Newsletter · 2026-08-19 · Ashutosh Agarwal

Critics Fuse the AI Financing Wave and the Private Credit Boom Into One Risk - Private Credit & Alternatives - Week of August 19, 2026

Private Credit & Alternatives for the week of August 19, 2026. The week's most credible skeptics stopped treating AI data-center financing and the private-credit boom as separate stories, Nvidia offered to absorb the first 25 percent of any loss on chip-backed loans, and the 2027 refinancing wall got a working timetable.

Private Credit & Alternatives

Week of August 19, 2026: Critics Fuse the AI Financing Wave and the Private Credit Boom Into One Risk

Last week ended on a hopeful note. Redemptions at the core of private credit were cooling, and the industry had found a giant new borrower at the frontier: the companies building artificial-intelligence data centers. Two separate stories, one calming down, one revving up.

This week the skeptics came back, and they did something more unsettling than simply repeat the bear case. They stopped treating the two stories as separate. In a run of unusually candid podcasts, some of the most credible critics in the business argued that the AI-financing wave is not a break from the private-credit boom, it is the same boom, wearing a new costume. The same investors are funding both: insurance companies, pension funds, and yield-starved savers. The same machinery sits under both: assets that are hard to sell, valued by a model rather than a market, sold on the promise of a smooth, high income stream. And the same underlying danger runs beneath both. As one guest put it, financial disasters are almost never caused by a risky asset behaving riskily, they are caused by a risky asset that everyone had been told was safe.

Two things also happened this week that filled the biggest gaps from last week's issue. First, the long silence from the insurance world finally broke, though not from the insurers themselves. A veteran bank analyst walked through a live bankruptcy that he says is a preview of "what the demise of private credit is going to look like," and explained why ordinary people holding annuities could be the ones who get hurt. Second, the vague warning about a "2027 refinancing wall" got a precise, operational timetable from someone whose actual job is managing that exact risk. And at the frontier, NVIDIA tried to answer last week's central worry, that the graphics chips backing all this new debt could lose their value, by promising to absorb the first chunk of any loss itself. That promise is clever. It also quietly piles an enormous amount of risk onto a single company.

The mood, in short, flipped back. Last week the relief was winning. This week the people who see trouble coming were louder, more specific, and, for the first time this cycle, telling one connected story instead of several scattered ones.

TL;DR

The insurance warning everyone was waiting for finally arrived, from a critic, not a company. On The Julia La Roche Show, analyst Chris Whalen called the bankruptcy of an investment firm called 777 Partners "what the demise of private credit is going to look like", a slow-motion unwind with "literally hundreds of affiliates" and "probably fraud." His bigger point: firms like Apollo and Blue Owl "buy an insurance company because they use it as a balance sheet for their private credit strategies," financed by the annuities and life-insurance policies ordinary people buy. Several insurers tied to 777 already failed, and "if there's any consumers out there who have annuities written by these companies, they could lose some or all of their money."
The frontier AI bet is really a private-credit bet, and it's a debt-fueled one, which is worse. On Monetary Matters with Jack Farley, the Financial Times' Robin Wigglesworth said the tech giants' off-the-books obligations, leases and purchase commitments to build AI data centers, jumped from roughly $1 trillion to about $1.5 trillion in a single quarter, citing Goldman Sachs's tally. He estimates the headline $500 billion NVIDIA financing package is "roughly 80% debt." His verdict: "This is a debt cycle… that actually makes me more worried," because debt-fueled building booms "quite often end in tears."
NVIDIA tried to kill the depreciation worry by eating the first 25% of losses itself, and investors promptly started buying insurance against its loans. On The Investopedia Express, Caleb Silver explained that Jensen Huang said NVIDIA "might in some cases use its own balance sheet to backstop as much as 25% of the cost of a project," a "residual value support mechanism." At the same time, "you are seeing a lot of investors buying credit default swaps against NVIDIA's loans", protection that "is spiking right now." Investor Vincent Daniel described the same 25% first-loss piece on RiskReversal Pod.

What's new

1. The insurance silence breaks, with a live bankruptcy as the warning. For two straight weeks, no one holding the industry's most opaque risk, the insurers, would come to a microphone. This week the silence broke, but from a critic looking in, not an executive explaining himself. On The Julia La Roche Show (Aug 15), analyst Chris Whalen, who writes The Institutional Risk Analyst, used the collapse of an investment firm called 777 Partners to argue that this is the shape of things to come. 777, he explained, "got involved in all kinds of things, insurance, reinsurance, sports teams. They own a bunch of soccer clubs. They were financing a Canadian airline… there was no limit to what these guys would get involved in, and that typifies private credit. You almost can't define it." Several of its principals were accused of fraud by the SEC and stepped away; the firm has now filed for bankruptcy in Texas, on top of other bankruptcies already in place for companies it owns, and it has "literally hundreds of affiliates."

His point is not really about one firm. "This is what the demise of private credit is going to look like," he said. "It's going to be a big mess, and it's going to take years to unwind. There's probably fraud involved in many of these aspects." Because these are private, not publicly traded, companies, "you vote on them" only when they run out of money and land in court, so the reckoning arrives slowly, in a courtroom, not suddenly, in a stock price.

Then the part that should matter to anyone with a retirement policy. Insurance companies, Whalen noted, "are supposed to be islands of stability. They hold assets at book value", meaning at the price they paid, not what the market would pay today. "But ever since the private credit guys got involved with it… you've had people like Apollo and Blue Owl and all the rest of them go out and buy an insurance company, because they use it as a balance sheet for their private credit strategies. And on the other side of the balance sheet, you have annuities and life insurance, which finances all of this." Translation: the private-credit firm buys an insurer, uses the insurer's money to make private loans, and pays for it all with the annuities and life policies sold to ordinary savers. "So if you're a consumer and you have a life insurance policy or an annuity written by a company that's controlled by a private credit fund, you may have something to worry about." He noted that several insurers tied to 777 have already failed, and that "if there's any consumers out there who have annuities written by these companies, they could lose some or all of their money." His frustration was that Congress isn't asking about it: members "are so clueless about just about everything having to do with the economy or finance that they don't ask any questions."

Whalen also flagged a second stress point in the mortgage world that shows the same machinery at work. He argued United Wholesale Mortgage had for years been "overstating the value of his mortgage servicing assets so he could go borrow money from banks," estimating those assets were overvalued "by about 20, 25%." Oaktree, now owned by Brookfield, put "a billion five" into the company at a 10% coupon to rescue it, and Whalen called Oaktree "the grim reaper of Wall Street… Oaktree only shows up when you're having problems." It's a colorful line, but the substance matters for this beat: the same distressed-credit specialists now central to the private-credit story are quietly cleaning up messes across the lending world, and the marks, the values people carry these assets at, keep turning out to be too high once someone looks hard.

2. The frontier bet is a private-credit bet, and Robin Wigglesworth put numbers on it. The single most important idea this week came from Monetary Matters with Jack Farley (Aug 16), where the Financial Times' Robin Wigglesworth, who has covered bonds for decades and has "been borderline obsessed with private credit for a long time", connected the AI buildout directly to the private-credit machine. His starting fact: the tech giants' off-the-books obligations have exploded. Lease commitments to build data centers "went from roughly $1 trillion to roughly $1.5 trillion" between the first and second quarters, using Goldman Sachs's numbers; separately, purchase commitments, promises to buy chips, cooling, and electricity, "also went from… roughly $1 trillion earlier this year to $1.5 trillion." Of the $1.5 trillion in leases, only about a third has actually started and shows up in the accounts; "a trillion dollars of that is for leases that haven't even started yet. And that doesn't appear except as a footnote."

The plainest example ties the AI story straight back to a private-credit giant. Meta is building a huge data center in Louisiana called Hyperion. Rather than pay for it directly, Wigglesworth explained, "they're only investing 20%… but they are guaranteeing that they will lease that data center for 20 years," in a joint venture with Blue Owl, and "they'll sell those bonds to other investors. So it's off balance sheet. It doesn't come up as a bond or a debt or a loan for Meta. But of course, it's on the hook for paying this lease for 20 years." Google, he added, has now disclosed purchase commitments of $800 billion, "almost half the total" for the whole group, of which roughly $200 billion is short-term, meaning due within about a year.

On the headline $500 billion NVIDIA financing package, Wigglesworth was blunt: he estimates it is "roughly 80% debt, 20% equity," and "this is a debt cycle… that actually makes me more worried." His historical read is the crux of the bear case: equity-funded booms (the dot-com bust) can pop and cause surprisingly little damage, but "debt-fueled capex cycles are very different. Like, even when… the underlying premise comes true and AI kind of transforms the world, quite often they end in tears." He urged listeners "to take press releases with a pinch, maybe even a fistful, of salt."

On private credit itself, he was nuanced and therefore more damning. He thinks it's "a fantastic asset class" that genuinely "does de-risk the financial system" by moving lending out of banks, but "way too much money" flooded in "too quickly," and it got "deployed poorly." His wake-up moment: as a working journalist, he started getting cold calls offering him private loans, "$1 million working capital, three, four percentage points above LIBOR", and thought, "nobody should lend any money to any journalist ever." When lenders are that indiscriminate, "you don't need to find high-quality borrowers. You just need to find borrowers to take the money so you can earn fees on it. It's very similar to what we saw in 2008." His forecast: "there is a default cycle that is probably going to be far worse than what the backward-looking numbers look like," though "not… catastrophic," and the asset class "deserves to survive and thrive" after it learns its lesson. He is especially wary of two things: unrealistic recovery assumptions (high-yield bonds might recover "70, 80 cents on the dollar," but a failed software company has "no plants and factories… there's nothing there for you as a creditor"), and payment-in-kind, where a strained borrower pays interest with more debt instead of cash, "it's unambiguous that lots of private credit funds have been using [it] as a way of deferring the pain." His bluntest structural view: illiquid loans should never be sold to savers in a semi-liquid wrapper, "if you invest in loans with a five-year tenor, then you should be locked up for five years", and business development companies (listed funds that lend to mid-sized firms) shouldn't be levered on top of already-levered borrowers. In a crisis, he said, "some of these public BDCs are probably going to go to 30 cents or 40 cents of net asset value."

3. NVIDIA answers the depreciation worry, by taking on the first loss itself. Last week's newest debate was whether a bond backed by a fast-aging graphics chip could ever be safe. This week NVIDIA offered an answer, and it is a big one. On The Investopedia Express (Aug 17), Caleb Silver explained that NVIDIA "will offer loans to its customers that are building out their data centers," backed "by the same AI hardware that is being bought or leased." More striking, Jensen Huang posted on X that NVIDIA "might in some cases use its own balance sheet to backstop as much as 25% of the cost of a project," which he called a "residual value support mechanism." In plain terms: if the chips backing a loan fall below a set value, NVIDIA promises to make up the difference, up to a quarter of the project's cost. It is the chipmaker volunteering to eat the depreciation everyone was worried about.

Silver's caution was that this makes NVIDIA look "more and more like GE back in the 1990s with GE Finance", a manufacturer turning itself into a lender to keep its own products selling, and that the whole edifice still rests on chip prices staying high. "At some point in time, chip pricing is going to peak and then it's going to start coming down… NVIDIA can out-innovate itself and come up with even cheaper, faster, better chips," and then "the guarantees on all these loans that NVIDIA is making against its own hardware… they're priced at yesterday's price." The market is already reacting: "you are seeing a lot of investors buying credit default swaps against NVIDIA's loans", insurance against those loans going bad, and that protection "is spiking right now." Silver's frame is worth holding onto: the mortgage bonds of the 1970s were also "a very good, reasonable idea," and BlackRock's Larry Fink has himself compared this moment to their birth, but the cleanup after 2008 "was very ugly."

Investor Vincent Daniel picked apart the same structure from the money side on RiskReversal Pod (Aug 14). NVIDIA is "basically the first loss… 25% on depreciation of GPUs," he explained; the alternative managers "raise the capital… from sovereign wealth, pension, insurance, family offices," package it into "securities that are collateralized by the GPUs," and sell it on, collecting a fee of "between 50 and 100 bps" plus a possible incentive fee, "just collecting a toll." Why does it work? Because, quoting a fellow investor, "there's an insatiable appetite… if you could promise them high single-digit yields with no volatility… the investor [market] is infinite." The scheme solves the bears' first fear, that there wasn't enough money to fund the buildout, by tapping "a new pocket of capital because the debt securities market was probably tapped out" and the tech giants' own balance sheets were "tapped out." His unresolved worry is the second bear fear: the return on all this spending. Enterprises are "heat-seeking to the lowest cost on their tokens," including cheaper open-source and Chinese models, which points to a price war, and "the 80% margins that we're seeing at the memory names and at NVIDIA are unsustainable." His base case: "at some point we will see a bust. This probably prolongs the boom." He also made the same Meta Hyperion observation as Wigglesworth, the Louisiana data center was funded "through an SPV and KKR and Blue Owl," "the same sort of institutions that are funding this facility." Two independent voices, one week, on the same 25% backstop and the same cast of lenders: the AI trade and the private-credit trade are now, unmistakably, the same trade.

4. The "2027 refinancing wall" gets a precise timetable, from someone who does this for a living. Last week Steve Eisman warned that private credit's real test would come when software loans start refinancing "sometime next year." This week that abstract fear got a working schedule. On The Credit Edge by Bloomberg Intelligence (Aug 13), David Seidowitz, head of U.S. liquid credit at ICG, a $126 billion alternative manager, and a former decade-plus Apollo portfolio manager, laid out the mechanics. The problem loans come from the "2021 LBO vintage" (buyouts done with borrowed money in the zero-rate era), and "25% of that cohort" at the riskier end "are software," with weakness spreading to "software adjacent or AI risk adjacent" services companies.

The timing is the part worth internalizing. "2028 and 2029 are meaningful walls," he said, not because of the total size, but because of the makeup: "over 50% of the 28 maturities are B3 or lower" (deep junk). Those are companies that locked in "a cost of capital in 2021 that was so attractive", near zero, "that they didn't bother to reprice or refinance… now that option is coming to an end." And the deadline is closer than the calendar suggests. "If it's a 28 maturity, January 28… if it's not refinanced by January 27, you risk downgrade to CCC, which means that you start losing two-thirds of your audience, right? Because the CLO market's two-thirds of the total loan market." In other words: a loan due in early 2028 effectively has to be refinanced by early 2027, or it gets downgraded into a rating that the biggest buyers, CLOs, which are bundles of loans, cannot hold. "We're kind of in the window in the next several months."

Seidowitz's overall read is measured, not apocalyptic: he thinks the wall can be handled "orderly and methodically" if sponsors move early ("loans are prepayable, usually at par… it's not a terribly expensive option to extend yourselves out"), and he's encouraged that new deals are pricing "at wider spreads" and, notably, "with tighter documentation", a small win for lenders after years of borrower-friendly loan terms. But he flagged that "JP Morgan has defaults going up next year… doubling if I recall," and that "seven times levered LBOs that are B3s" will get squeezed as bigger, higher-quality borrowers, like Oracle, "very close to falling into high yield" with more than $100 billion of debt, crowd them out of the market. On the AI financing itself, he was in wait-and-see mode: "we had heard about all this financing, but hadn't seen a lot of it… no two deals are exactly alike… counterparties are going to matter, structure is going to matter." And he does think "there's a bubble" in AI funding, "I don't know how big it is."

5. The marks-and-liquidity debate gets its sharpest data, in a conversation replayed this week. Money Stuff re-aired its interview with hedge-fund manager Boaz Weinstein of Saba Capital this week (Money Stuff: The Podcast, posted Aug 14). It is a re-run, not a fresh conversation, but it remains the most concrete walk-through of the liquidity trap at the heart of the retail private-credit boom, so it's worth surfacing. Weinstein's core charge is that BDCs and "interval funds" (semi-liquid funds that let investors redeem a set slice each quarter) "over-promised liquidity to their investors. You know, you have fire insurance in a sell-off that you can get out, but it doesn't work if there's actually a fire." His example of the fire: at one fund, Cliffwater, redemption requests went from about 1%, to 4%, to "the shocking number of 14 percent." And he laid out the doom loop, borrowing a term from George Soros: "the reflexivity of falling NAVs leading to larger outflows, leading to forced selling, leading to falling NAVs. And then you're out in three or four years."

Weinstein is not just talking; he's trading it. Saba (with a partner firm, Cox) is tendering, publicly offering to buy, shares of private funds run by the likes of Blue Owl and Starwood at steep discounts, bidding in a "20 to 35" range (about 27 cents on the dollar below stated value for the Starwood real-estate fund). Why bid at all if he's so bearish? Because at a discount your yield rises: "instead of… getting 7% at 100, you're getting 11%… at 65. You're clipping your checks." His most damning observations were about the marks, the values managers assign to their own loans. He said Apollo is "considered the most conservative on the marks," but "the marks are falling as we speak," and for some chunky positions the values are "off by… as much as 25 points" between one fund and another, or versus how banks like JPMorgan have already remarked the same exposures. Public BDCs, he noted, already trade at big discounts, an Apollo BDC "at a 30 discount," some names near "60 cents on the dollar", which makes it "really hard to justify private BDCs at much smaller discounts." He pointed to distribution cuts as the trigger for the whole outflow wave: savers were promised "1% a month," then got "84 basis points" as floating rates fell, "and that kind of started… the avalanche," compounded by the first defaults (he named First Brands and Tricolor). His two-word summary of the problem, borrowing Cliff Asness's phrase, was "volatility laundering", the way monthly, model-based marks make private credit look smooth and safe when the underlying loans are neither. The industry, he noted, is now "two and a half trillion."

6. The operator's rebuttal: Brookfield says it's a "healthy adjustment," and points to its own redemptions. Against all this, the loudest operator voice of the week pushed back, from inside a firm running the money. On Brookfield Perspectives (Aug 13), CEO Bruce Flatt's Q2 shareholder letter addressed the private-credit worry head-on. "Much has been written lately about the risks in private credit. Concerns center on direct lending to sponsor-led private equity-backed borrowers and exposure to software businesses." Brookfield's answer is that it simply isn't in those corners: its credit franchise is "concentrated in… real asset finance, asset-backed lending, and opportunistic credit," leaving it "with limited exposure to the sectors and investor channels currently under the greatest pressure." His verdict: "We do not… view today's environment as a systemic problem… What we are seeing is merely a healthy adjustment following a period in which abundant capital led underwriting standards to become too loose in parts of the market."

Flatt backed it with a hard number, the same redemption metric everyone is watching, but pointing the other way. "This quarter, redemption requests at Oaktree's… business development company dropped below 5%, even as many others saw redemptions climb into the high double digits." (Brookfield completed its acquisition of Oaktree, the distressed-credit specialist, in July, capping a seven-year partnership; more than 60% of its investment staff are now U.S.-based, which also positions it for inclusion in major U.S. stock indices.) He also planted Brookfield's flag firmly in the two growth stories: the 401(k) opening and the AI buildout. On retirement money, he called the roughly "$14 trillion" U.S. defined-contribution market a source of capital where "even modest allocations to private assets have the potential to create one of the largest new sources of long-term capital for our industry," and pointed to a collaboration with AllianceBernstein to put Brookfield strategies into custom retirement plans "alongside existing target-date funds." On AI, he tied credit to power: Brookfield increased its framework with Bloom Energy "five-fold to invest up to $25 billion" in behind-the-meter fuel cells for data centers, and highlighted a U.S. Department of Energy "$17.5 billion commitment" to finance equipment for "a further 10 new large-scale reactors" served by its Westinghouse nuclear business. This is the bull case delivered by an operator: the stress is real but contained to other people's books, and the two big secular tailwinds, retirement money and AI power, are ours to ride.

7. Retail's biggest on-ramp yet, a Blackstone-Wellington-Vanguard fund for $2,500. The retail push that has run under this whole cycle took its most concrete step yet. On Invest Like a Billionaire (Aug 18), the hosts (who run an alternatives sponsor themselves) dug into a new product from Blackstone, Wellington, and Vanguard, three of the biggest names in investing, designed to hand ordinary savers access to private markets. There are actually two funds, one "100% pure alts" and one blending public markets with "around 40%" Blackstone alternatives. The features are genuinely new for the retail channel: a "$2,500 minimum," "you don't have to be an accredited investor," a "daily NAV calculation" so buyers pay net asset value rather than a premium, and a liquidity sleeve offering "3% per quarter" redemptions on the private version.

Even these enthusiasts had two reservations, and both are central to the beat. The first is who's picking the investments, the classic fear, in one host's words, of "let's go get the Muppets to buy the junk we can't convince anyone else to buy." Their reassurance is structural: Wellington, "a fiduciary," is "solely in charge" of selection and "have to be convinced" to buy Blackstone's deals, so Blackstone "is not in the driver's seat." The second reservation is fees, and here they disagreed on the exact numbers, which is itself telling. One host cited an analysis (by Layla Cooney of Accredited Investor Insights) putting total annual expenses at "about 3.4% for class I and over 4% for class M"; another countered that his own read showed "under 2%", "1.39%… or 1.89%" for the blended fund and "1.43 or 2.28" for the all-privates version. Both agree that headline number sits on top of the underlying Blackstone fees, "management fees of 1 to 1.25%, plus the carry of 10 to 15%", so the product "has to significantly overperform… a traditional 60/40 portfolio" just to keep up. Their deepest structural doubt was about size: Blackstone is "so big" that "everything they do is these billion-dollar funds," which "disincentivizes really anything creative," versus the "Goldilocks zone" of hungrier emerging managers. Their bottom line captured the ambivalence exactly, "I rejoice and I also cringe."

8. The CLO desk's view: skip the private-credit version, and software isn't the monster it looked like. For the syndicated-loan and CLO plumbing beneath all of this, the week's clearest read came on The CLO Investor Podcast (Aug 12), where Sam Garza, a portfolio manager at DoubleLine, explained why his firm avoids the private-credit version of these securities. (A CLO, or collateralized loan obligation, is a bundle of corporate loans sliced into safer and riskier pieces and sold to investors.) DoubleLine's stance comes "from the top", from CIO Jeffrey Gundlach, and from Garza's own scar tissue: pre-2008, his middle-market loans looked fine right up until they "just came in one month" with "massive downgrades," precisely because they weren't marked to market. "You can truck along and truck along… and then only when things are bad enough, it sort of happens all at once." The extra yield on private-credit CLOs, about "40 bps" more at the safest AAA level ("160 bps over SOFR for private credit" versus "120" for the ordinary syndicated kind), isn't enough to tempt him, because in a downturn "that paper will be harder to move." Notably, he runs a lower recovery assumption than the industry standard, "60% in general" rather than 70%, which is exactly the too-optimistic-recovery worry Wigglesworth raised, here baked into a real firm's models.

Garza's more surprising message was reassuring on the topic that spooked everyone earlier this year: software loans. Software has grown into "the largest portion of the bank loan market," which by itself "there's probably going to be some problems there during the next cycle." But the panic has cooled. Back "in February and March, people were very, very concerned" and wanted to "throw out all software"; since then, a widely-read research report shifted the mood toward a name-by-name view. Managers now tell him they have "10% software loans" but are "worried about 1% or 2%… certainly not worried about 10%," partly because "switching costs are very high" for the software these loans funded, and the private-equity owners "have their own playbook" to use AI rather than be killed by it. Garza's own concern is "more about how big the sector is" than about AI specifically. He also echoed a structural complaint that keeps surfacing: "captive" CLO equity funds, vehicles set up mainly to keep issuing deals, have flooded the market with "deals that shouldn't exist," compressing returns for everyone else.

9. Secondaries and evergreens keep building, with a new buyer and an honest caveat. The exit-and-access plumbing kept expanding. On Private Markets 360° (Aug 13), Peter Aliprantis, head of private wealth for the Americas at EQT (which he notes is "the second largest private equity firm in the world," with no credit arm today), described the firm buying a secondaries specialist, Collar Capital, to add both private-equity and credit secondaries, a deal closing in "mid to late August." His pitch for credit secondaries, the business of buying investors' existing stakes in private-credit funds so they can cash out early, is a long runway: "$200 to $255 billion in activity" last year, "which only accounted for about two to three percent of the total private equity volume."

He was also candid about the "evergreen" vehicles (funds with no fixed end date that let investors buy in and redeem periodically) now carrying private assets to wealthy individuals. EQT runs three, a REIT, a private-equity operating company, and an infrastructure operating company, and "distribution partners… are raising more money from investors in the evergreen structures than they are in traditional drawdown structures," a shift that "started probably about a year or two ago." Crucially, he insisted "there is no separate origination team or deal team" for the evergreens, so retail buyers get "the same deals" as sovereign-wealth and institutional clients. The demand, in his telling, is structural: with U.S. public companies shrinking "from roughly 8,000… to around 4,000," and "roughly 90 percent of the companies that have $100 million in revenue or more" now private, "if you aren't looking at private market exposure, you're missing out on a huge portion of the global economy." It's the cleanest statement of the industry's growth thesis, delivered, as always, by someone selling the product.

The debate

Calm, or the quiet before the bill comes due? Last week the bulls got their data point, redemptions cooling. This week the bears took it back, not with panic but with a calendar. The steel-man for the optimists is still real and got a fresh, credible voice: Brookfield's Bruce Flatt, running the money, called this a "healthy adjustment," not a systemic problem, and pointed to his own Oaktree fund's redemptions "below 5%" while others climbed higher. If you're diversified into real-asset and asset-backed lending rather than software buyouts, the stress is genuinely someone else's. The steel-man for the pessimists is that the most operationally credible voices put the real test on a specific date. ICG's David Seidowitz, whose literal job is refinancing this debt, said the 2028 wall means a "January 27" deadline, "in the next several months," with "over 50%" of those maturities deep junk and JPMorgan's defaults "doubling" next year. Wigglesworth expects "a default cycle… far worse than what the backward-looking numbers look like." And Weinstein's reflexivity loop, falling values force selling that forces values lower, explains how a slow drip becomes an avalanche. The honest synthesis is unchanged from last week but sharper: the visible symptom (redemptions) is easing at the best-run shops, while the chronic one (a covenant-light, software-heavy, 2021-vintage loan book that must refinance into higher rates, with recoveries penciled in too high) is still building toward a date now marked on the calendar. Relief at the top of the quality ladder does not settle what happens at the bottom.

The GPU-bond question, now with a partial answer, and a new place the risk hides. Last week's open question was whether a bond backed by a depreciating chip could be safe. NVIDIA's 25% "residual value" backstop is a real, if partial, answer: the manufacturer will absorb the first slice of any value shortfall, which genuinely de-risks the senior lenders and helps explain the "insatiable" demand Vincent Daniel described. The bull case is that this is a proven playbook, planes, cars, and mortgages have all been financed this way, now applied to the most valuable computing asset on earth, with the vendor itself standing behind it. But the skeptics landed two blows. First, the backstop doesn't remove the risk; it concentrates it, onto NVIDIA's own balance sheet, at the exact moment its "80%… margins," in Daniel's view, are "unsustainable." And the market is voting: credit default swaps against NVIDIA's loans are, per Investopedia, "spiking." Second, Wigglesworth reframed the whole thing as an off-balance-sheet debt cycle, $1.5 trillion of hidden lease and purchase commitments, a $500 billion package that is "80% debt", and warned that debt-funded booms "end in tears" even when the technology delivers. The swing factor to watch is no longer just chip depreciation; it's whether NVIDIA's own credit stays pristine while it becomes the lender, insurer, and vendor to its entire customer base at once.

Are the marks real? This is the question that has run under the entire cycle, and this week it got its most concrete airing. The bear evidence is specific: Weinstein bidding 27-to-35 cents below stated value for private-fund shares and finding marks "off by 25 points" between funds; Whalen documenting servicing assets overvalued "20, 25%"; DoubleLine quietly using a 60% recovery rate where others use 70%; public BDCs already trading at 30%-plus discounts to the values their private cousins still carry. The bull rebuttal is that the best operators are conservative and are being tested and passing, Apollo "the most conservative on the marks," Brookfield's Oaktree fund honoring redemptions below its cap. Both can be true: the marquee names may be honest and still sit in an industry where the average mark is soft, and "volatility laundering", smooth, monthly, model-based valuations, makes the whole asset class look calmer than the loans underneath. The tell to watch is the gap between what public BDCs trade at and what private funds claim they're worth. It has not closed.

Access, or the Muppets buying the junk? The Blackstone-Wellington-Vanguard fund is the cycle's clearest test of whether private markets can be sold to ordinary people responsibly. The bull case, made even by skeptical hosts, is that the design is thoughtful: a fiduciary (Wellington) picking the deals so the sponsor can't dump its worst; daily pricing at net asset value; a $2,500 minimum that democratizes access. The bear case is a stack of fees, plausibly 2% to 4% all-in once Blackstone's own management fee and carry are counted, that demands real outperformance just to beat a plain 60/40 portfolio, and a liquidity promise (3% a quarter) that is exactly the "fire insurance that doesn't work in a fire" Weinstein and Wigglesworth warned about. Wigglesworth's line is the one to keep: don't sell an illiquid asset in a semi-liquid wrapper, because "retail investors… pull their money out when they're afraid, and these structures aren't built for that."

Where the risk actually sits. The most pointed thing said this week about the insurance balance sheets holding private-credit assets came from a critic: Chris Whalen warning that annuity holders at private-credit-owned insurers "could lose some or all of their money." Ticker-level BDC candor was again mostly secondhand: the one hard number came from Brookfield about its own Oaktree fund (redemptions "below 5%"), while the discounts and distressed marks came from an outside bidder, Weinstein, rather than from the managers carrying the assets.

The names in play

NVIDIA (NVDA): moved from counterparty to lender-of-first-resort this week, offering to backstop up to 25% of a project's cost through a "residual value support mechanism", and drawing a spike in credit default swaps against its loans. The single company now anchoring the AI-credit market, with margins two investors this week called unsustainable.
Blackstone (BX): on both frontiers at once, a named member of the $500 billion NVIDIA consortium, and the marquee sponsor behind the new Blackstone-Wellington-Vanguard retail fund ($2,500 minimum, no accreditation, 3% quarterly liquidity, layered fees). Also cited by Weinstein, from the short side, as the manager that has treated redeeming investors best.
Blue Owl (OWL): the recurring name on the risk side, the private-fund shares Saba is tendering for at a discount, the JV partner on Meta's off-balance-sheet Hyperion data center, and, per Whalen, one of the private-credit firms using an owned insurer "as a balance sheet."
Apollo (APO): named by Whalen as an insurer-owner using annuities to fund private credit; praised by Weinstein as "the most conservative on the marks" even as those marks fall; a named member of the NVIDIA financing consortium. Its stock, Vincent Daniel noted, has been "in the penalty box" over software exposure.
Brookfield / Oaktree (BAM/BN): the operator counterpoint of the week. Completed the Oaktree acquisition in July; reported its BDC's redemptions "below 5%"; called the environment a "healthy adjustment," not systemic; and is pushing into both 401(k) access (via AllianceBernstein) and AI power (a $25 billion Bloom Energy framework, plus Westinghouse nuclear). Oaktree also surfaced as the distressed rescuer ("the grim reaper") of United Wholesale Mortgage.
ICG (private / ICG LN): via David Seidowitz, the week's most precise voice on the refinancing wall, a "January 27" deadline for 2028 maturities, "over 50%" of them deep junk, and a bubble in AI funding of unknown size.
KKR (KKR), BlackRock (BLK), Goldman Sachs (GS): the other named members of the NVIDIA consortium. Goldman, notably, is making new highs by capturing the banking and advisory fees "without taking that lending risk," while the pure alternative lenders' stocks lag.
EQT (EQT): buying secondaries specialist Collar Capital to add private-equity and credit secondaries, and leaning hard into evergreen funds, the second-largest PE firm in the world making its move on the wealth channel.
DoubleLine (private): via Sam Garza and CIO Jeffrey Gundlach, the house that won't touch private-credit CLOs, runs a below-market 60% recovery assumption, and thinks the software scare has been overdone name-by-name.
Saba Capital / Cliffwater / Starwood (private): the live liquidity trade, Saba (Boaz Weinstein) bidding 27-to-35 cents below stated value for private-fund shares; Cliffwater the fund whose redemptions ran to 14%; Starwood's real-estate fund among the discounted targets.
Meta (META) and CoreWeave (CRWV) / Oracle (ORCL): the borrowers. Meta's Hyperion data center is the archetype of off-balance-sheet AI financing (20% owned, 20-year lease, Blue Owl JV). CoreWeave was called "the real king of debt," and Oracle "the weakest of the litter" among the tech giants and "very close to falling into high yield."
777 Partners (private/defunct): the bankruptcy Whalen holds up as the template for "what the demise of private credit is going to look like", hundreds of affiliates, alleged fraud, and failed insurers whose annuity holders may lose money. The most conspicuous cautionary tale of the week.
The insurers, Athene, Corebridge, Global Atlantic, F&G, Brookfield Reinsurance (various): the balance sheets carrying the most opaque private-credit risk, and now the subject of a public warning from Whalen about their annuity holders.

Read-throughs

The AI trade and the private-credit trade are now one trade, analyze them together. Two independent voices this week (Wigglesworth, Vincent Daniel) described the same structure: illiquid assets, financed with debt, marked to model, funded by insurance and pension and retail money, sold on a smooth high yield, with the same lenders (Blue Owl, KKR, Apollo, Blackstone) on both sides. Stop watching "private credit stress" and "AI capex" as separate dashboards; the risk, the capital, and the players overlap.
The refinancing wall now has a date, and it's sooner than 2028. ICG's rule of thumb, a January 2028 maturity must be refinanced by January 2027 or risk a downgrade that locks out two-thirds of buyers, means the pressure starts building over the next several months, not in two years. Watch for the pace and pricing of 2021-vintage software and services refinancings, and for whether the new discipline (wider spreads, tighter loan documents) holds.
NVIDIA's credit is now a systemic variable. By backstopping 25% of project costs, NVIDIA has made its own balance sheet the shock absorber for the entire AI-credit market, so the market has started buying protection against it. The single most important new indicator is the price of credit default swaps on NVIDIA's loans: if that protection keeps getting more expensive, it is the market pricing the "latter innings" Vincent Daniel warned about.
The marks-to-market gap is the master signal. Public BDCs trade at 30%-plus discounts; private funds claim far smaller ones; outside bidders like Saba will only pay 27-to-35 cents below stated value. Until that gap closes, either private marks fall or public discounts recover, assume the smooth, model-based "volatility laundering" is masking real losses, and treat any fund's own valuation as a starting point for negotiation, not a fact.
Depreciation and recovery rates are where the losses will actually show up. The two numbers quietly doing the heavy lifting across every structure this week are the same: how fast a graphics chip loses value (NVIDIA is now insuring the first 25%), and how many cents a defaulted loan actually returns (DoubleLine uses 60%, Wigglesworth thinks software recoveries could be near zero). Both are assumptions, not observations, and both are set optimistically. When they're tested, that's the loss.
The insurance channel is where the retail exposure sits. How a private-credit-owned insurer values its loan book, and what that means for the annuities funding it, is the least-explained link in the chain and the one Whalen argues reaches ordinary savers first.

What changed

Last week the story was relief at the core and a bigger bet at the frontier: redemptions cooling, even the loudest bear conceding the panic had "calmed down for now," and the industry unveiling a half-trillion-dollar plan to bankroll the AI buildout. The two developments felt separate, one a de-escalation, one an expansion.

This week the skeptics returned and fused them. The most credible critics stopped narrating two stories and started narrating one: the AI-financing wave is itself a private-credit phenomenon, funded by the same capital, built on the same illiquid-and-marked-to-model machinery, exposed to the same danger of risky things dressed as safe. Robin Wigglesworth put $1.5 trillion of off-balance-sheet numbers on it and called it "a debt cycle." Vincent Daniel traced the money from sovereign-wealth and insurance pockets through the alternative managers to the GPU-backed securities. And the two loudest gaps from last week closed: the insurance silence broke (Chris Whalen, warning that annuity holders could be the ones who lose), and the "2027 refinancing wall" got a working timetable (ICG's David Seidowitz, marking January 2027 as the real deadline).

At the frontier, NVIDIA tried to settle last week's depreciation debate by promising to eat the first 25% of any loss, and the market answered by buying insurance against NVIDIA itself. The operators, meanwhile, held their ground: Brookfield's Bruce Flatt called it a "healthy adjustment," pointed to his own fund's sub-5% redemptions, and kept building toward retirement money and AI power. And the retail on-ramp got its most concrete step yet, a Blackstone-Wellington-Vanguard fund open to anyone with $2,500, arriving in the same week the critics explained, in detail, why illiquid assets sold to skittish savers are exactly the wrong combination.

The cycle's question has evolved again. It began as "will private credit blow up," became "what does it look like when it stops growing," then "who's holding the asset when the professionals trim." Last week it added "is the industry diversifying into AI or just moving its valuation problem to a shinier collateral pool." This week it collapsed all of those into one: if the private-credit trade and the AI-financing trade are the same trade, same money, same machinery, same soft marks, then there is only one question left, and it is the one that has always mattered here. Are the values real? This week, more people with real credibility said, out loud, that they doubt it.