The CFTC positioning data released on Friday contained the forced seller the Treasury market has been looking for all week. Money managers cut their long position in ultra-long bond futures by about $27m per basis point of risk in the two weeks to 6 October — roughly $38bn of the current 10-year note — while the 30-year yield rose to a 24-year peak of 5.68% (Bloomberg). Open interest fell with it. This was a reduction, not a repositioning.

The mechanism is in the contract's fine print. A Treasury future is a promise to deliver one bond from a basket, and the seller picks the cheapest. Which bond is cheapest depends on the level of yields; as the 30-year pushed towards 6%, the cheapest-to-deliver started migrating to a longer bond. When it switches, the contract behaves like that longer bond and its duration jumps. An asset manager holding futures to meet a duration target suddenly holds more duration than its mandate allows, and must sell. The selling pushes yields up, which pushes the switch further along. Bloomberg flagged the hazard last Monday; Friday's data show it operating.

What makes this a lead is which side it is on. The leveraged basis trade is hedge funds short these futures against cash bonds; the asset managers are the longs on the other side. In March 2020 it was the leveraged shorts who were forced out. This time the data point at the unleveraged longs — a different, less-watched channel for the same loop we described on 1 October for mortgage hedgers.

Then the timing. The window ends on 6 October, which is where the week's yields peaked. Since then the $39bn 10-year and $22bn 30-year auctions both cleared midweek, and the 30-year closed Friday at 5.60%. Either the forced leg has run its course, or the data simply stop where they stop. Next Friday's release will show whether the selling continued into the auctions, and that is the number we want.

The Euro Stoxx banks index has fallen about 8% in two weeks, to its lowest since June, and entered a technical correction on Thursday. Société Générale, Crédit Agricole and Deutsche Bank are each more than 15% off their recent highs, and the premium France pays over Germany is the widest in more than a decade (Bloomberg).

On Monday we wrote that the French selloff had crossed a border and an asset class — into the euro and the periphery. By Thursday it had crossed into bank equity, which is the channel that matters. European banks hold their own sovereign's bonds as capital, as liquidity buffers and as collateral. When the sovereign's yields rise, those bonds are worth less, the bank's capital is thinner, and its own borrowing costs rise because the market treats the bank's credit as a function of the state's. That is the 2011 loop. The second channel is slower: Roberto Scholtes of Singular Bank told Bloomberg that yields "appear to have crossed a pain threshold", and that if rates rise as far as the curve implies, non-performing loans could rise significantly while lending slows.

Bloomberg's verdict is caution, not panic: balance sheets are stronger than in 2011, bank shares had tripled since 2022, and the sector was a crowded consensus long. All true. But the first leg of a doom loop and the unwind of a crowded trade look identical from the equity side. The thing that separates them is the sovereign spread, and whether it keeps widening once the positioning has washed out. Third-quarter results, which Bloomberg expects to reassert the sector's soundness, will also be the first set marked at these yields.

The S&P 500 hit a record on Tuesday, stumbled for two sessions on AI demand fears, bounced on Friday and closed the week 0.1% from its high with VIX at 14.8. Barely a third of its members are above their 50-day moving average. In the rate-sensitive Russell 2000 the figure is 27%, and the index has now fallen five weeks in a row. The 10-year approached 5.4% this week, its highest since 2002, Brent is above $100, and the weakest corporate borrowers are paying about 15% (Bloomberg).

The week's prices on our own list show the same split. The companies selling power to data centres were the week's winners: Talen up 18% to a high, Vistra up 15%. The companies buying it fell: IREN down 16%, Nebius 9%, CoreWeave 8%, Applied Digital 7%. The listed BDCs fell 2–4% and Prospect Capital 14% while high-yield spreads tightened on the week. Regional banks fell 2.4%; Jefferies and Goldman are 13–15% off on the month. Microsoft and Amazon closed at highs.

The mechanism is that the index measures the market value of the companies the bond market is financing, and the breadth measures the companies the bond market is charging. A 15% coupon is not a price; it is a refinancing wall. Any firm that has to roll debt in the next year at that cost is handing its equity to its lenders, which is why the median stock is sinking while the cap-weighted few rise. The AI weight is doing double duty: it is the reason the index is up and, through roughly $500bn of AI debt issuance this year, part of the reason the long end is up.

That is where the week left the machine. The index says conditions are fine. The bond market, the median member and our own tickers say they are tightening. Both cannot stay true for long.

Bonds of Virgin Media O2, the British telecom Liberty Global co-owns with Telefónica, have been in free-fall since last week and are now deep in distressed territory. On Thursday Liberty's executives drew a crowd at a Morgan Stanley leveraged-finance conference in London; some investors could not get into the room (Bloomberg).

The trigger was regulatory. On 2 October the Competition and Markets Authority provisionally flagged concerns over Nexfibre — a Virgin Media joint venture — buying rival Netomnia. That deal would have brought cash into Virgin Media; the final decision is due by 15 December. The company's reply is that its owners are "committed to the business and reducing leverage" and that no debt is due until 2029.

That last line is both the reassurance and the problem. A levered telecom's deleveraging plan runs on asset sales. Block the sale and the bonds reprice from "repaid at par from proceeds" to "restructured before 2029", which is what distressed means when nothing is actually due. Three years with no maturity is three years in which rates and competition can get worse and nothing forces a resolution. And because the bonds are a key component of the European high-yield index, the loss passes straight through to every passive junk fund.

The same week produced the other kind of leveraged-finance fight. Electronic Arts' creditors say the company defaulted by skipping a change-of-control payout after its $55bn buyout by Silver Lake, Affinity Partners and Saudi Arabia's PIF. EA's answer is an investment-grade rating from Egan-Jones on a subset of the bonds and a defeasance manoeuvre — buying Treasuries to cover future coupons so the bonds are secured and the downgrade trigger never fires. Creditors say an alternative rating cannot replace a public one and the documents bar a mid-process swap (Bloomberg). In both cases the money is being fought over in the document, not the business.

On Saturday the Reserve Bank of India announced its strongest currency action since the 2013 crisis. From Monday the three state oil companies — Indian Oil, Bharat Petroleum and Hindustan Petroleum — will buy their dollars directly from the RBI through a special window rather than in the market. Banks must set aside cash with the central bank against some foreign-exchange transactions, a tool borrowed from the People's Bank of China, and the limit on FX derivative transactions has been cut (Bloomberg).

The rupee closed Friday just above its record low of 96.965 per dollar. It is Asia's worst currency this year, and it has kept sliding despite rate rises and dollar sales that shrank reserves by $51bn in the four weeks to 2 October. Anil Kumar Bhansali of Finrex called the measures "a virtual shutdown of the market" and expects a 1% rally on Monday; offshore, the rupee rose 0.6% after the announcement.

The plumbing is simple. India imports its oil, Brent is above $100, and the oil companies are the largest single source of dollar demand in the spot market. Routing that demand through the central bank removes it from the price — but the dollars still leave the RBI's reserves, faster rather than slower. The cash-reserve tool makes hedging expensive, which deters speculators and deters exporters hedging real receipts in equal measure. This is what a defence looks like once rates and intervention have failed: not a better price for dollars but control over who gets them.

It fits the month. The dollar index is up 3.1% in 20 days, the Philippines burned $4.85bn defending the peso, and now the second-largest emerging market has gone to administrative measures. Monday's open in Mumbai is the test, and the number to watch is reserves, not the exchange rate.

Citi expects Scott Bessent to cut the 20- and 30-year auctions by $3bn each at the 4 November refunding, funded with more bills, and thinks the 20-year could be cancelled outright. The first hint may come in next Friday's dealer survey, which Citi's Jason Williams says could ask whether long-end demand is being "partly cannibalized" by high-quality hyperscaler issuance (Bloomberg).

Two mechanisms are in that paragraph. The first is the maturity trade. Three-month bills pay 4.06%; the 30-year pays 5.60%. Shifting borrowing to bills is cheaper today and must be rolled every quarter, which is to say the sovereign is taking on exactly the funding mismatch the private sector is being punished for. BNP's strategists doubt it would lower yields at all.

The second is crowding-out, and it is the first time a Treasury document may put the AI debt boom and sovereign funding in the same sentence. Pension funds have a finite appetite for 30-year paper. Oracle, Broadcom and SpaceX have sold them close to $500bn of long bonds this year at a spread over Treasuries. Williams does not think that supply has moved the level of rates, but does think pensions are "leaning into long-end IG" more than usual. If Treasury asks the question, it is because it is wondering the same.

The other policy development of the week sits on the day after the refunding. Trump has set up a committee — his economic adviser, the EEOC chair and the ethics office director — to examine Fed governor Lisa Cook at a White House hearing on 5 November and recommend whether there is "cause" to remove her (WSJ). Her lawyers call it a box-checking exercise; her term runs to 2038 and she sued to block the first attempt. A Treasury shortening its debt and a White House building a case against a governor, in a week the bond market asked for restraint, is policy risk of the slow-burning kind.

OpenAI's annualized revenue, the latest month's sales multiplied up to a full year, was approaching $50bn at the end of September, according to financial documents shared with investors and seen by the Financial Times. Late last month the same investors were being told, and the press was reporting, a figure of about $70bn, and OpenAI did not deny it at the time. A number that hundreds of billions of dollars of borrowing depends on had just shrunk by $20bn. The Nasdaq 100, the index of America's largest technology shares, fell 1.7% after the report, Nvidia fell 2.9%, Oracle almost 6%, and the companies that rent computing power to OpenAI, which the industry calls neoclouds, fell harder (CoreWeave 7.8%, Nebius 7.3%).

Nobody's sales actually fell. The gap is a matter of definition. Anthropic counts sales made through Amazon's and Google's clouds in its figure; OpenAI does not. Investors trying to compare the two "grossed up" OpenAI's figure, inflating it to match Anthropic's method, to about $40bn in July, then applied the company's own statement that revenue had grown more than 70% since, and arrived at $70bn. The new presentation shows roughly $30bn in July, and 70% growth from there lands near $50bn.

What matters is what sits on top of that number. OpenAI's revenue figure is the input to every downstream decision about money spent building things: Oracle's leases, the Broadcom and SpaceX chip financings, Nvidia's guarantees, the $1.4tn valuation OpenAI is currently seeking in a private fundraising. OpenAI is a private company and is not required to disclose anything. Yet the market has been pricing hundreds of billions of dollars of publicly traded debt and shares off a figure that can move by $20bn when someone changes how it is counted.

The 70% growth rate is intact, and it is still extraordinary. But the day's lesson is that the single most important measure of demand in the whole AI build is a number the company reports about itself, with no agreed definition, and the public market reacts to it as if an auditor had signed it. Yesterday's moves in the neoclouds, down 7% on a day when the demand they depend on did not actually change, are the price of that. Nothing was added to the risk here; we found out that a load-bearing number was softer than we had been treating it.

Ares, Barings, BC Partners and Churchill have all looked over troubled funds in recent months, according to Bloomberg, as managers try to sell off battered vehicles in the $1.8tn market for lending to companies by investment funds rather than banks. Listed funds that borrow money, lend it to mid-sized private companies and pass the interest to shareholders, including BlackRock's TCP Capital, WhiteHorse Finance and Investcorp Credit Management, have said they are weighing their options, which may include a sale. These funds are called business development companies, or BDCs, and several of them being on the block at once is not a normal state of affairs. "I can't think of another time like this where you've seen so many public BDCs up for sale," said KBW's Paul Johnson. The talks are preliminary.

A BDC publishes a net asset value, its own estimate of what its loans are worth. For most of this cycle the share price has been the only outside opinion of that estimate, and it has been a skeptical one. Prospect Capital is down 11% this week and 18% from its high, and FS KKR, Blue Owl Capital Corp and Ares Capital all fell 2 to 3% in a week when the extra interest paid by the debt of companies too shaky to be considered safe barely moved. The shares are falling while the loans they hold are supposedly fine, which is the market saying it does not believe the valuation.

A sale changes that. When a rival buys a BDC, or buys loans out of one, the price paid is a real price that somebody else set. Either the buyer pays something close to the fund's own valuation, which vindicates the books, or it pays the discounted share price, which tells every other holder of similar loans what they are worth. Bloomberg frames the trade as a "growth shortcut": with new lending volumes shrunk and investors pulling record sums out of the unlisted funds, buying loans from someone else is the only way to add them. Blue Owl's co-chief executives ruled out a buyout of their own firm earlier this week.

This is the test we have said must come. It is not arriving as a wave of borrowers failing to pay. It is arriving as consolidation, which is quieter, and which will put real transaction prices on loans that have only ever been valued by the people who own them. Which way that moves the odds depends entirely on the price that gets disclosed.

Investors pulled $2.4bn from Guggenheim Partners' four largest mutual funds in September, after reports that federal prosecutors and the SEC, the agency that polices American markets, are examining how chief executive Mark Walter was able to borrow from his own insurance companies without the firms disclosing the loans (Bloomberg). The Total Return Bond Fund lost $1.8bn, 6.1% of its assets, and is now below $29bn; Limited Duration lost $317m, Macro Opportunities $222m, Core Bond $65m. Funds gain and lose money every month, and this would not be a story if it were steady. It is accelerating: $103m left in July, $647m in August, $2.4bn in September. Guggenheim says each fund carries a four- or five-star rating from Morningstar, the firm that grades funds, and that its business managing money for institutions continues to attract clients.

Two things make this more than a story about reputation. First, these are funds that promise investors their money back on any day they ask, and they hold bonds in a market where the average interest rate paid by the safest corporate borrowers has crossed 6%, and where the dealers who normally stand ready to buy bonds have been cutting back how much they hold. A withdrawal of 6% in a month has to be met by selling into that. Nobody is complaining about performance; the exit still makes the fund a forced seller. Second, the thing under investigation is the link we keep returning to: the owner of an asset manager borrowing from the insurers he controls. Insurers' balance sheets, the lists of what they own and owe, are where a great deal of this cycle's lending risk has gone to sit quietly, and the question of who gets to borrow from them, on what terms, with what disclosure, is exactly the question regulators have not asked often enough.

This was a run on a name, not on a kind of asset. But runs on names are how runs on whole asset classes start, and the mechanism, which is to pull your money from whichever fund is in the news regardless of what it holds, does not care what the bonds are worth. A forced seller is the kind of thing that brings a spark closer rather than adding to the pile, and that is how we count it.

The Treasury Department, which borrows the money the government spends, sold $22bn of thirty-year bonds on Thursday to solid demand, matching the strong sale of ten-year bonds earlier in the week, and the interest rate the government pays to borrow for thirty years fell 0.06 percentage points to 5.61%, down from its highest level since 2002 (Bloomberg). "Demand held up," said TD's Molly Brooks. That matters because the question all month has been whether anyone would keep buying at these rates. Rates got a second push lower when Trump said he would hold off attacking Iran again until after the November 3 midterms and oil came back from a peak near $105; the rate on two-year government borrowing fell to 4.75%, and the bets traders place on the Federal Reserve, America's central bank, imply roughly a one-in-five chance it raises rates this month.

This is the first week since the selling began that these long bonds have been tested twice at these levels and held both times. We take it seriously: a rate the market previously would not accept has now been accepted, with real buyers, and that is what a floor looks like if one is forming. Friday morning has given a little back (the ten-year rate is 5.26% and the thirty-year 5.62%), but that is noise against a six-week move.

The relief came from the edge, though, not the center. What fell was the risk of a strike before the election, not the amount of debt coming to market. The Congressional Budget Office, Congress's own forecaster, put the deficit for fiscal 2026 at just under $2tn, up 12% on the year, the largest since 2021 and expected to top 6% of everything the economy produces; spending was $7.4tn, revenue $5.4tn, and the Supreme Court's tariff ruling forced the government to pay refunds (Wall Street Journal). Every percentage point on interest rates now compounds through $32tn of debt held by the public. Neither party is campaigning on cutting the deficit; the president is promising $5,000 checks.

So the week ends with the long bonds calmer and the reason for their trouble unchanged. The auctions tell you there is a price at which buyers show up. The deficit tells you the quantity that has to be sold at that price is $2tn a year and rising. For this week that is a point against a crash, and we say so plainly; it is not a point that survives the quantity.

The average rate on a thirty-year fixed mortgage reached 7.4% as of October 8, up 0.12 percentage points on the week and the seventh weekly rise in a row, the longest such run in three years, according to Freddie Mac data reported by the Financial Times. That is the highest since early November 2023. Rates were below 6% in late February, days before the strikes on Iran, and have risen 1.4 points since. The bond market's trouble has become a bigger monthly payment for anyone buying a house.

The connection is direct. A mortgage rate is the interest rate the government pays to borrow for ten years, plus the extra interest investors demand to hold mortgage bonds instead, and both have risen. The second part feeds the first: as rates rise, homeowners stop borrowing again to pay off their old loans, so mortgage bonds stay outstanding longer and lose more value when rates rise, and the people holding them sell government bonds to limit that loss. That is the loop we described last week. The household is at the end of that chain, and now it is paying for it.

The effect on the housing market is a stall rather than a crash. About one-fifth of listings had a price cut last month, the highest share in four years; the number of homes for sale and the number of agreed sales were both soft; sellers are waiting for rates to come back down. "Buyers are pulling back in response to ongoing affordability challenges, while sellers are showing flexibility," said Realtor.com's Hannah Jones. Traders see a 20% chance that the Federal Reserve, America's central bank, raises rates this month and a 76% chance of at least one more rise by the end of the year, so the direction is not changing soon.

The market's view of the lenders is oddly calm. Capital One and Synchrony rose 2 to 2.5% on Thursday, and Affirm is at its high and up 13% in a month. Shares in the companies that lend to consumers are priced for a household that keeps spending; the mortgage market is priced for one that cannot move house. Both can be true for a while. They have rarely stayed true together for long. None of this is a spark. It is weight added to households, slowly, which is the tinder side of the meter.

Britain's government paid more to borrow for thirty years on Wednesday than at any point since January 1998. The rate on its thirty-year bonds jumped to 6.036%, past the record it had set on October 1, and the ten-year rate climbed to 5.48% (up 0.13 and 0.10 percentage points on the day). Rates on US government debt rose too, to levels last seen in 2002 (the thirty-year up 0.07 points, the ten-year 0.06), as oil (Brent) went back above $100 a barrel on fresh attacks in the Strait of Hormuz (Reuters). A government paying 6% for thirty years is not a bad day. It is the market saying it does not want to lend for that long.

The timing could hardly be worse for Britain. John Healey delivers his first budget on October 28. On Tuesday he met the banks that deal in British government bonds to gauge their mood, and the day before that Bank of America's economists had told him his plans mean £15bn more borrowing in each of this year and 2027/28. The thirty-year bond was once the backbone of how Britain borrowed. It is now the one nobody wants.

France is drawing the obvious conclusion. Its finance minister, Roland Lescure, told the Wall Street Journal that France would be "strategic" about how long it borrows for: "the reasonable thing to do... would lead us to have a shorter maturity. But it's at the margin. We're not traders" (WSJ). In other words, France will borrow for shorter periods because that is where the lenders are. Its ten-year rate is 4.91% and its two-year 3.61%, and the gap between them has widened to 1.3 percentage points, which is what it looks like when lenders charge a steep premium for tying money up longer. The average piece of French government debt does not have to be repaid for eight and a half years, against about six for the US. Lescure says a recent ten-year sale went fine, but demand for thirty-year debt is thin.

Here is why that matters. A government that borrows for shorter periods pays a lower rate today and has to come back to the market sooner to borrow again. That is the mismatch we normally hunt for in banks, money owed soon against promises that run long, and it is migrating onto the books of governments. The steady buyers of thirty-year debt, pension funds and foreign governments, have pulled back, and the US has already started buying back its own long-dated bonds. Three of the seven governments in the G7, the club of big rich economies, are now reshaping how they borrow to fit a market that will not lend for long. The market is setting the terms of government finance. That has not been true until now.

This is fragility, not a spark. Nothing fails because a bond pays 6%. What stacks up is the number of governments that will have to borrow again sooner, and the next time one of them finds the buyers absent, that is the stress event. We count it in the pile, and the pile grew a little on Wednesday.

NatWest, a British bank, is giving up its role as one of the firms that must bid at every US and European government bond auction, in the middle of a rout in those bonds. It will cut up to 10 jobs and keep only its business in British government bonds, according to Bloomberg. The bank calls it "simplifying" its markets division, which will now focus on lending, advice and helping clients protect themselves against price moves (Bloomberg). A dealer walking away is not a sign of panic. It is a sign the job stopped paying, and that is its own kind of warning.

The numbers explain it. NatWest's bond-trading business brought in £32m in the second quarter, down about 28% from a year earlier, and the bank blamed choppy markets and a drop in customer trading. The markets unit employs about 1,600 people, roughly 70% fewer than in 2019. Dealing government bonds through a selloff ties up a lot of a bank's capital, and lately it has not made money.

Here is why one mid-sized dealer matters. The firms that hold that role are obliged to bid at every auction and to take the bonds nobody else wants. That obligation is what keeps the US government able to borrow on a bad day. On a good day it is invisible, and Wednesday was a good day. The government sold $39bn of ten-year bonds at a rate a sliver below where they were trading before the sale (0.0188 percentage points), with more bids for every dollar on offer than at any auction since 2016 (2.77 dollars bid per dollar sold). Investors other than the dealers took 97.5% of it, leaving the dealers holding a record-low 2.5% (Bloomberg). "Finally the deeper-pocketed guys find these levels appealing," said Monty Gandhi of SMBC, who thinks the big investors are buying back bonds they had bet against.

So the market did not need the dealers' money this week, which is exactly when a dealer can afford to leave. The question is who holds the bonds in the week it does. NatWest is a small player in US government debt, and its exit adds nothing to the stock of risk. But it is one more piece of evidence for the pattern we keep seeing: regulated banks keep stepping back from standing between buyers and sellers, and what replaces them is hedge funds borrowing heavily to profit from a small gap between two nearly identical prices, which is the money that leaves first when things get rough. That nudges fragility up, by a small amount. Nothing was added to the pile; a little of the cushion under it was taken away.

A listed fund that has bought bank risk-transfer deals for years, ArrowMark Financial, told investors last month that the remaining bonds from a 2021 deal with Bank of Ireland are worth about 34 cents on the dollar, down roughly 43% since the end of last year. The deal was originally $256m and sits on loans made to private equity firms, the buyers of companies with borrowed money, written just before interest rates rose. Separately, First Trust Alternative Opportunities Fund cut its valuation of a 2024 Bank of Ireland deal to 91 cents from over 100 at year-end, according to a filing with the SEC, the agency that polices American markets (Bloomberg). These are the first visible prices on a trade that has moved about $1tn of loan risk off bank books, and the first one says the losses are real.

The trade works like this. A bank keeps a portfolio of loans on its books but pays an outside investor to take on the first slice of losses. The investor puts up cash and collects a fat rate of interest; both these deals recently paid more than 13%. The regulator then treats the loans as protected, so the bank has to hold less capital against them, and the freed capital goes into new lending or buying back the bank's own shares. The trade has a name, significant risk transfer. Bank of Ireland runs a program of them, called Mespil, for loans that finance takeovers. Lenders have used the trade to shed risk on about $1tn of loans.

The design is that the investor eats the losses, and that is what these prices say is happening. Bank of Ireland is not the problem here; the people who sold it the protection are. Bloomberg notes that the structures are so opaque it is hard to know how often this is going on. We can see these two only because one holder is a listed company and the other publishes its holdings every month. Most of this paper sits with hedge funds, private credit funds and insurers that publish nothing. And the filing does not say whether either fund bought its position with borrowed money, which is the number that matters: a 13% return bought on credit becomes a forced sale at 34 cents.

This is the pattern we keep writing about, in a single trade: a loss that would once have shown up as a bank setting money aside is now a fund writing down a holding, in a market where the price is known to the two parties to the deal and the regulator considers the risk gone. It does not make a crash more likely this morning. The loans were bad before ArrowMark said so. What changed is that someone finally put a number on them, and the $1tn channel these losses sit in was already in our count. Our measurement got better; the world did not get worse.

It cost more on Wednesday to insure SpaceX's debt against the company failing to pay than ever before, after the Financial Times reported it wants to borrow $10bn from banks and $30bn from investors in bonds rated safe enough for conservative buyers, all to purchase Nvidia chips. The price of five-year protection rose to 1.94 percentage points a year, meaning $19,400 to insure $1m; when that insurance started trading in June it cost about 1.10 points. The extra interest SpaceX's bond due in 2056 pays over the US government widened to 2.36 percentage points, up 0.09 on the day and from 1.75 in June (FT). This is a company that raised $86bn selling shares to the public and $25bn from bond investors four months ago. Morgan Stanley's Adam Jonas asked a room of 40 clients who owned the shares, and no hand went up.

The queue behind SpaceX is not shortening. The Wall Street Journal reports Broadcom is arranging more than $50bn for OpenAI's own chip and has approached Apollo and Blackstone among others to lend it, while Oracle is talking to Apollo and Goldman about paying for its own chips (WSJ). Apollo, which is in all three conversations, is at bottom an insurer's pool of savers' money run by an investment firm.

Selling shares is getting harder. Firmus Grid, an Australian builder of data centers backed by Nvidia, stopped taking orders on Thursday for a $5.5bn sale of shares to the public at a $30bn valuation and has not said what the price is. Days earlier it had reported interest well above what it was offering; now the A$11 price has failed to attract enough support and the deal may be cut back or abandoned, according to Bloomberg. Firmus runs two data centers. The valuation rests on a pipeline of planned ones across Asia serving Meta and OpenAI, and about 58% of the shares would be free to trade on the first day (Bloomberg).

Put the three together. Lenders are charging more to a chip buyer they already lent to; share buyers are declining to fund a chip buyer that has not built anything yet; and the deals keep coming, more and more through private lenders rather than public markets. Buyers are pushing back at the edges, in prices. The supply is not responding. That makes a crash more likely, and it is the slow kind: every deal that closes on worse terms adds borrowed money to the pile, and a pile that keeps growing while its buyers get choosier is what turns a small fire into a large one.

India's central bank has spent $51bn of its foreign currency reserves in four weeks defending the rupee, and the currency is still just 0.2% from its record low (96.97 to the dollar). Reserves fell to $734bn on Friday from $785bn in early September, and a special scheme that pulled in $133bn of dollar deposits has already handed more than a third of it back. The first rise in interest rates in nearly four years did little. "At present, the central bank's intervention is the only factor that is keeping the rupee where it is," Ritesh Bhansali of Mecklai Financial told Bloomberg. Foreign investors have sold almost $30bn of Indian shares this year (Bloomberg). Spending roughly a fifteenth of your reserves in a month is not routine. It is what a central bank does when nothing else is working.

In the Gulf, Bahrain's dollar bonds due in 2028 now pay 8.99%, the most in six years and up 1.6 percentage points since late August, when the selloff in US government bonds began. Its dollar debt has lost 4.2% in that time and its sukuk, bonds structured to comply with Islamic law, 3.5%, and the shorter-dated debt has moved most. Bahrain runs a deficit of 10.5% of its economy, owes more than 140% of it, and is rated too shaky to be safe by all three of the firms that grade borrowers. It was last at these levels in 2018, when Saudi Arabia, the UAE and Kuwait put up more than $10bn. Fady Gendy of Arqaam expects Gulf support to "become more explicit and broad-based on any large and sustained market weakness" (Bloomberg).

These are opposite oil stories, India imports crude and Bahrain exports it with the war having cut its output, and they share one cause. The 5.3% the US government now pays to borrow is the floor under what every borrower of dollars pays, and a dollar near an 18-month high is the exchange rate that bill arrives in. When the US raises the price of dollars, the weakest borrowers of dollars feel it first, whatever their local economy is doing. Bahrain has rich neighbors. India has reserves, enough to pay for eleven months of imports, according to Governor Malhotra. The next borrower down the list may have neither.

This moves the odds of a crash up, and on the ignition side. Nothing new was borrowed this week. What changed is that strain which had been a rich-country problem is now draining reserves in Asia and pushing a Gulf government back toward the rates at which it last needed rescuing. A spark is more likely to come from the edge than the center, and the edge just got closer.

The minutes of the September meeting of the Federal Reserve, America's central bank, which sets interest rates, were released on Wednesday. They show all 19 members of its rate-setting committee backed the quarter-point rise to 3.75–4%, the first increase since July 2023, and that "most participants assessed that another increase in the target range for the federal funds rate would likely be appropriate by year end." Several said the economy had gained momentum. And many said that despite the rise in long-term borrowing rates, "financial conditions appeared to be supportive of economic growth, with equity prices having risen substantially this year and spreads on corporate bonds having remained narrow" (Bloomberg). A unanimous vote to raise rates is not unusual. A unanimous vote to raise them, with the government paying 5.3% to borrow for ten years, on the grounds that the stock market is at a record, is.

Read that last quotation carefully. In other words, the Fed can keep tightening because shares are up and companies are paying only a little more than the government to borrow. Those small gaps are the same ones we read as the market not yet pricing what is coming down the pipeline. Both readings can be true; they just end differently. The Fed is raising rates into a 5.3% ten-year because the stock market is at a record, which means the stock market is now part of how the Fed decides what to do, in a way it was not when rates were being cut.

The same minutes cleared up a smaller mystery. The New York Fed's part in July's joint intervention to prop up the yen, the first in nearly three decades, was done "acting purely as fiscal agent for the US Treasury," with the Treasury Department's money; the Fed's own holdings were not touched. The timing and size were not disclosed; Scott Bessent has called the US contribution "nominal". Japan alone spent a record ¥15.4tn ($97.5bn) in the month to August 26 (Bloomberg).

So the picture is this. The Fed is raising rates to lean on inflation, which strengthens the dollar. The Treasury Department, which borrows the money the government spends, is spending to weaken the dollar against the yen and buying back its own long-dated bonds to lean on borrowing rates. Two arms of American policy pushing in opposite directions, each doing a little. Strategists quoted by Bloomberg still see the yen heading back toward 160.

This leaves the odds higher, not lower. The Fed is reading the things we worry about, record share prices and cheap corporate borrowing, as permission to keep tightening, which means the one institution able to cut rates fast when something breaks has told us it will not move until something does. Nothing was added to the pile today. The hand that could take things off it has said it is busy.

Loans to American companies that trade below 60 cents on the dollar, meaning the market thinks a large part of the money is never coming back, now add up to $65bn. A year ago it was $40bn. That is the most since the first weeks of the pandemic (the highest since March 2020, according to JPMorgan strategists writing on Tuesday, via Bloomberg). The wider bucket of troubled loans, priced at or below 80 cents, is $139.8bn, up almost 90% in twelve months and $4bn short of the May 2020 peak. There are 141 borrowers in it, 35 more than a year ago.

Why loans rather than bonds, and why now. These are leveraged loans: company debt where the interest rate floats with the market instead of being fixed on the day it was signed. When the Federal Reserve, America's central bank, raised rates in September and the benchmark these loans track (SOFR) climbed, every one of these borrowers saw its interest bill rise within weeks. A company that sold a fixed-rate bond can sit tight until the bond comes due; a loan borrower cannot. The rate shock we have been tracking in US government bonds has arrived at the company level with no delay.

The second thing is where the trouble sits. Technology is 39% of the distressed pile ($54.4bn), with software names CDK Global, Qlik and Quest Software among the largest contributors, and more than $100bn of software debt needing to be borrowed again to pay off what is coming due. These are the same borrowers that fill the portfolios of investment funds that lend to companies instead of banks, where the loans are not traded and the lender decides what they are worth. The open market has now put a price on them. JPMorgan expects 4.5% of loan borrowers to fail to pay next year, double its 2.25% projection for this one.

The funds you can buy on the stock exchange say the same thing. Prospect Capital, a fund that borrows money, lends it to mid-sized private companies, and passes the interest to shareholders, is down 12.6% in five days; Blue Owl Capital Corp is down 4.4%; FS KKR 2.7%. Meanwhile the extra interest shaky companies pay on their bonds compared with the government is unchanged on the week (312bp, just over three percentage points). Bond investors are relaxed; the people who hold the loans are not. Put this number beside the one in the next story, $40bn being arranged so a single company can buy chips, and you have the day in two figures.