Pantheon paid 95 cents on the carrying value for a diversified BDC loan book. The stock market says those same loans are worth 61 cents. One of them is wrong by a factor of six.
The Dow fell roughly 700 points as the Treasury's buyback program failed to hold the long end — Secretary Bessent said the operation could exceed $4 billion and that Treasury would "make a market" in long-dated paper, and yields snapped back toward 20-month highs anyway. Walmart cratered 9.2% to $103.84 on soft guidance. Risk was for sale everywhere.
Everywhere except the one place you would expect it to be sold hardest.
| Tier | Ticker | Close (Aug 20) | Change |
|---|---|---|---|
| The fee stream (general partners) | OWL — Blue Owl | $11.40 | −3.14% |
| KKR | $107.02 | −3.11% | |
| APO — Apollo | $129.98 | −2.73% | |
| BX — Blackstone | $141.35 | −2.58% | |
| ARES | $140.23 | −2.01% | |
| The balance sheet (BDCs that actually hold the loans) | BIZD — BDC ETF | $13.29 | +0.23% |
| BXSL | $24.67 | +0.45% | |
| MAIN | $58.40 | +0.40% | |
| HTGC | $17.03 | +0.77% | |
| ARCC | $19.78 | −0.10% | |
| OBDC | $11.28 | −0.44% | |
| TCPC | $4.03 | −1.47% | |
| Benchmarks | SPY / QQQ / IWM / XLF | 762.60 / 710.93 / 297.67 / 56.95 | −0.84% / −0.72% / −1.34% / −0.92% |
The entities that hold the credit risk finished flat to green. The entities that merely collect fees on it fell two to four times the market. That is not how a credit scare normally trades.
My read (opinion): it is backwards because the BDCs already de-rated and the GPs did not. The Boston Fed measured it: the S&P BDC Index fell roughly 20% relative to the S&P 500 from early 2025 to mid-2026. Eighteen months of repricing is already in BDC equity. None of it is in the general partners, who are still valued on AUM compounding. Today the market began pricing the second one. The risk in private credit has migrated from the loan book to the fee stream, and that migration is about 5% complete.
Every private credit debate for a year has hit the same wall: the loans do not trade, so the marks cannot be falsified. On August 4 that changed. Per TCPC's own Q2 release (read the "Portfolio Sale Transaction Highlights" and "Recent Developments" sections):
Read that carefully, because both sides of the argument are inside it. The bears are right that marks are stale — a real buyer paid a 5% discount to a mark that was seven months old, and it still forced a double-digit NAV write-down. The bulls are right that the assets are money-good — a sophisticated secondaries buyer with full diligence access paid ~90–95 cents on carrying value for a diversified, mostly-first-lien middle-market book.
Management's own framing was the giveaway. CEO Phil Tseng said the sale realized "a substantial premium relative to the value implied by the Company's current share price." He is right, and that sentence is an admission that the equity market and the transaction market are pricing the same loans at wildly different levels.
If the loans were impaired, BDC creditors would know first. They are doing the opposite. In the week of August 17, Blue Owl Technology Finance priced $400 million against a $200 million target (a tap of its 6.500% 2029s) and Blackstone's BCRED raised $750 million against roughly $500 million. Blue Owl Capital sold $750 million the prior week into as much as $3.3 billion of demand. Earlier this year PIMCO bought the entire $400 million of OBDC's 6.4% 2028 notes. BCRED upsized $850 million from $500 million in April.
PIMCO's Lotfi Karoui put the split plainly in a recent note: BDC bonds have recovered most of their underperformance while BDC equities keep signaling doubt about reported NAVs. Both cannot be right. Bondholders sit senior, see the same tape, and are oversubscribed 1.5–4x at 6.4–6.5%. Equity holders are demanding a 21–39% discount to book. When senior and junior claims on identical collateral disagree this violently, the junior claim is usually the one mispriced — because it is the one owned by retail.
None of the above means the credit is clean. It is not.
| Metric | Reading | Source / date |
|---|---|---|
| US private credit default rate | Record 6.0%, and a new high again in 2Q26 | Fitch, via LSTA (Jun 15 & Jul 30, 2026) |
| Defaults at Ares / Blackstone / Blue Owl / Golub funds | Highest since at least 2021; Blue Owl flagship 2.8% in Q2 | WSJ analysis via InvestmentNews, Aug 10 |
| PIK share of BDC loans | ~6% (2022) → ~10% (early 2026), +67% in three years | Boston Fed CPP 26-6, Aug 5 |
| BDC lending spreads | Compressed ~1 percentage point in two years, to 4–5pp over SOFR | Boston Fed CPP 26-6 |
| Non-traded BDC redemption requests (top 10) | 13% of assets in Q1, 14% in Q2 2026 — most managers gated | With Intelligence, via InvestmentNews |
| Software/internet share of median BDC book | ~20% by loan count; interquartile range 15–30%, some >33% | Boston Fed CPP 26-6 |
The Boston Fed paper is the single best document in this debate and almost nobody has read it. Its central finding is a contradiction, stated plainly by the authors: PIK usage is rising while spreads are compressing. If borrowers increasingly cannot pay cash interest, lenders should demand more compensation. They are accepting less. The paper offers two explanations — implicit restructuring (cut the coupon to avoid the default) or simple competitive erosion of standards. Both are bearish. Meanwhile the ratio of reported fair value to cost sits stubbornly near 1.0.
The paper also hands you a tradable coefficient, which is rare in central bank research: a one-standard-deviation decline in a BDC's fair-value ratio predicts roughly 50 basis points of negative abnormal equity return the following quarter. The marks are soft, but they are not noise — they carry information, with a lag. Screen on the delta, not the level.
One thing the bears overstate: contagion to banks. Committed bank credit to BDCs grew from ~$10bn (2013) to >$50bn (2025), with ~$35bn drawn — but that is under 2% of large-bank Tier 1 capital, and banks sit senior secured. This does not break the banking system. It breaks pension LPs, PE-owned insurers, and retail BDC holders. Which is exactly where the losses are designed to land.
Every analyst is watching software concentration. That is the front door, and it is well guarded — everyone has counted it (~20% of the median book).
Here is what happened this morning, and I have seen no one connect it. Wingspire Equipment Finance closed a $140 million equipment financing with an unnamed private-equity-backed GPU cloud provider, funding high-density GPU servers. The borrower sells compute to "AI laboratories, enterprise customers and public-sector organizations." Wingspire Capital is a portfolio company of Blue Owl Capital Corporation (OBDC) — a listed BDC whose shares closed today at 0.79x book.
"AI infrastructure requires capital partners that understand both the equipment and the pace of the market. These are large, capital-intensive investments that require substantial lending capacity and structuring expertise." — Spencer Jakemer, VP, Wingspire Equipment Finance
Now overlay what Chicago Booth's Stefan Hepp published on August 7, drawn from his July 2026 working paper "The AI Infrastructure Debt Complex":
"There are even loans secured by GPUs, with the chips that run AI themselves pledged as collateral... But while most houses hold their value for decades, GPUs are being displaced every two or three years as faster ones are developed, and nobody yet knows what a used one is worth."
Hepp sizes specialist compute providers' GPU-secured borrowing at roughly $35 billion, inside an estimated ~$200 billion of private credit exposure to AI infrastructure. He stress-tests the whole complex at $60–140 billion of realized credit loss against a $10–14 trillion equity re-rating — and finds that first-loss sits outside the regulated banking system, landing on "private-credit funds and dedicated AI-infrastructure credit vehicles... limited partners in those funds — principally pension plans and endowments — with private equity–owned insurance platforms, and, to a lesser extent, with private investors of business-development companies."
Quantifying the Wingspire deal (my calculation): $140m against OBDC's $14.955bn portfolio is ~0.94% — immaterial today. But price the residual risk. Industry data cited by Hepp and by Sascha Steffen's August 14 analysis puts H100 three-year residuals anywhere from 50–70% of value (providers) to below 30% (critics). On a $140m book that is a $42m–$98m loss range in a severe scenario, or roughly 0.3–0.7% of OBDC's $14.26 NAV. Survivable alone. The point is the precedent and the trajectory, not this ticket.
Steffen's note on Nvidia's August 10 MOUs with Apollo, BlackRock, Blackstone, Brookfield, Goldman and KKR to mobilize $500bn+ contains the sharpest observation of the month. Nvidia's platform is marketed as resolving the circularity of a chipmaker financing its own customers. But: three of the six platform managers own or control the insurers likely to hold the resulting paper — Apollo owns Athene, KKR owns Global Atlantic, Brookfield has Wealth Solutions. "One layer down, a second circularity has been created." The entity underwriting the credit and the entity bearing it share a shareholder.
Steffen's structural point on why the ratings work: what makes CoreWeave's $8.5bn A3-rated facility investment grade "is not the chips; it is a take-or-pay contract under which an investment-grade hyperscaler owes the payments whether or not it uses the capacity." The rating describes the payer, not the collateral. Where the payer is Microsoft or Meta, that substitution is legitimate. Where the payer is an AI lab still raising the money it has already committed, "it is not credit substitution; it is credit deferral." Oracle sits in between: ~half of a $638bn backlog attributed to OpenAI, FY26 free cash flow of −$23.7bn, downgraded by S&P to BBB− in July 2026 — one notch above high yield.
Connect three facts that are usually reported separately.
One: Alphabet, Amazon, Meta and Oracle have issued ~$223 billion of bonds year-to-date as of August 20 per LSEG — more than double all of 2025. Two: Evercore ISI's Krishna Guha argued this week that "high quality hyperscaler debt is a close competitor for government debt and is also hedged in sovereign bond markets... this issuance is increasingly global across currencies, helping to explain the global nature of the move in yields." Three: Boston Fed: the overwhelming majority of BDC loans are floating-rate, and "companies that could comfortably service their debt at SOFR plus 5 percentage points when SOFR was near zero face a substantially greater burden when SOFR is higher than 4 percent."
So the chain is: AI capex → record corporate bond supply → higher long yields and a stickier front end → heavier debt service on middle-market BDC borrowers → more PIK → more non-accruals. The AI buildout is directly raising the cost of capital for the small companies BDCs lend to. Nvidia's capex boom is a credit event generator for Main Street borrowers who have never touched a GPU.
It runs the other way too, and this is the level that matters most for the trade. OBDC's weighted average portfolio yield is 9.9%. The 30-year Treasury touched a 19-year high of 5.33% on Tuesday. That is a risk premium of roughly 4.6 percentage points for illiquid, unrated, PIK-inflected middle-market credit — the thinnest of this cycle, and consistent with the Boston Fed's measured 1pp of spread compression. Meanwhile investment-grade hyperscaler paper pays 4.75–8%, with Morgan Stanley IM's Vishal Khanduja pegging long-end IG AI leaders near 6.5%.
Price: $11.28 (Aug 20 close, −0.44%) · NAV: $14.26 (6/30/26) · 0.79x book · 11.2% dividend yield
Entry zone: $10.80 – $11.50
Target: $13.20 (0.93x book — the Pantheon clearing level applied to OBDC's book, less a liquidity haircut). Stretch $14.00.
Invalidation: non-accruals at cost rising above 5.0% in the Q3 print (from 2.8% today), or a daily close below $10.52 (the April 2, 2026 52-week low).
Timeframe: 1–3 months to $12.50; 6–12 months to $13.20+.
Why (two independent signals): (1) Pantheon's arm's-length print at 95% of carrying value on comparable first-lien collateral implies a ~5% mark gap, not the 21% OBDC's stock demands. (2) PIMCO bought 100% of OBDC's $400m 6.4% 2028 notes and BDC IG deals are printing 1.5–4x oversubscribed — the senior claim is bidding aggressively while the junior claim trades a fifth below book.
Third signal — insiders: the transaction record shows 13 purchases and zero sales, including CEO Craig Packer buying 41,600 shares at $11.75 and director Logan Nicholson buying repeatedly, most recently on June 2, 2026 at $11.31 — above today's price.
Per the Q2 primary filing (Exhibit 99.1 to the 8-K filed Aug 5): NII $0.36/sh, adjusted NII $0.34, dividend $0.33, non-accruals 0.8% at fair value / 2.8% at cost, net leverage 1.11x, portfolio $14.96bn, weighted average yield 9.9%. CEO Packer: "credit performance continued to track in line with expectations."
The real risk, quantified — it is rates, not credit. Dividend coverage is $0.34 adjusted NII against a $0.33 dividend: 103%. Roughly 94% of the book floats. A 100bp decline in SOFR removes on the order of $0.05–0.07 per share per quarter of net investment income after the offsetting benefit on floating-rate liabilities. Against one cent of cushion, that forces a dividend cut — the same mechanism that already hit BXSL (coverage 97%) and prompted OBDC's own 16% cut earlier this year. The bear case for OBDC is Fed easing, not defaults. That is an unusual and underappreciated asymmetry: the macro event most people think would help BDCs is the one that cuts their dividend.
Price: $4.03 (−1.47%) · Stated NAV: $6.58 · Pro-forma NAV ~$5.90 after the disclosed 10.4%/$0.68 transaction hit · 0.68x pro-forma book
Entry zone: $3.85 – $4.15
Target: $5.00 (0.85x pro-forma NAV)
Invalidation: the KBW strategic review concludes with "reinvest the new leverage capacity" and no capital return announced by the Q4 print; or a dividend cut below $0.12/quarter.
Timeframe: 1–3 months (review outcome is the catalyst).
Why: this is the only name in the group where a board has explicitly put liquidation on the table while the stock trades at two-thirds of book. TCPC's release lists the options as reinvesting, share repurchases, "potential strategic combinations," or "completing an orderly realization of portfolio assets." At 0.68x book, orderly realization is the bullish outcome. Post-deal the balance sheet is pristine: net leverage 1.38x → ~0.4x → below 0.3x, unfunded commitments under $40m, $157.5m cash at 6/30 plus ~$152m of gross proceeds, and a $50m buyback authorized through April 2027 of which only $2.87m has been used (at an average $4.34).
Why only SPECULATIVE — the credit is genuinely bad. Non-accruals are 1.6% at fair value but 7.4% at cost, a 4.6x gap that tells you how much has already been written down. Revenue −27.4% year over year. Trailing ROE −19.08%. And the catch nobody is modelling: TCPC sold 48% of its debt portfolio and delevered to 0.4x, so net investment income falls hard from here — the current $0.17 dividend was covered 129% by Q2's $0.22 NII on a balance sheet that no longer exists. Assume a cut. You are buying the balance sheet and the catalyst, not the income.
Structure: long OBDC $11.28 against short OWL $11.40, roughly dollar-neutral.
Entry: OWL short $11.40 – $12.30
Target: 15–20 points of spread outperformance; OWL standalone objective $9.50
Invalidation: OWL weekly close above $13.00, or top-10 non-traded BDC redemption requests falling below 8% of assets in the Q3 With Intelligence data (from 14% in Q2).
Timeframe: 1–3 months.
Why: Blue Owl is the single most concentrated expression of every strand of this story — it gated a retail fund in February; its flagship fund's defaulted loans hit a five-year-high 2.8% in Q2 per the WSJ analysis; it is one of the six Nvidia platform MOU partners; it holds 80% of the Meta Hyperion JV; and its OBDC portfolio company just underwrote GPU collateral. Its fee stream is levered to AUM in a retail channel running 13–14% quarterly redemption requests. You are long the assets at 0.79x book and short the fees at ~14x distributable earnings on the same underlying loans.
Honest caveat: OWL has rallied ~25% and the sell-side is warm on it (real-assets revenue +44%, AUM $319bn as of 6/30). The short leg is the riskier half — treat the pair as the position, not OWL outright. Today's 3.7x-beta decline is one day of evidence, not a trend.
Price: $24.67 (+0.45%). Dividend coverage has fallen below 100% (≈97%) and NAV has declined five consecutive quarters. Q2 adjusted EPS $0.75 beat $0.65 but revenue missed at $320.5m vs $327.8m — a beat on the line that does not matter and a miss on the one that does. Management has, in the phrasing of multiple sell-side notes, telegraphed the reset.
Trigger to act: the actual cut announcement. History in this group (OBDC cut 16% and its coverage went to 116%; the stock bottomed afterward) says the cut is the buy signal, not the sell signal — the de-rating happens in anticipation. Set an alert; do not front-run it. Buy zone on announcement: $21.50–23.00. Do not chase above $25.50.
| Scenario | Prob. | Specific trigger conditions | Where it lands |
|---|---|---|---|
| Bull | 30% | 30Y sustainably back below 5.00%; Q3 non-accruals flat sector-wide; BDC IG issuance stays 1.5x+ oversubscribed; TCPC review returns capital | Discounts narrow from ~21% to ~10%. OBDC $13.20+, TCPC $5.00, BIZD +12–15% |
| Base | 50% | 30Y grinds 5.15–5.40%; non-accruals at cost drift up 50–100bp; two to three more dividend cuts (BXSL first); no AI-credit event | Discounts persist 15–25%. OBDC $11.00–12.50; total return is the 11.2% coupon. Nothing breaks, nothing re-rates |
| Bear | 20% | An AI-credit event: a neocloud misses a payment, or Oracle (BBB−) is cut to high yield, or a rating agency changes data-center residual-value methodology (Fitch has an open consultation). Insurance holders of A-rated SPV paper face higher capital charges and become forced sellers into a market with almost no secondary liquidity | Marks finally converge to transaction prices. OBDC ~$9, TCPC ~$3, OWL ~$8. GPU collateral gets a public print |
Note the shape of that distribution. The bear case is not "middle-market borrowers default." It is "an AI financing structure breaks and drags the marks with it." Private credit's fate in the next six months is more dependent on Oracle's credit rating and GPU residual values than on the health of the mid-sized American companies it nominally finances. That is a genuinely new fact about this asset class, and it is roughly two years old.