The Redemption Panic: Why BDC Gates Aren’t Defaults

The Redemption Panic: Why BDC Gates Aren’t Defaults

Non-traded BDCs are capping withdrawals again and the alarm is pointed at the wrong thing. The cap is the most honest sentence in the prospectus. The price it lets you exit at is set by the party you’re exiting, on a quarter’s lag, and cannot be audited from outside — except, it turns out, in one way that nobody markets.

The Headlines Against the Structure

Every few weeks another headline lands: a giant private-credit fund throttles withdrawals, gates investors, limits redemptions. Apollo, Blackstone, Blue Owl, Morgan Stanley, Ares — the biggest names in the asset class, all capping how much money investors can pull. It reads like the opening frames of a run.

The instinct is to assume the loans are going bad, and that instinct is wrong: a redemption cap is a statement about liquidity terms, not about whether borrowers are paying. But the reassurance usually offered in reply — gates are a feature, the credit is fine, ignore the noise — is wrong in the other direction, and it’s the more seductive error because its first half is true.

Both readings miss the same thing. The cap deserves credit rather than fear. The reason people are queuing has almost nothing to do with missed payments and almost everything to do with a suspicion that the net asset value they can redeem at is too high. If that suspicion is right, exiting at NAV is a free option, and the 5% quarterly cap is the only thing standing between that option and the investors who stay. So the cap isn’t what to examine. It’s what makes the examination urgent.

Why the Mark Is the Part I Take Personally

I’ve owned one position whose collateral value was restated on me without explanation, and it’s why this post is about the NAV rather than the gate. A tear sheet put loan-to-value at 4.3% to 5.9%. Fourteen months in, having received nothing, I got an update that set the unproven collateral aside and disclosed a “conservative LTV of 17.8%” — three times the original figure, same assets, same lender, no reconciliation offered. The number was never printed again. What failed first wasn’t the security package; it was the valuation, and the fight over priority afterwards was an argument about the size of the gap.

That was one small, idiosyncratic deal and it is not evidence about Apollo or Blackstone. What it bought me is a narrow reflex: when a number I can’t independently check is the number my exit is priced off, I want to know who produces it and what would make them revise it. That is precisely the position a non-traded BDC investor occupies, at $300 billion of scale, with a quarterly redemption right struck at the manager’s own mark.

The Wave Is Real, and It Grew

Second-quarter 2026 was the first quarter in which all three of the largest non-traded BDCs were visibly capped at once. Industry redemption requests ran roughly $15.6 billion against $5.9 billion actually met, versus $13.9 billion requested and $7.4 billion met in the first quarter — more than $12.7 billion returned across the half-year. Requests rose while payouts fell, so the unfilled backlog widened at both ends. New fundraising collapsed to about $500 million in May, roughly 75% below already-depressed January levels, and the first quarter was the first on record in which non-listed BDC outflows exceeded inflows.

Fund Requests (% of shares) Filled
Blue Owl Technology Income 38.1% ~13%
Blue Owl Credit Income 18.8% ~27%
Apollo Debt Solutions 16.8% ~30%
Cliffwater Corporate Lending 17% ~29%
Ares Strategic Income 14.4% ~35%
Morgan Stanley North Haven 11.6% ~43%
Blackstone BCRED ~10% ~50%
Oaktree Strategic Credit 4.5% 100%

Second-quarter tenders, except Blue Owl’s two figures, disclosed in April. All held a 5% cap except Oaktree, whose demand fell inside it. HPS, Monroe, BlackRock, New Mountain and Fidelity also reported elevated requests.

Two things in that table cut against the run narrative. The request percentages are partly manufactured by the cap itself: when you’re prorated, the rational move is to resubmit in full next quarter whether or not your need changed, so a headline percentage that recycles its own backlog is a queue length rather than a measure of fear. And the money leaving is discriminating — the largest ask by a wide margin belongs to a technology BDC, and at Apollo, onshore investors asked to redeem 4.3% while offshore requests hit 12.5%. That’s capital sorting by sector concentration and distribution channel, not a stampede.

The sponsors call this a rotation rather than a retreat, and because that is also their marketing, I went looking for whether it survives independent arithmetic. It does. Credit strategies raised about $11.9 billion against roughly $12.9 billion of redemptions — a net outflow near $1 billion, down 55% year over year. Over the same span, real-estate and infrastructure vehicles raised about $23.1 billion, up 33%, with infrastructure up 61%. Credit’s share of all alternative fundraising has fallen from more than half to roughly a third. Money is moving within alternatives, which is a materially different phenomenon from investors concluding private credit is unsound.

The Cap Is a Feature — and Sponsors Keep Declining to Use It

Non-traded BDCs are semi-liquid by construction: illiquid private loans on one side, capped quarterly repurchases at NAV on the other. The 5% cap is not an emergency lever; it’s a day-one contractual term that exists so a manager is never forced to dump loans at distressed prices to fund exits. The plumbing behind it is also better than the fragility literature implies. Refinancing and prepayment activity has historically run 25–30% a year and its worst reading since 2005 was still 12.8%, more than two quarters of redemptions. Roughly half of dividends get reinvested, worth about 1.25% a quarter. Cliffwater’s fund redeemed 12% across two windows this spring while running positive net cash flow, funded by subscriptions, reinvestment and maturities rather than sales.

So far this is the sponsors’ argument too, and arriving at the same place as the people selling the product is a reason to press harder rather than relax. Pressing turns up the part that undercuts it, and it isn’t isolated. When average redemptions across perpetual non-traded BDCs jumped to 4.8% of NAV in the fourth quarter of 2025 from 1.6% the prior quarter, five funds paid out above the standard 5% cap rather than prorate. BCRED then did it at scale, raising its cap to 7% in the first quarter and filling every request, funded by $400 million of fresh money from Blackstone and its employees, before returning to 5% the following quarter. Oaktree’s fund needed an affiliate purchase to clear over-cap demand in that same quarter.

That is the real weakness in “the cap protects you,” and it’s the opposite of the weakness the headlines describe. A shock absorber that bends whenever the sponsor’s brand is exposed is discretionary, not structural. The failure mode worth worrying about isn’t a manager enforcing the cap — it’s a manager who won’t, funding over-cap quarters from house capital until either the goodwill or the capital runs out. Blackstone spent $400 million to keep a gate from looking like a gate. Proration cuts the same way: because the queue rolls forward, being gated is itself a reason to join it, which is the reflexive loop the cap exists to prevent.

The counter-signal deserves equal weight, because it arrived in the same data. Oaktree’s second-quarter tender came in at 4.5% — under the cap, no proration, its first clean quarter since late 2025, at the fund that had needed help to clear the previous one. Whatever is happening here is not uniform across sponsors, and demand falling back inside a cap is the most encouraging single fact available.

The Credit Data Turned, and the Average Conceals Everything

The reassuring version of this story leans on a non-accrual rate around 1.4%. That was the fourth-quarter 2025 print and it no longer holds. Across a universe of 174 BDCs, non-accrual debt reached $9.98 billion at cost in the first quarter of 2026 — 2.01% of reported debt investments, up 40% from $7.12 billion and 1.45% the prior quarter. Normalise across every lender holding the same borrower, counting each tranche, and exposure rises to $16.04 billion, or 3.24%, about 60% higher than reported. Measured at fair value rather than cost, the normalised rate is more than double the reported figure. Separately, 32 borrowers representing $1.3 billion of principal switched some or all interest to PIK after paying all cash the prior quarter.

A falling PIK share tends to get read as de-risking. Cash-to-PIK conversion at that pace supports the opposite reading: a borrower that stops paying interest in cash is not a borrower getting healthier.

But the aggregate is close to useless, and this is the finding that should reorganise how anyone reads these numbers. The deterioration is not general. In the first quarter, large-cap BDC non-accrual rates jumped 72% and software-heavy funds more than doubled, while lower-middle-market funds held firm — fair-value marks flat sequentially at 99.2%, with net equity inflows in the quarter. Much of the sector’s pain traces to a handful of large-cap and software names. That concentration is the whole point: a single sector average misleads in both directions, and the fund-by-fund view is the only one that tells an allocator anything about their own book.

The Circularity Problem, and the One Way Around It

Here is the difficulty with every figure in the section above, and it disqualifies most of the metrics this debate runs on. Non-accrual status, PIK share and fair-value marks are all computed by the manager, off valuations the manager sets, on a quarterly lag. Using them to audit whether the manager’s NAV is honest is circular. They tell you what the manager has already conceded. They cannot tell you what the manager hasn’t.

There is one escape from that circle, and it’s the most useful thing I found in this data. Take the manager’s own marks — not as a valuation, which is the contested part, but as an input — and ask empirically what happens next to loans carried at particular prices. Someone has now done that across the sector:

Loan carried at Falls into non-accrual within 15 months
90 to 95 — a level the market treats as routine 13.8%
Below 90 20.2%

That converts a mark into a forecast the manager isn’t making. A loan at 92 is not being described as impaired by anyone, and roughly one in seven of them is in non-accrual inside five quarters. Below 90 it’s one in five. And non-accrual is close to terminal: those loans rarely leave the category without a restructuring.

Two numbers make that actionable. Loans priced below 90 were 8.4% of all BDC loans at the end of the first quarter. And in that single quarter, 135 borrowers had roughly $9 billion of principal marked down by more than 15%, with the top twenty names accounting for two-thirds of it. Apply the observed transition rate to the sub-90 population and you get a forward estimate of non-accruals that doesn’t depend on any manager agreeing with you — which is exactly the property every other metric here lacks.

The Real Question Is the Mark

Non-traded BDCs repurchase shares at NAV. Everywhere else the same kind of asset gets a second opinion, and the second opinion is consistently lower.

How the same broad asset gets priced Where it lands Who sets it
Non-traded BDC shares NAV, i.e. par The manager
Listed BDC shares 15–20% below NAV (14.7% average, 20.4% median) A daily auction
Secondary stakes in private credit funds 20–35% below stated NAV Saba and other bidders
Software loans in BDC books ~97% of par on average The manager
Software loans in the syndicated market ~85 cents, against ~96 for non-software Traders

None of those gaps proves the private marks are wrong. Different instruments, different liquidity, different borrower size, and listed BDC share prices carry leverage and manager quality that a NAV does not. But they establish something narrower and sufficient: the NAV is an opinion, and every party without a stake in it holds a lower one.

That matters because of what redeeming at an opinion does. If reported NAV sits above realisable value, every share repurchased at NAV hands the exiting investor more than their share of the fund and leaves the difference with the people who stayed. A June 2026 study of 59 semi-liquid BDCs names the mechanism directly: stale or inflated marks make redemptions a wealth transfer from remaining investors to redeeming ones, which gives everyone a reason to go first. That is a rational sequence rather than a panic, and it explains the queue far better than any story about borrowers missing payments. It also explains how Boaz Weinstein can bid 20–35% below stated NAV for fund stakes while saying he’d happily get in line and tender at NAV: at that spread, the queue is the trade.

Two pieces of evidence push this past theory. The same loan gets different numbers: PIMCO finds that for credits held by multiple BDCs, the spread between the most conservative and most optimistic mark on one loan has widened materially, and that price dispersion inside BDC portfolios has run an order of magnitude tighter than the syndicated market since 2021 — an implausibly narrow range for comparable credit risk. 9fin now sells a product whose entire function is showing how a single credit is marked across every BDC holding it. When an industry needs a tool to reconcile its own valuations, those valuations are not measurements. And the consequences aren’t hypothetical: BlackRock’s TCP Capital disclosed roughly 24% in cumulative NAV markdowns year-to-date, alongside a federal inquiry into its valuation practices and the departure of its chief executive. That is a listed, professionally managed BDC conceding in real time that its prior marks were off by a quarter of the fund.

The Case That I’m Wrong

The strongest rebuttal is empirical and runs the other way. If private marks were systematically too aggressive, realised losses would exceed the unrealised markdowns that preceded them. Historically the reverse holds: BDC unrealised losses have run roughly 2x realised losses, which says these managers have been too conservative, and that investors who redeemed to dodge future losses were the ones paying for the exit. Long-run credit losses in the Cliffwater Direct Lending Index sit near 1%, on a book that is overwhelmingly senior, secured and modestly levered. Perpetual BDC leverage was below prior levels at the end of March, cutting against the claim that funds are levering up to fund exits.

The sector I’ve singled out also has the better track record, not the worse one. Software assets in BDC books have historically produced lower loss rates than the average holding — roughly 50 basis points of non-accruals against about 135 across all sectors — and the BIS finds fewer than 1% of software loans behind on payments. Even the bearish end of the forecast range is careful: direct-lending default rates running near 5.6% could reach 8% against a 2–2.5% historical average, which the analysts making that call describe as significant but not systemic. And there’s precedent for the whole episode resolving quietly, since non-traded REITs went through an almost identical redemption-and-fundraising squeeze in 2022–23 and stabilised.

Taken together: the dilution mechanism is real and the incentive to go first is real, but the direction of the mark error is genuinely unresolved. Anyone telling you confidently that non-traded NAVs are 20% too high is doing what I did in 2018 — reading a number they cannot audit and treating the reading as knowledge.

Software: Paid Less to Carry More

If the marks and the fundamentals collide anywhere, it’s software, which is now approximately 32% of BDC investments at both cost and fair value on a bottom-up read of 175 portfolios. Definitions matter here — depending on where you draw the sector line the estimate has run from the mid-teens upward — but a third of the asset class is a different proposition from a fifth.

The finding that should bother a lender is about pricing rather than defaults. Through late 2025, as AI disruption risk climbed, spreads on software loans did not rise to meet it. They compressed and converged with non-software spreads, fastest on new issues, and were lowest of all at non-listed BDCs, the segment under the least market scrutiny. Lenders are being paid less to carry a risk that has grown. The exposure is also doubly concentrated: about 60% of software lending goes to firms borrowing from seven or more BDCs, up from under 10% in 2015, while the five largest BDCs hold roughly 37% of all software principal. A software-specific shock would surface on many balance sheets at once and land hardest on a few.

The reckoning is deferred rather than resolved. Roughly 35% of software principal still matures in 2028 and 2029, which is how marks stay near par while syndicated equivalents trade at 85 cents. Two signals inside the recent data point in opposite directions and both deserve weight: software cash-to-PIK conversions actually declined in the first quarter, which is genuine good news, while lenders reported sharper drops in software equity pricing — plausibly a leading indicator of loan write-downs to come, since the equity absorbs damage first.

But “software” is not one credit, and this is where lazy analysis fails in both directions. The per-borrower question is easy to state and hard to answer: is AI displacing this company’s product, or is the company deploying AI to entrench itself? Mission-critical enterprise software with deep integrations and high switching costs is a fundamentally different risk from a commoditised point solution a newer tool can leapfrog. About three-quarters of BDC software exposure sits in horizontal productivity, application and automation tools — the categories where substitution is most plausible. That is a more useful fact than any headline exposure percentage.

Conditions for new lending, meanwhile, are the best in years. With retail capital sidelined and software issuance collapsing from roughly 30% of new syndicated loan volume at end-2025 to about 9% in 2026, competition is thinner and lenders are being paid again: new-issue spreads widened around 25 basis points after years of compression, senior secured first-lien paper yields roughly 9–11%, covenants are tightening, and healthcare has displaced software as the leading destination for new direct-lending dollars. The asset class isn’t good or bad right now. It’s separating, along sector mix, vintage and PIK reliance, and that rewards the manager rather than the wrapper.

Would I Buy One Here? No — and Not Because of the Cap

I’m passing, but the popular reason for passing is the wrong one. The cap isn’t disqualifying; it’s the most honest sentence in the prospectus. What disqualifies it for me is that the price at which I’d exit is set by the party I’d be exiting, on a quarter’s lag, and I have already owned one position whose collateral turned out to be worth whatever the last memo said. I don’t need to believe non-traded NAVs are inflated. I only need to notice that I couldn’t tell if they were, and that everyone bidding for the same assets in an open market holds a lower opinion.

The tripwires usually offered for this asset class are non-accruals climbing toward some threshold, PIK rising as a share of income, and software markdowns turning into realised losses. Two problems with that list: a 6% non-accrual trigger is no longer the distant marker it reads as, at 2.01% reported and 3.24% adjusted, and all three are manager-produced, which makes them useless for the question that actually matters. What I watch instead:

  1. The sub-90 population, and the transition rate against it. One in five loans marked below 90 reaches non-accrual within fifteen months, and one in seven of those marked 90–95 does. Sub-90 was 8.4% of loans last quarter. This is the only forward-looking measure here that doesn’t require the mark-setter to agree with the conclusion.
  2. Realised losses overtaking unrealised. The cleanest falsifiable test available. Unrealised at roughly 2x realised means marks have been conservative; if that ratio inverts and holds, the marks were fiction and the redeemers were right.
  3. Cross-BDC dispersion on the same loan. When several funds carry one credit at materially different values, the number is an opinion — and the widening high-to-low spread is now observable, and sold as a product.
  4. A sponsor that stops supporting its own cap. Over-cap quarters funded with house money or affiliate purchases are now common enough to be the norm rather than the signal. The informative event is the reverse: a large fund letting a deep proration stand, or announcing that support is being withdrawn. That’s when the discretionary cushion is gone and the contract is all that’s left.
  5. Which side of the NAV gap closes. Either the listed discount narrows toward private marks or private marks fall toward the discount. That convergence is the referee.

And the fund-level questions, unchanged: what kind of software, how much 2020–22 vintage, what the liquidity actually costs, whether the book is large-cap or lower-middle-market — because those two cohorts just posted opposite quarters — and the one I’ll never skip again, what the collateral is really worth as opposed to what it’s carried at.

Bottom Line

The redemption wave is a valuation dispute wearing the costume of a liquidity crisis. The caps are the structure working; the request percentages partly recycle their own backlog; the money leaving is rotating into real assets rather than fleeing alternatives. The credit data has genuinely deteriorated — non-accruals up 40% in a quarter, adjusted rates above 3%, cash converting to PIK — and it still doesn’t describe borrowers failing to pay en masse. What it describes is a widening gap between books, with large-cap and software-heavy funds absorbing nearly all of the damage while lower-middle-market marks sat flat at 99.2%.

What’s left is a $300 billion asset class where the exit price is an opinion published by the party on the other side of your trade, and where the sponsors best placed to enforce that price have repeatedly chosen not to. Five funds paid over their caps in a single quarter; the largest spent $400 million to avoid using its own. That is survivable while the queue is a rotation and the marks are roughly right. Both of those are assumptions. Only one of them can be tested from outside, and the way to test it is to stop reading what managers have conceded and start counting the loans they’ve quietly marked into the nineties.

Sources

Commentary and personal experience — not investment, legal, or tax advice. Investing carries risk, including total loss of capital. Always do your own due diligence.