A 24-year-old's AI fund blew up. Jane Street ate it.

Archived reading, published Sat, 29 Aug 2026 10:19:46 UTC (22 days ago). This is not the current state of the meter.

See the live reading →

CRASH-O-METER

Cracking

how close are we

-1 since the last reading

Fragility90

how much tinder is stacked up — moves slowly

Ignition34

how close a spark is — moves fast

Crash Lab watches the machinery under the current boom: the debt paying for AI data centres, private credit, the leverage that has moved out of banks into places nobody has to mark, the bond market and the dollar.

Every four hours it reads the day’s reporting from 7 sources and rewrites this page.

Why it moved: Down one to 62, on the spark side. The window's one genuine forced sale, Situational Awareness's portfolio being liquidated and Jane Street absorbing the hit, happened in July and cleared without anyone else being dragged into it. That tells us how a spark behaves in the current market rather than lighting a new one. The broad market is unchanged from last time (the VIX, the market's gauge of expected turbulence, sits at 14.43, which is calm; the extra interest investors demand for lending to shaky companies is 2.63 percentage points above government rates, and for solid companies 0.79 percentage points). The dry-tinder score holds at 90: Jefferies' finding that a fifth of direct lending sits in software improves our measurement of how concentrated private credit has become. It does not add new concentration.

Reporting from 27 Aug to 29 Aug

Crash points

Eight places it could go, scored 0–10. Tap one for the explainer.

What’s moving the needle

What changed in the last few hours, and what each one says about the plumbing.

Private credit and BDCs29 Augmedium

Private credit's biggest bet is the thing AI eats

The same institutions funding the disruption hold the loans against the businesses being disrupted, and both exposures are valued by the same people.

From Jefferies' latest GREED & fear report, two numbers that belong next to each other. About 70% of private credit lending goes to companies owned by private equity firms. And an estimated 20–25% of the private credit market is exposed to software, with direct lending to software-as-a-service companies reaching $538bn, or 19% of total direct loans, by end-2025. Software was the perfect asset for this kind of lending. Recurring subscription revenue, profit margins in the eighties, almost no money spent building things, customers who never switch. That profile is why the firms that buy companies could pile six or seven times the company's earnings in debt on a software business and lenders would fund it at floating rates without promises in the loan contract that actually bite. Now consider what the AI spending boom is a bet on. Several hundred billion dollars of chips are being financed on the thesis that writing and running software gets dramatically cheaper, and that charging per seat for routine business software stops working. If that thesis is right, the loans behind a fifth of the direct-lending market are worth less than the books say. If it is wrong, the chips are worth less than the books say. Blue Owl is on both sides: $2.4bn of chip-backed lending for Iren this week, and a large book of loans to private-equity-owned software companies. The early-warning numbers are consistent with strain building rather than a break. The share of loans where interest has stopped arriving: 2.4% at Ares Capital, 2.8% at Blue Owl, 2.9% at Golub, 3.6% at Blackstone Secured Lending in the second quarter, against 1.5%, 1.4%, 1.2% and 0.3% in the first quarter of 2025. In a healthy book those numbers are close to zero. Quadrupling in five months is not healthy. Global private debt has gone from $0.9tn in 2020 to about $2.1tn. The publicly traded funds that hold these loans do not believe any of this. ARCC, OBDC and BXSL are all within about 1–5% of their recent highs.

Household credit29 Aughigh

Brazil moves on consumer credit; Washington moves the other way

Two regulators looking at similar consumer stress are moving in opposite directions, and only one of them supervises the largest credit market in the world.

Brazil's central bank is preparing measures to restrain high-cost household lending. The trigger, in data released Friday: the share of household income going to debt payments excluding mortgages hit a record 26.6% in June, up from 25.7% at the end of last year. That means more than a quarter of what Brazilian households earn goes to paying off consumer debt before they pay for housing. More than half of Brazilian credit-card users, 52.8 million people, carry revolving balances or installment loans. The interesting detail is the choice of tool. The IMF recommended a cap on how much of any individual's income can go to debt payments. Two people familiar with the discussions told Reuters that policymakers will instead act on lenders first, limits on the products they can offer rather than on the borrowers who take them. That is a regulator deciding that the supply of expensive credit, not the demand for it, is the thing to constrain. Set that against the US, where the same week the OCC, the regulator that supervises national banks, and the FDIC announced they are refocusing bank supervision on "material financial risk", issuing a narrower definition of "unsafe or unsound practice" and raising the bar for formal warnings. There is a real argument for this: the warnings sent to Silicon Valley Bank before it failed were, as one regulator involved put it, "a random assortment" rather than about the actual risks. But the direction is unambiguous: fewer supervisory findings, at a moment when one of the three firms that grade borrowers has subprime car-loan payments running three or more months late at 6.13% in July, a record for the month and more than twelve times the rate for borrowers with good credit. The New York Fed has 5.5% of all car debt three or more months late, the second-highest reading in more than twenty years. In a normal year that number sits around 3–4%. The lenders' shares reflect none of it. Capital One, Synchrony and Credit Acceptance are each within 5% of their recent highs.

Hidden leverage and shadow banking28 Aughigh

Jane Street stopped being a market maker

The firm that absorbed the AI selloff's biggest single casualty is a private partnership whose finances nobody outside it can see, and we only learned its size because it lost money.

Jane Street lost money over a month for the first time in a decade. The Financial Times reports a $15bn hit to trading revenues, driven mainly by two things: its stake in Situational Awareness, the AI hedge fund run by 24-year-old Leopold Aschenbrenner, and its own bets on many of the same AI-linked stocks (SanDisk, Nebius, Bloom Energy). Aschenbrenner was forced to sell his positions during July's selloff. The FT does not say whether the $15bn is money actually lost or a shortfall against expected revenue, and the distinction matters a great deal. We cannot resolve it from the piece. The part worth understanding is not the loss but the drift. A market maker holds inventory for seconds and gets paid for the gap between the buying price and the selling price. It is financed, cleared and supervised on the assumption that it ends the day close to flat. Jane Street now takes positions over weeks. "When you start seeing the kind of revenue these guys are putting up, the only way that's possible is when your holding period extends as well as the size," Bloomberg Intelligence's Larry Tabb told the FT. "The longer your holding period, the more chance that something you can't forecast occurs." In other words, the firm is no longer doing what its plumbing was built for. And the seed stake and the firm's own bets were the same trade twice. Owning a piece of a fund that holds Nebius while holding Nebius yourself is one exposure wearing two hats, invisible in any position report that treats them separately. A separate market note says the SEC, the agency that polices American markets, sent subpoenas on August 24 to Bank of America, Citi, Goldman and JPMorgan seeking information on Situational Awareness's trades and borrowed money, and that the fund faced demands from its lenders for more cash with its portfolio down roughly 67%. Funds have bad quarters. Losing two thirds of your money in a month is not a bad quarter. Four major banks were involved, one fund blew up, and nothing else broke. The system took it.

The dollar, gold and reserve status28 Aughigh

Japan has spent $170bn this year buying its own currency

The largest foreign holder of US government bonds is now a forced seller of them for reasons that have nothing to do with its view on American credit.

Japan's Ministry of Finance disclosed on Friday that it spent ¥15.39tn ($96.5bn) supporting the yen between July 30 and August 26, the largest single intervention round on record. The 2026 total, including April and May, is ¥27.13tn ($170bn), more than the whole of 2024's ¥15.3tn and the largest annual figure ever. Currency intervention on this scale is not normal. The last time Japan spent anything close to this was 2022, and that was treated as extraordinary at the time. Here is the awkward part of the plumbing. To buy yen you must sell dollars, and Japan's dollars are mostly US government bonds. Japan's holdings stood at about $1tn in June, down 16% from a $1.25tn peak in 2021. Selling those bonds pushes US borrowing costs up, which widens the interest-rate gap that is weakening the yen in the first place. Each intervention buys time at the cost of making the underlying condition marginally worse. The July 31 operation was a joint one with the US Treasury Department, the first in 28 years, executed out of euro-denominated assets rather than dollars, which was presumably designed to avoid exactly this problem. The more interesting number is domestic. The interest rate on Japan's own ten-year government bond has gone from 0.4% to near 3%, a 30-year high. For the first time in a generation, a Japanese institution has a home-currency alternative that pays. Last week Japanese investors were net sellers of ¥2.8tn of foreign bonds and equities after two straight weeks of buying more than ¥2.5tn. But August as a whole is still net buying of ¥2.2tn, on track for the largest month in 14 months. The trend has not turned. Meanwhile the dollar index, which tracks the dollar against a basket of other currencies, is up 0.83% on the week and gold is down 3.2% over five days. Whatever is under strain here, it is not the dollar.

The AI capex bubble28 Aughigh

Iren plans $30bn of spending and pays 9% for the money

Private lenders are now lending against equipment on a two-and-a-half-year timeline to fund a buildout whose payback assumptions run far longer.

Iren, a Sydney-listed bitcoin miner turned AI data center operator, told investors on Thursday it expects as much as $30bn of capital spending in the 2027 financial year. On Friday its shares fell 12.6%, to 21% below their recent high. The funding announced alongside it is the part worth reading twice. Blue Owl and Pimco led $2.4bn for Iren: $1.2bn in senior secured loans, $1.2bn in senior secured notes, both at 9%, drawable in stages, coming due two and a half years after funding. The money buys Nvidia Blackwell Ultra chips for a campus at Mackenzie, British Columbia. Blue Owl's Kurt Tenenbaum described it to Bloomberg as "a combination of equipment financing for GPUs and a direct lending underwrite", and one of the largest of its kind. Two and a half years is a short fuse for equipment whose economic life is the entire question in this cycle. For the loan to be repaid on schedule the chips have to be earning contracted revenue almost immediately and the market for new borrowing has to still exist in early 2029. If either fails, the lender's remedy is possession of depreciating silicon in northern British Columbia. Watch the pattern more broadly. Nebius, which upsized its debt raise to $5bn, has reportedly begun pledging its infrastructure and its customer contracts as security, having already borrowed $775m in July against existing chips and cash flows from a highly rated customer. First the company's own money, then the chips, then the promises the chips are meant to fulfill. There is not much left after that. The giant cloud companies are on the other side of this trade and priced accordingly: Microsoft sits at its high, Amazon rose 4% on Friday. The companies borrowing to compete with them did not.

Fed, Treasury and policy28 Aughigh

Treasury's own alumni cannot explain the buyback

The Treasury Department is shortening the government's debt maturity into a meeting at which the Federal Reserve, America's central bank, may raise short-term interest rates.

On August 19, mid-quarter and outside the usual schedule for announcing borrowing plans, the Treasury Department said it would at least double its purchases of older government bonds that do not come due for ten to thirty years, from $2bn to at least $4bn per operation, effective September 9 through November 4. The interest rate the government pays to borrow for thirty years had just touched 5.34%, its highest since 2007. Buybacks of this kind are not unusual in themselves. Doing one without warning, two weeks after publishing the schedule that was supposed to contain it, is. Three former senior Treasury officials have now told Nikkei what they make of it. Nathan Sheets, Citigroup's chief economist and a former Obama-era under secretary, was the bluntest: "I do not think there were any meaningful signs of illiquidity in the market, and it is not clear what bar was met that triggered the expansion of the buybacks at this time. It was an unusual action. I did not find it shocking, but I found it puzzling." He added that the biggest driver of long-term interest rates is fiscal policy and how much the government borrows, not how easily bonds can be traded. That matches what the market did. Interest rates fell for a few hours and were back where they started by the next afternoon. Analysis cited in this week's sweeps found no failed auctions, no dealers running out of room on their books, no forced sales, the classic signs of dysfunction were all absent. These buybacks do not reduce how much the government borrows. They change the shape of it. The Treasury buys bonds that do not come due for decades and issues short-term debt instead, which lowers the extra interest lenders demand for tying money up for a long time but raises the government's exposure to whatever happens to short-term rates. Treasury Secretary Bessent has floated funding them from the roughly $1tn the government keeps in its checking account rather than issuing new short-term debt. The short end is precisely where the risk now sits: Kevin Warsh told the Jackson Hole conference that inflation is not meaningfully slowing and financial conditions are not tight, the odds of a September rate increase moved above 50%, and the two-year rate went to 4.32%. The next inflation report lands September 11. The Fed meets on the 15th and 16th.

Signs of the times

Stuff you wouldn’t have believed was possible until 2026.

The co-founder came to the wedding

Rob Granieri, the 54-year-old co-founder of Jane Street, attended the August wedding of Leopold Aschenbrenner, aged 24, days after Aschenbrenner's hedge fund had been forced to sell its positions. Jane Street had invested heavily in the fund and held many of the same AI stocks. A former Jane Street employee works there.

leverageconcentrationai

28 AugFinancial Times

Bitcoin miner forecasts $30bn of spending

Iren, which mines bitcoin and has a logo patch on the Golden State Warriors' jerseys, told investors it may spend up to $30bn on capital expenditure in the 2027 financial year, and borrowed $2.4bn at 9% from Blue Owl and Pimco to buy the first batch of chips. The shares fell 12.6% that day.

ai capexgpu debtneoclouds

28 AugBloomberg

Last seen: Peloton and Beyond Meat

A measure of how much convertible bonds (bonds that can be swapped for shares) are behaving like stocks rather than debt is at its highest since 2021, when Peloton and Beyond Meat raised billions on 0% interest. AI companies have driven global convertible bond sales to $147bn this year, more than 50% above the same point last year.

convertiblesrisk appetite

28 AugBloomberg

$170bn to defend a currency

Japan has spent ¥27.13tn ($170bn) buying yen so far in 2026, more than the entire 2024 total, including a record ¥15.39tn in the four weeks to August 26. Its US government bond holdings, the source of the dollars, are down 16% from their 2021 peak to about $1tn.

interventionreservestreasuries

28 AugNikkei Asia

The rumour mill

What the crash-callers are saying, checked against real reporting.

Partly true

“Nvidia wants to give OpenAI a $125bn guarantee.”

The talks reportedly ran to $250bn; the deal announced was a $105bn guarantee, and the Wall Street Journal put Nvidia's revised exposure in the Ohio data center structure at less than $120bn. The $125bn figure is not one anyone reported.

Claimed by Wealthion

Confirmed

“New Street Research estimates Meta's return on invested capital on its AI capex is negative 2.2% unless it rents out its compute.”

Barron's and other outlets carry the New Street forecast: AI-related return on invested capital of −2.2% in 2026, falling to −3.9% by 2030, unless Meta sells access to its data center capacity. Meta shares rose 5.15% over the past week.

Claimed by Meet Kevin

Confirmed

“The median AI spend inside US enterprises is $12 per person per month.”

Ramp's August AI Index puts median US company spending at about $11.95 per employee per month, up from roughly $5. Companies have borrowed more than $400bn globally this year to build the capacity that revenue is meant to fill.

Claimed by Meet Kevin

Partly true

“Salesforce borrowed about $25bn in order to buy back about $25bn of its own stock.”

Confirmed in substance: $25.0bn of senior notes issued in March 2026, proceeds used for a $25.0bn accelerated share repurchase that took delivery of roughly 103 million shares at about $198. Total buybacks in the quarter were larger, around $27bn, so not all of it was debt-funded.

Claimed by Meet Kevin

Earlier developments

Dispatches from previous readings. The same argument, no longer the news.

Hidden leverage and shadow banking28 Aughigh

Banks pause Delaware Life annuity sales amid probe

It shows how quickly traditional banks will sever distribution ties with private-capital-backed insurers when asset valuations are questioned.

Truist and Fifth Third have paused the distribution of annuities from Delaware Life, the insurer controlled by Mark Walter's TWG Global. The halt follows TWG's disclosure that it held more than $20bn in loans that should have been marked as affiliated assets but were not, prompting S&P and Fitch to lower their outlooks on the insurer. This is the mechanism by which shadow-banking opacity transmits stress to the real economy. Delaware Life relies on regional banks to sell its annuities to retail investors, bearing the underlying market and insurance risk itself. When an accounting mismatch in the private portfolio threatens the firm's credit rating, the traditional banking channel immediately shuts down the retail funding pipeline. TWG insists there has been no fraud and no one has been harmed. But the speed with which the banks stepped back demonstrates how fragile the funding model is: the moment the marks are questioned, the cash flow stops.

The AI capex bubble28 Aughigh

Lambda borrows $1bn to buy GPUs for Microsoft

It is vendor financing one step removed, allowing hyperscalers to expand capacity via off-balance-sheet neocloud borrowing.

Lambda has raised roughly $1bn in short-dated private debt, arranged by JPMorgan, to purchase Nvidia GPUs that will be leased to Microsoft. The deal follows a separate $926m loan Lambda closed earlier this month for infrastructure buildout. This is the core circular trade of the AI boom. A neocloud borrows heavily in the private credit markets to buy depreciating hardware, using the hyperscaler's lease agreement as the implicit credit support. The structure keeps the massive capital expenditure off Microsoft's balance sheet while generating immediate, debt-funded revenue for Nvidia. We do not know the exact terms of the lease or the debt. But the plumbing requires the hyperscaler to honor the lease for long enough to amortize the hardware. If the AI revenue does not materialize and the lease is broken or not renewed, the debt rests entirely on the residual value of used chips.

Hidden leverage and shadow banking28 Aughigh

SoftBank seeks $10bn loan for its OpenAI stake

It demonstrates the scale of traditional bank leverage being used to prop up private AI valuations.

SoftBank is seeking a $10bn, two-year loan at roughly 275 basis points over SOFR via Mizuho to help refinance the debt it used to invest in OpenAI. The proceeds will partly repay a $40bn bridge loan secured earlier this year. This is leverage stacked directly on top of venture capital. SoftBank is using floating-rate syndicated bank debt to fund an illiquid equity position in a cash-burning AI laboratory. The structure requires either OpenAI's valuation to continue rising so the stake can be borrowed against further, or SoftBank to cover the interest from its other operations. Borrowing billions in the loan market to hold private tech equity is the definition of fragility. If the AI valuation bubble pops, the equity gets marked down, but the bank debt remains at par.

Bond market dysfunction28 Aughigh

Zero-coupon convertibles hit $72bn as safeguards vanish

Investors are waiving basic fixed-income protections to chase AI equity upside, leaving them holding zero-yield unsecured paper if the stocks turn.

Companies have issued $72bn of zero-coupon convertible bonds this year, putting 2026 on track to break the previous full-year record. The issuance is dominated by AI-linked companies and digital-asset treasuries taking advantage of high stock volatility. Investors accept zero cash interest in exchange for a call option on the issuer's equity. If the stock rises, they convert and profit. If it falls, they are left holding unsecured debt that pays nothing. This makes the debt highly equity-like, stripping away the income cushion that normally protects bondholders during a downturn. AI and crypto firms are using this structure to fund massive capital expenditures without cash interest burdens, shifting the downside risk entirely to yield-starved credit investors. It is the exact dynamic that burned buyers of pandemic-era tech debt when rates rose and valuations collapsed.

Fed, Treasury and policy28 Aughigh

Regulators narrow bank supervision to 'material' risks

Bank regulators are explicitly stepping back from early intervention, allowing more risk to build before they issue formal warnings.

The OCC and FDIC are rewriting their supervision rules to focus strictly on "material financial risk." The agencies are issuing a new definition for "unsafe or unsound practice" and raising the bar for issuing Matters Requiring Attention (MRAs) to banks. The overhaul is an explicit deregulatory push designed to stop examiners from treating supervision as a compliance exercise. Under the new rules, regulators will likely stop issuing notices for minor infractions and focus only on risks that significantly affect a bank's balance sheet or liquidity. While this reduces compliance friction, it also removes the early-warning system that slows down bank risk-taking. By the time a risk is deemed "material" enough to warrant an MRA under the new standard, the leverage is already fully embedded in the system.

Private credit and BDCs28 Augmedium

Loans that stopped paying tripled; the funds rallied anyway

The whole case for private credit is that its losses stay small and its valuations stay honest. The first half of that claim is now visibly failing while the share prices behave as though it is not.

The second-quarter 2026 numbers are now in, and they are the worst of this cycle. A note from Jefferies, the investment bank, summarized in market coverage, puts the share of loans that have stopped paying interest at Ares Capital at 2.4%, Blue Owl at 2.8%, Golub at 2.9%, and Blackstone Secured Lending at 3.6%. In the first quarter of 2025 those figures were 1.5%, 1.4%, 1.2%, and 0.3%. Across the ten largest listed funds that borrow money, lend it to mid-sized private companies, and pass the interest to shareholders, loans that have stopped paying are 3.95% of debt portfolios at cost, or 5.95% if you count every slice of any borrower with one impaired loan: $3.3bn of an $83.6bn book, with $772m of interest income at risk. When a lender stops counting the interest on a loan because it has stopped arriving, that is an admission it no longer expects to be paid. That matters more here than at a bank, because these funds pay their dividends out of the interest they collect. Two things cushion the gap. One is interest paid by adding to the debt rather than handing over cash: a Boston Fed study cited in the same coverage has that practice rising from about 6% of these funds' loans in early 2022 to about 10% by early 2026. The other is that the loans are valued by the manager itself, not by a market. And the share prices say none of this is happening. Over twenty days FS KKR is up 15.9%, Blue Owl Asset Management 19.1%, Ares Capital 6.3%, Blackstone Secured Lending 7.1%. Every one of those is within about 5% of its recent high while the credit underneath deteriorates on a straight line. Either the stock market thinks 3.6% of loans going bad is the peak, or it is buying the fee stream and ignoring the book.

Private credit and BDCs28 Aughigh

An A$3.5bn builder funded by retirees just failed

It is the clearest live demonstration that private credit's risk is not just the risk of not getting paid back, but the risk of money being locked up in the hands of people who were sold a product for the interest it pays.

Bathla, a Sydney property developer with more than 520 subsidiaries, about 2,000 homes under construction, and roughly 13,000 more in the pipeline, has gone into administration owing creditors more than A$3.5bn. Administrators at Teneo put the figure at about A$3.3bn excluding the deposits paid by thousands of apartment buyers, and warned the company would burn A$40m before year-end if it kept building. The interesting part is who lent the money. According to the ABC, the bulk of it came not from banks but from private credit funds, money raised from self-funded retirees and other retail investors and lent on to property developers. Australia's version of the asset class has attracted close to A$250bn. ASIC, the agency that polices Australian markets, ordered a deep dive into the sector last year covering asset valuations, fee transparency, and conflict management. It has now described what it is seeing as the "first significant cracks," citing the collapse of several large borrowers and limited redemptions at several major funds hitting liquidity constraints. This is the mechanism Crash Lab keeps pointing at, running in a smaller and therefore faster market. A construction loan cannot be sold in a day. A retail fund that offers monthly or quarterly withdrawals against a book of construction loans is making a promise it may not be able to keep: it owes investors cash on demand, but its assets cannot be turned into cash on demand. There is no government backstop behind it. When one borrower of this size fails, the loss is not the only problem. The line of investors asking for their money back forces the fund to sell whatever it can sell, which is never the loan that went bad. Australia is not the United States and A$3.5bn is not large enough to threaten any financial system. But it is the first sizable test in this cycle of retail-funded direct lending against assets that cannot be sold quickly, and it did not hold.

Hidden leverage and shadow banking28 Aughigh

The market maker and the fund it seeded owned the same trade

Hidden concentration is the defining feature of borrowed money you cannot see, and here the concentration ran through one of the largest trading firms in the world rather than around it.

The Financial Times has filled in how Jane Street, one of the biggest market-making firms in the world, lost money over a month for the first time in a decade. The firm had invested heavily in Situational Awareness, the AI-focused hedge fund run by 24-year-old Leopold Aschenbrenner. A former Jane Street employee works there. And Jane Street was itself exposed to many of the same AI-linked stocks the fund had bought, including Sandisk, Nebius, and Bloom Energy. When those positions turned in July, the stake and the direct positions moved together. The result was a $15bn hit to trading revenues, and a letter to employees conceding the firm was not protected against some of the moves. The important point is about correlation, not size. A market maker's traditional risk is the inventory it holds for seconds. Jane Street, as Bloomberg Intelligence's Larry Tabb puts it, has extended both holding period and size: "The longer your holding period, the more chance that something you can't forecast occurs." Investing in a fund that owns the same stocks you own is the opposite of protection. It is the same bet, twice, with the second copy financed by someone else. That someone else is now being asked questions. Market commentary reports that on August 24 the SEC, the agency that polices American markets, sent subpoenas to Bank of America, Citi, Goldman Sachs, and JPMorgan in the Situational Awareness matter, seeking information on the fund's trades and its use of borrowed money after demands for more cash from its lenders and a portfolio hit of roughly 67%. In an ordinary bad quarter a fund might lose 10 or 15%. Losing two thirds of your money in a month is not a bad quarter. Nothing broke. The loss was absorbed seven weeks ago without a crisis in the firm's funding, which is genuine evidence of resilience. What it establishes is that the banks that lend to trading firms financed a concentrated AI-stock position at a firm that is itself one of the largest financial institutions in the US, and nobody outside the chain could see the overlap.

The AI capex bubble28 Augmedium

The chips are the collateral, the tenant is the credit

If debt secured by AI chips is really a bet on the tenant paying rent, then a single large customer walking away from a contract reprices an entire financing complex at once.

The pipeline this week, all of it secured or structured: Nebius upsized its raise to $5bn ($3bn of private convertibles due 2030 and $2bn due 2034) toward as much as $25bn of spending on building things this year, and has begun pledging its infrastructure and customer contracts as security, having already taken a $775m loan in July backed by existing AI chips and cash flows from a customer rated safe by the agencies that grade borrowers. Applied Digital is gauging interest in roughly $3.5bn of bonds from a company too shaky to be considered safe, for its Delta Forge 1 campus in Louisiana. Goldman is sounding out $800m for two Novva campuses in Nevada, and is expected to lead $1.085bn of secured notes for Digital Drive's CoreWeave-leased campus, issued through a separate company set up to hold the debt off the parent's books, affiliated with PowerHouse Data Centers and Chirisa Technology Parks. More than $410bn has been borrowed this year for data centers and AI. The structure is consistent enough to name. The borrower has very little of its own money in the deal. The security pledged to lenders is either an AI chip, which loses value over time and has a resale market nobody has stress-tested at scale, or a building, which is only worth the lease. The actual creditworthiness being underwritten in almost every case is the tenant (Microsoft, CoreWeave, an unnamed customer rated safe) and that obligation does not appear on the tenant's books as debt. The stock market is pricing this distinction with some precision. Microsoft closed Friday at its period high, exactly 0% off. The people borrowing against it did not: Applied Digital fell 7.6% on the day and sits 18.9% below its high, IREN fell 12.6%, Core Scientific is down 20.1% over twenty days and 31.1% off its high, Nebius 24.7% off. The extra interest that the riskiest corporate borrowers pay compared with the government, meanwhile, tightened to 2.63 percentage points. Someone is wrong. The bond market is lending into this at rates that imply almost nothing goes wrong. The stock market is marking the borrowers down a fifth to a third.

Fed, Treasury and policy28 Aughigh

Regulators raise the bar for telling a bank off

The intensity of bank supervision is falling at the same moment banks' largest growth area is lending to the funds and structures that took the risk off their books.

The OCC, the regulator that supervises national banks, and the FDIC, which insures bank deposits, announced new supervision rules on Thursday centered on "refocusing supervision on what matters most — material financial risk." The overhaul writes a new definition of "unsafe or unsound practice," a term previously left open to interpretation, and sets a higher bar for issuing Matters Requiring Attention, the formal notices examiners send when they want something fixed. The agencies say they are removing "Biden-era distractions" that turned examinations into a check-the-box exercise, and the OCC will publish its policies and procedures manual for the first time. The rationale is not silly. The reference case, cited by a regulator involved in the action, is Silicon Valley Bank: the Federal Reserve, America's central bank, had issued several notices about governance, controls, and anti-money-laundering, and none about the interest-rate and liquidity risk that actually killed it. The notices "should have been about the actual risks to the bank rather than a random assortment." Under the new rules the agencies would stop issuing notices for minor infractions such as breaches of the IT handbook. The question is what "material financial risk" is judged to include. These formal notices are the pre-enforcement instrument, the thing that happens before a legal order, and often the only thing that happens. Raising the threshold means fewer early, cheap interventions and a greater reliance on examiners correctly identifying the one exposure that matters, in advance. That is precisely what did not happen at Silicon Valley Bank. The exposure worth watching is bank lending to non-banks: credit lines, warehouse facilities to private credit funds, and the chip-backed structures now being arranged by JPMorgan and Goldman. On any reading that is material financial risk. It is also, from a supervisor's chair, the hardest kind to see, because the bank's own loan looks senior and secured. The security pledged is somebody else's book.

Bond market dysfunction28 Aughigh

Three former Treasury officials call the buyback puzzling

Shortening the life of the national debt to hold down long-term interest rates converts a problem of being locked in at high rates into a problem of having to borrow again soon, exactly as the Federal Reserve considers raising rates.

The critique of the Treasury Department's buyback of government bonds that do not come due for twenty or thirty years has now come from inside the building. Treasury buybacks are not unusual in themselves. Doing one without warning, outside the regular schedule that was supposed to contain it, is. Nathan Sheets, Citi's global chief economist and a Treasury under secretary under Obama, told Nikkei: "I do not think there were any meaningful signs of illiquidity in the market, and it is not clear what bar was met that triggered the expansion of the buybacks at this time. It was an unusual action. I did not find it shocking, but I found it puzzling." He adds the obvious point that the biggest driver of long-term interest rates is fiscal policy and the volume of borrowing. Separately, in a New York Times op-ed, Geng Ngarmboonanant, now a JPMorgan managing director, formerly deputy chief of staff to Janet Yellen, notes that the doubling of buybacks to at least $28bn of long-term debt over two months came outside the regular quarterly process where the Treasury announces how it plans to borrow, which is where debt-management changes are normally made. Within 48 hours the ten-year rate was back above 4.70%, higher than before the announcement. What a buyback does is not reduce debt. It swaps bonds that do not come due for decades for short-term bills, shortening the average life of the national debt and taking risk off private investors' books. That works if the problem is a temporary shortage of buyers for long-term bonds. It is expensive if the problem is the deficit, because the government has quietly refinanced itself at the short end of the borrowing schedule just as the new Fed chair is telling markets that conditions are not tight enough and the odds of a September rate increase have gone above 50%. The long end is calm today: the ten-year rate at 4.67%, the thirty-year around 5.17%, the MOVE index (the same idea as the VIX but for government bonds rather than stocks) at 11.22, up nearly 25% in five days but still low by historical standards. The intervention is being read as insurance, not distress. The awkwardness is that once you defend a level, you own it.

Private credit and BDCs28 Aughigh

Direct lending against chips, not cash flows

Private credit is moving from lending against businesses to lending against hardware, on loans that come due before the hardware has paid for itself.

Blue Owl funds, with Pimco alongside, have led a $2.4bn debt deal for IREN, split between a $1.2bn loan and $1.2bn of notes, both secured against the company's assets, both at 9%, both due two and a half years after the money arrives. The money buys Nvidia Blackwell Ultra graphics processors for IREN's Mackenzie campus in British Columbia. Blue Owl's Kurt Tenenbaum described it as "a combination of equipment financing for GPUs and a direct lending underwrite for Iren" and one of the largest of its kind (Bloomberg). This is a different animal from most private credit. A normal loan from one of these funds is secured against a company that earns money. This one is secured, in large part, against hardware whose resale value depends on whether the next generation of chips arrives early or late. Three things have to stay true for the lender to get paid back: IREN's contracted revenue shows up, the chips hold their value if it does not, and the borrower can borrow again to pay this off in early 2029. IREN told the market on Thursday it expects up to $30bn of spending on building things in the 2027 financial year. Its shares fell 12.6% on Friday and are down 15.4% over five days. The filing does not say how much the lender would get back if it had to seize and sell the chips, or what happens to that security if a new chip generation makes these ones obsolete before the loan comes due. Blue Owl manages $319bn and has already financed data centers for Meta. Across the ten largest listed funds that borrow money, lend it to mid-sized private companies, and pass the interest to shareholders, the share of loans where the lender has stopped counting the interest (because it stopped arriving) reached 3.95% in the second quarter, up from 3.75% in the first. The industry is adding a new, faster-depreciating kind of asset to a book that is already slowly getting worse.

The AI capex bubble28 Augmedium

Lambda borrows $1bn to buy chips Microsoft will rent

The strongest set of books in the chain is the one carrying none of the debt.

Lambda, an AI cloud provider backed by Nvidia, has raised about $1bn of short-dated private debt arranged by JPMorgan and placed with private investors. The money buys Nvidia graphics processors. The processors will be leased by Microsoft (Bloomberg). It follows a separate $926m loan that closed earlier this month for the same purpose, and comes while Lambda is in talks for a pre-IPO round of up to $3bn. Trace the circle. The chip vendor holds a stake in the buyer. The buyer borrows in the private market to purchase the vendor's product. The end user, one of the best-capitalized companies on earth, takes the equipment on a lease rather than buying it, so the asset and the debt sit on the small company's books and not on Microsoft's. Each party has a rational reason for its position. The combined effect is that the credit risk of a giant tech company's computing expansion is held by private investors, not by the giant tech company. What the reporting does not tell us is the shape of the Microsoft lease: how long it runs relative to the debt, whether Microsoft has to pay regardless of whether it uses the chips, and whether it can walk away. That single fact determines whether the lenders are really exposed to Microsoft or to Lambda. Nvidia has separately withdrawn the guarantee program that used to sit behind deals like this, where the chip maker would promise to rent the equipment back if the buyer could not. So on current evidence there is no vendor safety net here. More than $400bn of debt has been raised globally for AI this year. The share of it that ultimately rests on one lease clause is the number nobody has.

Private credit and BDCs28 Augmedium

Two banks stop selling Mark Walter's annuities

Insurance-funded private credit shrinks from the funding side first, and this is the funding side.

Truist and Fifth Third have paused distribution of products tied to Delaware Life Insurance, the largest insurer controlled by Mark Walter's TWG Global, while a US probe of Walter's empire proceeds (Bloomberg). The banks were acting as sales channels through branches and advisor networks; the insurance and market risk sat with the insurer. The background is a June disclosure that TWG's two insurers held more than $20bn of loans that should have been marked as affiliated but were not. S&P and Fitch, two of the three firms that grade how likely a borrower is to pay you back, cut their outlooks on Delaware Life in July. S&P held an A- rating while flagging that it could go lower. TWG says there "has been no fraud" and "no victim here". The mechanism worth understanding is the direction of the money. A company that sells policies paying a saver a monthly income for the rest of their life takes the premiums and invests them in longer, harder-to-sell credit, increasingly credit created by a manager it is affiliated with. The policy sales are the funding leg. Cut the distribution and the investment side stops growing; it does not immediately shrink, because these policies are stickier than bank deposits and usually carry penalties for cashing out early. There is no reporting of policyholders rushing to cash out here, and we should not imply any. But this is the first thing in weeks that looks like a lender pulling back rather than a number being revised. Insurance company books are the largest single destination for the risk that left the banks, and the way that pipe narrows is exactly this: a distribution partner declines to keep selling while a rating is under review.

The dollar, gold and reserve status28 Aughigh

Japan has spent $170bn buying its own currency

The largest foreign holder of US government bonds is selling them to defend its currency, and a domestic interest rate near 3% gives its savers a reason to stop sending money abroad.

Japan's Ministry of Finance disclosed on Friday that it spent ¥15.39tn ($96.5bn) between July 30 and August 26 supporting the yen, the largest single round on record. Adding April and May, the 2026 total is ¥27.13tn ($170bn), the biggest annual intervention ever, nearly double 2024's ¥15.3tn (Nikkei). Here is why this belongs in a US newsletter about financial risk. To buy yen, Japan must sell dollars. To get dollars, it sells US government bonds. Japan's holdings of those bonds stood at about $1tn in June, down 16% from a $1.25tn peak in 2021. Every yen-buying operation is, mechanically, new supply dumped into a market where the interest rate the government pays to borrow for thirty years has been near 5.3% and the Treasury Department has doubled its buyback operations to at least $4bn each to hold rates down. Which explains the otherwise odd fact that Washington joined Tokyo in a joint yen-buying intervention on July 31, the first in 28 years. That is not routine. Governments almost never coordinate currency operations. Supporting the yen is a way of stopping Japan from having to sell more US bonds. Two governments are jointly managing one price to protect another. The second leg is quieter and matters more. The interest rate on Japan's own ten-year government bonds has gone from 0.4% to about 2.9%, a 30-year high. Japanese investors sold a net ¥2.8tn of foreign bonds and stocks last week, after two weeks of buying more than ¥2.5tn each. August as a whole is still net buying. One week is not a trend. But the decades-long arrangement whereby Japanese savings funded everyone else's bond markets only works while there is nothing worth buying at home.

Crypto and TradFi contagion27 Aughigh

The dead model is up 37% in a month

A structure can be both discredited and still capable of moving 37% in a month; the borrowed money does not go away because the story did.

The FT published an analysis this week finding that the 50 largest companies that hold bitcoin as their main treasury asset have gone from a combined $150bn of market value in July 2025 to $67bn; that 43 of the 50 now trade below where they were before announcing the pivot, with 35 down by at least half; and that Strategy alone accounts for $79bn of the decline from its peak. One researcher called the frenzy "dead" (FT). The market disagrees, at least about direction. Over the last twenty days Strategy is up 36.5%, Bitmine up 37.7% and SharpLink up 33.9%, against bitcoin up 19.3%. On Friday all three fell about 7% while bitcoin fell 3.5%. Both things are true and the reconciliation is the point. These companies are not a business model. They are a leveraged bet on a coin, and roughly a two-times leveraged bet at that, in both directions. That is precisely what the FT's numbers describe: a structure that works while the shares trade above the value of the coins, because selling new shares adds value, and collapses when they do not, because the only way to keep going is to sell coins or issue debt. The $83bn of destroyed value is what happened when the premium vanished. The 37% bounce is what happens when the coin moves 19% and the leverage is still there. So the honest reading is not that the market has refuted the FT. It is that the sell-off the FT documents has left behind a smaller, still-leveraged instrument, and $67bn of it is a fraction of the $150bn that has already been written down without incident.

Hidden leverage and shadow banking28 Aughigh

Convertibles with no coupon and no cushion

Investors are financing the AI buildout by selling insurance against volatility and taking no income for it, which works until volatility arrives.

Global issuance of convertible bonds (bonds that can be swapped for the issuer's shares if the stock rises enough) has hit $147bn this year, more than 50% above the same point last year, and $72bn of it pays no interest at all, already within a whisker of the $73bn full-year record set in 2025 (FT). Bloomberg reports that a standard measure of how stock-like the convertible market has become is at its highest since 2021 (Bloomberg). A convertible is a bond plus the right to buy the issuer's shares at a set price. The more volatile the shares, the more that right is worth, and the less interest the issuer has to pay to sell it. Nicolas Crémieux of Mirabaud put it plainly: "The companies coming to this market now are far more volatile than the ones they replaced. If you're selling something more valuable, you can offer less for it." At zero interest, the investor's entire return is the bet on the stock. There is no income to cushion a fall in the shares. If the stock goes nowhere for three years, the holder has lent money for free; if it falls, they own an unsecured claim on a company that chose this market because borrowing the normal way was expensive. ON Semiconductor sold $1.3bn at 0% in May, Ciena almost $2.9bn at 0% in June. The last time the market looked this stock-like, the issuers were Peloton and Beyond Meat. The defenders' argument, that 2021 broke because of interest rate hikes, not structure, is worth taking seriously, but the odds of a September rate hike are now above 50%. And the concentration is invisible one deal at a time: a portfolio of thirty convertibles can be one bet on AI volatility.

Fed, Treasury and policy28 Aughigh

The Fed chair says rates may need to go higher

Nearly every floating-rate loan in private lending and every AI construction project was priced on the assumption that rates were heading down. The new Fed chair just made a rate increase the market's best guess.

Kevin Warsh, in his first speech since becoming Federal Reserve chairman in May, told the Jackson Hole conference that "we must be confident that underlying inflation is moving to our objective, clearly and at sufficient speed. Otherwise, we have work to do." He added that borrowing conditions across the economy are not currently tight enough to slow inflation, and that interest rates are the Fed's "predominant tool." The numbers he cited: the Fed's preferred inflation measure running at 3.7% over twelve months and 4.1% over six, with 54% of the items in the index above 3% over a year and 49% over six months. The interest rate on two-year government bonds rose as much as 0.09 percentage points to 4.32%, while the rate on bonds that do not come due for thirty years fell slightly, according to Bloomberg. The implied odds of a rate increase in September went above 50%, up from around 36%. The next consumer-price report lands September 11; the Fed meets on the 15th and 16th. The mechanism worth holding onto is not the rate increase itself. It is that almost every structure we track was built on the assumption that the next move was down. Loans made by private lending funds charge a floating rate: the borrower pays a fixed margin on top of a benchmark rate (currently 3.64%) that moves with the Fed. A rate increase raises the interest payment on the same loan books where borrowers have already stopped paying at the highest rate in nine years. It improves the lender's income on paper and worsens the borrower's ability to pay in practice, at the same time. Debt financing AI data centers is being priced off today's interest-rate curve for assets meant to last fifteen years. And it collides with the Treasury Department, which borrows the money the government spends. Treasury Secretary Scott Bessent is buying back government bonds that do not come due for decades, trying to hold those long-term rates down, while the Fed prepares to push short-term rates up. The gap between two-year and ten-year rates sits at just 0.47 percentage points. There is not much room left in the curve to absorb a disagreement between the two arms of the state about which direction rates should go.

What would change our mind

The specific, observable things that would move the number - in either direction.

The extra interest investors demand for lending to shaky companies rising above 3.5 percentage points, or for solid companies above one percentage point. That would mean credit markets are finally treating AI equipment debt and loans to private-equity-owned software companies as the same trade, rather than ignoring the risk in both.

Would move the number

A US fund that lends to mid-sized private companies blocking or capping withdrawals the way MA Financial and Centuria Bass have in Australia, or one of those funds trading at less than 80 cents on the dollar of its stated value while borrowing costs stay low everywhere else.

Would move the number

A second disclosure at Jane Street's scale: a bank, dealer or market maker publicly restricting lending against baskets of AI-linked stocks, or another firm reporting a month of losses from the same positions.

Would move the number

On the other side: SoftBank placing its $20bn bond and clearing the $40bn OpenAI bridge loan, or an AI-linked private loan changing hands near full value in a visible secondary sale. Either would let us take the dry-tinder score down.

Would move the number