The biggest holders of US assets hedged 41%. Lowest since 2015.

Archived reading, published Thu, 03 Sep 2026 10:18:30 UTC (17 days ago). This is not the current state of the meter.

See the live reading →

CRASH-O-METER

Cracking

how close are we

Fragility91

how much tinder is stacked up — moves slowly

Ignition39

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 22 sources and rewrites this page.

Status: Held at 65. The market is where it was at the September 2 close we read at 06:00 UTC: the VIX, the market's gauge of how much turbulence traders expect over the next month, is 15.32 against 15.2 (under 20 is calm; it passed 80 in March 2020); the S&P 500, the main index of large American shares, is 1.7% below its high; the extra interest the shakiest companies pay compared with the government is 2.65 percentage points, less than a month ago; and the interest rate the government pays to borrow for ten years is 4.80%. The only overnight moves are the dollar index, which tracks the dollar against a basket of other currencies, down 0.35%, and the same 2.5% jump in gold we quoted four hours ago. So ignition, how close a spark is, stays at 39. Fragility, how much tinder is stacked up, holds at 91. Bloomberg's finding that pension funds and insurers in six markets protected only 41% of their foreign-currency holdings against those currencies falling is a better measurement of a position that has been in place since June 30, not new borrowed money. And Dell's $130bn order book says the demand end of the AI chain is real, which is not a reason to raise a number that counts stacked-up risk. Nothing was closed out or sold.

Reporting from 1 Sep to 3 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.

The dollar, gold and reserve status3 Sepmedium

The protection that isn't there

With only 41% protected, an exit from US assets, if it comes, shows up in the currency market before it shows up in the bond market, and the sellers would be the same institutions that buy the government IOUs that do not come due for twenty or thirty years.

Pension funds and insurers in six markets, Japan, Canada and Taiwan among them, had protected only 41% of what they hold in foreign currencies against those currencies falling, as of June 30. That is the smallest share since at least 2015, according to Bloomberg's review of their filings (Bloomberg). These are the largest owners of US government bonds and US shares outside America, and the dollar is down about 2.3% this quarter. A falling dollar with the protection off means the losses land on the very institutions that hold the most. A Japanese pension fund that owns US government bonds is holding two bets at once: the bond, and the dollar the bond is priced in. Protecting the second bet means agreeing today to sell those dollars back into yen at a future date. That costs money, roughly the gap between short-term interest rates in America and in Japan, and for most of the last decade it was money wasted. The dollar tended to rise when markets fell, which cushioned the bond losses on its own. So the protection came off. The arithmetic that made that sensible needs two things to stay true: the protection stays expensive, and the dollar stays the thing that rises in a panic. Bloomberg's point is that both are being tested at once. Money-market brokers put a 94% probability on the Bank of Japan, Japan's central bank, raising its interest rate to 1.25% on September 17–18 (Nikkei), which narrows the gap between the two countries' rates and so cuts the cost of protecting. And the European Central Bank's own data say gold overtook US government bonds as the world's largest reserve asset last year (FT), which is not what you would see if everyone still trusted the dollar to be the thing that goes up. The mechanism that matters for us: when an unprotected holder decides to protect itself, it does not sell the bond. It sells the currency. Laura Cooper of Nuveen, a large asset manager, puts it plainly: given the scale of foreign holdings, "it doesn't take a dramatic change in positioning to matter." In other words, these institutions own so much that even a small shift toward protecting themselves moves the market. The dollar index, which tracks the dollar against a basket of other currencies, is 99.2 this morning, down 0.35%. The market data does not show this yet, and we are not pretending it does. This belongs in fragility, the slow-moving pile of tinder: a large, concentrated, unprotected position that nobody had to report until the filings came due.

Bond market dysfunction3 Sepmedium

Five companies, a third of all new long-term borrowing

The AI building boom is not only a stock-market story. It competes with the government for the same buyers of debt that does not come due for twenty or thirty years, which is a direct line from companies' borrowing into what Washington pays to borrow.

Amazon, Microsoft, Alphabet, Meta and Oracle borrowed roughly $200bn from bond investors in the first half of 2026, and technology companies accounted for a fifth of all new borrowing by companies rated safe this year and more than 30% of the new debt that does not come due for twenty or thirty years, according to a note from Neuberger Berman, an asset manager, that turned up in our sweep (nb.com). Financing for the big cloud companies, data centers and chipmakers reached about $165bn before the midpoint of the year, $27bn more than in all of 2025. Five companies are now competing with the US government for the same lenders, in the same maturities, and the government is paying more because of it. This is the mechanism behind a number we have been quoting all week without fully explaining: the interest rate the government pays to borrow for ten years is 4.80% and for thirty years 5.27%, the highest since 2007, just before the last financial crisis. Long-dated company bonds and long-dated government bonds compete for the same buyers, pensions and insurers, the very institutions in the dispatch above. When a third of the new thirty-year paper on offer carries Microsoft's or Oracle's name, the Treasury Department, which borrows the money the government spends, has to pay more to place its own. Mark Cabana of Bank of America told Bloomberg the market for government debt "has not been able to hold any type of significant rate decline" (Bloomberg). In other words, every time rates start to fall, something pushes them back up. One reason is that every dip meets another data-center bond. Against that supply, the Treasury's answer, buying back at least $4bn of its own old bonds per operation starting September 9, is a rounding error. The government sold $797bn of debt in a single week in late August. We want to be precise about what this is and is not. Every source we have, including Kashkari at the Federal Reserve, America's central bank, and the dealers Reuters spoke to, says the market for government debt is working: no auction has finished with buyers demanding a sharply higher rate than expected, and none of the gauges of stress between closely related interest rates has widened. This is the market finding a price, not the machinery failing. But the price being found is the cost of financing the AI buildout and the deficit at the same time, in the same maturities, to the same buyers. That is fragility, not ignition: nothing is about to break, but the government

The AI capex bubble2 Sephigh

Dell booked $130bn. Its buyers are borrowing to pay.

The AI demand showing up in suppliers' results is real. The question we track is whether the debt paying for it can be borrowed again when it comes due, and the share prices of the companies renting out AI computing say investors are less sure than Dell is.

Dell raised its forecast for AI server sales in its 2027 fiscal year to $74bn from $60bn, reported a record quarter that beat what analysts expected (sales of $47bn against a $44.9bn estimate), and said it has booked more than $130bn of AI server orders in the past twelve months from 6,500 customers (Channel NewsAsia). The shares rose about 8% in after-hours trading and have more than tripled this year. The orders are real and the cash is real. What is not settled is where the cash comes from, because the companies placing the orders are paying with borrowed money. Start with what is fine. This is revenue, booked when the servers ship and paid in cash. None of it is a projection, and none of it is a backlog in the sense SB Energy uses the word. The AI buildout is buying real boxes from a real company. Now follow the cash upstream. Dell's chief operating officer named the buyers: companies that borrow to build data centers and rent the computing power out, a group the industry calls neoclouds, with CoreWeave and Nscale as examples; governments; and large corporations. The neocloud share of that is funded almost entirely with borrowed money. In the last fortnight alone Nebius enlarged a debt deal to $5bn and sold $5.75bn of bonds that can convert into shares, JPMorgan began marketing $5bn of debt for Volta, a company seven months old, Blue Owl arranged $2.4bn for Iren earmarked specifically for Nvidia's Blackwell chips, and Lambda closed a $926m loan (Bloomberg via sweep). Dell's sales are what comes out the far end of two markets: lending to companies by investment funds rather than banks, and the debt of companies too shaky to be considered safe. The prices say investors know this. Nvidia is up 7% on the week and 1.6% below its high. Dell has tripled. Their customers are moving the other way: CoreWeave is down 8% in five days and 25% below its high, Applied Digital 20% below, Nebius 26% below. The sellers trade like winners and the buyers trade like the ones carrying the risk. That is the correct ordering, since the seller gets paid on delivery and the buyer has to earn the money back over the life of the debt. It is also exactly the shape of a boom in which the buyers borrow to pay the sellers. Dell's order book is a fact about demand. It is not a fact about whether that demand can keep borrowing to pay for itself. For the odds, this adds nothing new to the pile today; the borrowed money was already there. What it does is confirm how much of the pile the

Crypto and TradFi contagion2 Sepmedium

Fifty-eight banks, two new tokens, $12.5m of takers

Digital dollars issued by banks would connect the market for short-term government debt directly to crypto trading, inside banks large enough that their failure would matter to everyone. It is a new channel, not yet built, that nobody has tested under stress.

Twenty-one financial institutions, Goldman Sachs, Bank of America, Citi, Deutsche Bank, UBS, Wells Fargo and MUFG among them, said they will set up a company this year to issue a digital token meant to be worth exactly one dollar, a stablecoin, in the first half of 2027, with a euro version to follow (Reuters via Livemint). They will compete with Qivalis, a separate group of 37 institutions planning a euro token this year. Some banks, including BBVA, are in both. Until now this business belonged to companies outside the banking system; this is the largest banks in the world deciding to run it themselves, which moves the risk inside the buildings regulators watch. What a stablecoin is, mechanically: a deposit that has left the bank and become a token, backed by a reserve of short-term US government debt, Treasury bills, whose interest the issuer keeps. Tether, based in El Salvador, has issued more than $180bn of them and made billions doing exactly that. That is $180bn of demand for government bills and $180bn of deposits not sitting in any bank. The banks' move is best read as defensive. If ordinary and institutional money is going to become tokens under the new American and European laws written for them (the GENIUS Act and MiCA), the banks would rather issue the tokens than watch the deposits leave. This cuts against our thesis in one respect, since it is risk moving toward the regulated banks' balance sheets, the lists of what they own and owe, not away from them. It confirms the thesis in another: the same week saw Circle create about 5bn of its USDC tokens, Coinbase and Better launch mortgages secured on bitcoin off a $260m waiting list, and Vanguard and Wellington test tokenized shares in money-market funds as something a lender will accept as security (sweep). The pipes between crypto and the market for government bills are being laid faster than anyone is measuring them. The contagion question is simple. A run on one of these tokens, everyone asking for their dollars back at once, is a forced sale of government bills. Done by Tether, that is a market event. Done by a vehicle owned by 21 banks, it becomes a question about the banks. For today this adds nothing to the pile, because nothing has been built; the day it carries real money is the day we move fragility. And there is one data point on demand: Société Générale, the only major bank that has actually issued a dollar token, has $12.5m of it in circulation.

Fed, Treasury and policy2 Sephigh

Two dates: September 9 and September 17

A Japanese rate rise on September 17 would test whether $4bn buybacks can hold up the price of thirty-year US government bonds while the single largest foreign holder of them has a new reason to bring its money home.

The market for US government bonds that do not come due for twenty or thirty years now has a calendar, and the two dates on it pull against each other. On September 9 the Treasury Department, which borrows the money the government spends, runs the first of its enlarged buybacks, at least $4bn per operation in bonds with ten to twenty and twenty to thirty years left to run, continuing to November 4 (sweep). On September 17–18 the Bank of Japan, Japan's central bank, meets, and money-market brokers price a 94% chance it raises its rate to 1.25%. Its governor, Ueda, said on Tuesday a rise will be "discussed thoroughly in every meeting, including the next one" (Nikkei). The first date is Washington trying to hold its long-term borrowing cost down; the second is the country that owns more of that debt than any other being handed a reason to take its money home. The two are connected by scale. Bessent, the Treasury secretary, announced the expanded buybacks on August 19. By September 1 the interest rate the government pays to borrow for thirty years was back at 5.27%, exactly where it stood before he spoke, and the ten-year rate was a tenth of a percentage point higher, at 4.80% (Bloomberg). The program is presented as help for older bonds that rarely trade, not as an attempt to hold rates at a level, and $4bn is what the Treasury sells in about an hour of a normal week. It is a backstop for stale bonds, not a defense of a price. Japan is the larger force. Japanese institutions hold more than $1tn of US government bonds. Japan's own government now pays 3% to borrow for ten years, the first time since 1996, and that is roughly the point at which a Japanese life insurer can earn a return above inflation at home without taking on any currency risk. Add the arithmetic on currency protection from this morning's dispatch and the route home for that money is open. Bessent's own undersecretary said as much on Tuesday. Bessent himself called Japan's era of stimulus over and christened the new one "Takaichinomics". This is policy pulling in two directions. The Treasury is spending its own balance sheet, the list of what it owns and owes, to hold long-term rates down, while its officials in effect endorse a Japanese rate rise that pushes them up. Warsh, who chairs the Federal Reserve, America's central bank, has meanwhile called 2% a "firm and fixed target" and offered no help. None of this is a mistake yet. It is a policy with no lever attached. For the odds, this is the

Signs of the times

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

Data centers asked for as much power as every home in America

Large power users, mostly data centers, have asked to connect more than 700 gigawatts to the grid across the Midwest, Mid-Atlantic and South, which is more than ten times what American data centers use today and roughly what every home in the country uses. Most of that cannot be real: in Texas, where requests jumped from 48GW to 474GW since 2023, the state has frozen new connections to work out which ones are.

ai_capexghost demandpower

1 SepReuters via Channel NewsAsia

58 banks, two digital dollars, $12.5m in circulation

Twenty-one global banks will launch a digital token meant to be worth exactly one dollar in 2027, and a rival group of 37 is doing a euro one this year. The only major bank that has actually issued one, Société Générale, has $12.5m outstanding against Tether's $180bn, which tells you how far the demand sits from the ambition.

stablecoinsbankscrypto_tradfi

1 SepReuters via Livemint

The Dutch moved gold that never traveled

The Dutch central bank moved 78 tonnes of gold from New York to London, citing 'geopolitical unrest', but physically shipped only 27 tonnes; it sold the rest in the Americas and bought fresh bars in London. Central banks do not lightly give up on a vault: gold overtook US government bonds as the world's largest reserve asset last year, and the line to get it out of New York now includes France and the Netherlands.

goldreservesdollar

2 SepFinancial Times

Dell tripled. Its customers didn't.

Dell has booked more than $130bn of AI server orders in twelve months and its shares have tripled this year; CoreWeave, the customer it named on the call, is 25% below its high and down 8% this week. The supplier is priced like a winner and the buyer, which pays with borrowed money, like the one holding the debt.

ai_capexvendorsneoclouds

2 SepChannel NewsAsia

The rumour mill

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

Partly true

“The 10-year Treasury yield crossed 4.8% and global yields are touching highs not seen since 2008.”

The interest rate the US government pays to borrow for ten years closed between 4.79% and 4.81% on September 1–2, its highest since late 2023, and a Bloomberg gauge of government borrowing costs worldwide is at its highest since mid-2008. Individual countries differ: Britain is paying its most since 1998 and Japan its most since 1996, so 'since 2008' understates some and overstates others.

Claimed by Meet Kevin

Partly true

“The Shiller CAPE ratio just reached 41 for the first time since 2000.”

Several outlets put the Shiller CAPE, a measure of how many years of average profits investors are paying for American shares, at 41.0 to 41.4 in August, the highest since September 2000. The 'first crossing' framing is loose, because it has sat around 40 to 41 since June. Directionally correct and worth knowing.

Claimed by Wealthion

Partly true

“CoreWeave has $5.5bn cash against $18bn debt and spent $14bn on capex versus $3.6bn operating cash flow.”

CoreWeave's second-quarter filing shows $5.52bn of cash and $13.5bn raised from new borrowing after repayments, but the figure for money spent building things is wrong in the wrong direction: in August the company raised its 2026 forecast for that spending to between $35bn and $39bn. The real number is more alarming than the claim.

Claimed by Meet Kevin

Partly true

“The US has struck Iran twice in three days, Iran says it hit American facilities in five countries, and two supertankers were hit leaving the Strait of Hormuz.”

The two rounds of US strikes and the tanker attacks are reported by neutral outlets. The claim that Iran hit American facilities in five countries is Iran's own, and NBC names four. This is the story behind gold's 2.5% overnight jump to $4,474, and it is one we will not treat as settled until the most reliable outlets confirm it.

Claimed by Kitco NEWS

Earlier developments

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

Private credit and BDCs2 Sephigh

Fifteen percent, forty lenders, five months of falling prices

Fifteen percent interest is what a borrower pays when no bank will lend to it, and Bathla shows what that rate buys the lenders when the flats stop selling.

Bathla Group, a Sydney apartment developer founded in 1997 by a former taxi driver, declared itself insolvent last week owing about A$3.3bn ($2.4bn) to more than 40 investment funds that lend where banks will not, Bloomberg reports. Many of those loans promised the funds around 15% a year, which is the rate a borrower pays when no bank will touch it, and it is only affordable while the flats being built keep rising in price. Sydney prices have now fallen five months in a row. The founder, Bhart Bhushan, personally guaranteed some of the debt, PAG, one of Asia's largest private investment firms, was among the lenders, and the company has 24 hours to agree a deal with its creditors. The way this kind of borrowing is stacked is worth spelling out. A developer borrows against the land, then borrows again against the building work, and then, once the flats are finished but unsold, takes a third loan against the unsold apartments themselves, which the trade calls a residual stock loan. Every layer is paid from the same place: selling apartments. At 15% the interest can only be met if prices are climbing, and the security on the last layer, the thing the lender gets to keep if the borrower cannot pay, is precisely the stock nobody is buying. More than forty lenders is the other problem. Nobody holds enough of the debt to run the negotiation, and each fund has its own investors asking for their money back. Bloomberg says a handful of Australian funds of this kind have already capped how much investors can withdraw. That is what a fund does when it cannot pay the people who want out, and a fund that lent at 15% against a half-built tower cannot raise cash from a half-built tower. Scale is the important question. Australia's market for this kind of fund lending is roughly A$200bn and, per ASIC, the country's corporate regulator, as much as 60% of it has gone into property. In North America the property share is 15 to 20%, according to MSCI. So this is not a rehearsal for the American problem we have been following in the filings of BDCs, funds that borrow money, lend it to mid-sized private companies, and pass the interest to shareholders, where the loans went to software companies. It is a cleaner experiment: one type of lender, one type of asset, and a price that has been falling for five months. On the crash question, this is a measurement rather than new danger. The loans were made long before last week and the prices have been falling for five months; what happ

The AI capex bubble2 Sepmedium

Four hundred data centers planned, a third in use

China built its AI data centers with government money and no customers signed up, and the question of how fast the chips wear out is the same question sitting under every American loan secured on those chips.

China had nearly 150 AI data centers, which it calls "intelligent computing centres", running by late 2024 and another 400 planned, and on average only about 30% of the servers inside them were doing anything, according to a Caixin commentary published Wednesday. Government bodies and the borrowing vehicles run by local governments put up about 35% of the money. Many facilities, in eastern and western provinces alike, run below 40%. Seven machines in ten sitting idle is the number that matters, because the machines lose value whether or not anyone uses them, and the same question sits under a growing pile of American loans. The piece is written for Chinese local officials trying to replace the money they used to make from selling land, which has collapsed, with AI infrastructure. One paragraph applies everywhere. The best chips are replaced by better ones every year, so the hardware in a data center realistically lasts three to five years. Some cloud providers, Caixin notes, spread the cost over six years in their accounts, which "masks the financial peril". That spreading of cost, which accountants call depreciation, is the quiet mechanism of this bubble. A chip bought for a dollar and written off over six years costs the company about 17 cents a year; over three years, 33 cents. Every operator's reported profit margin, and every lender's calculation of whether a borrower's income covers the interest on debt secured against chips (the GMI Cloud loan, the Iren notes, the CoreWeave book), rests on which of those two numbers is true. Nobody has to find out until the chips are retired. China is running the experiment with different money: government subsidies and special bonds rather than bond markets and the funds that lend to mid-sized private companies, and, as Caixin puts it, with facilities built "before securing clients". That is the real difference from the United States, where most of the capacity being financed is already under contract to somebody, and it argues against reading China straight across. What does not argue against it is the 30%. Use is what turns a signed lease into cash, and a 30% figure anywhere in the system is a reminder that the American contracts assume the demand will show up. Caixin's proposed fix, a red alert when a region drops below 30% use that halts new approvals and cuts its bond quota, tells you its authors think the number can fall further. Our confidence here is medium: this is an opinion piece, and the figures date fr

Crypto and TradFi contagion2 Sephigh

Raised $760m to buy bitcoin. Down 99%.

One company built to hold bitcoin has already collapsed without any lender losing money; the question is whether the bigger ones, now rising again, are doing it on borrowed money this time.

Nakamoto Inc., a company built by Trump ally David Bailey to do nothing but hold bitcoin, raised about $760m to buy the coin and has lost roughly 99% of its share price since the May 2025 merger announcement that created it, Bloomberg reports. Its shares now trade for less than the bitcoin it owns is worth. The losses fell on shareholders, and nothing in the reporting says a lender lost anything, which is the part worth your attention. "We bought Bitcoin at the top," Bailey said, "and then Bitcoin performed very poorly." Companies like this run on a premium. Sell new shares for more than the coins behind them are worth, use the money to buy more coins, and the number of coins per share goes up, which justifies the premium, which lets you sell more shares. It works exactly as long as investors will pay more for the shares than for the coins. Once the shares fall below the value of the coins the machine runs backward: selling new stock makes every existing shareholder poorer, and the sensible move is to stop buying coin and buy back your own shares instead. That is what Nakamoto is doing, putting buybacks ahead of bitcoin and hunting for a business that generates cash so it can keep buying without selling stock or borrowing. Two things are worth saying plainly. First, this is broadly fine. The people who lost money paid a premium for a coin they could have bought directly, and Bloomberg describes a rout across the sector that wiped that premium out without describing a single lender taking a loss. A bubble shrinking through its shareholders is how a bubble is supposed to shrink. Second, prices say the machine has restarted at the top of the sector even as it died at the bottom. Bitcoin is up 23% over 20 days. Strategy is up 25%, BitMine 25%, SharpLink 28%, all of them shares whose only business is holding a coin that has itself gone nowhere over five days, which makes them a magnified bet on it. Strategy is still 10% off its recent high. So the premium is being rebuilt, selectively, for the largest names. This does not move the odds of a crash: nobody was owed money, so nothing was left unpaid. What would move it is finding that the big names are rebuilding the premium with borrowed money this time, because a loan against a coin is a story about the lender rather than the shareholder. Nothing in this window answers that.

Bond market dysfunction2 Sepmedium

The Treasury Department says Japan is setting America's borrowing costs

When the department that borrows America's money blames its own interest bill on what Japan's central bank decides next, Japan's September 18 meeting becomes a day American bond markets have to worry about.

A senior official at the Treasury Department, which borrows the money the US government spends, said on Tuesday that Japan's bond buyers are what move America's borrowing costs. She said it two weeks after her own department doubled its purchases of its own long-dated bonds to hold rates down and, per Bloomberg, watched the whole effect erased. The interest rate Japan's government pays to borrow for ten years has crossed 3% for the first time in 30 years, while the US rate for ten years stands at 4.8%, a 20-month high, and Japanese institutions hold more than $1tn of US government bonds. A government blaming its own interest bill on a foreign central bank's next meeting is not something you hear often, and it turns that meeting, on September 17 to 18, into an American event. "Japan is such a large buyer of sovereign debt," Undersecretary Erin Browne told Nikkei, "it tends to have an impact, particularly on U.S. markets." The mechanism is simple. For decades Japanese insurers and pension funds bought American bonds because their own government's bonds paid almost nothing. At 3% they have something to buy at home with no currency risk. And when the Bank of Japan, the country's central bank, raises rates, the cost of insuring dollar bonds against swings in the yen goes up too, which makes the American bonds they already own less worth keeping. That rate rise is now near-certain. Governor Ueda said Tuesday a hike "will be discussed thoroughly in every meeting, including the next one", on September 17 to 18; Tokyo money markets price a 94% chance of a quarter of a percentage point, taking Japan's rate to 1.25%. Browne all but endorsed it: "market expectations oftentimes influence monetary policy." What came out of the G20 meeting in Asheville is a pair of promises about government borrowing. Finance Minister Katayama said Japan will "steadily reduce our ratio of debt to GDP". Browne said Bessent will soon release a plan to cut the US deficit. Neither has a date or a number attached. Ueda, for his part, named three causes for rates rising everywhere: inflation from the Middle East crisis, "strong demand for funding AI investment" and worries about government debt. The second ties this story directly to the one about the AI buildout. The thing we cannot yet say is whether Japanese money is actually coming home or is merely expected to. Nikkei's framing is that Japanese investors are "more likely" to pull money home; the interview cites no figures on money actuall

Hidden leverage and shadow banking2 Sepmedium

The SEC is asking the banks, not the fund

A $45bn fund was forced to sell almost everything and no bank reported a loss, which is the best evidence yet that the borrowed money behind AI stocks sits somewhere nobody measures until it breaks.

The SEC, the agency that polices American markets, has ordered major Wall Street banks to hand over records about Situational Awareness, an AI-focused hedge fund (an investment firm that manages money for the wealthy and can bet in ways ordinary funds cannot) whose lenders forced it to sell most of its shares during July's selloff, Bloomberg reported in late August. The fund's founder, 24-year-old Leopold Aschenbrenner, had grown it to $45bn on returns of 200% in 2025 and more than 400% in the first half of this year, "supercharging" those gains with borrowed money, according to a second Bloomberg piece. It lost more than two thirds of its assets in July. Funds have bad months; losing two thirds of your money in one is not a bad month, and yet no bank reported losing anything on it, which is why the regulator is asking the banks rather than the fund. The questions went to the banks, not the fund, and that is the point. A fund of that size borrows through the bank departments that lend hedge funds money against their holdings, known as prime brokers. When the holdings are concentrated in AI stocks and those stocks fall, the security behind the loans falls with them, the bank demands more cash right now, and the only way to raise it is to sell the same stocks into a market already falling because everyone else in the trade is doing the same. Ken Griffin's Citadel bought the bulk of what Situational Awareness sold. A regulator asking banks about a client's trading is usually asking two things: how much did you lend against what, and did you know how much everyone else had lent. The investor list is the other half. Most of the money came from wealthy individuals and tech founders, but the University of Virginia's endowment was in, and posted a 27% return for the year to June 30 partly on the strength of it, taking its assets to a record $18.8bn. July's losses, as Bloomberg puts it, are "next year's problem". This is the borrowed-money problem in miniature: one theme, a lot of debt, a seller who had no choice, and a bigger buyer who could wait. It resolved without a bank taking a visible loss, which is why markets never registered it. That is how it moves our reading: nothing was added to the pile this week, but we learned that the pile is measured badly. Whether the banks had properly sized what they lent to the fund is exactly what the SEC wants to know, and neither Bloomberg piece says how big it was.

Household credit2 Sepmedium

One dollar in eight on American credit cards is three months late

The pile of card debt that has stopped paying is growing while the amount lenders formally give up on each month is shrinking. That means losses are being pushed into the future, not avoided, and the lenders' share prices are looking only at the second number.

One dollar in every eight owed on American credit cards has gone at least three months without a payment, on Federal Reserve data for the second quarter of 2026 (12.8% of all card balances; CNBC, reading the same data, puts it at 12.92%). Less than four years ago the figure was not much more than half that (7.6% in the third quarter of 2022). The companies that lent the money did not treat this as bad news, and their shares rose on September 2. Either the market has decided these losses are already in the price, or it is watching a different number. The cards are not the only strain. Attom, which tracks property records, counts 227,500 homes entering foreclosure in the first half of 2026, up 21% on a year earlier and 28% on 2024. FICO, the credit-scoring firm, reports that 3.2 million student-loan borrowers, roughly one in seven of those with a payment due (15%), have been thirty days late at some point in the past six months. The lenders traded as if none of that had happened. Affirm rose 5.9%, Sallie Mae 3%, Capital One 2.5%, and Credit Acceptance closed at its highest level of the period. The different number exists. Each month the big card issuers report how much debt they have formally given up on, a figure called the charge-off rate, and in August it improved to 3.28% from 3.42% in June (Bread Financial at 6.60%, American Express at 1.7%). A charge-off and a ninety-day delinquency measure different things. The charge-off is what the lender wrote off this month. The ninety-day figure is the pile of balances that have stopped paying and have not yet been written off. When the pile grows while the monthly write-offs shrink, bad loans are sitting longer in the late column instead of being dealt with. That makes this quarter's profits look better and pushes the loss into a later one. The Federal Reserve's own second-quarter charge-off rate did not fall, it rose, to 3.70% from 3.66%, and the New York Fed notes the trouble is concentrated among lower-income borrowers. The question we keep coming back to is who ends up holding these loans. This week Reuters went through Bank of England filings and found British banks pledging store-card and vehicle-lease loans as security for central-bank borrowing, £17.8bn of it in the riskiest category the Bank accepts. When card debt stops paying, the loss does not necessarily stay with the company whose name is on the card. None of this adds new risk this week. It is a better measurement of risk that was already there: t

Private credit and BDCs2 Seplow

An offer at 26% off, and fewer than $5m of takers

An outside buyer's cash offer is the only real-world price a lending fund without a stock-market listing ever gets, and this one came in a quarter below what the managers say the loans are worth.

Cox Capital Partners has offered to buy up to $90m of shares in five lending funds that do not trade on any exchange, at prices averaging 26% below what the funds themselves say the shares are worth (individual bids ran from 14.1% to 37.1% below stated net asset value). Fewer than $5m of shares were handed over. The offer matters more than its size, because it is the only price these funds will ever get from somebody who does not run them, and it says the loans inside are worth about three-quarters of what the books claim. The same report puts the queue of investors waiting to get their money out of such funds at nearly $15bn, and separately says BlackRock has been sounding out buyers for TCP Capital's loan book. The funds in question are a type that borrows money, lends it to mid-sized private companies, and passes the interest to shareholders. They carry the label business development company, or BDC, and the ones here have no share price. Their value is whatever the manager says the loans are worth. The only way out is the fund's own quarterly buyback of shares, usually capped at 5% of the fund's stated value. When more investors want out than the cap allows, the queue lengthens ($15bn of it, on this account), and an outside firm steps in offering cash today at a discount. That offer is the closest thing these funds have to a market price. The low take-up cuts two ways. Holders may believe the manager's numbers and prefer to wait in the 5% queue. Or they may simply not have seen the offer. Either way, the manager is under no obligation to adopt Cox's price, so nothing in the fund's reported value changes. Two caveats. First, the sourcing is weak: the figures arrive through a single sweep that links to a forum post, and no major newspaper carried them in this window. Second, the BDCs that do trade on an exchange are not showing distress in their prices. Ares Capital is 0.8% off its recent high and FS KKR is up 11% in twenty days. Against that, a fact-checked dataset from PIMCO and The Lead Left has the average listed BDC selling for 75 cents on each dollar of stated value (0.75x NAV), a 25% discount arrived at in the open. If the Cox numbers hold, the private funds and the public ones are being marked down by about the same amount. The difference is that one of those discounts is published every day and the other had to be smoked out with a cash offer. This is a discovery, not a new risk. The loans are what they were last week; what we learned is what so

Crypto and TradFi contagion2 Sephigh

Twenty-one banks, one digital dollar token, a 2027 launch date

The banks want back the customer cash that walked out the door to unregulated token issuers, and the one big bank that has already tried has $12.5m to show for it.

Twenty-one banks and asset managers, Goldman Sachs, Bank of America, Citi, Wells Fargo, Deutsche Bank, UBS, Santander and MUFG among them, said on Tuesday they will set up a company this year to issue a digital token meant to be worth exactly one dollar, a stablecoin, in the first half of 2027, with a euro version to follow. The scale they are chasing belongs to Tether, based in El Salvador, which has issued more than $180bn of its dollar token and invests the money backing it in assets including US government bonds, the IOUs Washington sells to cover its deficits. The scale a bank has reached so far belongs to Société Générale, the first major bank to issue a dollar-backed token, which has $12.5m of it in circulation. The banks are trying to win back a business that used to be theirs, and the record to date says customers have not been waiting for them. They will compete with Qivalis, a separate group of 37 institutions planning a euro token this year. Some banks, including BBVA, sit in both camps. A stablecoin is a bank deposit that left the banking system and turned into a government bill held by somebody else. The interest on that bill goes to the issuer, not to the person holding the token, and the holder has no deposit insurance. Banks know this trade well, because it is their own business with the regulation stripped out, which is why they want it back. Whether the large firms that would use these tokens want one issued by a committee of 21 regulated institutions rather than a single fast company is the open question. Reuters notes there are few signs of demand for bank-issued coins. The same day, the London Stock Exchange said it will launch digital versions of London-listed shares with Payward, the parent of the crypto exchange Kraken, on its overnight trading venue next year, pending approval. The exchange says its tokens will carry full shareholder rights. Payward's existing xStocks give holders only limited ones. This runs against the direction we usually report. Most weeks the story is risk leaving the regulated banks for somewhere with less oversight. Here the regulated banks are trying to follow it out the door. Whether that makes the system safer or just changes the label depends on details nobody has published, what the tokens will be backed with and on what terms holders can cash out. Until those appear it moves the odds very little, and we are including it because the banks' answer, when it comes, will tell us something.

Bond market dysfunction2 Sepmedium

September 9 and September 18: two dates for the government's longest-term debt

The Treasury Department's bond buybacks and the Bank of Japan's likely rate rise land nine days apart, and both hit the same thing: the government IOUs that do not come due for twenty or thirty years. One is a rounding error next to the money the other could move.

The interest rate the US government pays to borrow for thirty years is back exactly where it stood before the Treasury Department stepped in to support the market (5.27% on September 1), which means the support announced on August 19 bought two weeks of relief and nothing more. Two dates now decide what happens next. From September 9 the Treasury will buy back at least $4bn of its own long-term bonds per operation, double the previous cap, running through November 4. On September 17 and 18 the Bank of Japan meets and is expected to raise its rates, which would give the largest foreign holders of American government debt a reason to bring their money home. The buybacks cover bonds with ten to thirty years left to run (two categories, 10-to-20 and 20-to-30 years), and Scott Bessent, who runs the Treasury Department, has said it "could do more". In Tokyo, Governor Ueda said on Tuesday that a rate rise "will be discussed thoroughly in every meeting, including the next one". Money markets there put a 94% chance on a move to 1.25%. The August 19 announcement pulled rates down for two weeks. By September 1 the thirty-year rate was back at 5.27%, and the ten-year rate, at 4.80%, is more than a tenth of a percentage point higher than before the announcement. The reason the tool is small is arithmetic. In a single late-August week the Treasury sold $797bn of new debt, $235bn of it in notes, the medium-term kind. Buying back $4bn against that is what the Treasury says it is, a way of helping old, hard-to-sell bonds find a buyer. It is not a purchase big enough to set the price. Japan matters because Japanese institutions hold more than $1tn of US government bonds, and their home alternative, Japanese government debt, now pays 3%, a level not seen since 1996. Treasury Undersecretary Erin Browne said outright that Japan's rising yields are affecting US markets. A Bank of Japan rate rise raises the reward for keeping money at home and tends to strengthen the yen, which is what Tokyo wants after the currency hit a 39-year low of 163 to the dollar in July. If Japanese money does come home, it comes out of exactly the bonds the buybacks are meant to support. What the evidence does not show is a breakdown in the machinery. Neel Kashkari at the Federal Reserve says the market is "functioning as it should". The sweeps we read found no auction where the government had to pay more than the market expected to get its bonds sold, and no blowout in the gap between two closely rela

Private credit and BDCs2 Sepmedium

An insurer, a housing agency and $20bn to the family

It is the running thesis in a single balance sheet: cheap public-backed funding on one side, unmarked related-party loans on the other, policyholders in the middle.

Two insurers controlled by Mark Walter, the Guggenheim chief executive and sports owner, owed more than $6bn to the Federal Home Loan Bank of Indianapolis at the end of June, a 36% jump in six months and roughly double the balance at the start of 2025, according to Bloomberg. Over the same period the insurers, grouped under a unit called Group 1001, have lent more than $20bn to other parts of Walter's empire. The plumbing matters. The Federal Home Loan Banks were created in the Depression to keep mortgage lending alive: they borrow cheaply on the strength of their government sponsorship and pass that funding on to members against pledged collateral. Insurers can join. Once in, an insurer can post bonds it already owns, draw cheap money, and invest the proceeds at a spread — a trade that has nothing to do with home loans. Bloomberg says insurers' borrowing from the system has tripled since 2013, and that Group 1001 is going further than peers in maximising what it can draw. So one side of these balance sheets is funded partly with quasi-public money; the other side holds loans to affiliated companies. The people ultimately exposed are policyholders, whose annuities are backed by whatever those related-party loans turn out to be worth. Cornelius Hurley, a former FHLB Boston director, calls it 'a pure and simple case where the taxpayers are subsidizing funds that are given to insurance companies'. What the reporting does not tell us: what the $20bn of affiliate loans are secured on, or how state insurance regulators value them. That gap is the point. This is the migration Crash Lab keeps returning to — risk moving from banks, where it is stress-tested and marked, into insurance balance sheets, where a related-party loan can sit at cost. The share prices of the listed insurance-plus-credit managers (Apollo 7.5% off its high, KKR 7.7%) show mild unease, not alarm. Nothing here says the loans are bad. It says nobody outside the group can check.

The AI capex bubble2 Sepmedium

Nvidia owns the landlord, the tenant and the guarantee

Investment-grade project debt is being issued against leases whose real credit is a chip vendor's guarantee — the vendor-financing thesis in its largest form yet.

SB Energy's IPO filing puts a $439bn backlog against a business that has not yet earned a dollar from data centres. Nikkei's read of the S-1 shows where the number comes from: 8.8 gigawatts of contracted capacity, none of it operating, with OpenAI expected to be the largest tenant. SoftBank's unit lost roughly $3.2bn in the first half of 2026 on $139m of revenue, almost all from its legacy energy business, and describes itself as 'substantially dependent' on OpenAI as both tenant and investor, per CNBC. It is expected to list at $50bn or more. The chain that holds this together runs through one company. Nikkei counts Nvidia's $3bn investment in SB Energy (the landlord), $30bn in OpenAI (the tenant), and a guarantee of up to $105bn to SB Energy if OpenAI cannot pay the lease on the Ohio site. OpenAI has put $500m into SB Energy and will receive warrants; SB Energy has promised to buy OpenAI's services. Building the contracted sites will cost more than $170bn, most of which has to come from outside investors — the same investors Nvidia and the banks are now courting for a $500bn AI-infrastructure raise. Here is the mechanism a reader should take away. Data-centre project finance is lent against the lease. The lease is worth what the tenant's promise is worth. The tenant's promise is backed by the chip vendor, who sells to the tenant and owns part of the landlord. SB Energy's chief executive said as much: Nvidia's presence 'helps us to unlock things like investment-grade financing'. Investment-grade debt on a company with no revenue is only possible if the buyers are, in substance, lending to Nvidia. We do not know from the filing how the $105bn guarantee is carried on Nvidia's own books, or the term of the leases behind the $439bn. The market is not asking: Nvidia is up 7% in five days and 1.6% off its high.

Hidden leverage and shadow banking2 Sephigh

A 24-year-old's margin call, an endowment's record year

A leveraged AI bet turned into a record endowment year and a two-thirds loss in the same twelve months, with the loss landing in a different fiscal year to the gain.

Situational Awareness, the AI-focused hedge fund founded by 24-year-old Leopold Aschenbrenner, lost more than two-thirds of its $45bn in July when margin calls forced it to sell the bulk of its public portfolio to Citadel. Bloomberg now reports that the University of Virginia's endowment was one of the few institutions invested. UVA's long-term pool returned 27% for the year to 30 June and reached a record $18.8bn. The arithmetic of that sentence is the story. The fund rose 200% in 2025 and more than 400% in the first half of 2026 'supercharging its returns with borrowed money'. Those gains were marked at 30 June, the endowment's year-end, so they flow into the 27% the university will report in the coming weeks. The collapse happened in July. As Bloomberg puts it, the rout 'will be next year's problem'. UVIMCO does not disclose the size or timing of its positions or the identity of its roughly 90 outside managers, so the actual exposure is not knowable from the outside. Two things about the plumbing. First, leverage in AI equities does not only sit in the visible places; it sits in funds like this one, financed by bank prime brokerages, and it shows up in institutional portfolios as a single return number. Second, when it unwound, the system handled it: the margin calls were met by selling to one large buyer within a day, and the SEC's subpoenas to the banks, reported 24 August, are the regulator asking what happened rather than a rescue. That partly cuts against the Crash Lab thesis — this leverage was inside the measured system and the measured system caught it. The unresolved question is how many other pools of borrowed money were long the same names in July, and whether the next forced seller finds a Citadel waiting.

Crypto and TradFi contagion2 Sepmedium

The bitcoin treasury company worth less than its bitcoin

When a treasury company trades below its coins the only way to grow is to stop being a treasury company — the model, not just the stock, has failed.

Nakamoto Inc., the public vehicle David Bailey built to stockpile bitcoin, has fallen about 99% from its peak since the May 2025 merger that created it, Bloomberg reports. Bailey had raised roughly $760m for the company; his hedge fund had returned 640% on similar bets. 'We bought Bitcoin at the top, and then Bitcoin performed very poorly,' he said. The mechanism that broke is worth spelling out. A digital-asset treasury company works only while its shares trade above the value of the coins it holds. At a premium, it issues stock, buys more coin, and each purchase is accretive to existing holders — a flywheel. Nakamoto now trades 'well below the value of its Bitcoin holdings', which means issuing shares to buy coin destroys value. Bloomberg's phrase is exact: 'the financing engine has broken down'. Bailey's response is a conventional company's response — buy cash-generating businesses, prioritise buybacks of the discounted stock over adding bitcoin, avoid new debt. Bloomberg describes this as part of 'a broader rout in digital-asset treasury companies that has wiped out the premium'. The tape supports that reading only partly. Bitcoin is $77,098, up 21.6% in twenty days; Strategy is up 25.4% over the same period, tracking its coin rather than re-rating. On the week, though, the ether treasury names have lagged badly: BitMine is down 7.4% and SharpLink 2.6% against ether-adjacent bitcoin down 0.9%. Equities falling faster than the coin they hold is what premium compression looks like from the outside. We do not have a current discount figure for Nakamoto, and Bloomberg does not give one. The contagion question for this beat is narrow and specific: which of these vehicles carry debt, and what happens to that debt when the equity that was supposed to service it trades below the collateral.

The AI capex bubble2 Sepmedium

China built 150 compute centres. They run at 30%.

It is the first place the AI buildout has run long enough to show utilisation numbers, and they are half of what the depreciation schedules assume.

A Caixin commentary offers a look at what the AI buildout looks like a few years further along. By late 2024 nearly 150 'intelligent computing centres' were operating in China with another 400 planned. Average utilisation hovers around 30%; many facilities in eastern and western provinces run below 40%. Government entities and local government financing vehicles — the off-budget borrowers that funded the property boom — account for about 35% of the investment. The author's diagnosis is that local officials 'build facilities before securing clients, prioritizing peak computing power over vital software and data ecosystems'. And then the sentence that should interest anyone holding a US data-centre bond: chips iterate annually, so the realistic useful life of AI hardware is three to five years, while some cloud providers 'stretch accounting depreciation to six years, masking the financial peril'. The proposed remedy is a red alert at 30% utilisation, halting new approvals and cutting special-bond quotas. This is commentary, not audited data, and China's centres are built for a different customer base than Ohio's. But set it beside the Reuters review of US grid requests — Texas data-centre power applications up from 48GW to 474GW, only 20 of more than 100 proposed Pennsylvania projects bothering to seek permits — and the pattern is the same in both systems: capacity is being committed ahead of demonstrated demand, financed by entities whose losses do not have to be marked promptly. In China that entity is the LGFV. In the US it is the project-finance vehicle lending against a lease. The question a Chinese compute centre at 30% utilisation poses to the US market is what a stalled hyperscaler tenant would pose to SB Energy's lenders: who eats the depreciation on a three-year asset financed over six.

Bond market dysfunction2 Sephigh

Treasury names the leak in its own market: Japan

The largest foreign holder of Treasuries now has a reason to sell, and the US Treasury has publicly acknowledged the channel.

A senior US Treasury official has said out loud what the rate move implies. 'Japan is such a large buyer of sovereign debt, particularly developed market debt around the world,' Undersecretary Erin Browne told Nikkei. 'It tends to have an impact, particularly on U.S. markets.' Japan's 10-year yield crossed 3% on Tuesday for the first time since 1996; the US 10-year is at 4.80%, a 20-month high, and the 30-year at 5.27% — back to the level moments before Scott Bessent expanded Treasury buybacks on 19 August, according to Bloomberg. The channel is simple. Japanese institutions hold more than $1tn of Treasuries. For decades the yield at home was close to zero, so owning US debt made sense even after the cost of hedging the dollar. As Japanese yields rise, the case for holding a foreigner's bond weakens and money comes home. Nikkei's phrasing: rising Japanese rates mean investors are 'more likely to withdraw funds from the American and repatriate capital'. Browne's admission that this is already 'impacting' the US market is the Treasury conceding that its own long end is partly priced in Tokyo. What happens next is scheduled. Money-market brokers price a 94% chance the Bank of Japan raises rates on 17–18 September; Governor Ueda says a hike is on the table at every meeting. Browne did not object — 'market expectations oftentimes influence monetary policy' — which reads as Washington accepting a hike that will push Japanese yields further up. Bessent, she said, will soon release a US deficit-reduction plan; Japan's finance minister has pledged to 'steadily reduce' debt to GDP. None of this is dysfunction. Bank of America's Mark Cabana says only that the market 'has not been able to hold any type of significant rate decline'. The 2s10s curve sits at 41bp. But the Treasury has just told you which buyer it is worried about losing, and that buyer's central bank meets in two weeks.

The AI capex bubble2 Sepmedium

Neocloud debt: $19bn in a week, chips as security

The AI buildout is now being funded by debt secured on the hardware itself, arranged increasingly by private credit rather than banks, at exactly the point where public equity has stopped paying up for the borrowers.

JPMorgan began sounding out lenders on 27 August for a $5bn debt package for Volta Infrastructure Holdings, a data-centre developer founded roughly seven months ago that raised $300m at a $2.4bn valuation in early August, with Nvidia, Andreessen Horowitz, Altimeter and Michael Dell among the backers. In the same two days, Nebius upsized a debt offering to $5bn on top of $5.75bn of convertible notes it had just sold; Blue Owl arranged $2.4bn for Iren — a $1.2bn term loan and $1.2bn of senior secured notes — earmarked for Nvidia Blackwell Ultra GPUs at a campus in British Columbia; Lambda closed a $926m term loan B; and Core Scientific finalised $600m, most of it a letter-of-credit facility. Call it $19bn of new neocloud debt in a week, on top of the $182bn of bonds PIMCO counts from hyperscalers and chipmakers this year. Two things about the plumbing. First, the collateral is the chips and the contracts to rent them. Iren's loan is not general corporate borrowing; it is staged purchases of Nvidia hardware, to be repaid from compute revenue. That is a cash-flow loan against an asset whose resale value nobody has tested at scale. Second, who is lending. Volta's package is a bank syndication; Iren's is Blue Owl — a private credit manager — which puts a GPU-backed loan on balance sheets that mark quarterly and never have to sell. Nvidia sits on both sides of the Volta deal, as shareholder and supplier. If Volta borrows the full $5bn against a $2.4bn valuation, its debt would be more than twice its equity value. That is our arithmetic, not a term sheet; the leverage only makes sense if tenant contracts are already signed, and the reporting does not say whether they are. The market is pricing the vendor and the borrowers differently. Nvidia is up 7% in five days and 1.6% off its high. CoreWeave is down 8% on the week and 25% off its high; Nebius is 27% off; Applied Digital 21%. Lenders are extending credit to the tier equity investors are selling.

Private credit and BDCs2 Sepmedium

Eight in ten BDC software loans marked down

Private credit's largest single bet — loans to software companies secured on their own subscriptions — is where the markdowns are concentrating, and the listed vehicles are being bid up regardless.

Reuters has gone through the filings of 44 US business development companies. Among ten with comparable disclosures, non-accruals — loans that have stopped paying interest — were 3.4% of portfolio cost at 30 June, up from 2.5% at year-end. About 4% of all borrowers were marked below 80 cents on the dollar, against roughly 1% through 2023–25. And 81% of software loans have been marked down this year, against 40% for everything else. The analysis does not say by how much, and we have not seen the underlying table. Jefferies' fund-level numbers say the same thing from a different angle: Blackstone Secured Lending's non-accruals at 3.6% from 0.3% in early 2025, Golub at 2.9%, Blue Owl at 2.8%, Ares Capital at 2.4%. FS KKR and Crescent both told investors a 'handful of investments' accounted for most of their markdowns. The software figure matters because of how these loans were built. Much of direct lending's growth went into private-equity-owned software companies on 'recurring revenue' loans sized against subscription income rather than assets. The collateral is the revenue. If customers churn, or AI lets them pay less for the same function, the loan and the security weaken together. A mark below 80 is usually where a lender has stopped assuming full recovery; at 4% of borrowers, that is the pipeline for next year's non-accruals. This is a re-measurement of the second quarter, not a new event, and fragility does not move on it. But note the tape: BXSL and ARCC are within 1% of their highs, FS KKR is up 12% in a month, Blue Owl's BDC is up 5%. Investors are treating the software markdowns as a contained handful. The below-80 bucket says the handful is growing.

Fed, Treasury and policy2 Sephigh

Japan will hike into a 3% ten-year

The world's cheapest funding currency is about to get more expensive again, in the middle of a global long-end selloff that its own savers have historically helped absorb.

Kazuo Ueda said on Tuesday that the Bank of Japan will consider a rate hike 'thoroughly in every meeting, including the next one', on 17–18 September. Money-market brokers price a 94% probability of a quarter-point move to 1.25%. It would be the sixth hike since March 2024, and it comes the same day the 10-year JGB yield touched 3% for the first time in 30 years. Ueda listed the drivers of the global yield rise as inflation from the Middle East, 'strong demand for funding AI investment' and fiscal worries — a central bank governor naming data-centre borrowing as a cause of sovereign yields. The yen has not helped. It fell to 163 per dollar in July, a 39-year low, and stayed weak after a joint US–Japan intervention on 31 July. Ueda met Scott Bessent on Sunday; Bessent said afterwards that Japan's era of aggressive stimulus 'is now over' and has entered 'Takaichinomics'. Bessent is simultaneously trying, and so far failing, to cap his own long end with buybacks. Why a 25bp move in Tokyo matters here. For two decades Japanese insurers and pension funds bought foreign bonds because domestic yields were near zero, and global investors borrowed yen at nothing to fund positions everywhere else. At 3% on a domestic 10-year, the hedged yield on a Treasury is no longer obviously better than staying home. We cannot see the flows in real time; the inference is ours. But if the largest foreign holder of Treasuries has less reason to add, that lands on a US long end already at 5.26% with a Treasury that has just doubled its buybacks. The yen ETF rose 0.9% today. That is the market beginning to believe Ueda.

What would change our mind

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

The Bank of Japan raises rates on September 17–18, Japan's ten-year borrowing cost clears 3.25% and America's thirty-year pushes through 5.5%: that would tell us Japanese money is coming home, and we would raise ignition.

Would move the number

The dollar index breaks below 97 in the same week that US government bonds fall in price, which is the signature of foreign holders without currency protection buying it; the two moving together is the thing to watch, not either alone.

Would move the number

The extra interest the shakiest companies pay compared with the government rises above 3.5 percentage points, or the VIX rises above 25, while one of the borrowed-money AI computing companies or a non-traded fund lending to mid-sized private companies stops investors taking their money out: that is a funding market tightening, and we would move ignition by more than a couple of points.

Would move the number

Conversely, if the September 9 buybacks coincide with the thirty-year rate holding below 5.2% for two weeks, and CoreWeave, Nebius and Applied Digital recover half of what they have lost from their highs, we would take the number down.

Would move the number