The bank said no to Aschenbrenner. A smaller broker said yes.

Archived reading, published Sat, 12 Sep 2026 13:41:34 UTC (8 days ago). This is not the current state of the meter.

See the live reading →

CRASH-O-METER

Breaking

how close are we

Fragility91

how much tinder is stacked up — moves slowly

Ignition45

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 19 sources and rewrites this page.

Status: Held at 68. It is Saturday, and the prices in front of us are the same September 11 close we read last run: the S&P 500 is 1.8% off its high; the VIX, the market's gauge of how much turbulence traders expect over the next month, reads 15.8 (under 20 is calm; it passed 80 in March 2020); the interest rate the US government pays to borrow for ten years is 4.975%; and the debt of companies too shaky to be considered safe paid 2.7 percentage points more than the government on Thursday. Only bitcoin has updated, flat at $77,290, so ignition holds at 45. Fragility holds at 91. JPMorgan ending its lending to Situational Awareness is a lender pulling back from a fund that had already sold down its borrowed bets, and the fund rebuilding them through Clear Street was counted last run. Nvidia's possible $10bn anchor stake in Anthropic's share sale is a plan, not a position. DeepSeek's cheaper model is a threat to the assumption behind the AI building spree that the market has not yet priced, and nothing was sold off.

Reporting from 10 Sep to 12 Sep

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.

Fed, Treasury and policy12 Seplow

The Treasury buys the long bonds. The Fed edges toward the short ones.

Banks running short of spare cash, while the Treasury is already stepping in to buy long-dated bonds, leaves the central bank less room to raise rates without breaking something in the short-term cash markets that keep trading desks running.

John Williams, who runs the New York branch of the Federal Reserve, America's central bank, which sets interest rates, said this week the Fed is “getting closer” to restarting bond purchases in order to keep the banking system supplied with cash, that the shrinking of its holdings will halt in December, and that banks should use the Standing Repo Facility freely, according to market commentary summarized here. That facility is the Fed's backstop window for lending banks cash overnight against bonds put up as security. The same source puts use of the opposite facility, where banks and money funds park spare cash with the Fed overnight, at $626mn on September 8, down from $702mn on September 3. The Fed's total holdings stood at $6.7tn on September 2. A central banker inviting banks to lean on a backstop is not a weekly occurrence, and it matters because two arms of the government are now moving toward buying government debt at once. We have not seen the speech text and this is a single second-tier source, so hold the specifics loosely. The mechanism is not in doubt. The parking facility is the overflow tank: when banks and money funds have more cash than they want, it sits there. When it is nearly empty, every dollar the Fed lets roll off its holdings comes directly out of the cash banks keep at the Fed, and the lending window becomes the pressure valve. A central banker telling banks to use it “freely” is saying the tank is empty before the pipes groan. Put that beside the Treasury Department, which borrows the money the government spends. It began an expanded program of buying back its own bonds this week, with weekly limits reported at $14.5bn, and its first large operation bought $5.19bn of a $6bn maximum while interest rates rose (FT). The Treasury buys the government IOUs that do not come due for twenty or thirty years and pays for them by selling short-term ones. The Fed, if Williams is right, will soon buy those short-term IOUs to top up bank cash. Two arms of the state buying their own paper at both ends, both insisting it is technical. Perhaps it is. Buying bonds to manage the cash in the system is not the crisis-era program known as QE, and Williams was careful to say so. But the Fed is being pushed toward buying at the moment traders put roughly 70% odds on it raising rates on September 16. A central bank raising the price of money while adding to the quantity of it is not an emergency. It is the position a central bank finds itself in when the g

Private credit and BDCs12 Sepmedium

The market has already sorted the private lenders

The gap between the falling shares of the firms that run these funds and the unchanged price of risky debt is the clearest measure we have of how much of private lending's value rests on nobody challenging the prices the funds put on their own loans.

Nine of the funds that borrow money, lend it to mid-sized private companies and pass the interest to shareholders (BDCs, a name that will recur) trade for more than their holdings are said to be worth, and they carry average non-accruals, loans the lender has stopped counting the interest on because it stopped arriving, of 0.9% of assets. Ten trading at more than a 30% discount to their stated worth average 2.6%, per a sector analysis summarized here. Investors, in short, have already decided which of these funds to believe. Morningstar estimates $772m of interest at risk from borrowers who have stopped paying, against $38bn of cash interest across the sector, about 2% of the income stream. KBRA, which grades borrowers, found the typical rate of stopped loans at fixed-life BDCs was 2.75% of cost in the second quarter, up from 1.81% in the first; among the twenty largest it is the highest since 2017. None of that is new risk; it is second-quarter data being published, so our measurement of the problem improved rather than the problem growing. What the numbers describe is a sorting. The discounts line up with the stopped loans, which is either the market being efficient or a prophecy that fulfills itself, since a fund trading at a 30% discount cannot sell new shares to grow its way out of trouble. The Loparex loan shows why a 1% figure earns little trust on its own. OBDC said its second-ranked loan to the company was worth 88 cents on the dollar at the end of 2025, 63 cents in March, and about 5 cents now (matterfact). Blue Owl's BDC has traded at a 22–25% discount all year while reporting stopped loans near 1%. Glendon Capital flagged “inconsistencies” in the prices the funds put on their loans earlier this year. Now look at where the price damage is. The funds themselves are down modestly: ARCC 2.3% off its high, BXSL 2.4%, OBDC 6.7%. The firms that run them are down far more: Blue Owl 16.5% in 20 days, Blackstone 13.9%, KKR 12.3%, Ares 11.9%. The debt of companies too shaky to be considered safe pays 2.7 percentage points more than the government, and HYG, the big fund that holds that debt, is 1% off its high. So investors are not yet pricing loan losses. They are pricing the fee stream, meaning fundraising, growth in money managed, and a queue of withdrawal requests at one fund, BCRED, running at 10% against a 5% cap on what it will pay out, all of which depends on nobody looking too hard at the loan prices. That is a bet on the business model, made in t

Hidden leverage and shadow banking11 Sephigh

JPMorgan closed the credit line. Clear Street opened one.

When the big regulated bank says no and the same bets reappear through a smaller firm, the borrowed money has not gone anywhere. It has moved to where fewer people can see it.

JPMorgan has stopped lending to Situational Awareness, the hedge fund run by Leopold Aschenbrenner, after the largest loss in that industry's history, the FT reports. A hedge fund is an investment firm that manages money for the wealthy and can bet in ways ordinary funds cannot, and the bank that lends it the money for those bets and holds them is called its prime broker. The fund has begun working instead with Clear Street, a specialist broker the FT describes as “far smaller” than JPMorgan. The bank walking away is the part that is supposed to happen. The fund finding a new lender within days is the part that matters. Morgan Stanley had talks with the fund since it launched and never opened an account. Goldman Sachs and Citigroup have had relationships with it. Read that first as the system working. A prime broker is where a fund's borrowed money comes from and where it gets measured: the bank decides how much cash the fund must put up against each bet, sees the whole book and can shrink the loan whenever it likes. JPMorgan looked at a fund that went from more than $45bn to about $10bn in July and decided not to be the lender standing behind the next attempt. Then read what happened next. Per CNBC, Situational Awareness spent late last week and early this week buying options, contracts that give the right to buy a share at a fixed price later, on AMD, Bloom Energy, CoreWeave, SK Hynix, SanDisk and the Roundhill Memory ETF, a fund that trades like a single share. CNBC could not establish whether that is new money or what was left after the fire sale to Citadel. Options are borrowed money with the borrowing built into the price of the contract, and the dealer who sells them protects itself by buying the underlying shares. So the exposure to those companies is back, routed through a smaller firm and the traders who sell such contracts rather than through a bank credit line. Clear Street's terms, how much cash it demands the fund put up, are not public, and neither is how much of the remaining $10bn is at work. What can be said is that the best-capitalized lender in the chain stepped away and the position was rebuilt anyway. That is the migration this site exists to track, in miniature and with a name attached. It adds nothing new to the pile; it moves part of the pile to where it is harder to see, which is a fragility story rather than a spark. The names on the ticket are memory chips, and on those, see the next dispatch.

The AI capex bubble11 Sepmedium

DeepSeek trims the memory bill. Memory makers fall.

Roughly $500bn of borrowing for AI this year rests on the assumption that computing power stays scarce. A model that needs less of the scarcest component attacks that assumption directly.

DeepSeek released V4.1 Flash on Thursday, a model it prices at a fraction of a cent per million tokens (the chunks of text a model reads and writes) and says beats Moonshot's Kimi K3. MiniMax and Z.AI fell more than 8% in Hong Kong, and Alibaba slid more than 2% (Business Times, citing Bloomberg). The share prices are not the reason to care. The reason is that DeepSeek says its model needs less of the one component the entire AI buildout is short of. The damage in Hong Kong is the smaller story. The larger one is a design detail. DeepSeek says it has trimmed the model's KV cache, the working memory a model keeps while it is generating, which cuts the high-bandwidth memory and solid-state storage needed to run it. It also says the model, which has 552bn parameters (the internal settings that hold what it has learned), switches on only a tiny part of its capacity at a time. High-bandwidth memory is the tightest bottleneck in the AI build. A chronic shortage of it has driven profits at Micron and SK Hynix, and it is one reason the money the biggest cloud companies spend building things is assumed to keep rising. Microsoft's spending hit $145bn in its last fiscal year, and Bloomberg reported plans to reach 38GW of data-center capacity by 2032, a figure Microsoft called inaccurate. A model that needs less of the scarcest input is a shock to demand aimed at the one place in the supply chain where the seller sets the price. Two cautions, both ours rather than the reporting's. Cheaper AI has historically increased total usage, so less memory per token need not mean less memory in total. And DeepSeek's benchmark claims are DeepSeek's. But the fund in the previous dispatch bought options on SK Hynix, SanDisk and a memory fund in the days before this release. Nvidia is down 4.4% on the week. Anthropic and OpenAI are preparing to sell shares to the public into a price war in which “good enough and nearly free” is the competitor. The AI building spree has always required that computing power stay scarce. This is what a test of that assumption looks like, and the US memory companies have not yet had a trading day to price it. Nothing has broken, so this is not a spark. It is the first serious test of the assumption underneath a very large pile of debt.

The AI capex bubble11 Sepmedium

Nvidia may anchor its own customer's $100bn share sale

A supplier funding its customer by buying shares is safer than doing it with loans, but it still means the customer's price tag and the supplier's balance sheet rise and fall together.

Nvidia is considering putting as much as $10bn into Anthropic's IPO, its first sale of shares to the public, as the anchor investor, the large buyer who commits early so that others follow, Reuters reported and Bloomberg relayed. Anthropic is seeking to raise up to $100bn at a valuation around $2tn, which would be the largest such sale ever. Anthropic spends much of its money on computing that runs on Nvidia chips, so the supplier would be putting up the money that pays for its own customer's purchases. Anthropic raised $65bn in May at a $965bn valuation; OpenAI raised $122bn in March at $852bn. The plans are under discussion and neither company commented. If Nvidia takes a tenth of the sale, that is the circle we have been tracking in loan form, now appearing in share form. The loan form: Nvidia guaranteeing its customers' leases, and the $500bn of outside capital pools announced so customers can “independently underwrite” infrastructure. Shares are the more honest version. They never have to be repaid and there is no due date to borrow again for. But it means Nvidia's balance sheet, the list of what it owns and owes, would carry a stake in a five-year-old company valued at the IPO price, and that price would be set partly by the anchor's willingness to pay it. The valuation has doubled in four months. Whoever buys the other $90bn (index funds, pensions, the Goldman and BofA wealth desks Bloomberg says are competing to manage Anthropic employees' proceeds) buys at a level a supplier helped set. One further connection. Situational Awareness sold its public holdings to Citadel in July but kept its private stakes, including Anthropic. A share sale at $2tn is a chance to turn that stake into cash, and the fund badly needs one. None of this is a loss, and none of it is hidden. It does not make a crash more likely this week. It is a measure of how much of the AI complex's money now comes from inside the AI complex, and how few independent buyers are setting the prices. That is tinder, not a spark.

Household credit11 Sephigh

Six and a half times income, at rates last seen in 1998

Loosening the limit on how much people can borrow, at the moment interest rates are the highest in a generation, puts the risk of a payment jump on the households least able to absorb it, in plain sight and with the regulator's blessing.

The number of British first-time buyers borrowing more than four and a half times their income rose by two thirds last year, to 45,800 (a 66% rise in 2025, according to data from the FCA, the regulator that polices British financial firms, obtained by the FT). Loans above five and a half times income rose tenfold, to 4,628. Coventry Building Society this month began lending 6.5 times income to first-time buyers earning at least £30,000. Nationwide's Helping Hand product, which goes to six times, saw a 57% rise in loans at or above five times income. Loans this large were until recently rare enough that the central bank limited them, and they are being written at the most expensive moment in a generation. The cause is a rule change. In 2014 the Bank of England, Britain's central bank, capped loans that large at 15% of all new mortgages; in July 2025 it loosened that for first-time buyers. The response took a year. A multiple of income is really a question about the monthly payment. At 6.5 times, a household on £30,000 owes £195,000, and whether that is affordable depends entirely on the interest rate. British fixed-rate mortgages typically reset to a new rate after two to five years. These loans are being written while the interest rate the British government pays to borrow is the highest since 1998 (the thirty-year rate sits at 5.89%, and the ten-year above 5.37%). So the borrower with the thinnest cushion is being handed the most rate risk at the top of the cycle. This sits beside our running thesis rather than inside it. The risk is not hidden: it is on the books of regulated building societies, and a regulator chose it. 45,800 loans is not a number that can bring down a system. But a policymaker deciding that first-time buyers should carry more debt, at the moment the price of that debt is the highest in a generation, is the kind of decision that looks obvious in hindsight and unremarkable in the trade press. The brokers quoted by the FT called the trend “positive”. It adds a little to the pile and nothing to the spark.

Signs of the times

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

86 tons of Dutch gold go home

The Netherlands moved 86 tons of gold out of the US and Canada last week as protection against “extreme systemic risks”; France has already moved all of its gold out of New York, Germany debated doing the same, and Norway's sovereign wealth fund is trimming its US government bonds. America's allies are treating the keeper of the world's reserve currency like the firm on the other side of a deal, who has to be good for it.

goldreservescustody

12 SepAxios

Largest loss ever; still up 80%

Situational Awareness fell from more than $45bn to about $10bn in July, the largest loss any hedge fund has ever taken, and its founder, who is in his mid-twenties, told investors the fund was still up 80% for the year. Both can be true only when the gains were made with borrowed money, and that is what borrowed money does on the way up.

hedge fundsleverageai

11 SepFinancial Times

“A little scary” in Japanese

Japan's finance minister Satsuki Katayama said the US Treasury secretary's warning to bond traders that “I am the house now” sounded “a little scary” when translated into Japanese; former officials of the Bank of Japan, the country's central bank, called his intervention in Japanese monetary policy “exceptional”. Finance ministers do not usually talk about each other like that in public, and the US Treasury is now a variable in another central bank's credibility.

policyyenbessent

11 SepFinancial Times

A mortgage at 6.5 times income

Coventry Building Society is now lending 6.5 times income to British first-time buyers earning at least £30,000, while the interest rate the British government pays to borrow for thirty years sits at its highest since 1998. The number of first-time loans above 5.5 times income rose tenfold in a single year after the regulator relaxed the cap.

mortgagesukleverage

11 SepFinancial Times

Top-tier AI for a fraction of a cent

DeepSeek's V4.1 Flash charges as little as a fraction of a cent per million tokens, the chunks of text a model reads and writes, and claims to beat established models while needing less of the scarce high-speed memory chips; MiniMax and Z.AI fell more than 8% in Hong Kong on the day. Artificial Analysis, which ranks these models, calls the region below DeepSeek's line of price against performance the “death zone”.

aipricingmemory

11 SepBusiness Times Singapore

The rumour mill

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

Confirmed

“Trump has proposed $5,000 cheques to all ~270 million US adults, conditional on Republicans holding the House and Senate, costing about $1.35tn.”

Reuters, AP and BBC all carry the pledge and the cost estimate; the FT ties it to Thursday's deepening worry about America's public finances, with the national debt above $40tn and the interest rate the government pays to borrow for ten years near 5%.

Claimed by Meet Kevin

Partly true

“Japan's foreign reserves fell $79.6bn in August, mostly through sales of US Treasuries to fund yen intervention.”

The $79.6bn drop, and its link to record intervention to prop up the yen, are established by Reuters using Ministry of Finance data. That the sales were mostly US government bonds is inference: the data show foreign securities as a whole, not US bonds specifically.

Claimed by Meet Kevin

Partly true

“South Korea is buying gold for the first time in 13 years while dumping US Treasuries.”

The Bank of Korea's first gold-linked investment in 13 years is a $250m position in a gold fund that trades like a share, per Yonhap and Bloomberg. There is no evidence it is 'dumping' US government bonds, though the bonds that central banks around the world keep in the vaults of the New York Fed are at their lowest since 2012.

Claimed by Palisades Gold Radio

Partly true

“Blackstone controls $275bn of digital infrastructure assets and has formed a forward-deployed-engineer consulting firm with Anthropic.”

Blackstone values its data-center platform at about $185bn, not $275bn; the Anthropic services venture (Ode with Anthropic) did launch this summer with Blackstone and other Wall Street firms.

Claimed by Meet Kevin

Earlier developments

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

Bond market dysfunction10 Sepmedium

Five per cent, and the stock market bought the dip

Rates at the highs with volatility collapsing is the market saying 5% is fine; whether it is fine is a question about the plumbing, not the tape.

The 10-year Treasury closed Friday at 4.975%, up from 4.94% on Thursday and the highest close in our eight-run trail. The 30-year settled at 5.354%, a touch below Thursday's 5.37%, which was a 19-year high. The three-month bill and the five-year both rose. And the S&P 500 gained 0.9%, the Nasdaq 100 gained 1%, and VIX fell 11% to 15.8. The proximate cause on the rates side was inflation: the August CPI, released Friday, put core at 0.3% for the month against a consensus clustered at 0.2% — a claim we checked and confirmed (see the rumour mill) — and Bloomberg had already reported traders pricing roughly 70% odds of a Fed hike next week, with two-year yields jumping 16bp on Thursday, the most since the April 2025 tariff meltdown (Bloomberg). The Treasury's first expanded buyback took $5.19bn of a $6bn maximum; DWS's George Catrambone called it 'a squirt gun to a firefight'. What is odd is the equity response. Stanford's Hanno Lustig, in a paper the Washington Post summarised this week, finds that since 2022 stock and bond prices have moved together — the flight to safety no longer works — and that Treasuries no longer command a premium over top-rated corporate bonds once credit risk is adjusted (Washington Post). Friday did not fit that pattern either: bonds fell and stocks rose, which is what a normal risk-on day looked like before 2022. So the market read Friday as growth, not as a funding problem. Against that: 30-year mortgages at 6.76%, the highest in over a year (WSJ), a net interest bill above $1tn that exceeds defence spending by 16% (Setser, FT), and a 10-year at the round number the FT calls 'an important threshold for Wall Street'. We do not think equities are right about this, and we have said so by holding fragility at 91. But the meter that tracks stress has to record what happened, and on Friday the stress measures fell.

Hidden leverage and shadow banking11 Sepmedium

The fund that lost $35bn in July is buying options

Leverage that was flushed out of the AI trade in July is returning in a form that shows up on dealer books rather than in margin loans.

Situational Awareness — Leopold Aschenbrenner's hedge fund, which shrank from more than $45bn at the start of July to about $10bn after leveraged AI bets went wrong — has been buying options in AMD, Bloom Energy, CoreWeave, SK Hynix, SanDisk and the Roundhill Memory ETF, according to sources cited by CNBC. The trades were placed late last week and early this week. CNBC could not establish whether the fund has raised new money or is deploying what was left after the rout (CNBC). Recall what happened in July. A rapid fall in AI stocks the fund had bought with borrowed money forced it to unwind its public book to Citadel in a fire sale. Jane Street, a market-maker with a stake in the fund, posted a $15bn loss that month — its first losing month in a decade — while Hudson River Trading, with no exposure, made a record $11.4bn (Bloomberg). That unwind was one of the few genuine reductions in system fragility this year. This is it partially reversing. Options change the form of the leverage rather than removing it. A margin loan against stock can be called and the collateral dumped; a bought call has a maximum loss equal to the premium, so there is no forced seller on the way down. But the leverage moves to the dealer who sold the call and hedges it by buying the underlying, buying more as the stock rises — the mechanics that make a concentrated options buyer move a stock more than a cash buyer of the same size. That is our inference about structure; CNBC gives no sizes. The names are the memory-and-neocloud complex: CoreWeave is 17% off its high, Nebius 19%, Applied Digital 16%. A fund whose last leveraged bet shook a market-maker is back in the same trade, in a form that does not get margin-called. The market-maker on the other side now, we do not know.

The dollar, gold and reserve status11 Sepmedium

If Bessent's yen trade works, Japan sells Treasuries

The buyer base that used to absorb Treasuries regardless of price is shrinking, and the policy Washington wants from Tokyo accelerates it.

Japan's Government Pension Investment Fund, a $2tn investor, could sell as much as $62bn of US Treasuries without formally changing its asset allocation, according to Santander analysts. The fund's strategic ranges give its managers enough flexibility to cut foreign bonds 'in the coming months without waiting for a formal strategic asset-mix review' — and, the note says, this is 'especially likely if the Bank of Japan is successful in reversing the weakness of the yen through back-to-back rate hikes' (Bloomberg). Those hikes are arriving. Nikkei reports the BoJ will raise its policy rate to 1.25% on 18 September, the shortest gap between hikes since the cycle began in March 2024 (Nikkei). The 10-year JGB is above 3% for the first time since 1996. A Japanese pension fund looking at 3% at home in a strengthening currency, against 5% in Treasuries it has to hedge or bear the dollar risk on, does not need a policy review to start rotating. The irony is who wanted this. Scott Bessent has spent weeks pressing for a stronger yen and telling bond traders 'I am the house now'; former BoJ officials told the FT his intervention is 'exceptional' and risks the bank's credibility (FT). The trade he is pushing — BoJ hikes, yen up — is the trigger Santander names for Japanese selling of the bonds he is simultaneously buying back. Brad Setser's numbers give the backdrop: price-insensitive holders of Treasuries have fallen from 76% of the market to 43% in nineteen years, and Japan's share from 18% in 2004 to 4% (FT). GPIF is exactly the sort of holder that number describes. Japan is still the largest foreign holder at $1.1tn. $62bn is a Santander estimate, not a decision, and the health minister says the fund is still 'mulling' whether a review is even needed. But the 10-year closed Friday at 4.975%, and this is the marginal seller the price has to find a replacement for.

Household credit11 Sepmedium

7.5 million borrowers, one deadline, 18% have moved

A payment shock scheduled for the fourth quarter lands on the same households whose credit is already deteriorating fastest.

The Education Department is moving 7.5 million federal student loan borrowers off the SAVE repayment plan. Anyone who has not chosen another plan within 90 days of being notified is placed automatically in the standard plan, which divides the whole balance into equal payments over ten years. Notices began on 1 July; the first 90-day window closes on 29 September. So far about 18% of the 7.5 million have left the plan, and borrowers report conflicting information, missing payment histories and a broken online calculator, though the department says the glitches are largely fixed (Washington Post). The number that matters is this one: nearly half of SAVE enrollees had incomes low enough to qualify for zero-dollar monthly payments, and payments for everyone on the plan have been postponed for two years. The default destination is the most expensive plan available. For several million households the change is from $0 to a full amortising payment, arriving in the fourth quarter, roughly all at once. Federal student loans do not sit on the balance sheets of the lenders we track — Sallie Mae is private loans — but a household budget is fungible. A dollar to the Education Department is a dollar not paid to Capital One, Synchrony or Ally, whose shares are down 5–7% over twenty days while the market is 2% off its high. The context is already soft: New York Fed data put 90-day auto loan delinquency at a record 5.5% in the second quarter, above the Great Recession peak, per a web sweep we have not independently checked, and Chewy's chief executive told investors this week the company is 'not assuming a meaningful consumer recovery' this year (confirmed; see the rumour mill). What we cannot say is how much of the 82% still on SAVE will act before their windows close, or how many are the zero-payment cohort. The department will have that data; it has not published it.

Crypto and TradFi contagion11 Sepmedium

Perpetual futures on Nvidia, from a prediction market

Leverage on single stocks is being offered a new venue outside the prime-brokerage system where it has historically been counted.

Kalshi, the prediction-market company, will file as soon as next week for permission to list the first single-stock perpetual futures in the United States, on Tesla, Apple and Nvidia to begin with. It wants the CFTC and SEC to regulate them jointly as security futures. Agricultural commodities and ETFs are to follow. 'It's about time that these products come to the US, under a regulated umbrella,' co-founder Tarek Mansour told Bloomberg (Bloomberg). A perpetual is a leveraged futures contract with no expiry. It is kept close to the spot price by a funding rate paid periodically between longs and shorts rather than by delivery, and because it never expires there is no roll and no natural moment at which a position has to be closed. In crypto markets — where the product was invented — perps are the dominant retail leverage instrument, and it is through their liquidation engines that price falls turn into cascades of forced selling. That is a general description of the product, not a claim about how Kalshi's would be built. The filing lands the same week Nasdaq put $100m into Payward, Kraken's parent, at a $21bn valuation, in a deal under which Kraken will distribute tokenised Nasdaq-listed stocks carrying the same voting rights as the ordinary shares. The New York Stock Exchange is building its own blockchain venue for round-the-clock trading of tokenised stocks and ETFs (Bloomberg). This is where the crypto-and-TradFi beat has been heading: not contagion from coin prices, but crypto market structure — 24/7 trading, tokens, perps — being wired into the core of US equities. For our thesis the relevant point is that single-stock leverage has lived in margin accounts at prime brokers, under Reg T, where it is measured. A perp on a prediction-market venue is a different home for the same exposure. Leverage caps, clearing arrangements and whether the SEC agrees are all unknown; it is a filing, not a product.

Hidden leverage and shadow banking11 Sepmedium

Aareal follows Deutsche: €2.5bn of property risk, insured away

Each SRT relocates first-loss property risk from a measured balance sheet to an unmeasured one, and the freed capital is being spent on growth.

Aareal Bank is working with Alvarez & Marsal on a significant risk transfer tied to around €2.5bn ($2.9bn) of commercial real estate loans — hotels, offices and industrial property across Europe, in euros, sterling and dollars — according to people familiar cited by Bloomberg. The bank, owned by Centerbridge and Advent, is simultaneously considering a bid for Hamburg Commercial Bank (Bloomberg). The structure: the bank keeps the loans but buys protection on the first slice of losses from investors, which lets it shrink its risk-weighted assets and hold less capital against the portfolio. Bloomberg is explicit about what banks do with the capital freed: acquisitions, 'more profitable new business or returns to shareholders'. Demand from investors is 'robust', drawn by 'double-digit returns'. Aareal's core equity ratio was already 15.6% at the end of June. This is the second European bank in five days to do this — Deutsche's €2bn SRT on Monday included data-centre project loans — and Aareal is the lender that suffered a slump in US property valuations in early 2024. The loans do not go anywhere. What changes is who takes the first hit if European hotels and offices sour: a regulated bank whose exposure is disclosed, or a credit fund whose own funding we cannot see. And the capital released is earmarked to buy another bank. Fragility at the system level does not move on this; the risk is the same size it was on Thursday. What moves is our thesis's tally of where it sits.

Bond market dysfunction11 Sephigh

'Emerging-market-type risks': the Treasury's adviser on the Treasury

The people who advise the Treasury on how to borrow have started describing it in the language reserved for shaky developing countries. Every other price in the system rests on the assumption that lending to the US government is the safest thing you can do, and that assumption is now being questioned out loud.

The interest rate the US government pays to borrow for ten years came within a whisker of 5% in Asian trading overnight before easing (4.97%, then 4.94% in London, per the Financial Times; our own reading this morning is 4.92%). The thirty-year rate has slipped back too (5.32%, from Thursday's close of 5.36%). That makes this the calmest morning in three days, which tells you what the previous three were like. It is also the morning after a member of the Treasury Borrowing Advisory Committee, the panel of dealers and investors that advises the government on how to issue its debt, said this on the record: 'We can't ignore the fact that there are emerging-market-type risks in some of the actions the US has been taking.' Ellen Zentner of Morgan Stanley Wealth Management meant the buyback and the yen intervention, and the fact that both were improvised rather than run through 'a formal, institutionalised process.' Emerging-market risk is the phrase used for countries whose governments might change the rules on their lenders at short notice. People on that committee do not use it about the United States. Why the buyback is part of the problem rather than the fix comes down to scale and signal. The market in US government bonds, the IOUs Washington sells to cover its deficits, is $32tn. The buyback Treasury Secretary Bessent tripled to $6bn, and then bought only $5.19bn of (per Bloomberg), is about 0.02% of that. It cannot move the price. What it can do is tell traders that the Treasury Secretary has a level of interest rates he does not want to see breached, which invites them to find out how much he is willing to spend defending it. George Catrambone of DWS called it 'a squirt gun to a firefight.' The deeper point is in a Stanford paper the Washington Post surfaced. Hanno Lustig finds the premium investors once paid to hold US government bonds rather than other rich countries' debt has almost entirely gone, and that since 2022 they have not even paid extra for them over the debt of the safest companies, once you adjust for the chance those companies do not pay. Stocks and bonds have fallen together since 2022, as they did on Wednesday. A safe asset is one that rises when everything else falls; that is the whole point of owning it. On that definition the market has stopped treating US government bonds as safe, and a bond that falls with everything else is what an emerging-market bond looks like, whatever the name on it. This is ignition rather than fragility. N

Fed, Treasury and policy11 Sepmedium

Japan's central bank raises rates into Bessent's yen bet

Japan's central bank moving faster, the Federal Reserve possibly raising rates too, and a US Treasury Secretary undermining the Japanese bank's credibility all point the same way: fewer automatic Japanese buyers of US government bonds, at the moment Washington most needs them.

The Bank of Japan, Japan's central bank, plans to raise its interest rate to 1.25% from 1% at its meeting on September 17 and 18, the Japanese business daily Nikkei reports. That would be the highest Japanese rate since 1995, and it comes only three months after the last increase, in a cycle that had been moving once every six months. A central bank doubling its pace is a central bank that thinks it is behind. Governor Ueda's stated reasons, from July: oil, a weak yen, and demand from artificial intelligence. The interest rate Japan's government pays to borrow for ten years is above 3% for the first time in thirty years. Two things make this more than a Japanese story. The first is timing. It lands in the same week traders put roughly 70% odds on the Federal Reserve, America's central bank, raising rates too, per Bloomberg. The world's two largest bond markets may tighten at once, into a selloff that already has the US paying 4.92% to borrow for ten years and 5.32% for thirty. The second is Bessent. The US Treasury Secretary has spent two weeks daring traders to bet against the yen, declared 'I am the house now', and claimed to know the Bank of Japan's intentions before it announced them. Former officials of the bank told the Financial Times this is 'exceptional' and risks the bank being seen as setting rates under outside pressure, in which case, one said, its future statements 'would be treated far more sceptically by markets'. Japan's finance minister said Bessent's line sounded 'a little scary' in translation. Finance ministers do not normally review each other's language in public. Why this matters to US government debt: Japanese institutions were once the largest automatic buyers of it, the investors who showed up at every auction regardless of price. Brad Setser's numbers in the Financial Times have Japan's share of the market falling from 18% in 2004 to 4%. Our inference, and it is ours rather than the reporting's, is that a rate rise pushing Japanese bonds past 3%, at the same moment the US Treasury is trying to strengthen the yen, makes home bonds more attractive to exactly the Japanese investors the US needs to keep. And a central bank whose credibility has been dented is a weaker anchor for the currency Bessent is trying to prop up. Two interventions, each working against the other. This sits on the ignition side. It adds no borrowed money to the pile. What it does is make a buyer less reliable at the moment the seller most needs him, and that

Private credit and BDCs11 Sephigh

Australia's regulator asks the banks about what it cannot see

A regulator has just found that the lending risk it is responsible for has moved to a place where it has no way to measure it, and has had to ask the banks to look for it. That is the clearest example this week of the thing this publication exists to track.

Australia's bank regulator, APRA, has asked the country's banks and pension funds what they have lent to, and what they own of, the investment funds behind Bathla Group, a Sydney housebuilder that called in administrators last month owing about A$3.4bn, most of it to lenders that are funds rather than banks, per Bloomberg. APRA's own statement says the banks' direct exposure to Bathla is 'negligible'. It is asking anyway, because what it wants to know is 'the interplay between regulated entities and private credit'. In plain terms: which banks lend money to the funds, and which pension funds own pieces of them. That is the whole Crash Lab thesis in one regulatory press release. A regulator can see a bank's balance sheet, the list of what it owns and owes. It cannot see the balance sheet of a fund that lends to companies in place of a bank, because it does not supervise those. So when a A$3.4bn borrower fails it has to work backwards, asking the institutions it does supervise how much of their money reached the failure by a second route. Last week Australia's corporate regulator, ASIC, said a lack of basic information about this kind of lending made the risk hard to assess. The country's central bank is 'monitoring'. Regulators do not usually admit in public that they lack basic information about a market they are responsible for. The pensions line is the one to watch. APRA's message to the trustees who run Australians' retirement savings is to ensure 'the valuations applied to investments are appropriate'. That is a reminder that a fund holding a loan to a collapsed developer may still be carrying it on its books at full value until somebody asks. None of this is an American event, and A$3.4bn would not register at one of the big US funds that lend to private companies. But the pattern is the one American investors are already pricing. Shares in the firms that manage this kind of lending have dropped hard in a week (Blue Owl down 10%, Blackstone 8%, KKR and Ares 5%), while the listed lending funds they run fell only 2 to 3%, and the extra interest the shakiest American companies pay over the government sits at 2.71 percentage points. Investors are selling the managers before anybody has admitted a loan is worth less. A regulator on the other side of the world has just shown why: the prices those funds put on their loans are the thing nobody outside can check. This is fragility, not ignition, and it is the third kind of answer. The world did not get more da

Private credit and BDCs11 Sepmedium

Nearly a tenth of the loans at listed private lenders now pay their interest in IOUs

Two accounting devices let a loan book report rising income while the cash coming in falls: interest paid by adding to the debt, and loans on which the lender has stopped counting interest at all. Both are now rising together across the whole listed sector, not in one bad name.

Across 44 US business development companies, the listed funds that borrow money, lend it to mid-sized private companies, and pass the interest to shareholders, the loans were valued below what they cost at the end of June, and the gap widened over the first half of the year (portfolios carried at $92.88bn against $95.19bn of cost, 97.57 cents on the dollar, down 1.68 percentage points, according to a Reuters analysis carried in several of the news roundups we read this week). Loans on which the lender has stopped counting the interest, because it stopped arriving, rose to about 3.4% of cost from 2.5% at the end of last year. Among the ten largest of these funds one dataset has that figure at nearly double its three-year average (3.95% against 2.1%). KBRA, one of the firms that grade how likely a borrower is to pay you back, saw its median for the non-perpetual funds, those with a fixed life, jump from 1.81% to 2.75% in a single quarter. Reuters found 81% of software loans marked down. The number that describes the mechanism best is payment-in-kind: interest paid by adding to the debt rather than handing over cash. Data from the Federal Reserve's Boston branch cited this week has such arrangements rising from 5.4% of these funds' loans in early 2022 to 9.8% in early 2026, and this kind of interest now averages roughly 8% of the funds' investment income. The borrower hands over nothing. The lender books the interest as income anyway, and the loan grows. Nothing is wrong with a single loan on those terms. A tenth of a whole portfolio paying in IOUs means a meaningful slice of the income these funds report is a promise rather than cash, and the value per share that ordinary investors cash out against contains it. One detail from the filings of one of the largest of these funds, OBDC, is worth flagging: about $235mn of junior preferred and second-lien paper, debt that only gets paid after the senior lenders do, was carried near 90 cents on the dollar while the same borrowers' senior debt traded around 78 (per one roundup). Junior debt priced above senior debt in the same company cannot both be right, and it is the junior price that is waiting for a test. Loparex, at 63 cents in March and 5 cents now, was one such test. These figures reach us through secondary compilations of Reuters, KBRA, Octus and Boston Fed data rather than the primary filings, so treat the decimals as approximate. The direction is not in doubt, and the market has noticed, in a lopsided way:

Household credit10 Sepmedium

Pick whichever credit score gets you the cheaper mortgage

Loosening the ruler rather than the money is how lending standards slip without anyone announcing it, and the risk lands on a government guarantee rather than on a bank.

This month the regulator of Fannie Mae and Freddie Mac, the two government-backed companies that guarantee American mortgages, told lenders they may pull a borrower's VantageScore 4.0 alongside, or instead of, the classic FICO score, and use whichever produces the better rate, the Wall Street Journal reports. Nick Maciunas of JPMorgan estimates that when both are run, roughly one in four buyers lands in a higher credit band. Jumping a band can cut about 1% in upfront loan fees, which he translates into roughly a quarter of a percentage point off the interest rate. One Michigan lender says it has qualified borrowers who previously failed. The context is a thirty-year fixed mortgage rate at its highest in over a year (6.76%), because the interest rates on US government debt have been rising for six months. So this is the government's response to expensive credit: not cheaper money, but a more generous ruler. How it works. Fannie and Freddie do not lend; they guarantee, and they price the guarantee through fees set by credit-score band. The borrower's actual risk is whatever it is. What has changed is that the fee is now set by whichever of two models flatters the borrower, and that is a one-way ratchet: nobody will choose the score that costs them more. If VantageScore is a genuinely better predictor for some borrowers, this is a correction. If the two models simply disagree at the edges, a quarter of new mortgages are being priced as safer than the old model thought, and the difference sits with the guarantor, which is to say with the taxpayer who stands behind Fannie and Freddie. The Journal reports results are 'all over the place' by borrower, and nothing in the piece says how Fannie and Freddie will track which borrowers stop paying by which score was used. That is the thing to watch. It is a small change on a single loan and a large one across the biggest lending market in the country, and it arrives in the same week that a record one in eighteen car loans is three months behind (5.5% ninety days late). This is fragility, not ignition. Nothing here breaks tomorrow. It adds to the pile slowly, in the way lending standards always slip: nobody decides to lend to riskier people, they change how risk is counted. We would call it a small nudge up, and small is the honest word.

Hidden leverage and shadow banking10 Sephigh

Aston Martin's name walked out of its bondholders' security

The assets pledged to public bondholders are migrating to private lenders, which is the same story as the AI and private-lending beats. A fund that bets with borrowed money has lost real money on it before anyone has failed to pay.

Arini Capital Management, a $22bn London fund that manages money for the wealthy and can bet in ways ordinary funds cannot, founded in 2021 by Hamza Lemssouguer, 35, is down more than 8% this year after returning 27%, 21% and 10% in the three years before, the Financial Times reports. Almost all of the loss came in July. Two of its top traders have left. Funds have bad months; losing 8% in one, at a fund that made 27% the year it opened, is not a bad month but an event. The loss is worth understanding not because one fund is down but because of how. Arini's method is borrowed money used to make concentrated bets bigger, in the debt of struggling companies. Its two worst positions were undone by the same maneuver: the borrower moving assets out of reach of the creditors who had lent against them. Altice International, part of Patrick Drahi's telecoms group, shifted its most valuable assets out of the pool of collateral pledged to bondholders at the end of last year. Collateral is something valuable pledged to a lender, which they keep if you cannot pay; take it away and the loan is a promise with nothing behind it. Arini joined a group of senior creditors that called the move a default. Then in July a smaller group, including Arini and Silver Point, pushed for a deal paying them more than other creditors in any restructuring, as a reward for being willing to sue. The bonds fell further on the news. Arini wrote the position down. Aston Martin did something similar in July, taking its naming and branding rights out of the collateral backing more than £1.3bn of debt and pledging them instead to HPS, a firm that lends to companies in place of banks, as part of a contested £550mn of new financing. Arini owns hundreds of millions of pounds of the old bonds. Read that one slowly. The carmaker's most valuable asset, its own name, left the public bondholders' security and entered a private lender's, in exchange for new money. The old creditors' bonds fell. That is the migration Crash Lab tracks, seen from the losing end: lending by funds is not only where new risk goes, it is where existing collateral goes when a borrower is desperate enough to pay for it. The fund on the wrong side of that trade, betting with borrowed money, lost 8% in a month. On its own this changes the odds very little. One fund lost money and nobody was forced to sell in a hurry, which is the difference between a loss and a spark. It is here because it shows the losses in this corner are real b

Bond market dysfunction10 Sephigh

The Treasury offered to buy $6bn of its bonds. It took $5.19bn.

The one tool Washington has used to hold up its own bond market has now been tried in public, and investors answered by selling more.

The Treasury Department, which borrows the money the government spends, ran its first expanded buyback on Thursday: it had said it would buy back up to $6bn of its own bonds due in ten to twenty years, and it bought $5.19bn (Bloomberg). Buying bonds back is meant to push their prices up and borrowing costs down. Instead, by the close, the interest rate the government pays to borrow for two years had risen 0.16 percentage points to 4.59%, the largest one-day jump since the tariff selloff of April 2025; the ten-year rate was up 0.12 points to just under 4.95%; and the thirty-year closed at 5.37%, its highest in 19 years. The lever was pulled and the price went the other way. Why would a buyer trying to push prices up buy less than it said it would? A buyback is an auction run backwards. The banks that trade government bonds offer to sell them, and the Treasury accepts the offers it judges fairly priced, up to a ceiling. Taking less than the ceiling means the Treasury refused to pay what some of them asked. That is sensible housekeeping for anyone managing debt. It is also a message, that this is not a buyer who will pay any price, and the market heard the message rather than the housekeeping. "Bessent is bringing a squirt gun to a firefight," said George Catrambone of DWS Americas. Two things about how the money actually moves. The buybacks are paid for by selling short-term bills, so they swap long debt for short debt without reducing the total, and no new buyer appears. And the same afternoon the Treasury sold $22bn of new thirty-year bonds at 5.308%, the highest rate at any such auction since 2001, to what the FT called strong demand (FT). Nothing failed. There are still buyers for American debt; they are charging more for it. A $5bn purchase set against $22bn of fresh thirty-year selling on the same day was never going to change the price. The sharpest move was not in the long bonds. The two-year rate jumped because Brent crude hit $107 a barrel, August prices charged by producers came in at 5.4%, and traders moved to roughly 70% odds that the Federal Reserve, America's central bank, which sets interest rates, raises them next week. So the Treasury is buying long bonds to bring their rates down in the very week the Fed is expected to push the short rate up. Whether the two arms of government are working against each other depends on whether long rates are rising because of inflation or because of the sheer amount of debt on offer. Thursday said both. Eit

Fed, Treasury and policy10 Sephigh

The lenders who never asked the price are down to 43%

If the people who own America's debt have really changed, every new dollar Washington borrows costs more than the one before, and no buyback undoes that arithmetic.

Nineteen years ago, three quarters of US government bonds (76%) were owned by buyers who did not care about the price: foreign central banks piling up dollar reserves more or less automatically. Today that share is 43% (FT), according to Brad Setser of the Council on Foreign Relations, and that number is why a $6bn buyback is small against the problem. The other 57% is households and funds who want a return for their money. The interest bill alone now runs above $1tn a year, 3.2% of everything the country produces, and is 16% larger than the defense budget. The mechanism is the important part. When most buyers buy regardless, extra debt can be sold without offering a higher rate. When most buyers care about the return, every extra dollar borrowed needs a little more interest to find a taker; the higher interest swells the interest bill, which swells the deficit, which means more borrowing. Setser's phrase is that more debt "seems to accelerate the rise in the rate demanded". The line curves upward rather than climbing at a steady slope. Christopher Waller, a governor of the Federal Reserve, America's central bank, made a related point last week: there is "no more premium for safe, liquid US government debt". In other words, buyers have stopped accepting a lower rate just because the borrower is Washington. Into that, on Wednesday night, the president promised $5,000 to every adult if Republicans keep Congress, a pledge the FT reports would cost more than $1tn. Pooja Kumra of TD: "US buybacks and growing credibility risks are pushing term premia higher." In other words, lenders are demanding more extra interest for tying their money up for a long time. Thursday's thirty-year rate of 5.35% is the market's first reading of the two things together. It is not only America. The British government's ten-year borrowing rate hit 5.38%, its highest since 2007; Germany's reached 3.5%, the highest since 2011; and the European Central Bank raised rates to 2.5% and said inflation would stay "well above" its target. An oil shock is doing part of this everywhere. What is specific to the US is the shift in who holds the debt. This cuts against our usual story in one respect. Government bond markets are the most measured, most regulated part of the whole system, and they are repricing in full public view. Most of this is fragility being measured rather than added, since the buyers did not change this week, Setser counted them; the $1tn pledge is the piece that adds to the p

Private credit and BDCs10 Sepmedium

Sixty-three cents in March. Five cents now.

The prices private lenders put on their own loans are being checked one loan at a time, and investors are punishing the firms that set those prices before they punish the loans.

A fund run by Blue Owl has admitted that a loan it valued at about 63 cents on the dollar at the end of March is now worth about 5 cents. The fund, OBDC, is one of a class that borrow money, lend it to mid-sized private companies, and pass the interest to shareholders; it has stopped counting the interest on its loan to Loparex, a packaging company backed by the investment firm Pamplona, and marked the junior piece of that loan at roughly 5 cents (Reuters roundup). Funds trim the value of loans all the time, by a few cents. A fall of 58 cents in one quarter is not a loan going bad this spring; it is the fund's price finally catching up with a loan that had already gone bad. Stopping the interest matters because the interest is where the dividend comes from. The same reporting has OBDC carrying about $235m of junior claims on other borrowers at roughly 90 cents while the senior debt of those same companies, the debt that gets paid first, trades around 78 cents. If the people who get paid first think they will get 78 cents, the people who get paid after them cannot sensibly be worth 90. That is not a price. It is a disagreement between the fund's valuation committee and everyone else. We have this from a secondhand summary, so treat the detail as reported rather than confirmed. The totals point the same way. Across 44 of these funds, loans were valued at 97.57% of what was lent on June 30, down from 99.25% at year-end; loans no longer paying interest rose to 3.4% of the total from 2.5%; 81% of software loans were marked down against 40% elsewhere; and interest paid by adding to the debt rather than handing over cash has grown from 5.4% of these funds' loans in early 2022 to 9.8% by early 2026, per Boston Fed data cited in the same reporting (matterfact). That kind of interest counts as income. It is not cash. Now what the market did. The lending funds themselves are down 2 to 5% on the month. The firms that run them are somewhere else entirely: Blue Owl fell 4% on Thursday, is down 10% on the week and 13% on the month, and sits 17.6% off its high; Blackstone is down 14% on the month. The extra interest paid by companies too shaky to be considered safe is 2.71 percentage points above the government, which is to say it has barely moved. The stock market is selling the managers' fee income, the money they earn for looking after other people's savings, for raising new funds, for running funds that have shut the exits and cannot take in new money as easily, befor

The AI capex bubble10 Sephigh

Vantage's banks have lent all they will. Now it is asking Pimco.

When banks stop lending to a company that has borrowed $48bn and investment funds step in, the risk moves from a place regulators watch closely to one they barely see.

Vantage Data Centers, which has borrowed nearly $48bn since the start of 2025 (per Infralogic), is negotiating a credit line of up to $2bn from Pimco, PGIM and other large investment managers, a deal called Project Baja, because the banks it has been borrowing from have lent as much as they are willing to (FT). Vantage is a major builder for Oracle's Stargate, the vehicle through which OpenAI is buying $300bn of computing power. Banks do not usually run out of room for a customer this size. That they have, and that the next dollar is coming from funds instead, is the story. The FT's phrasing is exact: the part of Wall Street that lends for big construction "traditionally funds the construction of airports and gas pipelines", and the AI build-out has overwhelmed it. Banks are also trying to cut how much they have riding on any one developer, by selling AI loans into the bond market and by paying other investors to take on the risk that the loans go bad while keeping the loans on their own books. So the next dollar comes from institutional money, which, as our inference rather than anyone's reporting, means largely insurance and pension savings managed by firms like Pimco and PGIM. Two features matter. First, this is a line Vantage can draw and repay across several campuses in Virginia and Nevada, not a loan against one building. The lender is backing Vantage the company, and what stands behind the loan is a portfolio of leases. Second, follow the leases up: Vantage builds, Oracle leases, OpenAI contracts with Oracle. A pension fund's exposure to whether OpenAI can pay $300bn now runs through three companies in a row, none of which is a bank and none of which puts a fresh price on that exposure every day. That is the thesis of this site in a single deal, and it is fragility: one more thing on the pile, though we already counted the migration, which is why the meter did not move for it. Demand is not the limit. Bloomberg reports Microsoft plans to more than triple its capacity to 38 gigawatts by 2032 from about 12 today, on top of $145bn spent building things last fiscal year (Bloomberg). Money is the limit. What the market did on Thursday: Oracle fell 5.3%, CoreWeave 6.1%, Nebius 5.1%, Applied Digital 5%. On the week all four are up 4 to 12%. In this group a 5% day is ordinary. But swings that size are themselves a form of borrowing getting dearer, because anyone asked to stomach them charges for it.

Household credit10 Sepmedium

A rate rise arrives on top of a record in late car payments

Household borrowing is where a rate rise turns into people not paying, and the share already behind is at a level that used to take a recession to reach.

Lenders to ordinary Americans had a bad week before the Federal Reserve, America's central bank, which sets interest rates, has done anything: Affirm is down 8.2% in five days, Capital One 4.4%, Synchrony 3.7%, and Sallie Mae 8.1% on the month. Falls that size are not dramatic on their own. What changed underneath them is that traders now put roughly 70% odds on the Fed raising rates next week, and the interest rate the government pays to borrow for two years rose 0.16 percentage points on Thursday. A rate rise is landing on borrowers who are already behind at levels that used to require a recession. The mechanism runs through pricing. Credit card rates move with the bank's benchmark rate, so a rise reaches the borrower within one monthly statement. Car loans are fixed, so existing borrowers feel nothing, but the lender's own cost of money (the deposits it takes, and the loans it bundles and sells to investors) goes up while what its car loans pay does not. Buy-now-pay-later debt is short and refinanced constantly, which is why Affirm moved most. The borrower is already stretched. New York Fed data for the second quarter, as summarized in our sweeps, show one in eighteen dollars of car loans three months or more behind, worse than the peak of the Great Recession (5.5% of balances 90 or more days late, against 5.3% then). Among car loans to borrowers with poor credit that were bundled and sold to investors, 6.13% were late in July, a record for that month; banks have given up collecting 3.82% of card debt; and some 3.6m federal student borrowers have defaulted since the payment pause ended (sweep). A Fed survey cited in the same reporting has 34% of adults earning under $25,000 failing to pay all their bills in full last month, against 7% of those earning over $100,000. None of this is new risk. We have carried the second-quarter figures for weeks, and the sourcing in front of us for the July numbers is secondhand, so we hold it loosely. What is new is the spark getting closer: a rate rise arriving on top, and investors deciding to sell the lenders whose profit margins shrink fastest. That is ignition, not fragility. One oddity. Credit Acceptance, which lends to the borrowers with the worst credit of all, is 0.5% off its high. The market appears to believe that a lender which already prices for people not paying is safer than the ones that price for people who do. In a downturn that has sometimes been right.

Bond market dysfunction10 Sepmedium

The Treasury is buying $4bn of its own bonds a session. It is not enough.

A program announced to hold down the government's long-term borrowing costs has now run for two days. Those costs rose on both, which tells you the government is not the one setting the price.

The US government's borrowing costs rose for a second straight day this morning and now sit above every level in our last eight readings (the ten-year rate at 4.92%, the thirty-year at 5.35%). This is the second day of a Treasury program announced specifically to stop that from happening. A government stepping into its own bond market to push down what it pays, and losing on both days it tried, is not routine, and it tells you who actually sets the price. The program itself: on August 19 the Treasury Department, which borrows the money the government spends, said it would double its buybacks of bonds coming due in ten to thirty years, from $2bn to at least $4bn per operation, running from September 9 to November 4. One tally puts September's total at about $38bn (sweep). The Treasury buys old long-term bonds from the dealers who hold them and pays for them by selling new IOUs that come due in weeks. The market ends up holding less of the debt whose price falls badly when rates rise, and more of the kind that barely notices. Bessent, who runs the Treasury, has called it a 'Treasury Twist'. The twenty- and thirty-year rates fell sharply when it was announced in August. They have risen since the buying actually began. Why might $4bn a session fail to move a $28tn market? Because the people who set the price at the margin are not the Treasury. Hedge funds are financing about $3tn of government bond holdings with cash borrowed overnight against bonds put up as security, and roughly $830bn of that is one trade: borrowing heavily to profit from a small gap between two nearly identical prices (Reuters). Japan sold down $88bn of foreign securities in August to pay for a record defense of the yen. One ally's monthly selling was twice the Treasury's entire September buyback. And Waller, one of the Federal Reserve's governors, said on September 3 there is 'no more premium for safe, liquid U.S. government debt'. In other words, investors no longer pay extra to hold it because it is no longer scarce, so the interest rate that neither speeds nor slows the economy is higher than it was (sweep). Two details are worth watching. The five-year rate rose more than the ten-year today, even though the buybacks target the longest bonds, which suggests the market is repricing borrowing at every length rather than reacting to how much long debt is on offer. And gold fell 2.2% over five days while the dollar rose. That is not money fleeing America. It is money demanding more interes

What would change our mind

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

If the Federal Reserve raises rates on September 16 with the ten-year rate through 5% and shaky companies paying more than 3 percentage points over the government, that would confirm higher rates are now feeding into what companies pay to borrow, and we would raise ignition sharply.

Would move the number

Monday's trading in US memory-chip names (Micron, SanDisk, the memory funds) after DeepSeek's release; a sustained fall of 10% or more would be the first market test of the assumption behind AI debt that computing power stays scarce.

Would move the number

A second big bank cutting off a large fund running on borrowed money, or a private lending fund other than BCRED halting withdrawals, would move ignition; a bounce in Blue Owl or Blackstone shares while shaky companies still borrow cheaply would not.

Would move the number

The VIX, the market's gauge of expected turbulence, back below 14 (under 20 is calm) and the ten-year rate back under 4.75% after the Fed and Bank of Japan meetings would justify lowering ignition to the high 30s; a Treasury buyback that takes the full amount with rates falling would count as a policy test passed.

Would move the number