to put fresh equity in when the loan gets into trouble., with somewhere between $25m and $250m of earnings. The loan floats over a reference rate, carries an original issue discountA loan advanced at less than its face value — lend $99, be repaid $100 — so the lender collects an extra slice of yield on top of the interest coupon. Quoted in points of the loan amount., and runs five to seven years. The spread is the margin the lender charges over that floating reference rate, and it is the price of the entire business: hold it and the lender earns, compete it away and there is very little else in the loan to earn from.

The pricing is knowable and it has moved. StepStone's benchmarks, drawn from Kroll data as of November 2025², put US contractual spreads in a band of 450–550bp with original issue discounts of 1.0–1.5%, both roughly 50bp tighter than eighteen months earlier; gross spreads — contractual spread plus one-third of the OID — compressed by about 75–100bp from a 600–650bp starting point. Lincoln International's Q1 2026 valuation data³, covering roughly 7,100 US portfolio companies and more than $250bn of principal, gives the same picture from the valuation side: spreads of SOFRThe Secured Overnight Financing Rate — the benchmark US short-term interest rate that replaced Libor. A loan priced at SOFR+5.00% pays five percentage points over whatever that benchmark happens to be, so the lender's income rises and falls with policy rates instead of being fixed.+4.75% to +5.50% on deals with $40–100m of adjusted EBITDAEarnings before interest, tax, depreciation and amortisation — a rough proxy for the cash a business generates before financing and accounting choices. It is the denominator for leverage in this market: a loan at "5x EBITDA" is five dollars of debt for every dollar of annual earnings. Adjusted means costs the borrower argues are one-off have been added back, which is why the adjusted figure is the flattering one., closing leverage of 4.75x to 6.25x, average enterprise value of 11.7x EBITDA against average debt of 5.1x. Apollo's own retail vehicle discloses the number at portfolio level: a weighted average spread over the reference rate of 4.8%⁴ across $25.4bn of floating-rate investments at 30 June 2026, up from 4.6% six months earlier.

Three provisions of the document matter more than the coupon, and they run in the borrower's favour. CovenantsPromises written into the loan document — keep leverage under a stated level, keep interest cover above one, report the numbers monthly — that let the lender step in early when a business deteriorates. A loan carrying few of them is covenant-lite, and the lender's first formal trigger is then a missed payment, which arrives much later. have thinned; payment-in-kind toggles — the borrower's option to stop paying cash interest and add it to principal instead — have spread; and portability provisions increasingly let a loan survive a change of ownership. StepStone records that borrower-friendly terms persisted through the second half of 2025 precisely because capital supply exceeded loan demand, with the leniency concentrated among large-cap and upper-middle-market borrowers. That is the competitive dynamic showing up in ink.

A broadly syndicated loanA large corporate loan arranged by a bank and then sold in pieces to dozens or hundreds of investors, who trade it among themselves afterwards. Because it has many holders and a live secondary market it carries a public rating and a price that moves daily — the two things a private loan does not have. is cheaper, marked daily, and requires the borrower to satisfy a ratings process and a syndicate. A unitranche costs more, closes in weeks rather than months, can be sized to a plan the ratings agencies would not credit, and — this is the part the borrower pays for — can be renegotiated in a room with three people in it. The premium is a price for speed, certainty and future flexibility. Whether that premium survives after fees and losses is a different question, and the answer sits in Section 5.

2. How big — three official measurements that disagree, and why

Nobody knows the size of this market. The disagreement is definitional, and there is real money on both sides of it.

The Financial Stability Board's May 2026 report⁵ puts global private credit at $1.5tn to $2tn as of end-2024, using the narrow definition — non-bank direct lending to medium-sized companies negotiated bilaterally — with the United States at roughly $1tn of it. The Federal Reserve's May 2026 Financial Stability Report⁶ counts $1.4tn of US private credit loans in the second half of 2025, equal to about 10% of total US non-financial corporate debt and roughly one-third of below-investment-grade debt excluding bank loans. The Congressional Research Service⁷, writing in March 2026, calls it a "nearly $2 trillion U.S. market." Widen the definition to include asset-based finance, specialty lending, real estate credit and infrastructure debt and the number reaches into the high single-digit trillions; the FSB's own Graph 1.1 shows industry estimates spanning an order of magnitude depending on which activities are counted.

The FSB is blunt about why this matters: without a harmonised definition, "the scope and size of private credit can vary significantly." For an analyst, the practical rule is to fix on the narrow number when discussing corporate credit risk and to name the wider one only when discussing where growth is going. Both are used below, always labelled.

Lending to the funds and BDCs, at the Office of Financial Research's estimate of $410–540bn, equals 29% to 39% of the Fed's $1.4tn of private credit loans outstanding. And semi-liquid vehicles, at $241bn of net assets, are roughly 20% of the net assets under management in private credit vehicles — the Fed's own figure. Neither number is large. Both sit at the pinch points.

3. Why the lane opened — regulation, the sponsor, and the price of speed

The standard story is that Basel IIIThe post-2008 international bank capital rules, phased in over the following decade. They oblige a bank to hold more shareholders' equity against riskier loans, which made keeping a leveraged middle-market loan on the balance sheet expensive relative to what it earned. and the leveraged lending guidance pushed banks out of middle-market lending and non-banks filled the gap. The story is true and incomplete, and the incompleteness matters for how durable the resulting economics are.

The FSB's account lists four drivers: post-crisis bank capital rules, borrower demand for tailored financing and fast execution, a prolonged low-rate environment that pushed investors toward yield, and the expansion of private equity itself. Two of the four — the rate environment and the regulatory perimeter — are conditions. Conditions reverse. The other two — the sponsor's demand for certainty, and the lender's ability to underwrite a bespoke structure quickly — are closer to a service proposition, and a service proposition is defensible only for as long as competitors cannot copy it.

There is also a less flattering driver, and the official sector has now put it in writing. Research cited by the FSB finds that a bank's return on lending to a BDC can exceed its return on lending directly to the same size of middle-market company, "largely driven by lower risk weights, historically low default rates, and reduced operating expenses." The bank did not withdraw from mid-market credit risk on prudential grounds and leave it to someone braver. It re-entered the same risk one rung higher, where the capital chargeThe amount of its own shareholders' equity a bank must set aside against a given exposure, as determined by the regulator's risk weighting. A lower charge ties up less equity, so two loans paying the same interest rate can be very different businesses depending on how each is weighted. is lower and the monitoring cost is somebody else's. The migration of the loan out of the bank was partly an accounting migration.

What the borrower gets is real. Speed is the product. A sponsor bidding in a competitive auction needs financing certainty before it signs, and a syndication process introduces market risk between signing and funding. The unitranche removes that risk, and the sponsor pays a spread premium for it. But the premium exists because underwriting capacity was scarce. As Section 20 shows, it is no longer scarce.

By the cutoff, the only DC pilot this study could verify was Great Gray Trust's Panorix target-date series¹², launched with BlackRock's custom glidepath¹⁰ and Wilshire liquidity management. That is movement, but it is pilot-scale rather than the broad recordkeeper adoption implied by the market narrative.

4. The vehicle layer — six wrappers, six different promises

The same loan can sit inside six structures, and the structure, not the loan, determines what happens in a stress.

A drawdown fund takes commitments from institutions, calls capital over three to five years, and returns it over seven to ten. There is no redemption right. The FSB records that most US private credit funds retain this closed-end form, and that in the euro area only around 20% of private credit funds are open-ended. Three of the six wrappers are business development companies, a US fund structure built to channel investment into smaller companies. A BDC pools investors' money, lends it out, is capped by statute in how much it may borrow, and distributes most of its income to shareholders. What separates one BDC from the next is not the loans inside it but who is allowed to own it and how they are allowed to get out. A private BDC is a business development company sold to accreditedInvestors who meet an income or net-worth test and are therefore permitted to buy products that are not registered for sale to the general public. The legal theory is that they can absorb the loss and evaluate the risk without the protections a retail offering carries. and institutional investors, typically with limited or no repurchase facility. A listed BDC trades on an exchange: an investor who wants out sells to another investor, and the fund's assets are untouched. That is a crucial distinction and it is where the industry's most honest information comes from.

Then come the two structures that carry the mismatch. A perpetual non-traded BDC raises capital continuously at monthly NAV and offers to repurchase up to 5% of NAV per quarter, at the board's discretion. An interval fund is a registered closed-end fund that must accept at least 5% of repurchase requests at stated intervals. That is the liquidity mismatch in one line: the investor can ask for cash on a quarterly timetable, while the assets standing behind the promise are five-to-seven-year loans nobody trades. The arrangement holds for as long as more money is arriving than leaving. The Fed sizes them precisely: perpetual-life BDCs held $306bn of gross assets and $161bn of net assets; interval funds $119bn gross and $80bn net; together $425bn gross, $241bn net, about 20% of net assets in private credit vehicles. Finally, the insurance general account and the separately managed accountA portfolio a manager runs for one client under a bilateral contract, with the assets owned directly by that client rather than pooled with anyone else's. There is no fund to redeem from and no other investor to be diluted by — the client simply terminates the mandate, and typically pays a lower fee for the privilege of scale. hold the loan against a liability rather than against an investor's option.

A seventh wrapper is where the insurance channel meets the loan. Private credit loans are pooled and securitised into private credit CLOsCollateralised loan obligations. A pool of loans is dropped into a company that issues its own bonds in ranked slices: the senior slice is paid first and rated highest, the equity slice is paid last and absorbs the first losses. The point of the exercise is to turn a bundle of sub-investment-grade loans into a large tranche of highly rated paper that an insurer or a bank can hold against a small capital charge. — the FSB puts the market at $155bn outstanding as of October 2025, about 16% of the $977bn US CLO market. These structures carry a wrinkle that cuts both ways: rating agencies assign lighter-touch "credit estimates" rather than full ratings to the underlying loans, which is a transparency problem; but issuers typically retain the entire equity tranche, which is thicker than in a broadly syndicated CLO, at roughly 30–40% in the bank-financed variant. The FSB's own conclusion is that private credit CLOs have greater loss-absorbing capacity for senior investors than traditional CLOs. That is a genuine rebuttal to the lazy securitisation analogy and it should be conceded.

5. The fee stack — what each wrapper costs the person whose money it is

Most of the argument about this asset class is settled by arithmetic — the gap between what the assets earn and what the investor keeps.

Blackstone publishes the cleanest disclosure of the gap. In its second-quarter 2026 results¹³, the Private Credit strategy returned 6.8% gross over the trailing twelve months and 3.9% net. The difference is 2.9 percentage points — 43% of the gross return — consumed by fees, expenses and financing before the investor sees anything. That is the industry's cost structure, disclosed by the industry's largest participant.

The retail wrapper adds a second layer on top of that. Blue Owl's July 2026 shareholder update for OCIC discloses returns by share class. Class I, the no-load institutional share, has compounded at 9.23% since inception. Class S, the share an adviser sells with a maximum 3.5% upfront loadA sales commission taken off the top before the money is put to work — a 3.5% load means $96.50 of every $100 reaches the fund. It pays the adviser who sold the product, not the manager who runs it, and the investor never sees it as a line item afterwards. and an ongoing shareholder servicing fee, has compounded at 7.59%. The difference — 164 basis points a year — is the retail distribution cost, and it is borne entirely by the retail investor. Over five years the gap is 171bp.

Blue Owl states that OCIC Class I has outperformed leveraged loans by 325bp since inception. Take the manager's own comparison at face value and it implies a leveraged loan return of about 5.98% over the same period. Class S, the share the retail client holds, therefore beat leveraged loans by roughly 161 basis points, not 325. The illiquidity premium the product is sold on is, for its actual buyer, about half the advertised figure — before any adjustment for the fact that the leveraged loan index is marked daily and this one is not.

What the investor keeps depends on two variables and nothing else much: the gross yield on the assets, and the realised credit loss.

6. The funding stack — whose money it is, and what each owner can do about it

Five owners sit behind the loan, each with a different liability structure and therefore a different behaviour in a drawdown.

Pensions and sovereigns commit capital they cannot recall. The Office of Financial Research estimates uncalled capital commitments to private credit funds at roughly $300bn, with pension funds accounting for about $100bn¹⁴. This is the most stable money in the system, and it carries a contingent obligation running the other way: in a protracted downturn, capital callsA fund does not take an investor's cash up front. It takes a commitment, then calls the money down in instalments as it finds loans to make — on short notice, and not at the investor's convenience. arrive at exactly the moment the pension least wants to sell liquid assets to fund them.

Insurance general accountsThe pool of assets a life insurer holds against the promises it has made to its policyholders. The insurer keeps whatever those assets earn above the rate it credits the policies, so the general account is at once an investment portfolio and the collateral behind the contracts — which is why what sits inside it is a regulator's business as much as a shareholder's., affiliated and third-party, are the second pillar and the more important one. The FSB's proxy analysis suggests around 10% of North American life insurer portfolios may be private credit against roughly 3% for non-life; the NAIC counts $276.8bn of CLOs held by US insurers at year-end 2024, of which life insurers hold 82%¹⁵. Private-equity-linked insurers hold structured securities at about 27% of their portfolios against 12% for other large insurers. And the scale of the ownership change is stark: PE-backed insurers control close to $900bn of US insurance liabilities, up from $67bn in 2012, with 35% of new US annuity sales in 2023 going to them.

Offshore reinsurance is the plumbing beneath that. Asset-intensive and funded reinsuranceReinsurance is one insurer taking on another's obligations. In the asset-intensive or funded variety the object is not to shed death or longevity risk but investment risk: the annuity promises move across to the reinsurer along with the assets backing them, and from that point the reinsurer's capital regime — frequently another country's — is the one that governs them. moves annuity liabilities and their backing assets to Bermuda and Cayman entities affiliated with the manager. The NAIC has responded with Actuarial Guideline 55, requiring additional disclosure and testing of asset-intensive annuity reinsurance under moderately adverse conditions, alongside AG 53¹⁶, which since the 2022 annual statement has required life insurers to document and sensitivity-test the complex assets — structured securities, CLOs, and assets "originated by the company or affiliated or contracted entity" — standing behind those liabilities. This channel is no longer un-policed, and 2026 delivered the first real bill: Apollo recorded a one-time tax expense of $1.7bn in the first quarter of 2026¹⁷ on the revocation of ACRA's election under Bermuda's Corporate Income Tax Act. A structure built partly on tax geography just paid a large, one-off price for that geography.

The wealth channel is the quarterly-option money. The retirement system is the channel that has not yet opened. On 31 March 2026 the Department of Labor published a proposed rule, Fiduciary Duties in Selecting Designated Investment Alternatives¹⁸, at 91 Fed. Reg. 16,088, implementing Executive Order 14330¹⁹. It creates a six-factor process safe harbourA defined procedure which, if followed, shields the decision-maker from liability for how the decision turns out. It does not certify that the investment was a good one; it says the fiduciary cannot be sued merely for having selected it. — performance, fees, liquidity, valuation, benchmarking and complexity — for fiduciaries selecting investments with alternative-asset exposure, with private credit explicitly named among the covered classes and the comment period closing 1 June 2026. It sets no allocation cap, and points toward alternatives offered inside diversified vehicles such as target-date fundsThe default option in most US workplace retirement plans: one fund labelled with a retirement year, which shifts automatically from equities toward bonds as that date approaches. The participant chooses the year, not the contents — which is precisely why the industry wants its product inside one. rather than as standalone options.

Around that last channel every interest points the same way, and away from the beneficiary. The manager gets permanent, fee-bearing, price-insensitive flow. The recordkeeper gets a differentiated product. The plan sponsor gets a litigation safe harbour it did not previously have. The target-date manager gets a higher-fee sleeve. The participant gets an asset whose valuation is produced quarterly by the party earning the fee, inside a daily-valued account, with a fiduciary standard that now expressly permits it. Nobody in that chain is acting improperly. Nobody in that chain is the natural advocate for the participant's liquidity either.

7. The insurance channel — the only genuinely permanent capital, and what it costs

Athene is the template. Almost every strategic move in this industry over the past decade has been an attempt to reproduce it.

The mechanism, at the scale Athene's own Form 10-K²¹ discloses: sell fixed and indexed annuities — $28.1bn and $16.7bn of 2025 premiums and deposits respectively — and take in pension risk transferA company with a defined-benefit pension scheme hands the entire obligation to an insurer and pays a premium for the insurer to write the cheques from then on. It arrives as one very large, very long-dated liability that no individual has any power to call back. business; hold the resulting liabilities, which have contractual durations of years and surrender chargesA penalty the policyholder pays for cashing an annuity in early, usually tapering away over the first several years. It is what converts a legal right to exit into an economically unattractive one — and it is the machinery behind the word "permanent" in this business. that penalise early exit; invest the general account in assets originated by the affiliated manager, earning a spread between the credited rate and the asset yield; and charge the manager's fee on the same assets. The manager earns twice — once on the spread, once on the fee — and the capital never leaves, because the person whose money it is holds a policy, not a fund share.

Apollo reported total AUM of $1,047bn at 30 June 2026, of which $621bn — 59% — is perpetual capital¹⁷, and it discloses the composition: Athene $416bn, Athora $121bn, other perpetual vehicles $84bn. Seventy per cent of fee-generating AUM is perpetual. Spread-related earningsWhat the insurer keeps between the yield it earns on its investments and the rate it credits to policyholders. This is balance-sheet income rather than a fee: the manager is paid for carrying the investment risk itself, so the stream moves with credit losses and interest rates rather than with assets under management. were a record $877m in the quarter, against fee-related earnings of $785m. Retirement Services took in $22bn of the quarter's $60bn of organic inflows.

KKR runs the same template one size down. Its second-quarter results²² show perpetual capital of $334bn against $796bn of AUM — 42% of AUM and 50% of fee-paying AUM — with Insurance Operating Earnings from Global Atlantic of $288m in the quarter. Carlyle, Ares, TPG and Blue Owl have insurance relationships but nothing on this scale: Carlyle's perpetual fee-earning AUMAssets under management the firm is actually charging a fee on. Headline AUM also counts money committed but not yet drawn, and assets in vehicles past their fee-paying life, so the two figures can differ by hundreds of billions at the same firm. is $113bn, 34% of fee-earning AUM²³; Ares' perpetual capital is $223.3bn against $671.3bn of AUM, 33%²⁴.

The growth story tends to skip the cost side. This structure has rebuilt maturity transformationFunding long-dated, hard-to-sell assets with money that can be withdrawn sooner. It is the core function of a bank — and the reason banks are given deposit insurance, a central bank and a supervisor. Performed elsewhere, it reproduces the usefulness and, on occasion, the fragility. outside the banking perimeter: long-dated, illiquid, largely privately-rated assets funded by liabilities that are contractually sticky but not infinitely so, supervised by state insurance regulators using a solvency framework designed for bond portfolios. The NAIC is explicit that it is retrofitting: a principles-based bond definition effective 1 January 2025 requiring classification on economic substance rather than legal form; a 45% risk-based capitalThe insurance regulator's counterpart to a bank capital requirement: the equity an insurer must hold against each asset, scaled to how risky that asset is judged to be. Raise the charge and the insurer must put more of its own money behind the same holding, which lowers the return it earns on it. charge on CLO residual tranches; annual CLO stress testing; and from 2026 reporting, granular disclosure of private placements including fair value, Level 2 and Level 3The accounting hierarchy describing how a valuation was arrived at. Level 1 is a quoted market price. Level 2 is derived from observable market inputs. Level 3 is a model, run on inputs the firm supplies itself — where a privately originated loan with no market price necessarily sits. exposure, payment-in-kind interest and private letter ratingsA credit rating commissioned by the issuer or the manager and shown only to the parties who need it — the insurer, its regulator — rather than published. It does the same regulatory work as a public rating, without the public record or the market's scrutiny.. The NAIC also states it is "developing a due diligence framework for credit rating providers" and has adopted a process to challenge ratings used for regulatory purposes. Each of those is a small tightening. Together they are the beginning of a repricing of the capital arbitrage that made the channel attractive.

8. Leverage on leverage — four layers, and who absorbs the first loss at each

Ask most people where the leverage is in private credit and they will point at the fund. That is the least of it.

The FSB names the layering directly: leverage exists "at the level of the portfolio company, at the private credit fund, at the sponsor level, and investor financing," and warns that this layering "may amplify losses during market stress." The borrower carries 4.75x to 6.25x of debt to EBITDA at origination, and the FSB notes that with EBITDA add-backsCosts a borrower asks its lenders to disregard when calculating earnings — restructuring charges, items described as one-off, and savings projected from actions not yet taken. Each add-back raises reported earnings and so lowers the reported leverage multiple, without changing the debt by a dollar. the true figure may be closer to 7x. Above that sits fund-level leverage — modest on average, and the official data is reassuring on this point: the OFR finds a median leverage ratio of about 1.0x across private credit funds in Form PFThe confidential return US private funds file with the SEC on their size, leverage and exposures. It is not public — which is why regulators can see a distribution of fund leverage that the funds' own investors cannot., effectively none, though the 95th percentile exceeds 3.5x and that right tail alone accounts for $81bn of borrowing, roughly 23% of estimated private credit fund borrowing. BDCs run higher, around 0.7x to 1.2x debt to equity, and are capped by statute. Above that sit investors who themselves borrow to fund commitments.

Fund-level leverage being low on average is a genuine comfort about the fund layer and no comfort at all about the system, because the leverage that matters in a run is the leverage on the collateral, and that sits inside the portfolio company at six or seven turns. A fund with no debt holding loans to companies at 7x is a levered position by any economic reading.

9. The banks' senior position — the number that keeps growing while everyone argues about something else

Three official measurements of bank exposure to private credit exist and they differ by a factor of five. The spread is itself the finding.

The Federal Reserve's own staff, writing in May 2025²⁵, counted $95bn of committed credit lines to BDCs and private debt funds at end-2024, $56bn utilised, roughly 7% of banks' regulatory capital on average, with about 60% concentrated among five US global systemically important banksThe handful of banks whose failure regulators judge would damage the world financial system — G-SIBs. They carry extra capital requirements and closer supervision in exchange for the scale and cheap funding that put them on the list in the first place.. Their conclusion was that a full drawdown would reduce the CET1 ratio by two basis points and that "the financial stability implications seem limited." The OFR, with access to both Form PF and the Y-14 confidential collections, published a much larger number in March 2026¹⁴: total bank and non-bank lending to private credit entities of $410bn to $540bn, plus a further ~$300bn of uncalled capital commitments. The FSB, surveying members, reports $220bn of drawn and undrawn credit lines but concedes commercial estimates "could be more than twice as large," and cites separate work placing the largest banks' drawn loans at ~$300bn with $155bn of undrawn commitments and $91bn from US branches of European banks.

The OFR explains why its own Y-14 figure ($123bn committed, $30bn of it to BDCs, against over $1.6tn of Tier 1 capital at the same banks) understates: BDCs borrow through special purpose vehiclesA subsidiary company set up to hold one pool of assets and one set of borrowings, legally walled off from its parent. Useful for ring-fencing collateral — and also the reason an exposure can be real, disclosed and still invisible to anyone counting by borrower name., one SPV per bank lender, and the standard obligor-identification method omits them. Regulators cannot see this exposure cleanly in their own data. That is a supervisory fact, and it is worth more than any single point estimate.

The bank filings themselves are the best available check. JPMorgan's wholesale credit exposure by industry at 30 June 2026²⁶ shows Asset Managers at $176,789m — its third-largest industry exposure, behind only Real Estate and Individuals, and ahead of Consumer & Retail, technology and healthcare. Six months earlier it was $152,848m. That is 15.7% growth in half a year, and the exposure equals 58% of JPMorgan's $302,619m of CET1 capital. Of it, $137.0bn is investment grade and $39.2bn non-criticisedA supervisory grading. A bank moves a loan onto its criticised list when the borrower's condition has weakened enough to warrant special attention — well before a payment is missed. Non-criticised means the bank has not yet seen anything wrong.; criticised non-performing is $11m. Citigroup, by contrast, reports "asset managers and funds" at just 4% of its corporate credit portfolio²⁷, up from 3%. Goldman's disclosure runs the other way round — it reports $381bn of alternatives in supervised assets plus $142bn of non-fee-earning alternative assets, the latter explicitly including "credit facilities collateralized by fund assets."

This is not a JPMorgan quirk. Bank of America²⁸ states it plainly: "Asset managers and funds, our largest industry concentration with committed exposure of $244.9 billion, increased $10.6 billion, or four percent, during the six months ended June 30, 2026, which was primarily driven by investment-grade exposures" — the largest single industry concentration in a $1.5tn commercial book, at a bank with no meaningful private credit fund business of its own. Wells Fargo²⁹ reports "financials except banks" as its largest commercial and industrial category, at $208.8bn of loans, 20% of total loans, against $318.9bn of commitments, of which asset managers and funds alone account for $73.5bn drawn and $127.2bn committed. Non-accrual loans in that asset-manager book: $2 million. Add the three disclosed asset-manager figures together and the largest US banks have roughly $549bn of committed exposure to the fund complex — against an official-sector estimate of $410–540bn for lending to private credit entities specifically.

The "banks versus private credit" frame obscures the second-order consequence entirely. If a private credit fund takes losses large enough to matter, the bank sits senior and over-collateralised, and loses nothing on the loans. What happens instead is that the borrowing baseA lending formula rather than a fixed loan amount: the bank will advance up to a set percentage of the current value of the collateral. Mark the collateral down and the permitted borrowing falls with it, so the fund must either post more collateral or repay — which in practice means selling assets. falls, the facility deficiency must be cured, and the fund must post collateral or sell assets into a market with no bids. The bank's loss, if it comes, arrives as a counterparty event and a forced-sale externality, not as a credit loss on a corporate loan. Supervisory stress tests, as the FSB observes, "cover lending to private credit funds through stresses on banks' credit books" — that is, through the credit-loss channel, which is the channel least likely to fire.

10. The valuation machinery — the substitute for a market

No secondary market prices these loans. Something must therefore stand in for a price, and what stands in is a quarterly model produced under the manager's control.

The mechanics, as the FSB describes them after engaging market participants directly: the primary approach for performing loans is a discounted cash flow or yield analysis using a specified discount rate, supplemented by public-market proxies such as leveraged loan indices; valuations are "calibrated at origination of the loan, with adjustments made over time." Under US rules the fund's board designates the adviser as "valuation designeeUnder SEC rules a fund's board may hand the day-to-day work of determining fair value to the investment adviser, which then becomes the board-appointed valuation designee. The board retains oversight; the adviser sets the number.," and the adviser engages an independent third-party valuation firm whose work it reviews. The independent firm is paid by the manager. The audit committee reviews. The board approves.

None of this is improper, and the industry's defence has real content. Stakeholders told the FSB that private credit valuations are less volatile than public markets "due to protective features in loans, such as interest rate floors and tighter covenants," and that lenders and borrowers work toward resolution more closely than public markets allow. Both are true. A first-lien loan at 40% LTV genuinely should not move like a high-yield bond.

But the FSB also records, in the same section, that "some members have also identified practices where the income-based approach grants managers significant discretion," that discrepancies arise "with examples made in comparable cases having different valuations across managers," and — the sentence that matters most — that "perceived or actual stale valuations may create a first-mover incentive during stress events, leading investors to exit a fund before asset values are potentially marked down." It goes further: "managers may have potential incentives to manage valuations of their funds in a way that minimises the appearance of volatility."

That is a regulator describing a run mechanism in a product sold on the absence of volatility. One number sharpens it. Cliffwater's August 2026 release reports that private loans in its index were priced at $0.98 per dollar against $0.96 for public loans. The illiquid, unmarketable, single-lender instrument is carried at a higher price than the tradeable one. There are defensible reasons — different credit mixes, structural protections, no forced sellers. There is also the obvious one.

One supervisor has already examined this in detail, and its findings are more pointed than the FSB's summary. In the Financial Conduct Authority's multi-firm review of private market valuation practices³⁰, the FCA looked at governance, independence, conflicts, methodology and the use of third-party advisers across managers running private equity, venture, private debt and infrastructure. It found real strengths — quality of investor reporting, consistent methodology, third-party advisers introducing independence — and one structural problem it names without euphemism. Where a fund's management fee is calculated on net asset value, "managers face a conflict of interest that could dissuade them from taking appropriate write-downs to avoid lost management fees." The FCA extends the point to marketing, noting that using unrealised performance to raise the next fund "can create an incentive to present favourable valuations." And it identifies the specific gap that matters most for 2026: "Many firms did not have defined processes or a consistent approach for ad hoc valuations to revalue assets during market or asset-specific events," which is precisely the circumstance in which a stale mark becomes a first-mover advantage. Only a few firms, in the FCA's assessment, demonstrated the awareness and control over valuation conflicts it expected.

Every firm had identified the fee conflict; almost none had documented the others — marketing, secured borrowing, asset transfers, redemptions and subscriptions. The conflict that is obvious is managed. The conflicts that matter in a redemption wave were, at the time of the review, largely undocumented.

The 2026 stress produced a natural experiment. Lincoln International, which values loans for lenders and sponsors rather than for a fund selling a product, reported for Q1 2026 that its private market index fell 2.2%, that software enterprise values fell 8.8%, and that "the average fair value of loans to software businesses valued by Lincoln declined by 1.9% of par." Vertically integrated software multiples fell to 15.0x from 15.7x, horizontally integrated to 13.4x from 14.4x — the latter falling further, Lincoln notes, given "greater vulnerability to replacement by general purpose AI solutions." Cliffwater records Q1 2026 software loan markdowns of about 3%, moderating to under 1% in Q2. Set those against public software equity, which the BIS records as having "collapsed by almost 30%" between October 2025 and February 2026. The marks moved. They moved by roughly a tenth of what the public comparable did.

Whether that is prudent smoothing or deferred recognition cannot be settled from outside. What can be settled is what it does to behaviour, and Section 14 shows that.

11. Income quality — the gauge that moves first

Default rates are a lagging indicator in every credit market. In this one the lag is longer, because the lender controls the amendment, the toggle and the mark, and each of those defers the moment a loss becomes a statistic.

Payment-in-kind is the clearest signal because it is disclosed. The FSB reports that PIK is used in approximately 12% of loans with toggles accounting for about half, that usage of both PIK notes and toggles has risen sharply since 2022 and shows no sign of declining, and — citing work by Rintamaki and Steffen on US BDC loans — that loans carrying a PIK toggle go delinquent the following quarter 1 to 2 percentage points more often, against an unconditional probability of 3%. Lincoln puts numbers on the composition: 11% of the loans it values paid PIK interest in 2025, up from 7% in 2021, and 56% of those were "bad PIK" — companies that paid no PIK at closing and now do, a share up roughly 20 points since Q4 2021.

The listed vehicles disclose their own. Blue Owl Capital Corporation's total PIK income reached 10.7% of investment income in the June 2026 quarter, from 9.1% a year earlier³¹; Blue Owl Credit Income's reached 6.6%, from 6.0%³²; Blackstone Secured Lending's payment-in-kind interest income was 6.6% of total investment income in the quarter. These are not alarming levels. They are rising levels, in a year when reported non-accruals at the non-traded vehicles did not move.

The counter-evidence is good. StepStone's 1H26 update, on Kroll data as of June 2026³³ reports that PIK income at BDCs declined to 6.83% of income in the first quarter of 2026, from 7.31% in the fourth quarter of 2025 — though it remains above the 5.22% long-term average. What rose instead was the conversion rate: the trailing four-quarter share of transactions amended into PIK reached 2.21% in 1Q26 from 1.98%, slightly above its 1.99% long-term average and well below the 2024 and pandemic peaks. StepStone adds a detail that cuts both ways: "average marks on PIK term loans also fell, indicating that the market continues to be cautious about such loans by booking impairment charges ahead of potential underperformance." Lenders are marking PIK loans down before they fail. That is the valuation process working, and it is the strongest single piece of evidence in this study that the marks are not simply frozen.

Blue Owl's technology fund makes the trend hardest to dismiss, because it is the fund with the most exposed borrowers: Blue Owl Technology Income's Form 10-Q³⁴ reports total PIK income at 9.8% of investment income in the June 2026 quarter, from 8.1% a year earlier. That is a portfolio 64% weighted to software, reporting 0.2% non-accruals, in which nearly a tenth of income is now interest the borrower did not pay in cash.

Amendment activity is the second gauge and it points the same way. Lincoln reports that in Q1 2026 the covenant default rate fell to 3.1% while amendment activity rose 8% quarter over quarter. A falling default rate alongside rising amendments is one fact, told twice.

The third gauge is the one that is hardest to argue with, because it records completed transfers of ownership rather than opinions. Lincoln observed $39.4bn of debt foreclosed in 2025 and 2026, against $13.6bn in the preceding three years combined — nearly a threefold increase against three years of base — with almost 75% of those change-of-control transactions relating to 2021 and 2022 vintageThe year a loan or a fund's investments were made. Vintage matters because underwriting standards, leverage and pricing move with the cycle: loans written at the top of a market behave as a cohort, and they fail as one. deals. Lenders are taking the keys at a rate the default statistics do not show, on loans written at the top of the cycle.

Amortised cost is what the lender paid for a loan, adjusted for fees and repayments; fair value is what it now says the loan is worth. Which denominator a bad-debt figure is quoted against decides how large it looks. A non-accrual percentage measured against fair value has already absorbed whatever write-down was taken, so the same troubled loan can read twice as big at cost as at value — and the gap between the two is the markdown itself.

One listed vehicle moved the other way. Oaktree Specialty Lending³⁵ reported six investments on non-accrual at 30 June 2026, representing 4.2% of debt investments at cost and 1.8% at fair value — down from ten investments at 6.5% and 3.0% at 30 September 2025. A manager with a large legacy book worked its problems down while newer portfolios were working theirs up. Credit deterioration in 2026 is not uniform, and treating it as a single sector-wide event would be wrong.

Golub Capital BDC, a lower-middle-market lender with a long record, gives the same picture from a different angle: non-accrual loans at fair value of $73.3m at 30 June 2026 against $27.3m at 30 September 2025³⁶, with five preferred equity positions on non-accrual at $85.3m fair value against one at nil nine months earlier. Its weighted average investment income yield fell to 9.2% from 10.3% a year earlier. Rising losses, falling yield, same portfolio.

12. Defaults and recoveries — four series, four answers, one definition problem

Ask what the private credit default rate is in mid-2026 and you can be handed any number between 1% and 8% without anyone lying.

The FSB states the reconciliation plainly: outright defaults "remain low at approximately 1%, but this figure rises to around 5% when certain restructuring transactions, known as 'selective defaults', are included" — where a selective default is a restructuring on non-arm's-length terms, the lender effectively bailing the borrower out to avoid a default. A single lender holding a whole unitranche can execute that in an afternoon. A syndicate of forty cannot. Moody's, running the same arithmetic in April 2026³⁷, puts private credit's 2025 default rate anywhere from 1.6% to 4.7% depending on the treatment of distressed exchanges — and finds distressed restructurings were roughly 65% of all defaults that year. The definitional choice is now most of the answer.

RecoveriesHow many cents on the dollar a lender ends up getting back once a loan has actually gone bad and been worked out. The default rate multiplied by one-minus-recovery is the loss the lender wears — which is why a market can have a high default rate and be perfectly sound, or a low one and not be. are the second half and the evidence is thinner. KBRA's work implies average recoveries of roughly 53% in the US and 40% in Europe. The marketed history reads differently: Cliffwater reports realised losses from defaults running at about 0.5%, half its 1.0% long-term average across nearly 22 years and an index of more than 23,000 loans worth $553bn. Both can be true: a low loss rate in a period when almost nothing has been forced to liquidate, and a mid-50s recovery when things are.

13. Workout — what happens when one lender owns the whole loan

The unitranche changes the restructuring in ways that cut both directions, and the honest reading is that it is better for the borrower, better for the reported number, and unproven for the ultimate recovery.

On the good side: a single lender can act fast, inject liquidity, waive a covenant, and avoid a value-destroying bankruptcy that a fragmented syndicate would blunder into. The FSB heard exactly this from market participants — that "lenders and investors may work closer towards a resolution in private credit compared to public markets." The evidence supports it in one specific respect: FSB-cited research finds that loans to sponsor-backed companies show a positive correlation with PIK usage and are less likely to progress from delinquency to outright default, because sponsors inject liquidity. The sponsor's equity is a real cushion and it is exercised.

On the other side: the same concentration means nobody independent tests the lender's judgment about whether the company is worth saving, and the lender's fee stream depends on the loan continuing to accrue. The Lincoln foreclosure data — $39.4bn across 2025–26 against $13.6bn in three prior years, three-quarters of it from 2021–22 vintages — is the sound of that judgment being exhausted two to four years after origination rather than immediately. The delay is a design property, not a fraud signal. This asset class defers recognition, and the deferral has a term structure of roughly three years.

What recoveries the covenant-lite unitranche ultimately produces is genuinely unknown. There is no completed default cycle in this asset class at its current size, composition or documentation. The marketed history — Cliffwater's 22 years averaging 9.5% with a single negative year in 2008 — describes an asset class that was a tenth of today's size, lent at lower leverage with tighter covenants, and never faced a technological repricing of its largest sector. It is data. It is not a precedent.

14. The 2026 episode — the first test, and what broke in it

Between October 2025 and March 2026 the market discovered that private credit's largest sector was software, and that software's revenue model had a new problem.

The BIS put the exposure on the record in its March 2026 Quarterly Review: outstanding private loans to software-as-a-service firms grew from roughly $8bn in 2015 to over $500bn by end-2025, about 19% of total direct loans; one-third of private credit funds have SaaS exposure; BDCs allocated over 15% of their loans to SaaS in 2025 and account for roughly a fifth of all US direct loans. When software equities fell almost 30% between October 2025 and February 2026, BDC stock prices fell about 10% on average, and BDCs with high SaaS exposure underperformed peers by around 5 percentage points. Discounts to net asset valueA listed fund's shares change hands wherever buyers and sellers meet, which need not be the value the fund itself puts on its assets. When the shares trade below that figure the fund is at a discount — the market saying, in the only language it has, that it does not believe the stated value. deepened — which the BIS reads as potentially "worries about underlying valuations." The Congressional Research Service put SaaS exposure at around $500bn as of December 2025 and named the mechanism: automated coding by AI reducing software company revenues.

Then the redemptions came. The Congressional Research Service tabulated them as of March 2026, and its table is the clearest single piece of evidence in this study.

Source: Congressional Research Service, Private Credit Funds Redemption Restrictions: Market Context and Policy Issues, IN12674, 27 March 2026, Table 1. Fund sizes and caps as reported by CRS from media reports hyperlinked in the original; some funds enforced the cap and others met requests irrespective of it.

The Federal Reserve confirmed the aggregate: redemption requests rose from low levels in Q4 2025 and accelerated in Q1 2026, "when a number of funds received requests that were much larger than 5 percent of NAV," most managers capped at 5%, inflows weakened, and accepted redemptions exceeded new inflows in the first quarter of 2026, the first time this has occurred since these vehicles were created.

No wave of defaults arrived. Reported non-accruals at the affected vehicles were mostly in the 0.2% to 0.4% of fair value range, with BCRED at 1.1% in its June 2026 letter. Apollo's Debt Solutions BDC reported non-accruals unchanged at 0.8% of cost. Ares Strategic Income had none at all at 31 December 2025 and 0.3% six months later. Blue Owl Credit Income's non-accruals actually fell, from 0.6% of amortised cost to 0.3%. The credit did not break. The belief in the mark broke, and the redemption right was the only instrument available for expressing that.

Two things cut the other way, and both are real. First, the managers met the cap and did not suspend: Blue Owl states OCIC provided over $1.9bn of liquidity across two quarters, roughly 43% of repeat tenderers'A shareholder who has asked the fund to buy their shares back. These vehicles do not let investors sell to one another; each quarter the fund offers to repurchase a limited number of shares and shareholders tender — put their shares forward — into that offer. If more is tendered than the cap allows, everyone is cut back in proportion. original requests, without selling a private loan, funded from $2.7bn of ordinary-course portfolio repayments running at 1.4x coverage of repurchases. Second, the pressure is easing — OCIC's requests fell from 21.9% to 18.8%, OTIC's from 40.4% to 38.1%, and approximately 90% of OCIC's 90,000 shareholders remained invested. The Federal Reserve's judgement, on data through Q1 2026, is that "risks to financial stability from further redemption requests appear limited and manageable," resting on the observation that for the ten largest perpetual BDCs — 80% of the sector's assets — available bank credit and cash cover at least three calendar quarters of net redemptions at the 5% level.

The Fed's comfort about a retail liquidity mismatch rests on bank credit lines. The senior lender is the liquidity backstop for the retail wrapper. The analysis is correct on its own terms. Taken at face value, it is also the clearest possible statement of where the load in this system actually sits.

15. The stability debate — which fragility conditions are actually present

Four conditions make a credit system fragile: runnable funding against illiquid assets; leverage chains that turn a mark into a forced sale; correlated exposure across nominally separate holders; and a valuation regime that defers recognition. Test this system against each and the answer is genuinely mixed, which is why both of the popular verdicts are wrong.

Runnable funding. Mostly absent, meaningfully present at the margin. Eighty per cent of the asset class is closed-end or insurance-funded. Twenty per cent — $241bn of net assets — carries a quarterly option, and that option was exercised in 2026 at three to eight times the rate the vehicles will honour. The FSB's framing is the right one: liquidity mismatch here is a growing vulnerability rather than a present one, and it grows with every dollar of the marginal channel.

Leverage chains. Present but shallow at the fund level and deep at the borrower level. The OFR's median fund leverage of 1.0x is genuine comfort; the 95th-percentile 3.5x tail holding $81bn is genuine concern; borrower leverage near 7x adjusted is the real amplifier. Crucially, a private credit fund does not face margin callsA lender's demand for more collateral, immediately, because what it already holds has fallen in value. Margin calls are the mechanism that turns one fund's losses into everybody's: meeting them means selling, selling pushes prices down, and lower prices trigger the next call. the way a hedge fund does — the borrowing-base facility revalues, and the cure period gives days rather than hours. That is a real structural improvement over 2008 and it should be conceded without qualification.

Correlated exposure. Present and under-appreciated. The FSB reports that roughly half of companies borrowing from private credit funds also have bank revolvers; that five large asset management groups account for about a third of aggregate loan commitments across the private credit and equity industry; that the top five banks provide 63% of committed loans to BDCs and the top ten about 84%; and that BDCs maintain relationships with an average of 5.5 banks. Concentration on both sides of the same exposure.

Deferred recognition. Present, and it is the distinguishing feature. Everything in Sections 10 through 13 says the same thing: this system's shock absorber is time.

The Bank of England's judgement, in its July 2026 Financial Stability Report³⁸, is that vulnerabilities in risky credit markets including private credit "remain, and some have become more pronounced since the December 2025 Financial Stability Report." On the retail channel its language is precise: "Redemption requests have been elevated in several retail funds, with some limiting redemptions, underlining both liquidity mismatch and valuation concerns." Its second system-wide exploratory scenario is examining how stress propagates through these markets. That is the right question and nobody has the answer yet. Known unknown

The same report reaches the capital cycle running alongside private credit: "AI-related companies' use of credit markets has accelerated rapidly, including in public markets, private credit, leveraged and structured finance, and is set to increase further as financing needs continue to expand. This pace of investment is unprecedented historically." Private credit is simultaneously the sector most damaged by AI's effect on software borrowers and the sector most eagerly financing AI's build-out. Section 20 takes that up.

16. The conclusion, stated first — and what the consensus is missing

Two things are worth owning in this industry and neither is a loan.

The first is a liability that cannot be recalled by the person whose money it is. Apollo has $537bn of it in annuity form. KKR has Global Atlantic. Nobody else has anything comparable, and the gap between "perpetual" as Apollo means it and "permanent" as Blue Owl means it is the single most mispriced distinction in the sector. The second is origination breadth — the ability to write a $500m data-centre lease financing, a receivables facility, a European unitranche and a fund-finance line from the same platform. That is what allows a manager to keep deploying when one province of the asset class stops paying, and it is why Apollo can report $317bn of origination over twelve months while the direct-lending spread compresses beneath it.

Everything else in the value chain is either being competed away or was never rent to begin with. Direct-lending scale is a treadmill: it wins deals, it does not hold price. The wealth channel is a distribution business dressed as an asset-gathering moat, and 2026 has just shown what its liabilities do under pressure. The fee rate is falling. Incentive income has already stopped.

Where the consensus is incomplete. It reads AUM growth as evidence of pricing power — but Ares grew AUM 17% to $671bn while holding a record $170bn of undeployed capital, which is the definition of capacity outrunning opportunity. It reads permanent capital as a homogeneous category — but the composition determines everything. And it reads the 2026 redemptions as a liquidity event to be survived, when they were a valuation event that the liquidity terms merely made visible. If the marks were believed, nobody would have asked to leave a fund reporting 0.2% non-accruals and a 9.2% return.

17. Permanence — one word doing four different jobs

Every manager in this study reports a "permanent" or "perpetual" capital figure. Every one of them defines it differently, and the differences are the investment case.

What the word is meant to signify is simple enough. An ordinary fund has a life: money is raised, invested, returned, and the manager must go out and raise it again, so the fee stream has an expiry date built into the contract. Capital called permanent or perpetual has no such date — it stays, and it keeps paying a fee, until something specific removes it. The whole question is what that something is, and whether the person whose money it is can bring it about at will.

Sources, each the manager's own second-quarter 2026 disclosure: Apollo¹⁷; Blue Owl³⁹; Blackstone¹³; KKR²²; Ares²⁴; Carlyle²³. Percentages of AUM computed by AIR from each manager's stated perpetual capital and total AUM; Carlyle discloses perpetual capital on a fee-earning basis only, so its figure is not directly comparable to the others and is labelled accordingly.

What "contractual" means can be measured, and Athene measures it. Athene Holding's own Form 10-Q⁴⁰ discloses that at 30 June 2026 approximately 87% of its deferred annuity liabilities were subject to penalty upon surrender, up from 85% at year-end; roughly 36% of net reserve liabilities were generally non-surrenderable — pension group annuities, funding agreements, payout annuities and guaranteed investment contracts — with a further 54% surrenderable only with a penalty; and about 69% of policies carry market value adjustmentsA clause that adjusts what a policyholder receives on early surrender to reflect what has happened to interest rates since they bought. If rates have risen, the payout is cut — so the policyholder bears the loss the insurer would otherwise take from selling bonds early to fund the exit. that further discourage early exit. Net invested assets grew to $314.1bn from $292.4bn over six months, driven by $33.1bn of net organic inflows against $16.2bn of net liability outflows. That is what a liability looks like when it genuinely cannot leave: it is disclosed, it is quantified, and in the worst quarter this industry has had, it grew.

Blue Owl is the instructive case, and the point is not that it has done anything wrong. Its disclosure is complete: it tells you the $225bn is "permanent," it tells you 85% of management fees come from it, and it separately tells you — in the funds' own shareholder letters — exactly how many shares were tendered. Both facts are on the record. The problem is that a reader who takes the first number as a proxy for fee-stream durability, as the sell-side convention does, is reading a redeemable liability as a contractual one. Blue Owl carries the highest headline permanence ratio of any covered manager and the lowest genuine one.

The test came in November 2025 and it was unusually clean. Blue Owl announced on 5 November that OBDC II would merge into the listed OBDC, with an exchange ratio set at NAV-for-NAV whenever OBDC traded at or below its own NAV, and noted that OBDC II "does not anticipate conducting additional tender offers prior to the merger." Fourteen days later, on 19 November, the two boards terminated it⁴¹, citing "current market conditions," with the CEO stating that "we are no longer pursuing the merger at this point." OBDC II shareholders, who had been told the tender was pausing, were told the merger was off and the tender would be reinstated in Q1 2026. Permanence that can be restructured twice in a fortnight is a commercial arrangement, not a liability structure.

18. Fee-related earnings versus incentive income — the cycle already ran this test

The standard argument for a premium multiple on these firms is that management fees are recurring while carried interest is not, so the earnings mix has improved. That argument is correct. It is also less protective than it sounds, and 2026 has just shown why.

The two streams are paid for different things. A management fee is rent on the assets — a fixed annual percentage, owed whether the loans perform or not, and what remains of it after the firm's costs is what "fee-related earnings" measures. Carried interest and incentive fees are a share of the profits, taken only above a minimum return and worth nothing at all in a year that fails to clear it. One is a subscription. The other is a bonus.

Blackstone Secured Lending's income-based incentive fee fell to $1.69m in the June 2026 quarter from $34.72m a year earlier — a 95% collapse⁴² — while its management fee rose from $34.6m to $35.3m. Blue Owl's GAAP performance revenues fell 90%, to $102,000 for the quarter from $979,000. Carlyle's net accrued performance revenues fell to $2.4bn from $2.9bn at the end of 2025. TPG discloses the endpoint directly: fee-related earnings were 83% of pre-tax distributable earnings in Q2 2026, and 106% in Q1⁴³ — the "all other sources" line was negative. Distributable earningsThe cash-basis profit these firms present as available to pay out to shareholders, after costs and taxes. It is a non-GAAP measure each firm defines for itself, which is why the definitions are not identical between them. is the payout measure these managers report alongside GAAP, so a share expressed against it says how much of the payout is coming from fees rather than from anything else.

So the mix shifted exactly as advertised: fee-related earnings held, incentive income went to nothing. Apollo's FRE grew 25% to a record $785m; KKR's grew 37% to $1.21bn; Ares' was $491m; Carlyle's was a record $358m; Blue Owl's was $392m. Every single one grew.

The multiple argument does not address the complication. A management fee is only as durable as the asset base it sits on, and for the semi-liquid vehicles the asset base is now shrinking. Blue Owl Credit Income repurchased $955.0m of shares in the June 2026 quarter and $963.1m in the March quarter, against $464.2m in the June 2025 quarter — the fee base shrinking by roughly two billion dollars in six months at one fund, against year-to-date gross inflows of about $1.2bn. Blackstone's Credit & Insurance segment raised $31.0bn in the quarter but only $1.0bn of it was BCRED equity; the growth came from insurance separately managed accounts, at $7.5bn. The fee stream did not break. It rotated, from the highest-fee wrapper to a lower-fee one.

That is the fee-compression mechanism nobody is modelling: a mix shift driven by the highest-fee channel closing itself, with no price war required. Ares' management is explicit about the offset — a record $170bn of dry powder and "our largest ever forward investment pipeline" — and that is a real answer, because undeployed capital converts to fee-paying capital on deployment. Blue Owl quantifies the same effect: $31.1bn of AUM not yet paying fees, representing approximately $380m of expected annual management fees once deployed. Both are genuine. Both depend on deployment happening at spreads that justify it.

BlackRock's private markets base fees were $639m in the June 2026 quarter⁴⁴, up 28% year on year, out of $7.08bn of total revenue. Blue Owl's total management fees for the same quarter were $661.6m. Those are not apples-to-apples line items: BlackRock is reporting private-markets base fees only, while Blue Owl is reporting total management fees across the firm. The analytical point is narrower and better supported: wrapper mix and fee rate matter more than headline AUM, which is why Blue Owl's higher-fee wrappers still monetize more fee dollars per unit of private-credit capital than BlackRock's wider private-markets platform does today.

19. Origination as the binding constraint — and where the capital went instead

In private credit the scarce thing is the borrower, not capital. Control of the borrower is what earns the rent, and the question is what happens to that rent when capacity doubles.

Capacity has already doubled and the answer is visible. Ares holds a record $170bn of dry powder. Blackstone holds $228.1bn. KKR's uncalled commitments reached $486bn across the firm. Against that, sponsor deal volume has not kept pace — Lincoln records that total M&A transaction volume declined even as average deal size rose above $1bn, a flight to quality that concentrates competition on fewer, larger, better-covered credits. The price response is exactly what theory predicts: gross spreads down 75–100bp, original issue discounts 50bp tighter, and the historic premium for lower-middle-market lending converging toward 500bp alongside the upper market. StepStone's conclusion is that "private lenders have limited scope to tighten pricing further while remaining competitive relative to public markets."

Then, in the first half of 2026, the price turned. StepStone's 1H26 update, on Kroll data as of June 2026³³ records median US first-lien gross spreads rising to approximately 554bp in June from 525bp in March, with Europe at 632bp from 615bp and OIDs widening alongside. That is the single most important piece of evidence against this study's compression argument, and it deserves to be taken at face value rather than explained away.

Deal volume did not recover — US sponsored middle-market volume fell to roughly $46bn in the first half of 2026 and middle-market M&A to about $25bn, both below long-run averages, with StepStone attributing the weakness to "the AI scare and geopolitical uncertainty." Closed-end fundraising ran at $133bn through July against $186bn for all of 2025 and $213bn in 2024. And BDC net flows turned negative. The spread widened because supply of capital contracted faster than demand for it, led by the retail evergreen vehicles retrenching into liquidity management. That is a price improvement produced by the channel this study calls least durable withdrawing — which supports the structural read rather than refuting it, and which reverses the moment the channel comes back. StepStone's own conclusion is that "meaningful spread tightening appears unlikely in the second half," a forecast of stability rather than of a new premium. Direct lending's trailing twelve-month total return had already fallen to 8.1% by March 2026 while its income return held at 9.9% — a 180 basis point gap that is unrealised loss, arriving in the marks at last.

When the price of a product falls because too much capital is chasing it, the capital moves. Two destinations are absorbing it.

Asset-based finance. Lending against pools of cash-flowing assets — receivables, equipment, mortgages, royalties, aircraft — rather than against a corporate's EBITDA. The FSB records asset managers "acquiring origination platforms themselves, including those in the consumer and residential mortgage space," and notes the emergence of forward-flow agreements in which managers continuously purchase newly originated loans from bank platforms. Apollo's MidCap FinCo, at $14bn, is exactly this. So is Blackstone's $9.9bn of quarterly inflows into "infrastructure and asset based credit strategies." The attraction is honest: a larger addressable market, less sponsor competition, and collateral that is diversified rather than single-name. The risk is equally honest, and First Brands demonstrated it — an asset-based structure is only as good as the existence of the assets, and the largest asset-based lending fraud in recent memory involved $2.3bn of fabricated receivables pledged more than once.

AI infrastructure. The FSB's Box 4 sizes it: AI infrastructure capital expenditure of $2.9tn between 2025 and 2028, of which $1.5tn requires external capital including roughly $800bn from private credit, on Morgan Stanley estimates cited by the Bank of England. The FSB records that AI-related deals reached 34% of private credit deals in 2025, against a 17% average over the prior five years. This is asset-based finance at industrial scale: long-dated leases to investment-grade tenants, financed off-balance-sheet, at spreads that beat corporate direct lending. The Bank of England's July 2026 report calls the pace of investment "unprecedented historically."

The AI capital cycle runs through this business in both directions. Private credit's largest existing sector, software, is being repriced downward by AI. Private credit's fastest-growing new sector, data centres, is being priced upward by the same force. The same manager can be short AI's effect on borrower revenue in one book and long AI's capex in another. Nobody is netting those exposures, because they sit in different funds with different investors — but they sit on the same underwriting committee, in the same house, marked by the same process.

20. The valuation layer is the product — what the wealth channel is actually buying

What these managers sell to the wealth and insurance channels is reported smoothness, not yield. The yield is available elsewhere, in liquid form, at lower fees. What is not available elsewhere is a 9% return that never shows a drawdown on a client statement. That property is manufactured in the valuation process, and it is therefore the valuation process — not the credit — that is the load-bearing component of the whole edifice.

The consensus view this contradicts is the industry's own and it is stated by every manager in this study: that smoothing reflects the genuine economics of a senior secured, covenant-protected, held-to-maturity loan whose value really does move less than a traded bond. That view has support. The FSB heard it from participants and reported it without dismissing it. Interest-rate floors are real. A 40% loan-to-value is real. Being the only lender at the table genuinely does change the outcome.

Four pieces of disclosed evidence sit against it, and they are not opinions.

First. In the quarter when 18.8% of Blue Owl Credit Income's shareholders and 38.1% of Blue Owl Technology Income's asked to leave, both funds reported non-accruals of 0.2% of fair value and no new non-accruals in the prior quarter. Investors sold an asset they were told was performing. Either they were irrational, or they did not believe the number.

Second. Cliffwater's index prices private loans at $0.98 against $0.96 for public loans. When the illiquid instrument is carried above the liquid one, the difference is not a liquidity premium; it is a marking convention.

Third. The listed BDCs, which are marked by a market rather than by a model, disagree with the models by a wide and widening margin. FS KKR's net asset value fell to $18.30 from $20.89 in six months, a 12.4% decline⁴⁵ — and between 1 July and 5 August 2026 it repurchased 3,348,353 shares at an average of $10.75, which is 0.587 times its own reported NAV. Ares Capital's dividend reinvestment planA standing arrangement under which shareholders take their dividend in additional shares rather than cash. The fund can supply those shares two ways: issue new ones, or buy existing ones in the market. Which it does depends on whether the shares are trading above or below net asset value. bought shares in the open market at an average of $18.27 against NAV of $19.35⁴⁶, a 5.6% discount — where a year earlier the same plan issued new shares at $22.06. Two managers, the same asset class, the same quarter; the market takes one at 94 cents and the other at 59.

Fourth. One disclosed episode does the work of a hundred arguments. The Congressional Research Service records that "BlackRock reportedly wrote down a private loan from 100 cents on the dollar to zero within three months," and draws the conclusion this study draws: it illustrates "both the difficulty of valuing illiquid assets and the potential speed of asset quality deterioration." A loan carried at par one quarter and at nothing the next was not correctly marked in the first quarter. Single reported instance; the underlying credit is not named in the CRS note

Fifth. The BIS documented the transmission directly: software equities down almost 30%, BDC prices down about 10%, high-SaaS BDCs underperforming peers by five points, discounts to NAV deepening. Lincoln's valuation of software loans over the same window: down 1.9% of par.

The obvious reading of that fourth point is that BDC equity is overreacting, as equity does. It is a fair reading and it may be right. What complicates it is the third point: the discount is not uniform. The market applies a 6% haircut to Ares Capital and a 41% haircut to FS KKR. A market simply panicking about the asset class would not discriminate that finely between two lenders holding overlapping portfolios. This one is discriminating between managers, and what it is discriminating on is the credibility of the marks.

The consequence runs one step further. If the product is reported smoothness, the competitive threat is transparency, not another lender undercutting the spread. A private credit exchange-traded fund, a secondary market in LP interests, or a monthly independently-calculated index all attack the same asset — and all three are arriving. Lincoln International — the firm that values roughly 7,100 private companies for the lenders themselves — has partnered with S&P Dow Jones Indices to publish the S&P/Lincoln U.S. and European Senior Debt Indices³ from 2026, independently calculated benchmarks for the fair value of private loans. The BIS, writing in July 2025, saw the direction of travel: private credit ETFs "may introduce price signals that make the opaque private credit market more transparent, especially during downturns when discounts to net asset value could be large and persistent." The industry has spent a decade arguing that its assets deserve a valuation regime of their own. The infrastructure to test that claim is now being built, by the people who supply the current marks.

21. The managers — eight structural reads

Each read below rests on that firm's own second-quarter 2026 disclosure, opened and cited. The verdict is about competitive economics, not about the stock.

22. The incentive map — who is paid to see what

Most of what looks puzzling about this industry resolves once the payments are written down: who is paid by whom, and for what. The exercise is unflattering to nobody in particular and clarifying about everything.

Three consequences fall straight out of that grid and none of them requires anyone to behave badly.

First, every party with information about the mark is paid more when the mark is higher and smoother, and the only party with a direct interest in a lower mark is a shareholder who wants to leave. That asymmetry is why the FSB's observation about first-mover incentives is not theoretical: the redemption request is the only vote against the valuation that anyone in the chain can cast.

Second, the general partner's fee accrues on committed and deployed capital while the credit cycle lands on the limited partner, the policyholder and the fund shareholder. That is the master asymmetry of the field, and it reaches everywhere — it is why dry powder keeps rising when spreads are falling, why amendments are preferred to defaults, and why the highest-fee wrapper was pushed hardest into the least sophisticated channel.

Third, and least discussed: the private ratings that let insurers hold this paper at favourable capital charges are commissioned by the manager and never published. The FSB records that "in some jurisdictions, the reliance on smaller lesser-known agencies to provide private ratings for private credit assets has grown significantly, accompanied by reports of ratings inflation." The NAIC's response — a due diligence framework for rating providers and a formal process to challenge ratings used for regulatory purposes — is a supervisor saying, in the politest available language, that it no longer takes those ratings at face value.

23. The banks — what they are actually short

JPMorgan, Citigroup and Goldman Sachs sit in the most durable layer of the chain, and each has arranged its participation differently.

JPMorgan is the balance-sheet principal. Asset Managers is its third-largest wholesale industry exposure at $176.8bn, up 15.7% in six months, 96.9% investment grade or non-criticised, with $11m criticised non-performing. Against $302.6bn of CET1 capital that is a 58% ratio — a number that means nothing about expected loss and a great deal about how central this client set has become. The rent here is genuine: 200–400bp over the reference rate, first-lien security on 86% of the book, a securitisation risk weight on the borrowing-base structures, and none of the monitoring cost of lending to the underlying middle-market company. The banks are writing it on their own terms, too: in the Federal Reserve's July 2026 senior loan officer survey⁵¹ they reported, on balance, that standards on every queried category of lending to non-depository financial institutions — business credit intermediaries and private equity funds among them — sit at the tighter end of their range since 2011. It is the best risk-adjusted lending business JPMorgan has added in a decade.

Bank of America and Wells Fargo run the same trade at similar scale, and neither is usually described as a private credit participant. Bank of America's largest single industry concentration is asset managers and funds at $244.9bn committed; Wells Fargo's largest commercial and industrial category is "financials except banks" at 20% of total loans. Wells Fargo is explicit about why the risk looks contained, and the explanation is the whole thesis of this section: these loans "are generally secured and have features to help manage credit risk, such as structural credit enhancements, collateral eligibility requirements, contractual re-margining of collateral supporting the loans, and loan amounts limited to a percentage of the value of the underlying assets." Contractual re-margining against a percentage of asset value is a borrowing base. It is safe for the lender and it is the mechanism that forces the fund to sell.

Citigroup took the partnership route. The $25bn direct lending programme with Apollo, announced on 26 September 2024⁹ with participation from Mubadala and Athene, monetises Citi's origination without consuming Citi's balance sheet — the bank keeps the client relationship and the fee, Apollo's capital takes the credit. Citi's own disclosed exposure to "asset managers and funds" remains just 4% of its corporate credit portfolio, up from 3%. That is a materially different posture from JPMorgan's and, on the evidence of this cycle, a better-positioned one. It is also a smaller business. Progress on the programme's deployed volume is not separately disclosed in Citigroup's Form 10-Q

Goldman Sachs is both principal and manager, which makes it the hardest to read. It reports $381bn of alternatives inside supervised assets and a further $142bn of non-fee-earning alternative assets — the latter explicitly including "credit facilities collateralized by fund assets," meaning Goldman discloses lending to funds as an asset-management line item rather than a credit line item. Its disclosed effective fee rate on credit direct strategies is 75bp; blended across fund-of-funds it is 58bp. Those are the lowest fee rates in this study and the honest benchmark for where alternative credit fees settle when a bank rather than an alternative manager sells the product.

A fund-level loss event does not hit the bank as a charge-off. It hits as a borrowing-base deficiency, which the fund must cure by posting collateral or selling assets. If several funds cure simultaneously into the same illiquid market, the sale prices become observable — and observable sale prices are the one thing this asset class's valuation regime has never had to accommodate. The bank's senior position converts a fund's loss into a system-wide price discovery event. It protects the bank's principal and destroys the industry's central product feature. That is the transmission channel nobody rehearses, and it is the reason the "banks are senior, so it's fine" argument is only half right.

24. Steelman — the strongest version of the case against this study

Here is the bull case as its best advocate would put it, with the evidence that supports it.

The 2026 episode was a sentiment event in a sliver of the market, and the machinery worked. Semi-liquid vehicles are $241bn of net assets in a $1.4tn asset class — under a fifth. Every gate held. No fund suspended. Blue Owl's OCIC met a $958m tender without selling a single private loan, funded by $2.7bn of ordinary-course repayments running at 1.4x coverage, with $11.6bn of liquidity on hand and net leverage of 0.84x against a 0.90–1.25x target. OTIC held nine dollars of liquidity for every dollar being redeemed. The Federal Reserve, with access to the actual data, concluded that risks from further redemptions "appear limited and manageable." The caps did exactly what they were designed to do: they stopped a first-mover advantage from becoming a run. That is a structure passing a test, not failing one.

The credit is fine, and the credit is what matters. Cliffwater's index — 23,000 loans, $553bn — reports realised losses from defaults at roughly 0.5%, half the 1.0% long-term average, with interest coverage ratios rising, non-accruals and PIK income "stable and well below long-term averages," and a 7.7% trailing-year return that beat every major fixed income index. Lincoln reports 69.8% of private companies growing revenue in Q1 2026, up from 68.6%, and 62.6% growing EBITDA, up from 61.7% — with software companies actually outperforming the average at 74.9% growing revenue. Interest coverage on new deals has recovered to 2.2–2.4x. Loan-to-value sits at 38–42%. The FSB itself notes that private credit CLOs carry more loss-absorbing capacity for senior investors than broadly syndicated CLOs, with issuers retaining the entire equity tranche.

And the marks are not the fiction the critics need them to be. Blue Owl's OCIC has lost an average of 14 cents per year for every $100 invested since inception, against 24 cents of net credit losses for the senior secured middle-market loan index over the closest comparable period. OBDC II delivered a 9.3% annualised net return since 2017 with a 23-basis-point loss rate and satisfied every quarterly tender in full until the merger paused it. If the marks were systematically inflated, the realised losses on eight years of matured loans would eventually reveal it. They have not.

Underwriting is getting better, not worse, and the caps are already unwinding. StepStone's mid-2026 update records US total leverage at underwriting of 4.62x in June 2026, down from 4.99x a year earlier, with interest coverage on new US transactions close to 2.5x against 2.18x a year earlier and the biggest leverage reduction in information technology — the sector under scrutiny. Payment defaults "remain contained and below their long-term averages." On the redemption side, the requests are falling everywhere: OCIC from 21.9% to 18.8%, OTIC from 40.4% to 38.1%, and trade reporting of early third-quarter tenders suggests further easing. There is also a mechanical explanation for the headline numbers that this study must concede: when investors expect proration they rationally over-request in order to secure a larger fill, and that reflex unwinds once the gates stop binding. Some unknown share of the 18.8% and the 38.1% is queue management rather than an intention to exit. Third-quarter tender data is trade-reported and post-dates several of the filings used here; treated as indicative only

The permanence is real where it counts. Perpetual BDCs are a minority of a minority. The overwhelming majority of private credit sits in seven-to-ten-year drawdown funds and insurance general accounts, neither of which can run. The OFR's finding that the median private credit fund carries a leverage ratio of 1.0 — no leverage at all — removes the amplification mechanism that made 2008 what it was. There are no daily margin calls here. There is no repo chainRepo is short-term borrowing secured on securities and rolled over daily. A chain of it — each lender funding itself by pledging what it has just bought — is the amplifier the 2008 comparison in this section refers to: when the collateral falls, the whole chain must be unwound at once.. There is a borrowing base with a cure period.

That case is serious and three of its four legs hold. Here is where it breaks.

The first leg treats "no suspension" as proof of resilience. But a gate is not a shock absorber, it is a queue, and the queue is now two quarters long and lengthening for repeat tenderers — OCIC has met 43% of repeat tenderers' original requests over two quarters, OTIC 26%. An investor who wanted out in December 2025 is still, in August 2026, mostly not out. What the caps prevented was a forced sale. What they created is a growing population of shareholders whose expressed preference is to leave and whose capital is nonetheless still paying a management fee. Calling that capital permanent is a description of the contract, not of the intent behind it.

The second leg is where the case is weakest, because the index doing the reassuring is not independent of the thing it measures. The Cliffwater Direct Lending Index is constructed from BDC holdings valued by the same managers whose products Cliffwater's affiliated business sells, and Cliffwater's own $33bn Corporate Lending Fund received redemption requests of 14% against a 7% cap in Q1 2026. That does not make the data wrong. It does mean an index reporting that "non-accruals and PIK income remain stable and well below long-term averages" while Lincoln independently records PIK rising from 7% to 11% of loans with 56% of it unplanned, and lender takeovers tripling, is not a neutral referee. Use it and label it as such; it cannot settle a question about the marks.

The third leg — realised losses vindicating the marks — proves less than it appears because of survivorship in time. The 2021–22 vintages are the ones with 75% of the lender takeovers, and they are three to four years from origination on five-to-seven-year paper. The losses that will test the marks have not matured yet. Cliffwater's own release records what happened when they started to: software loan markdowns of 3% in Q1 2026.

The over-request point is the strongest single objection to this study's reading and it cannot be dismissed. It can, however, be bounded. Over-requesting explains why a number is large; it does not explain why the number exists at all, and it does not explain the composition. Blue Owl reports that the tendering shareholder base "remained largely unchanged with limited new participation" — the same investors are asking again, quarter after quarter, having received 43% of what they first asked for. That is a queue being worked through, not a queue being gamed. And the mechanical explanation says nothing about why the requests began in a quarter when reported credit was pristine.

The fourth leg stands. Fund leverage genuinely is low, there genuinely is no repo chain, and the cure period genuinely does convert a margin call into a negotiation. That is the strongest single fact in the industry's favour and it is why this study's malign scenario is a grind rather than a break.

25. Alternatives considered and rejected — with the condition that reopens each

26. The insurance-channel risk case — carried back to the names it touches

The insurance leg is where the durable rent in this business actually lives. It is also the leg carrying the largest single unhedged risk, and the two belong next to each other.

The mechanism of failure is not a run. Policyholders cannot run in any meaningful sense; surrender charges and tax treatment see to that, and the general account is term-matched by design. The mechanism is capital. An affiliated insurer holds private credit and structured credit at risk-based capital charges set by the NAIC, calibrated in part on ratings the manager commissioned. If those charges rise, or if the ratings are challenged, the insurer must hold more equity against the same assets. That reduces the spread available to the manager, reduces the insurer's capacity to write new annuities, and — because the manager's fee sits on the same assets — reduces fee-related earnings as well as spread-related earnings. One regulatory decision compresses both income streams at once.

The tightening is already underway and it is measurable. The NAIC's principles-based bond definition took effect 1 January 2025, forcing classification on economic substance rather than legal form — aimed squarely at rated feeder notes, the structure that turns a fund interest into something an insurer can hold as a bond. The RBC charge on CLO residual tranches is now 45%. Annual CLO stress testing is in place. AG 53 tightened asset adequacy testing assumptions for long-duration liabilities, and the NAIC's year-end 2025 guidance document⁵² collects every AG 53 filer's CLO exposures tranche by tranche, with the rating agency's rating on each; AG 55 added disclosure and moderately-adverse testing for asset-intensive annuity reinsurance — that is, for exactly the Bermuda structures the model relies on. From 2026 reporting, insurers must disclose private placements at fair value with the Level 2/Level 3 split, PIK interest and private letter rating information, insurer by insurer. And the tax dimension has already delivered a bill: Apollo's $1.7bn one-time charge in Q1 2026 on Bermuda's revocation of ACRA's corporate income tax election.

A second, quieter concentration sits underneath, and the insurer measures this one itself. Athene's Form 10-Q⁴⁰ reports total related-party gross invested assets of $58.5bn, 14.1% of gross invested assets at 30 June 2026, up from 12.7% at year-end — comprising $33.4bn of securitisations of unaffiliated assets where Apollo is the manager, $11.2bn invested in Apollo funds, $9.6bn in asset origination platforms and $4.2bn in retirement services platforms. Athene is careful to note that in the largest bucket the underlying collateral and borrowers are generally unaffiliated. The disclosure is exemplary. The trend is the point: the share of the insurer's assets connected to its own manager rose by 1.4 percentage points in six months. Every increment strengthens the manager's fee stream and tightens the loop that supervisors are examining.

The exposure is uneven name by name. Apollo carries the most concentrated by a wide margin — $537bn of insurance-sourced perpetual capital, 87% of its permanence, and spread-related earnings of $877m that exceed its fee-related earnings. A material RBC change hits Apollo harder than anyone. That is the price of owning the best liability in finance: you own the regulator too. KKR carries the same exposure at roughly two-thirds the scale through Global Atlantic, with $288m of quarterly insurance operating earnings and more diversification outside credit to absorb a hit. Blackstone is exposed differently and more comfortably: it manages insurance separately managed accounts — $7.5bn raised in the quarter — without owning the insurer, so a capital charge falls on its client rather than on its balance sheet, though it would still slow the flow. Carlyle, Ares, TPG and Blue Owl have insurance relationships too small to move their economics either way; for them the NAIC is a headline, not a P&L item.

This is a slow, disclosed, calibrated tightening by a regulator that has explicitly said it is "not treating CLOs or private credit as inherently inappropriate for insurance portfolios," with roughly 80% of insurer CLO holdings investment grade and 39% AAA. This is the gradual removal of the capital arbitrage that made the structure unusually profitable, well short of a shutdown, and it will show up as compressing spread rather than as a crisis. Inference — the direction is documented, the magnitude is not

27. The retail and retirement risk case — and which names actually carry it

This mechanism is behavioural rather than regulatory, and 2026 has already run the first act.

Retail money in private credit is $280bn globally, 13% of AUM, from virtually nothing a decade ago. In the US it sits mostly in perpetual BDCs and interval funds — $425bn of gross assets, $241bn net. The BIS names the design tension precisely: these vehicles "marry shares traded in a liquid market to illiquid underlying private assets that barely trade," and warns that the mismatch "could lead to the emergence of steep and persistent discounts." That was written in July 2025. Nine months later, requests to redeem ran at two to eight times the caps at every large fund.

Ranked by redeemable retail credit assets rather than by AUM, the ordering of the names inverts. Blackstone carries the largest single position — BCRED at $82.2bn of investments at fair value, with non-accrual issuers up from nine to fifteen in six months. Blue Owl carries the largest position relative to its own size: OCIC at $35.7bn and OTIC alongside it, inside a $319bn firm where 85% of management fees come from capital it calls permanent. BlackRock inherited HLEND at $26bn through the HPS acquisition and faced 9.3% tender requests against a 5% cap in its first stress as owner. Ares ($10.7bn ASIF, 11.6% requested) and Apollo ($15.1bn ADS, 11.2% requested) carry real but proportionally smaller positions. KKR's K-Series at $42bn is diversified across asset classes rather than concentrated in credit. Carlyle and TPG are barely exposed.

The forward question is whether the defined-contribution opening replaces this money with something stickier. On the mechanics, it should: a participant in a target-date fund does not choose the private credit sleeve and cannot redeem it separately. On the practicalities, it is harder than the retail BDC was. A target-date fund strikes a daily NAV and must transact daily for job changes, loans and retirements; a quarterly-gated, model-marked asset does not fit inside that without a liquidity buffer, and a liquidity buffer dilutes the very yield that justified the allocation. The DOL's proposed safe harbour requires fiduciaries to consider liquidity, valuation and benchmarking explicitly — three factors on which this asset class is, by its own regulators' account, weakest. The safe harbour reduces litigation risk; it does not answer the engineering problem.

Which is why the fifth prediction in Section 29 is that this channel opens slowly. The industry's growth narrative treats the whole defined-contribution market as an addressable pool. The realistic near-term prize is a small sleeve inside a diversified vehicle, at institutional fee rates, sold to a fiduciary who now has six documented factors to fail on. That is a good business. It is not the one being priced.

28. Second and third-order effects — the parts that are not yet priced

Suppressed default signals extend the boom and deepen the workout. A lender that can amend rather than default keeps writing new loans against a portfolio the market believes is performing. That extends the fundraising cycle, which raises dry powder, which compresses spreads, which lowers the return available to absorb the losses being deferred. Lincoln's data captures both ends: covenant defaults down to 3.1%, amendments up 8%, foreclosures at $39.4bn against $13.6bn in three prior years. The deferral is not free; it is borrowed from the recovery.

Bank seniority converts a fund loss into a counterparty event. Section 23 set out the mechanism; as a forward claim it runs like this. The first genuinely large private credit loss will not appear in a bank's charge-off line. It will appear as a borrowing-base deficiency, a collateral call, and a sale of loans at a price. That price is the thing the industry has never had to publish.

Retail liquidity terms, once tested, reprice the whole channel at once. The 2026 episode has established a public precedent: in a stress, a perpetual BDC pays thirteen to twenty-seven cents on the dollar of what is asked. Every adviser now knows that. Every prospectus disclosure of a 5% cap now carries a lived example. The channel's cost of capital has risen permanently even if flows recover, and the most likely commercial response — longer notice periods, lower caps, or hard lock-ups on new share classes — makes the product less attractive precisely as the DC channel opens.

The DC opening arrives with the wrong instrument. Executive Order 14330 and the DOL's proposed safe harbour point toward alternatives inside target-date funds. A target-date fund is valued daily and must transact daily for participants who change jobs, take loans, or retire. Putting a quarterly-gated, model-marked asset inside a daily-valued wrapper is a harder engineering problem than the retail BDC, and it is being attempted immediately after the retail BDC's liquidity terms failed their first test. The most likely resolution is a small sleeve with a liquidity buffer — which means the fee opportunity is smaller than the industry's growth narrative implies, and the valuation question moves into a vehicle with a fiduciary standard and a plaintiffs' bar attached.

Transparency infrastructure is the real competitive threat, and the industry is building it. Preqin inside BlackRock, S&P/Lincoln senior debt indices, private credit ETFs, and a developing secondary market in LP interests all point the same way. The BIS states the consequence directly: retail vehicles "may also give impetus to the creation of secondary markets for currently illiquid private loans, which could erode the benefits of private credit as an asset class." The benefit being eroded is the absence of a price.

And the AI trade sits on both sides of the same balance sheet. Private credit's largest sector is being disrupted by AI; its fastest-growing sector is financing AI. The FSB puts AI-related deals at 34% of 2025 private credit activity against a 17% five-year average, and flags the specific vulnerability: a data-centre collateral value that depends on electricity supply and on demand forecasts for AI services. If those forecasts disappoint, a manager could take losses on software borrowers displaced by AI and on data centres built for AI, in the same year, from the same thesis being wrong in opposite directions.

29. Falsifiers and dated predictions

Dated, testable predictions. Each is falsifiable against a specific disclosure.

1. 1By 31 December 2027, at least one of the six largest perpetual non-traded BDCs will have suspended repurchases entirely for at least one quarter, or converted to a wind-down or drawdown structure — as OBDC II did in 2025–26. Test: the funds' 8-K filings. Moderate confidence
2. 2Aggregate non-accruals at the six largest listed BDCs will exceed 4% of amortised cost by 31 December 2027, from a June 2026 range of 2.4%–4.0%, as the 2021–22 vintages resolve. Test: Forms 10-K/10-Q. Moderate confidence
3. 3Apollo's perpetual capital will remain above 85% insurance-sourced through 2028, and no other listed manager will reach 50% insurance-sourced permanence over the same period without an acquisition. Test: quarterly perpetual capital composition disclosures. High confidence
4. 4Blended fee rates on new US direct-lending capital will fall by at least 15 basis points from 2025 levels by year-end 2028, driven by mix shift toward insurance separately managed accounts and DC-channel vehicles rather than by headline fee cuts. Test: management fees divided by average fee-earning AUM, disclosed quarterly. Moderate confidence
5. 5Defined-contribution assets in private credit will remain below 1% of US DC plan assets at 31 December 2029, against roughly 0.1% in alternatives overall today — the channel opens, and it opens slowly, constrained by daily valuation and fiduciary liability. Test: DOL Form 5500 aggregates and recordkeeper disclosure. Moderate confidence
6. 6Bank credit exposure to asset managers and funds at the top five US G-SIBs will keep growing faster than their loan books through 2028, and will become a named supervisory focus in a Federal Reserve or OCC publication. Test: 10-Q industry exposure tables; supervisory publications. Moderate confidence

30. Research frontier — what could not be settled from outside

Four questions carry real weight in this analysis and none can be resolved with public disclosure. They are recorded here as open, not as findings.

What the 2021–22 vintages actually recover. KBRA's implied US recovery of roughly 53% and Cliffwater's realised loss history of 0.5% a year describe different worlds, and only completed workouts will settle which one this vintage inhabits. Lincoln's $39.4bn of lender takeovers is the leading edge of the answer, and it will take until 2028 to read.

How much of the valuation gap is scope and how much is judgment. Cliffwater's $0.98 for private loans against $0.96 for public has a defensible explanation — different credit composition, different structural protection, no forced sellers — and an indefensible one. Distinguishing them requires loan-level data on matched borrowers, which exists only inside the managers and the valuation agents. The S&P/Lincoln indices are the first public instrument capable of narrowing this, and they are months old.

The true scale and terms of fund portfolio financing. The FSB describes borrowing-base facilities structured as investments in senior securitisation tranches with equity layers of 30–40%, treated by some banks as securitisations for capital purposes. The OFR states that the standard method for identifying private credit obligors in Y-14 "omits borrowings via SPVs," which is how most BDCs borrow. The gap between the Fed's $95bn and the OFR's $410–540bn is largely this. Neither supervisors nor investors can currently size it precisely, and it is the layer where a genuine leverage chain would form if one forms.

Whether smoothed marks reduce forced selling or merely relocate it. The Bank of England's private markets system-wide exploratory scenario is designed to answer exactly this — how valuation behaviour develops in a stress, and whether lenders, investors and sponsors hold mismatched expectations of what an asset is worth when it must be sold. Interim findings are due later in 2026. Until then the strongest honest statement is the one this study makes in Section 20: the smoothing is real, its economic justification is partly real, and its behavioural consequence in 2026 was a run on the wrapper rather than a run on the credit.

Sources, dated and tiered

Tier 1 — official sector and regulators

Tier 1 — SEC filings, quarter ended 30 June 2026 unless stated

Tier 2 — index, valuation-benchmark and professional analysis

Not verified / not load-bearing

The following items appear in the surrounding commentary on this subject but could not be confirmed against a source opened in the preparation of this brief. None carries any conclusion here, and none is used in a figure.

• Closing market prices and price-to-NAV ratios for listed BDCs on any specific date. Aggregator data showing an average listed-BDC discount of roughly 21.6% against 30 June 2026 NAVs, and per-name figures reported in trade coverage, were not confirmable against a primary source. Only two price-to-NAV ratios are used in this brief and both come from transacted prices disclosed in the funds' own filings — FS KKR at $10.75 against $18.30, and Ares Capital at $18.27 against $19.35.
• Share-price performance of the listed managers. Reports that Blue Owl's shares fell more than two-thirds from their January 2025 peak were not verified against a primary source and are not used. This brief takes no view on any security's price.
• Deployed volume under the Citigroup–Apollo $25bn programme. Neither Citigroup's nor Apollo's second-quarter filings disclose progress separately. The programme's existence and announced size are sourced; its execution is not, and no conclusion rests on it.
• Broad recordkeeper adoption beyond the Great Gray / BlackRock Panorix pilot. Verified movement exists, but broad adoption across major recordkeepers and target-date managers is not yet confirmed. The report now names the pilot it could verify and treats larger rollout as not verified.
• First Brands and Tricolor loss attribution. The $2.3bn of fabricated receivables at First Brands and the named lender exposures circulating in trade coverage were not confirmed against a filing or court document. First Brands is referenced once, as an illustration of asset-based lending's specific failure mode, and no figure from it is used.
• Third-quarter 2026 tender outcomes. Trade reporting suggests early Q3 tenders eased materially — one manager's fund clearing in full below the cap, and the first funds to report showing requests near 4.6% of NAV against 7.9% for the same funds in Q2. None of this was confirmed against a filing before the evidence cutoff. It is referenced once, in the steelman, explicitly labelled as indicative, and no verdict rests on it.
• Q3 2026 developments generally. The evidence cutoff is 30 August 2026. Third-quarter tender results, September filings, and any BoE SWES interim findings published after that date are outside this brief.
• Aggregate private credit AUM of "$1.2tn in vehicle net assets." Used once, in the derivation of the semi-liquid 20% share, and taken from the Federal Reserve's own statement that $241bn represents "about 20 percent of net assets under management in private credit vehicles." The implied denominator is the Fed's, not an independently sourced figure, and is flagged where used.