The illusion that we live in a bull market is dangerously misplaced. Yes, the market goes up. But that does not mean the underlying market structure is robust. Rising asset prices often hide deep structural weaknesses in a financial market. A bull market can persist even when the underlying foundations are fragile.

What investors believe as “robust liquidity” is, in reality, an unmapped mutation of systemic technology risk shifting into the shadow banking sector. Capital allocation is moving deeper into an opaque, multi-trillion-dollar credit ecosystem. This tectonic shift represents a gray swan risk that is increasingly vulnerable to becoming a full-blown systemic event.

By moving highly interconnected financial liabilities out of public/regulators’ view, this creates a heavily leveraged architecture where a single, unforeseen market downturn could trigger a systemic collapse.

The Four Horsemen of this economic season are: AI financing, private equity, private credit, and the insurance balance sheets.

The Private Equity Zombies

Private equity firms and their portfolio companies are facing more than $330 billion of software and technology debt coming due through 2028 across high yield, leveraged loans and BDC-held loans, of which roughly $142 billion falls in 2028 alone. The primary entities hitting this refinancing wall are not the PE funds themselves, but rather the mega-leveraged software-as-a-service (SaaS) and cybersecurity portfolio companies acquired during the peak valuation era.

To give some color, across the United States, 4,600 PE-backed companies have been held for five years or more, and general partners are sitting on more than $860 billion in buyout net asset value tucked inside funds older than seven years, according to PitchBook’s Kyle Walters. The industry has a name for these aging, cash-flow-positive but unsellable holdings: zombies. They are not DEAD. They are not profitable either, and it is becoming very expensive for the investors.

That is what it looks like from a distance. Up close, in the loan books we go through, one name kept coming up.

Thoma Bravo Has Software-Credit Cooties

Thoma Bravo’s name keeps turning up in the private credit loan books we are going through. It is interesting that nearly three of its companies have hit the same maturity wall in four months. Let us take Medallia, the company went to its lenders in April. Then, we have Proofpoint - the company was extended in July at 150 basis points and 40+ documentation changes. Recently, Sophos. Lenders denied Sophos refinancing in August. Thoma Bravo has nearly $9 billion of portfolio debt maturing by the end of 2028.

Most of Thoma Bravo's other credits are direct loans, and those only surface when a BDC files a schedule of investments. The BIS puts BDCs at about a fifth of US direct lending. The rest has maturity dates that appear in no public filing at all. Imagine that…

Who is Thoma Bravo?

Well, Thoma Bravo is the world’s largest software-focused private equity firm. It manages over $172 billion in assets and acts as a massive corporate consolidator that buys out, restructures, and scales enterprise software and cybersecurity companies.

What they do?

Thoma Bravo is famous for its "buy-and-build" strategy. Instead of investing in small, risky startups, they execute massive take-private buyouts of established, public tech companies that have stable recurring revenue but need operational fixing.

What is all the Hoopla?

In mid-2026, Thoma Bravo handed control of the customer experience software company Medallia directly to its lenders (including Apollo, Blackstone, and FS KKR), completely forfeiting its $5 billion equity investment.

What happened?

The debt had outgrown the company: The loan started at $1.8 billion in 2021, underwritten against recurring revenue rather than earnings. It had a PIK toggle, so Medallia could pay interest in more debt instead of cash, and it permitted add-ons for acquisitions. By the handover, the balance was roughly $2.8 billion. A billion dollars of growth in the claim without the business getting better. Yes, that is a problem.

Blackstone-Effect: Blackstone declined to extend in late 2025 and the PIK toggle expired. That converts the position from deferrable to cash-pay, and there was no refinancing available. Sophos is the proof of that: a better credit, offered 225 basis points over the existing coupon, and private credit still passed. More on that below.

By this time, Thoma Bravo had two options - inject equity or handover the company. Thoma Bravo had said in March that it overestimated the company's growth prospects and paid too much and It chose to handover.

Now you have the backstory, this week, Thoma Bravo did it again.

Behold, Sophos.

Who is Sophos?

Sophos, a major UK-based cybersecurity firm purchased by Thoma Bravo for $3.9 billion in 2020, faces a $2.1 billion to $2.5 billion debt maturity wall coming due in March 2027. When Thoma Bravo went to private credit markets to refinance this debt, offering significantly higher interest rates (yields) to sweeten the deal, multiple major lenders walked away and flatly refused to fund it. This was a massive shift from previous years when lenders fiercely competed to fund software firms.

Why though? A.I is Amazing! No?

The primary reason lenders passed on the deal is a growing fear across Wall Street referred to as "AI displacement risk". Sophos primarily sells endpoint and network security software to businesses. Credit investors are increasingly terrified that traditional, subscription-based software companies will see their business models cannibalized or completely replaced by rapid advancements in AI. Lenders are unwilling to lock themselves into multi-billion dollar, long-term loans for companies facing this kind of tech displacement.

Adding to the drama, Moody's Ratings downgraded Sophos' credit rating to B3, while S&P Global Ratings revised its outlook to negative. Ratings agencies warned that if Sophos couldn't resolve its massive debt maturities quickly, a further credit downgrade would follow. This happened shortly after Sophos pulled off a major $859 million acquisition of Secureworks, which triggered later workforce layoffs.

Well, with private credit markets effectively locked, Thoma Bravo hired Goldman Sachs to negotiate directly with existing lenders to "amend and extend" the current loans. However, this triggered an aggressive standoff.

Creditors want Thoma Bravo to inject fresh equity capital into Sophos, pay off a portion of the existing debt immediately, and accept highly restrictive loan covenants.

Now, Thoma Bravo and its portfolio company Sophos stems from a massive, high-stakes battle in the credit markets over $2.5 billion in looming debt, driven largely by investor anxiety over artificial intelligence (AI) disruption had two ways to keep its lenders happy on Sophos. Put more equity in or give up contract rights.

Thoma Bravo said outright it wasn’t putting equity in. Sophos went to existing lenders after private credit passed on a $2.5 billion refinancing, and the sponsor gave up documentation rights while stating it does not intend to inject additional capital.

Thoma Bravo owns Proofpoint too, and in July it paid the same way there. It extended $4.3 billion of a $5.035 billion term loan by two years and agreed to about forty changes to the documentation. The company gave up the right to move collateral beyond lenders’ reach and the right to buy back its own debt in privately negotiated deals. Interest went up roughly $60 million a year. The sponsor put in nothing. NOT A DAMN THING.

Guess covenants are cheaper than cash. Go figure.

The situation is especially tense because Thoma Bravo recently suffered a highly publicized blow with another portfolio company, Medallia Inc.. In that case, private credit lenders completely took control of the company away from Thoma Bravo during a debt restructuring, wiping out roughly $5 billion in equity that the private equity giant and its co-investors had put into it. Lenders are using the Medallia precedent to squeeze Thoma Bravo for harsher terms on Sophos.

Read more

In Part I, we analyze the Q1 numbers, and Part II will focus on the Q2 numbers. We see that the software marks are consistently declining.

Subscribe now

Blue Owl Technology Income Corp (OTIC) is a private, perpetual-life non-traded business development company (BDC) managed by Blue Owl Capital. It focuses on direct lending and equity investments in established, high-growth software and tech-related companies primarily in the United States.

Blue Owl Capital faced $4.7 billion in withdrawal requests from two of its flagship private credit funds in the second quarter, though redemption pressure modestly eased from the prior three-month period. So Blue Owl maintained a 5% quarterly cap on redemptions to avoid liquidating the corporate loans in its portfolios.

Rightfully so, wealthy investors have pulled billions of dollars out of non-traded private credit vehicles in recent months, driven by concerns about lending standards and worries over artificial intelligence-driven disruption at software companies that borrowed from direct lenders.

Today, we’ll focus on Blue Owl’s technology-focused fund, Blue Owl Technology Income Corp (OTIC), which has faced even higher redemption pressure. Reuters reports that withdrawal requests at OTIC fell to 38.1% in Q2, down from 40.4% in Q1. The fund's concentrated exposure to software, which equals 64% of its net assets, has made it particularly vulnerable to investor exit requests amid tech sector uncertainty.

It is about to get way worse. Below, we analyzed OTIC’s Q1 loan book. In Part 2, we will analyze Q2 and how software loans deteriorated further.

NOTE: Barracuda went public in 2013, was taken private by Thoma Bravo, and was sold on to KKR in a deal that closed in August 2022. Blue Owl holds three loans against it, costing $104.8m, carried at $52.0m.

That is what much of this book is: software companies owned by private equity firms, sold between private equity firms, and financed with layered first- and second-lien debt. Roughly half of this book is not lending to software companies so much as lending against software buyouts. Of the 83 software borrowers, 41 are named Bidco, Topco, Parent, Acquisition, Purchaser, or Buyer, holding 46% of the software cost base, and the schedule lists six co-investment vehicles by name.

The filing never names a sponsor, but the borrowers name themselves. Where the debt was raised to pay a purchase price, the loan carries the software company’s business risk plus whatever leverage the buyer chose to add. When the debt behind one of them is marked at half, the question is whether the price paid was ever supportable, and that question does not stop at one borrower.

Read more

What’s interesting about those losses though? A few things.

Read more

is leading the ‘circle-jerk’ team is nothing new. There were teams like that before.

has been arguing since November 2025 that hyperscalers are understating depreciation by extending the assumed useful life of AI hardware. He has taken real pushback for it, including from CoreWeave’s Michael Intrator, who says chips get re-rented on successive contracts so the longer schedules are justified. Dr. Burry is right about the extensions, and the filings show it. This piece takes that as the starting point and asks the next question.

Dr. Burry follows the useful-life assumption into the income statement, and asks whether the earnings are real. The same assumption also sets what the hardware is carried at, which sets the loan-to-value a lender underwrote when it financed the purchase. Follow it in that direction and you get a different question. Who is standing behind the paper, and what happens when it comes due.

That question has been answered in the past, on three separate scenarios.

The ‘Circle-Jerk’

When a company that cannot borrow on its own credit borrows anyway, somebody made that possible. Who?

That is not a new question. It has been asked three times in the last hundred years, in three very different settings, and it actually produced the same answer each time.

The Structure

A buyer wants something and cannot pay for it. A lender will not lend against that buyer, or against the asset being bought. A third party steps in and guarantees the paper. Now the lender is lending against the guarantor.

This is ordinary and mostly benign. Trade finance works this way. So does insurance. So do export credit agencies.

It becomes dangerous under one condition. When the guarantor’s solvency depends on the same activity the guarantee is enabling.

At that point the guarantee is not risk transfer. It is a circle. And a circle holds until it breaks.

What does not break a circle (based on the past)

A short report. A valuation argument. Bad press. An earnings miss that does not touch a covenant. An impairment that does not accelerate a maturity. A credible bear thesis that is simply correct and early.

None of these force settlement. A circle is indifferent to being understood. It is only sensitive to being tested.

What broke the circle in the past?

There are only two ways a financing circle ends. Either you ask for new money to repay old money and are refused, or you already promised money you cannot now withhold.

Insull was refused loan in 1932. The Reichsbank was refused loan in 1938, on the fourth try, after three had cleared. Nortel had signed commitments it could not escape and funded into the collapse while writing off what it had already lent.

In all three, demand for the underlying product was intact. Electricity, telecom equipment, armaments. None of it stopped being wanted. The failure was at the settlement date, not the demand curve.

Nvidia now backstops GPU purchases for buyers who could not otherwise borrow. Much of the sector’s debt matures between 2026 and 2028. This piece works through what the historical record says to watch, and what it says not to bother watching.

In the past three cases (as you would see below) this is how the circle broke.

Chicago, 1932

Samuel Insull built the American electric utility industry. He worked out demand-based pricing in the 1890s, which kept generating plants running closer to capacity, which drove unit costs down, making electricity affordable. By 1930 his utilities produced roughly one tenth of the nation’s electricity, serving 5,000 communities in 32 states.

Then he built the financing on top.

Middle West Utilities was formed in 1912, the first of his holding companies. It owned small stakes in Commonwealth Edison and other operating utilities. It was itself publicly traded, and by 1912 it controlled utilities in 13 states through relatively small shareholdings. He designed it so that a tiny personal investment produced complete control.

He then stacked more holding companies above it. Each layer borrowed against the shares of the layer below. He ended up controlling more than eighty-five corporations through a pyramid of highly leveraged entities holding shares in one another. Bankruptcy auditors later found the two top companies had each invested millions in the stock of the other.

He funded it by selling stock and bonds to his own electricity customers and his own employees. In Illinois alone, the number of stock and bondholders rose from 50,000 to nearly 500,000 between 1919 and 1921.

Nothing in the structure was ever repaid. It was refinanced.

In 1932 refinancing stopped being available. Middle West Utilities could not meet $10 million in short-term notes. Insull went to the New York bankers and the credit was gone. A Chicago printing firm owed a small bill pushed the company into receivership.

What Happened

The power plants never stopped running: Commonwealth Edison, Peoples Gas, and Public Service of Northern Illinois all survived. Only the financing above them died. Insull was tried and acquitted. The regulatory answer was the Public Utility Holding Company Act of 1935.

New Jersey, 2001

Lucent was the equipment side of the old Bell System, spun out of AT&T. It was incorporated in November 1995, went public in April 1996 at $27 a share, and became fully independent on September 30, 1996. It began life with Bell Labs, more than $20 billion of annual revenue, and 137,000 employees.

Independence meant it could finally sell to AT&T’s competitors. A wave of new telecom carriers wanted switches and fiber gear. They could not pay for it.

So the equipment makers lent them the money. Lucent, Nortel, Alcatel, and Cisco all did it.

The seller was funding the buyer. Every sale produced revenue, and under the accounting rules it was revenue. But the cash leaving and the cash arriving were the same cash.

Lucent’s version got prosecuted. Per the SEC’s complaint, a software pool arrangement with Winstar let Lucent recognize $135 million in the fiscal year ended September 30, 2000, reduced to $125 million after a credit was documented. Post-dated side letters were involved. In aggregate the SEC charged Lucent with improperly recognizing roughly $1.148 billion of revenue and $470 million of pre-tax income in fiscal 2000. Lucent paid a $25 million penalty and was not required to restate.

But the fraud is a distraction. Nortel’s filings show the structural failure without any fraud at all.

From the FY2001 10-K:

Now, read the second row first. That is the exposure. Those were contractual obligations to lend more money to customers that were already failing.

Now, read the first row. The drawn balance fell by more than half in a single year. Not because anyone repaid. Because the loans were written down.

The asset side went at the same time. Nortel recorded an intangible assets write-down of $12,422 million for the year ended December 31, 2001.

Three things reversed simultaneously:

• The receivable.

• The revenue that receivable had produced.

• And the goodwill paid for the capacity to serve that revenue.

Berlin, 1938

This example is here because it is the most extreme documented version of the same structure, and because it ended in a way that is directly instructive. Nothing in it suggests any current participant resembles that regime in any respect other than the structure of the paper.

The comparison is to a financing mechanism and to nothing else.

The mechanics of the structure comes from the Nuremberg trial record, digitized by Yale Law School.

Armament contractors drew bills of exchange on a limited liability company, Metallurgische Forschungsgesellschaft. It was a dummy organization with nominal capital of only one million Reichsmarks. The bills ran six months, with provision for extensions running consecutively for three months each.

Any German bank would discount them, and those banks could rediscount at the Reichsbank within the last three months of earliest maturity. The amount outstanding was a guarded state secret. Schacht conceded that the arrangement let the Reichsbank lend to the government indirectly what it could not lend directly under law.

The system ran until April 1, 1938. Twelve billion Reichsmarks of bills were outstanding at that date.

It is critical to note that It was not a debt burden. Germany’s domestic debt had been destroyed by the 1923 hyperinflation, and academic work on the period shows the debt-to-income ratio stayed roughly flat from 1932 to 1936 despite heavy deficit spending.

The problem was access. The government had begun default proceedings on its long-term foreign debt in 1933, frozen foreign accounts, and blocked convertibility. There was no market to borrow in. The guarantee existed to manufacture one.

And here is the part that usually gets left out: The Bond market said NO

On March 31, 1938, Schacht stopped the program. He then tried to term out the debt. He offered long-term bonds to the bill holders in place of cash and they would not take them. In October 1938 he brought four packages of long-term bonds to the public at 1.5 billion Reichsmarks each. Private savers and insurance funds bought the first three. The fourth failed in late November.

That was the period the market finally declined to refinance, in Berlin, in 1938, under a government with every coercive tool available to it.

Next, from April 1938 until January 1939 Schacht authorized an increase of roughly 2.6 billion Reichsmarks in banknotes to discount armament paper. On January 7, 1939, he submitted a report signed by the Reichsbank directors calling for drastic curtailment of armament spending and a balanced budget as the only way to prevent inflation. On January 19, Hitler dismissed him.

The financing device did not run for the life of the regime. It ran about four years and was terminated by the man who designed it, after the bond market said no. What extended the arrangement past that point was the printing press and the seizure of foreign resources, neither of which is available to a private company.

The common ending

Now

The guarantee has been formalized: On July 1, 2026, Nvidia announced a program it calls the AI Compute Partnership. It acts as a financial backstop for neocloud customers’ GPU deployments and takes a recurring share of the resulting cloud revenue. It agrees to rent back unused GPUs at a fixed rate. Firmus and Sharon AI were early adopters. A Nvidia spokesperson confirmed the program to The Information, which reported that guaranteeing to rent unsold capacity makes it easier for buyers with weaker credit to obtain loans.

The debt matures soon: CoreWeave reported total debt of $35 billion as of June 30, 2026. Net interest expense reached $640 million for the quarter, more than double the $267 million a year earlier. Backlog stood at $104 billion. Much of the sector’s debt matures between 2026 and 2028.

The credit market: Oracle five-year CDS has traded near 200 basis points against roughly 78 for Nvidia and 93 for Meta. Oracle announced plans to raise $45 to $50 billion in debt and equity this year.

Behind those numbers sits a lender base that is not the banking system. Private credit funds originate most of the data center debt. Outstanding private credit loans to AI-related companies have gone from near zero to over $200 billion, and Morgan Stanley projects another $800 billion over the next two years. A February 2026 Chicago Fed study found bank exposure to AI-adjacent industries averaged 0.8 percent of total assets, while noting that additional exposure runs through channels not captured directly.

The collateral

Loans are being written against hardware. The value of that hardware over time is an assumption, and the assumption is disclosed. So it can be checked.

It turns out the issuers do not agree.

Three things follow:

First, the range is 4 to 6 years on comparable hardware. That is a 50% spread in the annual expense recognized against the same class of asset.

Second, Meta and Amazon moved in opposite directions within weeks of each other, in early 2025, on the same question. Meta added six months. Amazon took a year off a subset of its fleet and said why: technology is moving faster. Dr. Burry made this same point in July, with the figures attached, and drew the right conclusion from it: depreciation cannot move in two directions at once if it is measuring something physical.

Third, and this is the point that matters for the paper. The operator with the longest assumed life is the one whose hardware is pledged as collateral. CoreWeave sits at six years. Nebius, running a comparable business, sits at four.

One caveat worth mentioning here: The hyperscaler extensions in 2022 through 2024 were made at the level of servers and networking equipment as a class, in fleets that were still heavily CPU-based at the time. For us, the question is not whether those firms extended GPU lives. It is whether they revisited the assumption once the fleet mix shifted toward GPUs.

A useful-life assumption is not a footnote in this structure. It sets reported earnings, it sets the carrying value of collateral, and it sets the loan-to-value the lender thought it was underwriting. When comparable issuers disagree by 50% on that assumption - that is the signal.

The takeaway

The Lucent question: Is the customer paying with money it generated, or with money the seller supplied? If it were the latter, then, the revenue may not exist.

The Insull question: If the customer genuinely pays, does the financing above the asset still require periodic refinancing? Because that fails independently of whether the business works.

The Berlin question: Who is the guarantor, and does the guarantor’s health depend on the activity it is guaranteeing?

On that third question, one company is currently the chip supplier, an equity holder in its customers, the guarantor of collateral residual value, and the buyer of last resort for unsold capacity. Simultaneously.

What actually has to go wrong

Not that AI proves useless. AI is here to stay, whether one likes it or not.

Here is what is going to happen. The collateral loses value faster than the loan documents assumed. And the paper matures into a market where lenders have become cautious about pricing residual value themselves.

That is sufficient. It was sufficient for Insull, whose generating plants never missed a day.

The order we would expect this to transpire:

First, unrated privately held facilities and deals with weaker tenants. These do not print, so the evidence is a financing that quietly fails to clear rather than a spread that widens. Second, guarantee concentration turning from stabilizer into transmission line, at which point the guarantor’s own credit becomes the sector’s discount rate. Third, private credit marks, inside vehicles with periodic redemption windows, held by investors who believed they were buying yield rather than GPU residual value risk.

That third one is the Insull mechanism rather than the Lucent one. The assets keep producing and the structure above them fails anyway.

One data point already exists. In February 2026, CoreWeave fell as much as 12 percent intraday on a report that Blue Owl had failed to secure sufficient funding for a $4 billion data center project in Pennsylvania. Nebius and Applied Digital fell with it.

Watch the lenders, not the customers. A financing that does not close is the leading indicator. A revenue miss is the lagging one.

Tripwires

• Contingent commitment disclosure in Nvidia’s 10-Q. The dollar amount of backstop obligations, if disclosed, is the cleanest number in this structure.

• Undrawn commitments at neocloud operators. Exactly the line that sank Nortel. Ask whether anyone discloses it.

• DSCR covenant headroom on rated GPU facilities.

• Take-or-pay counterparty concentration. Hyperscaler-backed paper is a hyperscaler credit, not an operator credit, and should be analyzed that way.

• Useful-life assumptions, and any change to them. The single most checkable number in this structure, and it sits in the filings. Watch in particular for an extension at an operator whose hardware is pledged, and for anyone following Amazon down rather than Meta up.

Where this analogy fails

Lucent and Nortel sold into a customer base that did not exist. Data center capacity today is utilization constrained rather than demand constrained, and the largest offtake contracts sit with the strongest balance sheets in the world. That is a real difference from 2001 and we are not going to dismiss it.

The German case had coercion behind it. No private guarantor can compel anyone to hold its paper. That is why Insull broke in three years and Lucent in about eighteen months, and it is also why the ending here is bankruptcies and equity destruction rather than a currency event.

And the assets are not consumed. Armaments have zero residual by design. Data centers depreciate on a contestable schedule, which is a genuine problem, but it is a different problem.

The bear case does not require AI to fail. It requires the residual value assumptions and the refinancing window to be wrong at the same time. Those are separable questions from whether the technology works, and they should be analyzed separately.

Minsky had a word for this

Hyman Minsky was an American economist who spent most of his career at Washington University in St. Louis and ended it at the Levy Economics Institute at Bard. He was largely ignored while he was alive. He died in 1996 and became famous in 2008, when a lot of people went looking for someone who had explained what had just happened.

His central idea is called the financial instability hypothesis. The compressed version of it is four words: stability is destabilizing.

The three postures

Minsky said any borrower sits in one of three positions, and the position is defined by where the money to pay the debt comes from.

Hedge: Your cash flow covers the interest and the principal. You can pay off the loan out of the business. You do not need anyone’s cooperation.

Speculative: Your cash flow covers the interest but not the principal. You can service the debt, but you cannot retire it. When it matures you have to roll it. You need a lender to say yes.

Ponzi: Your cash flow covers neither. You are paying interest with borrowed money, or with proceeds from selling appreciated assets. You need a lender to say yes and you need prices to keep going up.

Minsky’s term “Ponzi” is a technical classification, not an accusation. He was not saying these people are frauds. He was saying their debt service depends on something other than their operations.

In the good times and the bad

Minsky said in good times, an economy moves from mostly hedge toward more speculative and more Ponzi. And it does that without anyone deciding to take more risk.

The mechanism is simple. Every year that nothing goes wrong is a year of evidence that the caution was excessive. Lenders who required a margin of safety watch competitors write thinner deals and get paid. Borrowers who levered up were right. The people who were prudent underperformed. So the standard moves, one deal at a time, and each individual move is defensible.

That is what “stability is destabilizing” means. Calm periods do not just precede fragility. Calm periods manufacture fragility.

The second half, which people forget

A borrower can move from speculative to Ponzi without doing anything at all.

Rates rise. Or the asset gets remarked. Or the refinancing window opens into a nervous market. Yesterday the interest was covered. Today it is not. Nobody changed their behavior. The classification changed.

This is why the useful question is never “are these people being reckless.” It is “what would have to move for a unit that is currently speculative to become Ponzi, and how far away is that.”

A note on the term you have heard

“Minsky moment” is not Minsky’s phrase. Paul McCulley of PIMCO coined it in 1998, describing the Russian financial crisis. It has since been applied to August 2007 and the crisis that followed.

There is nothing “sudden” about what is coming. It has been a process for quite sometime.

The Three Cases

Insull’s operating utilities were hedge units. They generated real cash from real customers. The holding companies above them were speculative moving toward Ponzi. They serviced debt by refinancing and by selling more stock, never out of operations. When the refinancing was refused in 1932, the classification did not change. It was simply revealed.

Nortel’s carrier customers were Ponzi units from the beginning. No cash flow, no ability to service, entirely dependent on new capital. The distinctive feature is that the new capital came from the vendor. Nortel was funding the units whose survival its own revenue depended on.

The Reich was a Ponzi unit in the technical sense. The bills could not be serviced from tax revenue. They were rolled. When Schacht tried to convert them into long bonds in October 1938, three packages cleared and the fourth did not.

All three were units that could not settle, only refinance. That is the single characteristic they shared, and it is why all three ended at a maturity date rather than at a moment of disillusionment about the underlying product.

NOW

The AI buildout is not one thing and it does not have one classification. The hyperscalers funding capex out of enormous operating cash flow are hedge units by any reading. That is not in dispute and it should not be blurred.

The question is about a specific layer. An operator carrying $35 billion of debt, with quarterly net interest expense that doubled year over year, whose maturities fall between 2026 and 2028, will not retire that debt out of cash flow. It will refinance it.

That is the textbook definition of speculative finance. It is the ordinary posture of most capital-intensive infrastructure businesses in a growth phase.

What makes it worth watching is the second half of Minsky’s point. The move from speculative to Ponzi does not require anyone to behave badly. It requires two things that are both currently in motion and both currently disclosed: the cost of refinancing, and the carrying value of the collateral.

The collateral question is already open in the filings. Comparable issuers assume useful lives ranging from four years to six on the same class of hardware, and in early 2025 Meta extended while Amazon shortened. Somebody is wrong about how long these assets last. Whoever is wrong has been booking the difference as earnings.

Minsky’s argument was that the guarantee, the backstop, the lender of last resort does not remove the fragility. It relocates it. Intervention stops the immediate failure and validates the risk-taking that produced it, which sets the terms for the next cycle. That is what a private backstop does too.

Nvidia’s program makes marginal deals financeable today. What it also does is move the question from “is this operator creditworthy” to “is the guarantor good for all of them at once.”

The Question

When the debt comes due, where does the money come from?

If the answer is operations, it is fine. If the answer is a new lender, it is fine .. until that lender says no.

That has been the whole answer for a hundred years.

Sources

Primary regulatory filings and trial records

• Lucent Technologies Inc., Form 10-Q/A FY1997. Incorporation November 29, 1995; IPO April 10, 1996 at $27; distribution September 30, 1996. SEC EDGAR.

• Lucent Technologies Inc., Form 8-K, May 18, 2004. $25 million penalty, no restatement required. SEC EDGAR.

• SEC v. Lucent Technologies Inc. et al., Complaint No. 18715, D.N.J. Winstar software pool transaction, $135M and $125M figures, post-dated letters.

• Nortel Networks Corp., Form 10-K FY2001. Customer financing table; $12,422 million intangible assets write-down. SEC EDGAR.

• Issuer useful-life disclosures. Microsoft FY2022 Q4 earnings call and subsequent filings. Alphabet Q4 2022 earnings release dated February 2, 2023 and Q4 2023 release dated January 30, 2024. Oracle fiscal 2025 Q1 disclosure. Meta Q4 2024 disclosure. Amazon Q4 2024 earnings release dated February 6, 2025. CoreWeave Form 10-K FY2025. Nebius. Each figure originates in the respective issuer’s own disclosure.

• Avalon Project, Yale Law School. Nazi Conspiracy and Aggression, Vol. 2, Ch. XVI Pt. 12 (documents EC-436, EC-438) and the IMT judgment on Schacht.

• Ritschl, A. (LSE), “Deficit Spending in the Nazi Recovery, 1933–1938.”

• Journal of Economic History (Cambridge), on the Borchardt hypothesis. June 1933 initiation of foreign debt default proceedings, citing Tooze.

• Gettysburg Historical Journal, Vol. 16 Art. 5 (2017). The October 1938 bond packages and the failure of the fourth, citing Nuremberg documents.

• Haas, Reimer, Guter-Sandu and Murau, “The Mefo Operation,” SSRN.

• Britannica, Harvard Business School, EBSCO, Encyclopedia.com. Insull biography, Middle West Utilities formation in 1912, pyramid structure, scale of operations, acquittals, PUHCA 1935.

• Wharton, “Accounting Games Companies Play,” 2004. SEC aggregate charge figures.

• Company-Histories / FundingUniverse. Lucent opening revenue and headcount.

• Novel Investor; chicagology (reprinting 1932 press). Insull shareholder growth, Middle West note failure.

• Data Center Dynamics, July 2026, and The Information via Michael Parekh. Nvidia AI Compute Partnership mechanics and spokesperson confirmation.

• Capacity, 2026. CoreWeave debt, interest expense, backlog.

• Reuters via Yahoo Finance. Oracle, Nvidia and Meta CDS levels.

• CNBC, February 2, 2026. Oracle capital raise plan.

• Business Insider via Finviz, February 20, 2026. Blue Owl Pennsylvania financing shortfall.

• Quinn Emanuel client alert, March 2026. Private credit origination, $200 billion figure, Morgan Stanley projection, Chicago Fed study.

• The point that hyperscaler useful-life extensions were made at the server-class level in CPU-dominated fleets, and that the sharper question is whether they were revisited as the mix shifted to GPUs. Raised by the Deep Quarry newsletter.

• Michael Burry, "Short Thoughts July 8, 2026," Cassandra Unchained. Amazon and Meta useful-life divergence with dollar figures; the argument that depreciation is an economic lever rather than a physical measurement. Also his X posts from November 2025 on hyperscaler depreciation generally.

We have spent this year inside the private credit loan books - mapping and analyzing, and the ABS trusts one at a time. What the data shows is not reassuring.

The more you know…

Privately placed bonds, as classified under the NAIC’s broad bond framework, reached 48.4% of total life industry bonds at year-end 2025, up from 37.4% five years earlier. How much of that is private credit rather than ordinary bilateral corporate debt is exactly what nobody agrees on.

Moody’s put US life insurers’ private and illiquid bond holdings at $807bn at year-end 2025, or 20% of the industry’s $4tn fixed income portfolio, up from $685bn and 18% a year earlier. The holdings are not evenly spread. The ten largest life insurers hold $352bn, or 44% of that total, while accounting for only 24% of total industry fixed income. Barclays found private credit holdings across US life insurers grew by more than 20% in 2025, with exposure above 15% at some private equity affiliated insurers including Apollo-backed Athene and KKR-backed Global Atlantic. The basis for that 15% is not stated in the reporting we reviewed.

The credit quality of the illiquid portfolio is weaker than the broader book. It carries 43% in NAIC 2, the Baa-rated tier, and 9% below investment grade, against 36% and 5% for the overall $4tn fixed income portfolio. Moody’s noted that in a downside scenario, impairment rates in the private credit segment could exceed those of the liquid portfolio.

PIK Your Poison - The OBDC Version - I

Now, a few weeks ago, we dissected the Blue Owls’ March loan book.

We are just curious as to why the whole book is deteriorating and software looks incredible in comparison.

To us, that means software is also crap.. they just refuse to mark it. Also OBDC is traded.. if they’re marking the software down lets say 50% to where some of these companies are clearing, like the bending spoons case, then it would be at closer to 6 than 11.

We parsed all 351 positions in the schedule. What that surfaced: the structures where one borrower shows up as three separate line items. The holdco PIK notes sitting above cash-paying operating loans. The full non-accrual roster, every position. What happens to a borrower when the fund that lent to it has to sell.

We’ll share what seems to be a secret about the industry deterioration with you.

And since this is a critical part I, we offer a special discount for anyone who is upgrading to founding members (Confidential Insights) by today:

Read more

Data providers never tell you what is coming. They surf. They ride the wave up, they ride it down, and they pivot the moment the direction is absolutely clear. If you were on the Titanic, they would have pivoted around the time you could no longer hear the band. The water had reached it.

So I opened Trepp’s latest report expecting a signal that had just popped. It had not. The signal is the share of lending going to acquisitions rather than refinancing, and Trepp says that when it approaches 30%, buyers have stopped paying for current income and started paying for momentum. It flashed across every major property type in 2007. It went off again in 2021 in multifamily and lodging, at levels never before seen in the dataset.

In multifamily the share hit 47% in 2021 and peaked at 50% in 2022. Look further back and it first popped in 2017 at 31%.

The headache has arrived.

Read more

Private equity firms have been buying life insurers to get at what the industry calls patient capital. Your retirement annuity is patient capital. The money sits there for decades, which makes it ideal for funding loans that cannot be sold. Critics say the arrangement puts policyholders at risk by obscuring default risk, inflating asset ratings, and pushing the losses from private lending onto state backstops.

Last week we published Duration Is Not a Defense. It ended on a simple point. When one of these loans goes bad, it can be a very long time before anyone outside the deal finds out.

Here is why: The borrower misses a payment, and the lender lets it slide, adding the missed interest onto the balance instead of calling it a default. Nobody has to write the loan down, because there is no market price to write it down to. The value on the books is whatever a model says it is. The credit rating on the paper is private, and the evidence says private ratings move late. The capital charge the insurer carries only changes once that rating changes. So the loan is worth less than it says on the page, and every mechanism that would normally tell you so is either delayed or discretionary. Imagine that!

Now, how much money are we talking about? That depends entirely on who is counting, which was the whole point of Part I. Published estimates of US life insurer private credit exposure run from roughly $800bn to something close to $1.8tn. They all have the same data. They just draw the line in different places.

Regulators have noticed this anomaly. In 2024 the NAIC’s Capital Markets Bureau published a study on private ratings among insurer bond investments, reporting that such ratings had nearly tripled in five years. The report was later pulled from the NAIC website, with the group saying it needed to clarify the findings. It has not reappeared. We have not read it, for the obvious reason.

In May, the Treasury Department said it would convene a series of meetings with state insurance regulators to review recent market events, emerging risks and the outlook for private credit. That is Treasury asking state regulators what is on their balance sheets, which tells you something about who currently knows.

Are you frustrated yet?

Treasury knows there is a trouble.

That leaves an obvious question. If nobody sees the loss while it is building, who is holding it when it finally shows up?

So this part follows the money, first one way and then the other.

Going out: your premium arrives, gets handed to a manager the sponsor owns, gets packaged into a structure that decides how much capital the insurer has to hold against it, and comes out the far end as a loan to a mid-sized company you have never heard of.

Coming back: what happens to that loss once somebody finally has to put a number on it.

We know who is at that end of the chain, because a Federal Reserve Bank published it. It listed which vehicles issued this paper and how much each one put out. Athene, Global Atlantic, Everlake and Resolution Life are all in there. It also published the figure that matters most to us: how much of that paper ended up with insurers outside the family that created it.

Now the loss coming back the other way. And this is where it stops being the carrier’s problem and starts being everyone else’s. A life insurer that fails does not go through bankruptcy. The other insurers in that state get a bill. And in 34 states they can work most of that bill back off their taxes over the following five years. Almost nobody outside the industry knows that, and it is the reason a private credit loss inside one company does not stay inside one company.

We also set out what would change our minds, because a position that cannot be wrong is not worth holding.

Read more

Firstly, the whole process is pretty opaque. When talking to those who put together these deals, they often only understand one component and are blind to the entire process. In the real estate industry, dealmakers don’t talk to servicers. Once the deal is done, originators walk away without a care in the world. If the debt is securitized, you are often unable to find out who exactly will get hit, as most of the information is not public. Unless you pay for pricey subscriptions, web sleuthing is your only recourse to try and put together a history of a building that could have been securitized several times over with a dizzying array of market participants. Each of these properties has a story that we now find is oft repeated, fueled by the strongest gateway drug out there. Slowly but surely, we are chalking the outline of this rather large debt bubble that also includes datacenters.

Total debt against income-producing commercial property stood at roughly $5 trillion at the end of last year. Banks hold $1.89 trillion of that, or 37.5%, and picked up 1.2 percentage points of market share. GSEs hold $1.14 trillion. Insurance companies $806 billion. Securitized structures, CMBS included, $753 billion.

Pension plans hold under 1%.

Trepp’s growth chart puts pension plans first, ahead of securitized lenders and GSEs. The chart has no printed data labels, so take the ranking and skip the decimal.

Growth rates flatter small books. Banks added 3.6% to $1.89 trillion, nearly 40% of it in the fourth quarter alone. Pension plans grew faster on less than one fiftieth of that balance. The money went where the money already was.

The maturity schedule is where this gets interesting. Against year-end 2025 positions, $797.1 billion of income-producing CRE loans come due in 2026. Banks carry $371.4 billion of it, about 20% of their book. Securitized lenders carry $232.2 billion, about 31% of theirs. Insurance: $69.7 billion. GSEs $44.6 billion.

Everything else $79.2 billion. The near-term wall is a bank and securitized problem.

The 2031-and-beyond row totals $1.65 trillion. Banks hold the biggest dollar slice of that too. As a share of each book, though, the picture flips: 45% of the GSE balance and 41% of the insurance balance sit out past 2030, against 30% for banks.

Trepp attaches a caveat; a lot of those near-term maturities are not economically binding. Plenty of 2026 and 2027 loans carry borrower extension options or have already been modified, usually tied to a paydown, extra reserves, or a performance test. The stated schedule overstates what actually has to refinance in any given year.

None of that makes the risk disappear. It spreads it out. Good loans extend, bad loans don’t, and it happens one at a time instead of all at once. That is why everything looks “normal” when it is NOT.

Take out the loans that can extend and the picture gets a lot tighter. Private-label CMBS is a separate universe from the $5 trillion above. Its 2026 maturities run $146.2 billion, and $76.6 billion of that is hard, meaning the extension options are gone. Of the hard balance, 36%, or $27.3 billion, carries a debt yield of 8% or less. Trepp calls that the level where “refinancing math historically becomes challenging without paydowns, equity infusions, or lender flexibility.” And 39% of the hard maturities land in the fourth quarter.

Trepp names the same new risk three times in three pages:

1. opaque risks inside private credit

2. private-credit exposures pressing on risk appetite and capital deployment

3. and private-credit stress feeding through to the rate path and collateral valuations.

So who is holding the first loss?

Blue Owl Capital bought the horizontal risk pieces of eight SASB deals in the first quarter. Face value $322.41 million. That’s 17.24% of all the risk retained in the new-issue CMBS market, roughly double Rialto’s volume. Two deals made up 52% of it: the $1.87 billion Extended Stay America Trust 2026-ESH2 and the $1.5 billion SLG Office Trust 2026-OMA. Horizontal strips are the first-loss position by construction. Whoever holds them takes the first dollar of loss.

KKR Real Estate Credit Opportunities took four deals. Blackstone one. Monarch Alternative Capital three. Prime Finance two. Harrison Street, Eightfold, Waterfall, Kayne Anderson, Elliott and TPG are all on the list.

Now, New York State Teachers retained $60 million of CMBS risk in the first quarter of 2025. Healthcare of Ontario Pension retained $26.75 million. Both sit at zero this quarter. That’s risk retention only, and it says nothing about what either plan holds in CRE more broadly.

Dodd-Frank says somebody has to keep 5% of every deal. The question is who volunteers, at what price, in a year with 39% of the hard maturities stacked into the fourth quarter.

The full piece maps the risk retainers Trepp identified for the quarter by deal count and dollar amount, sorts them into alternative managers and bank dealers, and traces one Pittsburgh office complex across two securitizations, repeated modifications and a foreclosure, up to a sponsor whose Tel Aviv-listed bondholders voted to accelerate about $168 million.

Author

Read more

The earnings season is always a nightmare to live with. The euphoria of the banks rallying as if there is no care in the world is nauseating. And here is why: Bloomberg reported that banks worldwide are bolstering their use of SRTs to free up capacity for new lending.

As we plow through earnings season, we are speaking with sources deep within the insurance-private credit space. Which brings us to a whole new level of cluster mess.

Here, we want to highlight exactly what is being done - as told by a big private credit firm, portfolio manager, in an interview he gave recently.

“There’s an efficiency that comes from doing things on a pooled basis. You know, if I’m an insurance company and I go out and I make, you know, 100 or 1000 or 5000 loans in a particular area, right.

And any one of those things has a risk of default that’s contained in that RBC charge. If I hold those at the balance sheet, you know, without any structure around it, you know, I’ve got a certain level of risk, but it’s generally fairly punitive for the insurance company and not not capital of, efficient for them to make those kind of loans.

If I pull those things into a structure, you know, the risk of the entire portfolio defaulting at the same time is effectively negligible, right? It’s you got the law of large numbers, you’ve got diversity, you’ve got all those kind of things. And so a certain amount of that portfolio then becomes IG and becomes very ratable. And you still can hold a residual position in the pool if you want to hold the entire pool.”

PCs KNOW WHAT THEY ARE DOING. THEY ARE AWARE OF THE RISK.

RISK IS BEING MOVED TO THE MIDDLE CLASS (INSURERS).

Let us translate that: The loans are the same loans. Pooled and tranched, a large share of them becomes investment grade and ratable (this is MENTAL) , and the capital charge falls accordingly. Nothing about the borrowers changed. The wrapper changed.

Now, I agree with our analyst when he said, “pooling is a mixed blessing; if your pool is full of crap, it doesn’t matter how much water you’ve got in it to dilute it. Once some know there’s poop in that water, they won’t get in… or, in this case, they will get in because there’s poop in it.”

On that smelly note, what our analysis argues is this:

We have read two credible bodies of work that reach opposite conclusions about the same sector. One reports that private credit is roughly 6% of life insurer general-account assets, and that insurers holding more of it show no higher estimated insolvency risk. The other argues that private equity has rebuilt life insurers into holding vehicles for affiliated private credit, and that the state guaranty fund system pushes the resulting downside onto competitors and, in most states, onto taxpayers.

They are not contradicting each other on the facts. They define the exposure differently, they study different populations, and the quantitative work is fitted to a history in which this exposure barely existed. Now, the industry defense leans on duration matching, the idea that a long-dated liability makes a long-dated illiquid asset safe to hold. That keeps an insurer from having to sell at a bad price. It does nothing about a borrower who stops paying.

Read more

Ordinarily, the investigation would have ended there. At most, it would have resulted in a complaint, a refund, a bad review or a decision not to return.

Instead, we asked a different question.

Who owns Panera?

Following that ownership chain led away from bread, coffee and consumer brands and into life insurance, annuities, reinsurance, asset management, private credit, asset-backed finance, collateralized loan obligations and mortgage lending.

What began as an examination of a disappointing meal became an investigation into how consumer businesses can sit at the visible edge of a much larger financial system.

This paper does not argue that poor mac and cheese is evidence of credit impairment. Nor does one disappointing meal prove that Panera is experiencing wider operational or financial deterioration.

The objective is to show how quickly an ordinary consumer interaction can lead into the structures through which insurance liabilities are transformed into privately originated, structured, and increasingly illiquid financial assets.

The question is no longer simply who made the mac and cheese.

The question is: How Deep Does the Mac and Cheese Go?

The short answer is that the chain does run deep, but not in the simple form of “Panera is just a private equity play.” Panera sits inside JAB Holdings’ fast-casual restaurant platform; JAB, in turn, has deliberately diversified from a consumer-focused holdings company into a hybrid consumer-and-insurance group; and the insurance arm is now connected to a growing ecosystem of affiliated asset management, reinsurance, and alternative-credit vehicles. By the end of 2025, JAB reported $72 billion of assets under management, with consumer and insurance assets split 57%/43%. JAB Insurance alone had about $33 billion of assets* after the acquisitions of Prosperity Life Group and Family Life.

Read more

On that note, this week on the menu is BlackRock Private Credit Fund (BDEBT).

The BlackRock Private Credit Fund (BDEBT) is a non-diversified, closed-end management investment company structured as a non-traded business development company (BDC). This is a private credit fund that is open to both eligible individual and institutional investors.

From our viewpoint, this fund should not be open to individual investors. Not because the loans are fake; they’re real, and real companies pay the interest.

Because of everything between the investor and those companies: an illiquid wrapper the prospectus itself says to consider illiquid, an exit capped at 5% a quarter that just prorated for the first time, values that are mostly monthly estimates rather than market prices, and a borrower list where 158 of 287 names carry acquisition-vehicle labels, Holdings, Parents, Bidcos, paperwork stacked on paperwork.

An institution can price that complexity. A household that qualified with a $70,000 income cannot, and it shouldn’t be asked to. That is the problem.

A $2.35 billion paper that BlackRock's own prospectus tells investors to consider illiquid, lent to borrowers that are mostly not companies at all but the acquisition entities private equity firms stack above companies. 158 of the 287 names in the schedule are Holdings, Bidcos, Buyers, Merger Subs.

The exit is a quarterly window the board can close as it sees fit, and this June, for the first time, investor withdrawal requests exceeded the quarterly cap and were prorated. An investor qualifies for this with a $70,000 income and $70,000 of net worth?

We think the product and the buyer don't match, and the fund's risk disclosures agree with us.

What the filings show

BlackRock Private Credit Fund is a non-traded business development company that commenced operations on 1 June 2022. As of 31 May 2026, it reports $2.67bn in total assets, a net asset value of $23.65 per share, 295 portfolio companies, 99.96% first lien exposure, 0.49x leverage, and a 9.80% distribution rate on institutional shares. The fund’s own performance history shows NAV per share at $24.98 in March 2024 and $23.65 in May 2026, including a -0.70% institutional-class monthly return in February 2026.

We reviewed: 469 investment tranches across 287 unique issuer names, $2,391.1M at cost, $2,354.5M at fair value, and $36.6M of net unrealized depreciation.

Our findings:

1. 158 of 287 issuer names, 55%, carry acquisition-vehicle tokens like Holdings, Bidco, Buyer, Parent, Merger Sub, Finco. $309.7M of fair value, 13.2% of the book, is lent to other financial firms, including a credit rating agency, a wealth-platform take-private in which BlackRock itself committed to hold minority equity, a fund administrator, and a proprietary trading firm.

2. 294 of 469 positions, 63% are carried below cost. 26 tranches across 12 issuers carry contractual PIK components, $120.6M of fair value, 5.1% of the total portfolio at fair value. And 83% of the book by fair value matures between 2029 and 2032, while the only exit for shareholders is a quarterly tender capped at 5% of shares outstanding that the board may amend or suspend.

   ---

What one needs to understand is that the retail investor buying BDEBT shares holds the top of one chain. The borrower sits at the bottom of another. Between them stand the BDC, its credit facilities, the sponsor’s holding stack, and, in the cross-border cases, at least one additional jurisdiction. The investor at the top does not have a clue what they are investing in.

From our analysis we can say that, PC funds borrow money from banks and other PC funds; then, buys wealth management firms and RIAs to acquire their investor accounts. What?

Does this then feed back into PC investments? Also, the rating agencies that rate these funds are listed as borrowers - so, as an investor, you ask yourself, how kosher are the ratings?

Read more

Here is the problem. The chips made by NVIDIA, owned by the guy who signs women’s chests, do not age with time. They age with replacement chips. Which are made and shipped by itself. Imagine the farce.

So, Jensen Huang is the Casanova of GPUs? Every chip is the love of his life until the next one ships?

The Problem We Keep Seeing In the PC Loan Books - And the Real Suckers in All of This:

As we continue to review the private credit loan books, guess the name that keeps popping up? NVIDIA (among others).

Here is what we also realized: an NVIDIA chip doesn’t get old with time. It gets old when the next NVIDIA chip ships. NVIDIA sells the asset, controls the upgrade cycle, and gets paid in cash. The clients who couldn’t pay cash carry the debt, and private credit carries the clients.

No, it is not that simple.

NVIDIA sells chips for cash. Its most ‘poorly-rich’ customers, the neoclouds and xAI, can’t pay cash, so private credit funds lend it to them, secured by the chips the loans are buying. The collateral is borrowed into existence. The loan buys the chips. The chips back the loan.

The “genius” lenders are Apollo, Blackstone, Magnetar, Coatue, Carlyle, Macquarie. There are more than $20 billion in GPU-backed debt, according to analyst estimates, and we think that’s likely low. CoreWeave alone carries $24.9 billion of total debt, added $8.5 billion in one quarter, and pays 26 cents of every revenue dollar in interest.

NVIDIA sits on every side of the trade: it makes the collateral, controls its depreciation through the annual release cycle, and puts equity into its own borrowers, $2 billion more into CoreWeave last quarter, an anchor LP in the vehicle buying its own GB200s for xAI.

For us, the whole structure rests on one unanswered question: what is a repossessed GPU worth?

The lenders’ own industry debate runs from 10% to 60% residual value after three years. To be blunt, the 10-60 spread isn't just about pricing. It's about whether there's anything there at all.

The debt reaches retail through a listed BDC (Great Elm) in CoreWeave’s syndicate at origination, a $14.4 billion wealth-channel fund (ADS) where we traced the VCI exposure, and annuity accounts buying GPU lease paper.

We went through Apollo’s ADS filing. The warning is right there. Markdowns wiped out more than the entire quarter’s income. The distributions went out anyway. PIK tripled. And money is now leaving the fund faster than it comes in.

Who loses when it cracks: not NVIDIA, which was paid up front. Not the managers, who earned fees building the exposure. Not the hyperscalers, who offloaded the capex and will buy distressed compute cheap. The loss lands on the capital furthest from the deal, with the least information about holding it: the annuity holder, the wealth-channel client, retail investors, retirees, and the private credit shareholders.

Here, in the realm of this private credit, risk migrates toward whoever can’t see it. That is usually the retail investors and retirees.

Behind the paywall: the full deal-by-deal roster with sourcing status on every claim, both sides of the residual value evidence, the transmission math back to NVIDIA, and the fresh ADS 10-Q showing a $14.4 billion retail fund in net outflow while it marks down its book. Position-level work on the VCI exposure comes next, for subscribers first.

Read more

For us, what is happening in private credit, CRE, consumer credit, and the behavior of the 99% and the 1% is not random. It follows hidden patterns and tells a story. That has been our signal.

On July 7, columns buckled on the 21st floor at 219/235 East 42nd Street, the former Pfizer headquarters. The project is slated to become the largest office-to-residential conversion in New York City history when it finishes in 2027. Nine surrounding buildings were evacuated. East 43rd Street between Second and Third Avenues is still closed.

According to Gothamist’s review of hundreds of structural drawings, city-approved plans required steel reinforcement on the columns that buckled. Department of Buildings officials declined to say whether that work was completed.

The developer is Nathan Berman of MetroLoft. The lender is Madison Realty Capital (a PE firm, anyone surprised at this point?), which provided the loan in a financing arranged by IPA Capital Markets. Within seven days of the columns buckling, MetroLoft closed on its next conversion in the Financial District, with Apollo lending $72M toward the $104M purchase price.

A structural failure at the flagship. A new acquisition closing the same week. A lender with $24B under management writing some of the biggest construction loans in the country. What the heck is this mess?

Behind the paywall: the 467-m tax program that can cut property taxes by up to 90% for as long as 35 years, Madison’s $630M loan for the Bentley Residences in Sunny Isles Beach, the litigation at the Aston Martin tower, a Texas state court finding that Madison engaged in manipulation and fraud on a $585M construction loan, and why Miami had the largest share of de-listings of all 86 markets we track in June, at over 29%.

Note: Please note that all our articles are for Confidential Insights (founding members) only. Substack did not let us remove basic membership, so we set it to the minimum amount.

Read more

Subscribe now

When we think of shadow lenders, private credit, private equity, and the rest, two things come to mind. One is Hyman Minsky and his warning about Ponzi finance, the stage where borrowers can no longer cover interest from cash flow and survive only by borrowing more or selling what they could. The other is an old phrase: “turtles all the way down.”

The phrase “turtles all the way down” describes infinite regress, the problem of every explanation demanding a further explanation. It comes from an apocryphal story, often pinned on Bertrand Russell, though nobody has ever traced it to him. The story goes that Bertrand, a scientist, finishes a lecture on the Earth’s place in the cosmos. An elderly woman in the audience objects. The Earth, she says, rests on the back of a giant turtle. The scientist asks what the turtle stands on. Another turtle, she says. And beneath that one? She smiles. “It’s turtles all the way down.”

The story shines light on a real problem. An explanation that never reaches solid ground is not an explanation. It is a pile of crap.

That is what a private credit’s SOIs look like. The loan is owed by a Bidco. The Bidco sits under a Holdco. The Holdco sits under a sponsor fund. Ask what any layer rests on and the filings answer with another shell. Go ahead, ask; we will wait.

In BXSL’s March 2026 schedule, 224 of 388 borrower names are financing shells, 54% of fair value that cannot be traced to a named operating business using public filings alone. The one thing that is supposed to be at the bottom of the stack, the actual company earning actual cash flow, is THE thing the disclosure never shows.

In the story, the old woman at least committed to her turtles. Private credit asks investors to fund the stack without ever seeing the bottom.

Which brings us to Blackstone’s $BXSL. A pile of shells all the way down. Here is our breakdown of the financials and the loan book.

Blackstone Secured Lending Fund ( ) carries $13.9 billion of investments at fair value as of March 31, 2026, spread across 695 positions in 388 borrowers.

Of the 388 borrower names, 224 (58% of the roster and 54% of fair value) are not operating companies. They are Bidcos, Holdcos, Buyers, Purchasers, Parents, and Intermediates. The fund’s single largest borrower exposure, $338 million at fair value, is a company identified in the SEC filing only as Snoopy Bidco, Inc. The filing tells you the coupon; however, it does not tell you who Snoopy is.

So we traced two of them all the way down. One shell, formerly named NMC Crimson Holdings, has appeared in the loan books of BXSL and at least three separate New Mountain credit vehicles. Another, Amerilife Holdings, LLC, is owned by two private equity firms and distributes annuities and Medicare products to American retirees through more than 300,000 agents. Blackstone’s shareholders are several steps removed from the retiree buying the annuity. The interest payments travel that entire chain.

Then there is the NAV question: We always wondered whether NAV stood for “No Asset Value”?

BXSL raised $1.0 billion of fresh equity through its at-the-market program in 2024. In 2025, that fell to $291 million. The shares ended 2025 trading below net asset value, the issuance math stopped working, and in February 2026 the board authorized a $250 million buyback at prices below NAV. Between December 31, 2025 and March 31, 2026, the portfolio’s net unrealized depreciation widened from $181 million to $355 million. Nearly doubled in one quarter.

In the full report: the complete sector map, the borrower shell census, the two nesting case studies with every source named, the non-accrual roster and the nine borrowers now carrying the distress footnote, the Medallia write-down that preceded an April restructuring, the NAV arithmetic with and without inflows, and what a stalled equity machine means for borrowers holding $1.8 billion of BXSL’s unfunded commitments.

SUBSCRIBER CONTENT BEGINS HERE —CONFIDENTIAL INSIGHTS

We are not recommending any private credit firms as a short to our clients. What we are doing is mapping the opaque private credit market one loan book at a time.

Read more

Subscribe now

This is a free article for all. It is critical for EVERYONE to read and understand the interconnectedness of our global financial system.

We are not going to open this piece with “we told you so.”

When Tricolor failed, we said it was not an isolated incident. When First Brands failed, we said it was not an isolated case. When Market Financial Solutions (MFS) filed for administration in February, we said the same thing, and then we mapped it.

On May 30, we published our second MFS deep dive and traced the chain: a UK borrower, a US warehouse platform, a UK G-SIB loss, a Singaporean exit, a Gulf substitution. Five regulatory perimeters in under three weeks, and no single regulator with visibility into the full mechanics. NO ONE.

This month, the data finally caught up with the reality and what we have been mapping.

Kroll’s half-year insolvency report, covered by City AM on July 14, counts 649 UK companies entering administration (a rough US translation is Chapter 11, in the sense that it’s a formal insolvency process) between January and June 2026. That is a 6% increase over the 610 recorded in the first half of 2025. Now, when we dug into the numbers, we found 49 financial services companies, up from 30 a year earlier. A 63% jump in one sector, in one half, in one country.

Kroll attributes the rise in financial company insolvencies to increased cost and regulatory pressures and to the implosion of MFS in February 2026. The Telegraph’s read of the same report goes a tad further: almost half of the financial services firms that failed were related to the fall of MFS, whose network of lending companies followed it down. All it took was one lender to skew a national dataset.

Now, Kroll’s restructuring co-head, Sarah Rayment, told City AM that many of the failed firms are intermediaries and brokers collapsing as a consequence of MFS, and that this is “not necessarily a systematic issue” with the industry.

STOP. Re-read that sentence again.

MFS failed in London. The losses surfaced in New York, Tokyo and Sydney. Nearly half of the sector’s failures follow it into administration. Two G-SIBs takes nine-figure charges. A US warehouse platform puts two facilities into default. A hedge fund holds roughly £200 million of exposure. The Bank of England asks lenders what else they are holding. And Ms. Rayment’s response is: not systemic.

Well, Ms. Rayment is wrong.

This is what we have been documenting for a year. Every failure gets reclassified as idiosyncratic after the fact. Tricolor was one bad auto lender. First Brands was one bad supply-chain borrower. MFS is one fraud. Now, you connect the dots; the pattern that emerges connecting them is the system.

The statement that none of this is systemic is such a farce.

What MFS was

Market Financial Solutions (MFS) was a Mayfair-based specialist lender founded in 2006 by Paresh Raja. Bridging loans and short-term buy-to-let finance for property borrowers who cannot meet high-street criteria. Company accounts showed a loan book of about £2.4 billion at the end of 2024, and the firm had borrowed more than £2 billion from institutional lenders. It employed nearly 150 people, had recently launched a new buy-to-let range with what it described as an additional £1 billion in committed funding, and, per press reporting on its final accounts, carried a clean March 2025 audit alongside record profits.

MFS took no deposits. It funded its loans by borrowing from banks and other lenders. Some of the lending was linked to high-value London property through companies connected to hundreds of homes in Knightsbridge, Belgravia and Mayfair.

The funder list, per court filings and press reporting: Barclays, Santander, Jefferies and Wells Fargo among the banks, alongside private credit and hedge fund firms including Elliott Management, Castlelake, Avenue Capital and Apollo’s Atlas SP Partners, with SMBC and Macquarie also reported to hold exposure.

Bloomberg drew the parallel within a day of the administration order: like Tricolor, a nonbank filling a gap the big banks had shunned while borrowing from those same banks to do it; like First Brands, lenders comforted by tangible collateral until double-pledging accusations arrived. We had drawn the same parallel months earlier, in the other direction. As always, ahead of the curve.

How it broke

Here is the public timeline, assembled from the court records.

According to The Telegraph, Barclays began blocking some transactions linked to MFS months before freezing the lender’s accounts in early January 2026.

Barclays knew. They always do.

On February 20, MFS applied to the High Court for administration, describing its crisis as a procedural matter with its primary banking provider. Two creditors, Amber Bridging Ltd and Zircon Bridging Ltd, took over the application, alleging “real and serious concerns” about mismanagement of the company. Per press reporting and Companies House records collated by researchers, the firm’s independent directors had departed in the weeks before the filing and Raja’s wife, a co-director, resigned days before it. By the administration date, Raja was the sole remaining director.

On February 25, the High Court approved the administration, and AlixPartners was appointed.

Then the numbers came out one by one. Per claim documents from Zircon and Amber, and from our critical sources, double-pledging may have resulted in an unaccounted-for deficiency of more than 80% on £1.2 billion of debt. Creditor estimates reported in the press put verifiable collateral at roughly £230 million.

The initial shortfall: £930 million.

First Brands, with land registry entries instead of invoices.

The escalation since, per court filings: a worldwide freezing order of up to £1.3 billion against Raja, sought by the administrators and granted by courts in London and Dubai, with a reported travel ban and a £5,000 weekly spending limit without consent. In a High Court claim filed May 5, the administrators describe the “systematic plundering” of £1.3 billion. Court filings alleged that eight companies presented as borrowers from the MFS group were in fact linked to individuals associated with Raja. Raja’s lawyers say mistakes were made, but there was no intention to defraud, and that he was not the beneficiary of any shortfall. The FCA opened an enforcement investigation on March 20. These are allegations. The legal process is still ongoing.

You can read about our MFS coverage in our newsletter.

What is not an allegation: from the creditors’ side, the money is gone until recovered, the intermediary network is in administration, and losses have already on bank income statements.

The exposure map

These are reported exposures, compiled from insolvency documents. These are just exposures, not final losses. Recoveries might reduce them, if there are any.

An unnamed Australian credit fund has also disclosed $80 million of exposure, per Contextual Solutions. Per administrator filings, two MFS units carried about £1 billion of debt from Atlas-controlled vehicles.

However, Barclays and HSBC deserve a more detailed look than the rest.

Barclays is the largest single creditor and was also MFS’s banker. It saw the anomalies first, blocked transactions, froze the accounts, and still took a £228 million charge. Its CEO described MFS as “a fairly deep and sophisticated fraud” and told analysts the bank will now restrict lending to structured finance counterparties that cannot demonstrate robust financial controls.

REALLY?

This is the same Barclays that disclosed a £110 million loss on Tricolor last October. Same bank, same product shelf, two collateral frauds, five months apart. No one cared.

HSBC is the one that proves our thesis. It is critical to understand that HSBC did not lend to MFS. Its exposure was tied to a credit arrangement with Apollo-backed Atlas SP, per Reuters, and surfaced on May 5 as a $400 million charge, described in its own disclosure as a fraud-related secondary securitization exposure with a financial sponsor in the UK. The shares fell 6% on the day.

HSBC’s CFO told reporters the issue was “idiosyncratic.”

There is that word again.

WE HATE THE WORD “IDIOSYNCRATIC.”

TEN WEEKS. That is how long it took a G-SIB to learn it was holding a nine-figure piece of a fraud it had no direct lending relationship with.

The risk had been warehoused and re-securitized until the institution holding it could not see the borrower. This is the mechanism from our SRT series: risk transferred on paper does not leave the banking system. It re-enters one layer removed, with the visibility stripped out.

HSBC discloses $111 billion of private-markets-related exposure, of which $22 billion is private credit. The $400 million charge is not the story. The $22 billion of exposure that surfaced its first fraud loss through a channel the bank would not name until sources did is the story.

The perimeter problem

MFS was not a regulated lender in any meaningful sense. Per the FCA’s own statement, MFS was an Annex 1 firm: registered with and supervised by the regulator solely for compliance with the money laundering regulations, not authorized or subject to wider FCA regulation.

Every institution in the table above lent, directly or indirectly, to a firm whose entire regulatory footprint consisted of an AML registration. Per HKA’s review of the FCA’s published Annex 1 supervisory work, the regulator had identified, before MFS collapsed, recurring weaknesses in this population including discrepancies between registered and actual activities and financial crime controls that did not keep pace with growth.

The red flags were not hidden either. In January 2026, weeks before the administration, the High Court handed down judgment in FP Redhill Ltd v Patel, a case concerning a £20.5 million property investment by the Fakhruddin brothers. Mr Justice Marcus Smith found that Raja’s response to a draft invoice showed “an obvious intention to create deceptive documents to justify illegitimate payments.” Two caveats belong next to that quote. The claimants had settled with Raja and MFS before judgment, on terms unknown to the court, and the judge acknowledged he was making findings about the state of mind of persons no longer before the court. The finding was untested by Raja at trial. It was also public, in a published High Court judgment, while institutional warehouse lines to MFS remained open.

As reported by the Telegraph, the PRA has requested information from lenders that backed MFS, is concerned about insufficient risk assessment and due diligence, and is examining whether banks carry indirect exposure through lending to private capital groups that backed MFS independently. Andrew Bailey, writing in the FT, has pointed to the interlinkages among private credit, banks, insurers, and private equity, and to the layers of leverage involved, and called for deeper scrutiny. The Bank of England has a system-wide exploratory scenario exercise on private markets underway.

Now, juxtapose that against “not necessarily a systemic issue.” The central bank governor is describing interlinkage and layered leverage. The restructuring industry is describing idiosyncratic broker failures. One of them is looking at the problem. The other is in the la-la land of denial.

The downstream

Here is what the Kroll data measures. MFS extended loans through a network of connected companies. Beyond the entities creditors allege were linked to Raja sat the legitimate tissue of the UK specialist property market: packagers, brokers, intermediaries, small lending shells. When the warehouse lines died, their revenue died. Per the Telegraph’s read of the Kroll data, they account for close to half of the 49 financial services administrations in the first half.

These are not headline names you will see in the news. They will not file 8-Ks. Their employees, their trade creditors, and their borrowers mid-completion absorb the loss silently.

The irony

Everything in this piece was knowable in advance. Most of it was known. It could have been prevented.

Dimon warned about cockroaches in credit after Tricolor and First Brands.

Well, with all due respect, Dimon should not be the one to take that moral high ground and talk about “cockroaches.”

The FCA had published its Annex 1 findings. A High Court judge had made findings about deceptive documents connected to Raja, in a published judgment, weeks before the collapse. Barclays itself was blocking MFS transactions, while other institutions were still funding them.

And the money kept flowing in. On January 13, 2026, five weeks before the administration order, Interbridge Mortgages publicly announced a new senior warehouse facility of up to £300 million from Citi. Citigroup agreed the funding line, £100 million of it, for the Raja-backed lender in January, just as MFS was hurtling toward collapse. There is no evidence Interbridge has been impaired by MFS’s failure. That is not the point. The point is that the diligence process of a US G-SIB signed a warehouse line into the same sponsor’s orbit weeks before the sponsor’s flagship filed, with a High Court judgment about that sponsor already in the public record.

The Mapping

We built a team for private credit mapping because of exactly this sequence. The model we are constructing does one thing: it takes the marks that funds report and tests them against the collateral, the borrower chain, and the underlying funding structure and more. MFS carried a clean March 2025 audit and recorded profits into the year it collapsed, against what creditors allege was an 80% collateral deficiency.

And here is the part we have made peace with, or tried to. The warnings do not stop the money. Funding into the sector continued after Tricolor. It continued after First Brands.

The next MFS is being warehoused right now, somewhere, with a clean audit and a record year. When it surfaces, the same institutions will take the same charges, the supervisors will request the same information, a CFO will call it idiosyncratic, and a restructuring firm will explain it was not necessarily a systematic issue.

We will have mapped it by then. That is the only part we could control.

We are getting requests from banks, funds, and others for access to our mapping. For now, we are offering only to limited individuals.

Unicus Research LLC publishes independent credit research for institutional clients only. Nothing here is investment advice or a recommendation to any retail investor. laks@unicusresearch.com. Unicus Research LLC © 2026

Sources

1. City AM, “Financial services bankruptcies rise as MFS collapse ripples through sector,” July 14, 2026: https://www.cityam.com/financial-services-bankruptcies-rise-as-mfs-collapse-ripples-through-sector/

2. The Telegraph, “Shadow bank collapse triggers surge in insolvencies” (Kroll H1 data, “almost half” attribution): https://news.google.com/read/CBMimAFBVV95cUxQMWZlRG5nWFZpTDBYcU55aUFRNkt4WG9aTlhNWlJKMEVmTUxDeEJXREl3eGtLdWw2aDBCcHNYMFZQVWRzbHptSWxRUTI0cDV4ZW9jN0NNTENBeHpJVEVJc1dFRWRwVlhCZ0pUaDJvQm1raG9CUGp6RmdzR0lUallqQjNUSk1veW5COVh3UjZnczN1M1lyWk9oZw?hl=en-US

3. Bloomberg, “MFS’ Collapse Ensnares Lenders Already Burnt by First Brands,” Feb 26-27, 2026: https://www.bloomberg.com/news/articles/2026-02-26/a-new-credit-blowup-in-london-has-wall-street-chasing-billions

4. Bloomberg, “MFS Creditors Warn of £930 Million Shortfall in Collateral,” Feb 27, 2026: https://www.bloomberg.com/news/articles/2026-02-27/mfs-creditors-warn-of-930-million-shortfall-from-double-pledges

5. Reuters, “HSBC shares slump 6% on surprise $400 million hit linked to MFS collapse,” May 5, 2026: https://www.msn.com/en-gb/money/other/hsbc-shares-slump-6-on-surprise-400-million-hit-linked-to-mfs-collapse/ar-AA22qecz

6. Bloomberg, “HSBC Profit Misses on Charges Related to the UK, Middle East,” May 5, 2026 (disclosure language; “idiosyncratic”; Atlas-vehicle debt per administrator filings): https://www.bloomberg.com/news/articles/2026-05-05/hsbc-profit-misses-on-charges-related-to-the-uk-middle-east

7. Bloomberg, “Citigroup Extends £100 Million Credit Line to Interbridge Mortgages,” March 19, 2026: https://www.bloomberg.com/news/articles/2026-03-19/citi-gave-100-million-credit-line-to-lender-backed-by-mfs-raja

8. The Intermediary, “Interbridge Mortgages secures up to £300m Citi warehouse facility,” Jan 13, 2026: https://theintermediary.co.uk/2026/01/interbridge-mortgages-secures-up-to-300m-citi-warehouse-facility-to-support-growth/

9. The Intermediary, “High Court case involving MFS founder raised questions weeks before lender’s collapse” (FP Redhill v Patel coverage, settlement and absent-party caveats): https://theintermediary.co.uk/2026/03/high-court-case-involving-mfs-founder-raised-questions-weeks-before-lenders-collapse/

10. FP Redhill Ltd v Patel [2026] EWHC 77 (Ch), Marcus Smith J (judgment text via ICLR): https://www.iclr.co.uk/document/2026000382/2026ewhc77ch_TNA/html

11. Reuters via Global Banking & Finance Review (FCA enforcement investigation, March 20; Annex 1 status): https://www.globalbankingandfinance.com/uk-regulator-launches-probe-collapsed-lender-mfs/

12. CNBC, “Why a small UK lender has major U.S. credit firms on edge,” May 18, 2026 (exposure figures citing FT-reviewed insolvency documents): https://www.cnbc.com/2026/05/18/mfs-private-credit-insolvency-banks-failure-collapse-barclays-mortgage.html

13. Mortgage Introducer, “FCA opens probe into collapsed lender Market Financial Solutions”: https://www.mpamag.com/uk/mortgage-types/bridging/fca-opens-probe-into-collapsed-lender-market-financial-solutions/569331

14. Mortgage Introducer, “Barclays to curb some lending after huge mortgage-related hit”: https://www.mpamag.com/uk/mortgage-industry/market-trends/barclays-to-curb-some-lending-after-huge-mortgage-related-hit/573304

15. BusinessCloud, “FCA opens investigation into collapsed MFS”: https://businesscloud.co.uk/news/fca-opens-investigation-into-collapsed-mfs/

16. HKA, “When confidence outpaces control: lessons from the MFS collapse”: https://www.hka.com/article/when-confidence-outpaces-control-lessons-from-the-mfs-collapse/

17. The Telegraph via Yahoo Finance, “Bank of England investigates collapse of £2bn shadow bank”: https://finance.yahoo.com/news/bank-england-investigates-collapse-2bn-190712226.html

18. Paul Quinn / theesk.org, “The Collapse of Market Financial Solutions”: https://theesk.org/2026/05/20/the-collapse-of-market-financial-solutions-lessons-to-be-learned-by-football-regulators-and-lenders/

19. Contextual Solutions, “MFS Collapse: A £2.4B Warning”: https://www.contextualsolutions.de/blog/market-financial-solutions-collapse-private-credit

20. The MFS Files (aggregator; director-departure timeline per Companies House records): https://www.mfsfiles.org/

21. Unicus Research, “Private Credit’s Global Contagion,” May 30 / June 10, 2026: https://contrarianunicus.substack.com/p/private-credits-global-contagion

We actually tried to remove the basic option entirely so no one would subscribe to a tier with little content. Substack does not offer a way to do that. We recognize this may have caused confusion for some of you, and we apologize for the inconvenience.

To make it right, we are offering a one-day discount (just today) on the founding member tier. If you upgrade today, you lock in the discounted price instead of the current $1,200 annual rate.

One more thing worth knowing. The founding member price will hold at $1,200 for the next few years. Whatever rate you join at, you are grandfathered in. It will not increase on you.

If you have questions about what founding members receive, just reply to this email.

Thank you for reading,

Laks Ganapathi

Founder, Unicus Research LLC

Well, last week we reviewed the loan book. BCRED, Blackstone’s $45 billion credit machine, prorated redemptions for the first time in its history. We read all 859 pages of its latest filings and parsed all 1,288 positions in its investment schedule to find out why.

A Bit of a Background

Five years ago, Blackstone launched a fund that let ordinary investors do what banks do: lend money to hundreds of American companies and collect the interest. It worked spectacularly as a fundraise. BCRED became the largest fund of its kind on earth, with $80 billion in loans to 695 companies, funded by $45 billion in shareholder capital and $35 billion in debt.

Then, this spring, the shareholders started asking for their money back. In the first quarter of 2026, they tendered 7% of the fund. The Board lifted its own 5% quarterly cap to pay everyone in full, $3.2 billion out the door, and Blackstone and its employees put their own money into the fund to help meet the wave; press reports of how much range from $150 million (CNBC) to roughly $400 million (AltsWire), and the fund itself has quantified it nowhere. One quarter later, requests hit 10%. This time the fund held the line at 5% and, for the first time ever, told the other half to get back in the queue.

The filings behind that decision are the most revealing documents the private credit boom has produced so far, and we went through them line by line. A few of the things sitting in there:

The fund’s biggest exposure to any single company is not a private equity buyout. It is $1.6 billion across two loans to Dropbox, a publicly traded company, under a facility that Dropbox itself said could fund its stock buybacks. While BCRED’s own investors were pulling $3.2 billion this quarter, the fund funded roughly $870 million of additional Dropbox draws, because it had promised to.

The share of the portfolio that has stopped paying quadrupled in a single quarter, from 0.6% to 2.4% of cost, as the borrower count nearly doubled from 9 to 17. One of them, a software company called Medallia, is carried at 62 cents on the dollar of its cost, and the recapitalization its lenders signed in June is set to wipe out roughly $5 billion of Thoma Bravo's equity in one of the largest private credit restructurings ever. BCRED is the biggest BDC holder of its debt.

The fund earned $0.54 a share this quarter. Markdowns took back $0.54. It paid a $0.60 distribution anyway.

Nearly a third of the borrower names in the loan book are not operating businesses. They are shells: Bidco, BorrowerCo, Purchaser, Topco. And nested inside the fund, off its balance sheet, is a second loan fund, levered at roughly 2.6 times, with a bad-loan rate already double the parent’s.

The full report traces who actually borrowed the $80 billion, sector by sector and name by name, and works through the question nobody in this market wants asked out loud: what does this fund’s NAV look like when the new money stops coming in, and what happens to 695 companies, and their employees, when the lender they built their balance sheets around goes into run-off?

The numbers are not reassuring, and it probably applies to every one of BCRED’s peers. We know, because we have already dissected them: Apollo’s ADS, Cliffwater’s CCLFX, now, BCRED.

Read more

Subscribe now

There is a new narrative forming in multifamily, and it goes like this. The Sunbelt is overbuilt, so the smart money should rotate to the Midwest. Columbus. Indianapolis. Kansas City. Cap rates that were never compressed by the institutional tidal wave, steady occupancy, modest but real rent growth.

Trepp just published a report making the case, and the numbers behind the pitch are real. Austin is sitting at 14% vacancy. Phoenix landlords are offering 3.5 months of free rent. Dallas-Fort Worth delivered 38,640 units in 2024 alone. Meanwhile, Columbus is posting 3 to 4% rent growth, and Milwaukee is running 95% occupancy.

Here is our problem with it. Our real estate expert analyst, , watched this exact play run in single-family residential in 2024 and 2025, and inventory in Indianapolis, Kansas City, Cleveland, Columbus and Cincinnati is already climbing in a way that looks very familiar to anyone who watched the South two years ago.

In this week’s full column:

1. Why is Midwest serious delinquency in CMBS almost matched by the Sunbelt, despite the friendlier going-in economics?

2. What the 2025 Census migration data shows, and who is behind it. Ohio swung from net domestic migration of -32,482 in 2021 to +11,926 in 2025, Michigan from -28,290 to +1,796. Our expert’s read is that a lot of those are pandemic movers coming home, plus temporary data center construction labor. If that is right, the gains are far more fragile than the headline suggests.

3. The risk hiding inside a 5.5% going-in yield on 1970s and 1980s vintage stock, and why you are going to be hearing the term “value-add” a lot.

4. The data center dependency. The Blackstone QTS cancellation in Virginia shows how quickly these projects can vanish.

And where our expert parts ways with Trepp entirely: re-entry timing in the South, the cities they recommend, water in Arizona and Nevada, heat, and the grid. And most importantly, how much is Blackstone’s QTS exposure to data centers?

Multifamily is a perilous proposition right now. Stay vigilant and stay frosty.

Read more

Subscribe now

On November 6, 2025, BlackRock TCP Capital reported a net asset value of $8.71 per share, unchanged from the prior quarter, and declared a $0.25 dividend. Three days earlier, its borrower HomeRenew Buyer, Inc., the entity behind the Renovo Home Partners remodeling brands, had filed Chapter 7 in Delaware. Bloomberg reported that funds managed by BlackRock were still marking the Renovo debt at par at the end of September. Weeks later, the mark was zero.

Eleven weeks after that dividend declaration, TCPC’s NAV printed $7.07, down 18.8% in a single quarter. The regular dividend was cut 32%, from $0.25 to $0.17. By March 31, 2026, NAV was $6.72. Since December 2021, NAV per share has fallen 53%. In May 2026, Bloomberg reported that the Manhattan US Attorney’s office has been examining how the fund valued its loans. The company’s own 10-K discloses a securities class action whose class period ends January 23, 2026, the week of the writedown disclosure.

This report goes beyond the headlines. We took the fund’s Q1 2026 SOI apart position by position: 323 positions, $1.39bn at fair value, $1.54bn at cost. We traced who the borrowers actually are behind names like Homerenew Buyer, Crewline Buyer, and AA Acquisition Aggregator. 42.3% of the portfolio’s fair value sits with borrowers named for the deal, not the business. We follow one loan from the fund through a private equity roll-up to 2,500 terminated employees and homeowners whose deposits are now unsecured claims. And we explain what TCPC’s forced shrinkage, from $1.9bn in total assets to $1.5bn in nine months, tells you about every private credit vehicle that can be redeemed.

If private credit or private equity were forced to liquidate their “assets” at scale tomorrow, where would they actually mark them to market? $0.10 on the dollar? $0.50 on the dollar? Zero? AND MORE IMPORTANTLY: WHO IS THE NATURAL BUYER CAPABLE OF ABSORBING THESE TOXIC ASSETS AT SCALE?

Private Credit has a brutal game theory problem coming. The first credible seller gets to set the clearing price while there is still buyer capacity; the last seller discovers that "NAV" stands for No Asset Value and Net Asset Value was only a model. First movers may take a haircut and survive. Late movers face an empty bid stack, adverse selection, and distressed buyers setting the terms. At the end, it starts to look less like orderly price discovery and more like the "we are selling to a willing buyer" scene from Margin Call. Except the buyer is only willing at a level that exposes the whole book.

The full analysis, including the borrower-level markdown ledger, the equity-for-debt restructuring lifecycle, and the valuation timeline the prosecutors will be reading, is below the paywall. And most importantly, this has a tremendous impact on the regular household.

Read more