Bad loans in private credit jumped 40% in three months

Archived reading, published Tue, 18 Aug 2026 10:49:51 UTC (8 days ago). This is not the current state of the meter.

See the live reading →

CRASH-O-METER

Cracking

how close are we

+3 since the last reading

Fragility80

how much tinder is stacked up — moves slowly

Ignition40

how close a spark is — moves fast

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

Every four hours it reads the day’s reporting from 19 sources and rewrites this page.

Why it moved: Up three: the median non-accrual rate at the twenty largest listed BDCs jumped from 2.0% to 2.8% in a single quarter and Fitch logged a record month of private credit defaults in July — the first hard, measured deterioration rather than more tinder. Partly offset by high-yield spreads still tightening to 267bp and a market that has cut the odds of a September Fed hike from 75% to one-in-three.

Crash points

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

What’s moving the needle

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

Private credit and BDCshigh

Non-accruals went from 2% to 2.8% in one quarter

Non-accruals are the one number in private credit that isn't a matter of opinion, and they moved sharply in a single quarter.

Loans placed on non-accrual status by the twenty largest publicly traded business development companies climbed to a median 2.8% of cost in the second quarter, up from 2% at the end of March. That is the highest since 2017, when the industry was digesting the oil crash. Fitch says private credit defaults hit a record in July. (FT) What non-accrual means matters here. A BDC lends to a mid-sized company at a floating rate and books the interest as income whether or not cash arrives. Putting a loan on non-accrual is the fund conceding that the interest is not coming — either the borrower has stopped paying or default looks imminent. It is one of the very few things in private credit that is not a matter of judgement. The valuation of the loan itself is a judgement. The non-accrual flag is close to a fact. So a 40% relative jump in one quarter is the clearest signal yet that the credit cycle the industry spent two years denying has arrived. Golub Capital's David Golub said as much to investors: "We're in a credit cycle. Others denied it for a while. I don't think there's a lot of denial any more." Two further details. FS KKR reported 7.1% of its loan book troubled — far above the median. And funds run by KKR, Blue Owl and Apollo's MidCap Financial shrank in the quarter, with repayments outrunning new lending. A shrinking loan book is a lender choosing not to lend. The important caveat: these are the listed vehicles, which must disclose. The much larger non-traded and evergreen funds report on their own schedule, to their own marks.

The AI capex bubblehigh

Nvidia agrees to stand behind OpenAI's rent

The AI buildout is being financed by a chipmaker's credit rather than its customers' cash flows, and the rules requiring lenders to retain risk have just been waived for it.

Nvidia has agreed to guarantee up to $105bn of OpenAI's lease payments on a data-centre campus in Pike County, Ohio, being built by SoftBank-backed SB Energy. It will also invest $1.5bn in SB Energy. OpenAI is contracting for roughly 8 gigawatts of capacity — a gigawatt powers as many as 750,000 US homes — with the first 800MW due in 2028. Nvidia's backing phases in from 2028 to 2030 and runs for twenty years. The $105bn covers only the first 4.25GW; it may take on the rest. (Bloomberg) The plumbing: SB Energy needs to borrow to build. Lenders will lend against the lease. The lease is only as good as OpenAI, a company that does not make money. So Nvidia — which will sell the chips that go inside — puts its own balance sheet behind the rent, and the lenders are effectively lending to Nvidia. The same purchases could be worth about $600bn of Nvidia revenue through 2030. Jensen Huang says this is not circular financing. Read that literally: Nvidia is not handing OpenAI cash to buy chips. It is doing something subtler — converting its equity-market credibility into third-party debt capacity for its own customer. This is the week's second such structure. Nvidia has also signed MOUs with Apollo, BlackRock, Blackstone, Brookfield, Goldman and KKR to mobilise over $500bn of compute financing. And last month SEC staff agreed with Latham & Watkins that some data-centre securitisations are not asset-backed securities, and so escape Dodd-Frank's requirement that sponsors keep skin in the game. (CNBC) It is a staff opinion, not a rule. It is enough.

Hidden leverage and shadow bankingmedium

The debt that isn't on the balance sheet is bigger

The most heavily analysed companies on earth have more debt in their footnotes than on their balance sheets, and the footnotes are where risk goes to be unmeasured.

Nikkei went through the filings of Alphabet, Microsoft, Amazon, Meta and Oracle and totted up the obligations that sit in the footnotes rather than the balance sheet. The figure is $1.65tn — more than the $1.35tn of debt these companies actually report. Four years ago it was roughly an eighth of that. Meta's off-balance-sheet total is about $420bn, nearly triple its recorded debt. Oracle's is $273.3bn as of end-May, a thirty-fold rise in four years. (Nikkei) None of this is improper. Accounting rules say that a lease on a data centre that is not yet operating, and a purchase commitment for GPUs and servers not yet delivered, are disclosed in the notes and not capitalised on the balance sheet. The obligation lands the day the lease commences. Until then it is real money owed and invisible to anyone reading the summary statements — which is most people. The reason it has exploded is structural. Hyperscalers keep initial cost down by having an operator supply the land, the building and the power, and then leasing it for twenty years. Meta's $27bn Hyperion financing reportedly sat in an SPV, backed by that twenty-year lease promise, and so did not appear on Meta's balance sheet. The counterweight, which is genuine: this group will earn around $1tn in operating profit this year. Bloomberg's tally puts total commitments through debt and contracts as high as $4.2tn, against $1.6tn of reported liabilities as of June — nearly double two years ago. (Bloomberg) Solvent companies with enormous cash flows and hidden fixed obligations are exactly how a corporate credit cycle starts quietly.

Bond market dysfunctionhigh

Thirty-year yields at 5.31% and the AI bid for capital

The cost of long money is being set jointly by the US deficit and five companies' capex plans, and neither is slowing.

The 30-year Treasury yield hit 5.31% on Monday, the highest since 2007. Last Thursday's $25bn auction of the same bond cleared at 5.22%, the highest auction yield since 2001; Wednesday's 10-year cleared at 4.683%, the highest since 2007. (Bloomberg, FT) The usual explanation — near-$2tn deficits, inflation above target for five years — is most of it. The new part is on the other side of the ledger. US investment-grade issuance hit $145.2bn in August, a record for the month with two weeks to run, and $1.46tn for the year, running 8.5% ahead of the 2020 record pace. Nomura reckons the roughly $200bn borrowed by the biggest tech companies alone equals about a quarter of the Treasury's net note and bond issuance to private investors — five times the 2025 share. That is crowding-out running backwards. The classic theory says government borrowing squeezes companies out. Here a handful of AI borrowers are competing with the sovereign for the same finite pool of duration demand, and everyone's yield goes up. "Whoever's issuing... is now competing with more borrowers," as Nuveen's Tony Rodriguez puts it. (Bloomberg) It is now visible abroad too. Alphabet raised C$8.5bn in Canada, then Amazon raised C$14bn. Canada's AA corporate index — the top rung, where both sit — has widened past the lower-quality A index. "They basically come in and reprice all of the high-quality curve," says AllianceBernstein's Souheir Asba. Alphabet is now marketing up to A$5bn in Australia, after a 100-year sterling bond in March. (FT) Small credit markets are being repriced by borrowers who do not live in them.

The AI capex bubblehigh

A $14bn data centre with $450m of property cover

When the insurance market won't take a risk at any sensible price, the risk doesn't disappear — it ends up with the lenders, unpriced.

Meta and BlackRock's Sopaipilla campus in El Paso costs $14bn and will draw one gigawatt. On the advice of Marsh, the world's largest insurance broker, it has bought all-risk property cover of up to $427mn during construction and $450mn once operating, for a premium of about $5mn a year. Terrorism cover is capped at $645mn, delay-of-completion cover at $218mn, and general liability at $50mn per event and $50mn in aggregate. The project is not insured against total loss. (FT) Run the ratio. Roughly 3% of the build cost is covered against property damage. Everything above that sits with the venture and, if the venture cannot pay, with its lenders. The reason is not carelessness. It is capacity. Insurers will write data-centre risk, but few want a concentrated exposure to a single site worth tens of billions, especially one packed with chips whose replacement cost and lead time nobody can pin down. Full cover has become, in the FT's phrase, prohibitively expensive. For smaller data centres lenders demand it. At this size they have learned to accept a fraction. This is worth noticing because insurance is where a hazard gets priced by somebody with money at stake. When the insurance market declines to write the risk, the risk does not vanish; it migrates to whoever holds the debt. Those holders are increasingly bond funds, private credit vehicles and insurance company general accounts — the same balance sheets absorbing the data-centre ABS the SEC has just exempted from risk retention. We do not know what covenants the lenders extracted in exchange, and that is the part that would tell you how bad this is.

The dollar, gold and reserve statusmedium

Gold up 9% in a month while nothing is wrong

Gold rising 9% in a month with equity vol at 16 means someone is buying it for a reason that has nothing to do with the business cycle.

Gold is at $4,448, up 1.5% on the week and 9.3% over twenty days. Over the same period the dollar index fell 1.35% to 99.63 and the Bloomberg Dollar Spot Index hit its weakest since 15 May. Traders have cut the odds of a Fed hike next month to one-in-three from about 75% in late July, after soft employment, inflation and retail sales data. Only about 36bp of further tightening is priced for the cycle. (Bloomberg) A weaker dollar and lower rate expectations explain part of a gold rally. They do not comfortably explain 9% in a month while the S&P sits 0.7% off its high and VIX is at 16. Something else is in the bid. The candidate is official demand. An OMFIF survey of reserve managers reports 74% expect the dollar's share of global reserves to fall over five years and 89% expect central bank gold holdings to rise over the next year — described as the first such survey in which more managers plan to cut dollar exposure than add to it. That is a survey of intentions, not flows, and should be read as such. The texture item is better than the survey. Venezuela's government and its opposition — who agree on nothing — have jointly demanded that the Bank of England return $4bn of Venezuelan gold. (FT) The reserve-status question is not being settled by anyone selling Treasuries. It is being settled by a slow preference for the one asset that is nobody's liability, and by sovereigns deciding they would rather have the metal at home.

Signs of the times

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

Crypto perpetual futures on Hyperliquid price Chinese humanoid-robot maker Unitree at $40.5bn, four and a half times the $9bn valuation of the IPO it prices at on Wednesday — an offering that was 5,526 times subscribed. An unregulated offshore derivative is now the price-discovery venue for a Shanghai listing that has not happened yet.

Bloomberg

Jane Street lost $15bn in July — its first losing month in a decade — on directional bets, and is still on course for a record year, having already booked over $40bn of net trading revenue. In its latest debt offering the firm described its strategy as evolving to include longer-term positions, which is a market-maker's way of saying it is a hedge fund.

Bloomberg

Anthropic's backers expect it to float in October at $2tn or more, which would be the largest IPO ever and eclipse SpaceX. One investor's arithmetic: "If Anthropic is growing 800 per cent a year, you'd think at the incredibly low end they would trade at 30 times revenue. That would make them a $3tn company." The company is five years old.

Financial Times

Taiwan's economy grew 12.9% in the second quarter on AI chip demand, filling government coffers so effectively that President Lai will hand every eligible resident NT$10,000 ($314) next year. The stock market is up almost 60% in 2026 and many of the island's 23 million people report feeling left out of it anyway.

Bloomberg

An NBER survey found bond investors, ordinary voters and finance graduates all put the average probability of a US debt crisis within ten years at close to 50%. Among concerned investors, 72% had made no change to their portfolios.

Financial Times

The rumour mill

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

Confirmed

“Tokyo began intervening to prop up the yen on 30 July and the US joined the next day — the first American participation in a yen-support operation since 1998.”

Confirmed by Japan's finance ministry and reported by Reuters and the WSJ; Goldman estimates roughly $85bn of intervention over 30–31 July. Nikkei reports Japan can now draw dollars from the Fed against its $1tn of Treasuries rather than selling them, which is why the intervention has not shown up as Treasury supply.

Claimed by Mark Moss

Partly true

“Alphabet spent $44.9bn on capex last quarter and posted negative $5.9bn of free cash flow — its first negative free cash flow quarter as a public company.”

The Q2 2026 numbers check out and multiple outlets describe it as the first negative quarter since the 2004 IPO. Worth holding next to the forecasts that aggregate hyperscaler free cash flow turns negative around 2026–27; those are consensus estimates, not facts.

Claimed by Coin Bureau

Partly true

“Nvidia has lined up more than $500bn of AI-buildout financing from Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, and lost about $130bn of market value on the announcement.”

The six firms and the $500bn figure are from Nvidia's own release, but they are memorandums of understanding to mobilise capital over time, not committed financing. The share fall is reported; the causal link to the announcement is not established.

Claimed by Coin Bureau

Confirmed

“Senator Rand Paul went underground at Fort Knox this month and confirmed roughly 147 million ounces of gold are there.”

He visited on 10 August and posted that the gold is there, all approximately 147 million ounces; the US Mint's figure is 147.3 million troy ounces. Filed under things that would only be newsworthy in this particular year.

Claimed by ITM Trading (Daniela Cambone)

Earlier developments

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

The AI capex bubblehigh

An investment-grade bond that pays like junk

When bond buyers price an investment-grade rating like a single-B credit, they are telling you the rating does not capture the risk in the structure.

A bond sale nicknamed "Project Odyssey" is being upsized to about $3.9bn after drawing more than $8bn of orders. The borrower is not Microsoft. It is two special-purpose subsidiaries of QualityTech LP, part of Blackstone-backed QTS Realty, raising five-year money for a data-centre venture in Georgia that Microsoft will use. The paper is expected to be rated investment grade. It is being marketed with a coupon around high-6% and a discount that lifts the yield to low-to-mid 7% — roughly where single-B junk bonds trade (Bloomberg). That gap between the rating and the price is the whole story. The rating agencies are looking at a long lease from one of the most creditworthy companies on earth. Buyers are looking at everything the lease does not cover: whether the building is worth anything if the tenant walks at expiry, what the kit inside is worth in year six, and how many more of these deals are coming. Investors who "initially rushed to snap up the offerings" are now showing exhaustion, which is why the deal needed a discount to clear at all. The structural point matters more than the pricing. The debt sits in an SPV, so it is not on Microsoft's balance sheet; Microsoft's obligation is a lease, which shows up in footnotes until the term begins. Meta did the same thing with the $27bn Hyperion financing. Analysts quoted by the Telegraph put big tech's off-balance-sheet commitments near $1tn; other tallies reach $3tn. We cannot verify those aggregates. What we can verify is that the market is charging junk prices for AAA-tenant risk, and getting paid to do it.

Bond market dysfunctionhigh

Japan's insurers hold $194bn of paper losses

Japan's life insurers are the marginal global buyer of very long bonds; if rates force them to shorten duration, the long end everywhere loses its anchor.

Unrealised losses on domestic bonds at Japan's major life insurers reached ¥30.86tn ($194bn) at the end of June, up 60% year on year. For the first time in this cycle they exceed the same firms' unrealised gains on Japanese stocks, which rose 48% to ¥30.03tn (Nikkei). The cause is arithmetic: 30-year JGB yields were in the 3.9% range at end-June, about 2.7 percentage points higher than in mid-2023, and bond prices fall when yields rise. The official answer is that unrealised losses vanish if you hold to maturity. Two things break that. First, impairment: if a bond's market value falls 50% below its purchase cost, insurers may have to write it down whether they sell or not. Nippon Life booked ¥44bn of impairments in April–June, Meiji Yasuda ¥25.3bn — small numbers, but they exist because bonds bought in the late-2010s zero-rate era have fallen that far. Second, duration mismatch. Insurers match asset duration to liability duration; where assets are longer than liabilities, rising rates cut net assets, and the fix is to shorten duration by selling bonds. That converts paper losses into real ones. There is also a liability-side leak. Sony Life's lapse and surrender rate rose 0.2 points to 1.4% in April–June, partly because a weak yen prompted customers to cancel foreign-currency policies. Surrenders force sales at the worst moment. This is the SVB problem relocated to insurance, where marks are softer and the disclosure is annual rather than daily. Japanese life insurers are among the world's largest buyers of long-dated bonds. If they become sellers, the buyer of last resort for global duration goes missing — with US 30-year yields already at 5.31%, the highest since 2007.

Private credit and BDCsmedium

Private credit funds queue up in the public bond market

Private credit's safety claim rests on being unlevered and unrunnable; funding itself in the public bond market chips away at both.

August set a record for US investment-grade issuance: $145.2bn, beating the $136bn sold in August 2020, making it the third consecutive record month. Year-to-date supply is $1.46tn, 8.5% ahead of 2020's record pace (Bloomberg). Buried in Monday's twelve deals were two from private credit funds — vehicles run by Blackstone and Blue Owl selling high-grade bonds. That detail is worth pausing on. The pitch for private credit has always been that it is patient, unlevered capital: money from insurers, pension funds and wealthy individuals, locked up, holding loans to maturity and therefore immune to runs. Funds that borrow in the public bond market are not quite that. They have bondholders, covenants and leverage ratios calculated off the value of loans that nobody marks daily. The timing is awkward. Median non-accruals at the twenty largest listed BDCs rose to 2.8% of cost in the second quarter from 2% in March, the worst since 2017; FS KKR reported 7.1% of its book troubled; Fitch says private credit defaults hit a record in July; and the biggest listed BDCs shrank again as repayments outran new lending (FT). Golub Capital's David Golub: "We're in a credit cycle. Others denied it for a while. I don't think there's a lot of denial any more." A shrinking loan book funded partly with public debt is a squeeze in slow motion. Loan yields are fixed by old deals; new borrowing costs are set by today's crowded market. Nothing about that is a run. It is margin compression that shows up before losses do.

Hidden leverage and shadow bankingmedium

Nobody in the survey mentioned the basis trade

The Treasury market's biggest hidden buyer is leveraged and funded overnight, which makes a funding squeeze, not a fiscal event, the likelier trigger for disorder.

Economists at USC surveyed bond investors, ordinary voters and people with economics or finance degrees about the US debt load. The average stated probability of a US debt crisis within ten years was near 50% in all three groups. Asked what actually supports the Treasury market, FT Alphaville noted that nobody named the one answer a plumbing specialist would give: demand for synthetic duration, otherwise known as the basis trade (FT). Here is what that means. A hedge fund buys a cash Treasury and simultaneously sells the corresponding futures contract. The gap between the two is tiny — basis points — so the fund finances the bond in the repo market, borrowing near the overnight rate (SOFR is 3.62%) and repeating the trade many times over. Done at enough leverage, a few basis points become a real return. Done at scale across the market, these funds become one of the largest marginal buyers of Treasuries, which is how a market absorbing nearly $2tn of annual deficits keeps clearing. The trade's dependency is not the direction of yields; it is the stability of repo funding and margin. Raise haircuts, and the position must shrink; shrink it in a hurry and the buyer becomes a seller into a falling market. Bloomberg reports traders are now hunting a related wrinkle: the "switch option", where a sharp yield move changes which bond is cheapest to deliver into the futures contract, handing shorts a profit. With long yields above 5% for the longest stretch since 2007, that option is live — and the same volatility that makes it valuable is what threatens to disrupt the futures market it depends on.

The dollar, gold and reserve statusmedium

Japan can intervene without selling Treasuries

The plumbing that lets Japan defend the yen without dumping Treasuries is now official policy, which removes one forced-seller channel and adds a political one.

Tokyo has a new tool. A COVID-era repo facility has been repurposed so the Bank of Japan can borrow dollars from the Federal Reserve using its Treasury holdings — worth more than $1tn — as collateral. US Treasury Secretary Scott Bessent announced the arrangement on 3 August and President Trump publicly backed it. It follows the first joint US-Japan yen intervention in 28 years, on 31 July (Nikkei). The mechanism matters more than the diplomacy. To buy yen you need dollars. Historically Japan raised them by selling Treasuries, which pushed US yields up at exactly the moment Washington least wanted it — currency defence and bond-market stability worked against each other. Repoing the Treasuries instead of selling them severs that link: Japan gets dollars, the bonds stay in the portfolio, and the US long end is spared the supply. Japan's top currency diplomat, Atsushi Mimura, called it "a US-Japan currency union". Analysts in Tokyo reach for "a new Plaza Accord". The yen has since given back about half its post-intervention gains, and officials are seen as ready to act well before 162.80, the level that triggered the last round. Futures price a 100% chance of a BoJ hike to 1.25% by the end-October meeting. Read this as risk being moved, not removed. Japan avoids realising losses on its Treasuries; the Fed takes collateral risk; the pressure returns to the JGB market, where 10-year yields are the highest since 1996. Meanwhile the dollar is at a three-month low as traders cut September Fed hike odds to one-in-three from about 75% in late July, and gold is up 9.5% in twenty days at $4,459.

Household creditmedium

Ten straight quarters worse than 2008

Household stress is concentrated exactly where there is no collateral, and it is now big enough to bend the spending data and the Fed's path.

The New York Fed's second-quarter household data has a number that deserves more attention than it got. The share of credit-card balances flowing into 90-day-plus delinquency was 6.97% — the tenth consecutive quarter above the roughly 6.75% level that marked the onset of the Great Recession. Card balances stand at $1.263tn, up $21bn on the quarter and $54bn on the year, and 12.92% of outstanding card balances are 90 days or more delinquent (NY Fed). Serious auto delinquency is the highest since 2010, on a $1.71tn book that grew $28bn in the quarter. And yet the headline household delinquency rate fell, to 4.7% from 4.8%, and student-loan delinquency improved sharply — 7.83% against 12.88% a year earlier. Both things are true. The aggregate is stable because most household debt is a mortgage held by someone with a fixed rate and equity. The distress is concentrated in unsecured revolving credit and subprime autos, where the borrower has no cushion and the lender has no collateral worth repossessing. That split is now showing up in spending. July retail sales fell 0.6% against expectations of a small rise — the first drop in nine months and the largest since May 2025 (we checked this claim; see the rumour mill). Sales excluding autos fell 0.3%. The market read that as dovish: odds of a September Fed hike collapsed to one-in-three. Note what that implies. Consumer weakness is currently being priced as good news for asset prices. It is only good news if it stops there.

Crypto and TradFi contagionhigh

Strategy is running the flywheel backwards

The template dozens of crypto treasury companies copied only works while the share premium exists, and the original is now demonstrating what happens after it goes.

Strategy sold $333.7m of common stock last week. It bought no bitcoin. It has bought none since mid-June. The proceeds went to repurchasing preferred shares and building a cash reserve, now $4.8bn. Over five weeks the company has sold roughly $2.1bn of common, bought back about $347m of its STRC preferred, and sold $213.3m of bitcoin. It still owns around $58bn of the coin. The shares are down about 73% over the past year (Bloomberg). The original machine worked like this: issue shares above the value of the bitcoin behind them, buy more bitcoin, report a higher figure per share, watch the premium justify itself. It requires the premium. Once the stock trades at or below the value of the coins, issuing equity to buy bitcoin destroys value rather than creating it. What is left is the liability side. The preferred shares carry fixed dollar dividends that must be paid in dollars. Bitcoin generates no cash. So the company is now doing what any leveraged holder does when the cheap financing dries up: selling equity and, at the margin, selling the asset to retire the most expensive claim on it. Our fact-check of one channel's account found Strategy sold 1,690 bitcoin at an average $64,262 and used the $108.6m to buy back 1,152,020 STRC shares. This is an orderly deleveraging, not a fire sale — $213m against $58bn is a rounding error, and bitcoin at $64,194 is flat over twenty days. But the direction has reversed. The largest corporate buyer of bitcoin is now a small, steady seller of it.

The AI capex bubblehigh

Nvidia has agreed to guarantee OpenAI's rent

Vendor guarantees turn a customer's credit risk into a supplier's contingent liability, and contingent liabilities are the part of the balance sheet nobody prices until they are called.

Nvidia has agreed to guarantee up to $105bn of OpenAI's lease payments on a data-centre campus in Pike County, Ohio, being built by SB Energy, and to invest a further $1.5bn in SB Energy itself, which is backed by SoftBank (Bloomberg). The campus is meant to reach roughly 8 gigawatts. Nvidia's backing covers the first 4.25GW phase, begins in stages between 2028 and 2030, and runs for twenty years. Here is the mechanism. A developer cannot borrow cheaply against a building whose only tenant is a five-year-old company with enormous losses. It can borrow cheaply if someone with a trillion-dollar balance sheet promises to make the rent payments if the tenant does not. That promise is what converts an OpenAI credit risk into something close to an Nvidia credit risk — and Nvidia sells the chips that fill the building. Nvidia says it is not guaranteeing the whole site or all of OpenAI's obligations, and Jensen Huang says this is not circular financing. What we do not know is the accounting. A guarantee of this kind may sit in a footnote as a contingent obligation rather than as debt. Bloomberg Opinion counts $1.6tn of liabilities across Nvidia and the hyperscalers as of June, nearly double two years ago, against roughly $1tn of operating earnings this year — and notes that many AI commitments do not appear in financial statements at all (Bloomberg). The thing that has to stay true is OpenAI's revenue growth for twenty years. If it does not, the loss does not vanish. It moves to Nvidia, on a schedule nobody currently marks.

Fed, Treasury and policymedium

The SEC quietly waived a Dodd-Frank rule for data centres

The single most important post-2008 constraint on securitisation has just been declared inapplicable to the fastest-growing category of securitised asset.

Last month the SEC agreed with the law firm Latham & Watkins that some data-centre debt is not an asset-backed security, and therefore is not subject to the risk-retention rules written into Dodd-Frank after 2008 (CNBC). Risk retention is the "skin in the game" rule. If you package loans and sell them to investors, you have to keep a slice — usually 5% — so that you care whether the loans are any good. It exists because in 2005 nobody who originated a subprime mortgage kept any of it. "The SEC basically agreed," Seth Messner of Katten told CNBC. Another securitisation lawyer, Orion Mountainspring of Orrick, was blunter about what it does: "It gives them the opportunity over time to push down the required equity in the deal." Less required equity means more leverage per dollar of building. The timing is not coincidental: Nvidia has signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilise more than $500bn of third-party capital for AI infrastructure. It is not confirmed that those platforms will use securitisation — but the lawyers CNBC spoke to think the guidance reads as if written for exactly this. Two caveats worth keeping. This is a staff opinion, not a rule, and staff opinions can be withdrawn. And a data centre is a real asset with a real tenant, which a 2006 mortgage pool often was not. The parallel is not the collateral. The parallel is the mechanism: a structure whose sponsor no longer has to hold the first loss.

Private credit and BDCshigh

Non-accruals at 2.8%, and a $367bn firm under subpoena

Private credit's defining trick — funding illiquid loans with insurance money — is now being examined by prosecutors at the firm that invented it.

Loans placed on non-accrual by the twenty largest listed business development companies rose to a median 2.8% of cost in the second quarter, up from 2.0% at the end of March — the highest in nearly a decade, on FT analysis of data from Solve (FT). A BDC is a listed fund that makes loans to mid-sized private companies. "Non-accrual" means the fund has stopped booking interest income from a borrower, because the borrower has stopped paying or is expected to. It is the closest thing private credit has to an honest number: unlike the mark on the loan, which the manager estimates, non-accrual status is a binary admission. FS KKR reported 7.1% of its loan book troubled. Fitch says private credit defaults hit a record in July. David Golub of Golub Capital: "We're in a credit cycle. Others denied it for a while. I don't think there's a lot of denial any more." The second story is uglier. Prosecutors and regulators are probing the empire of Mark Walter, chief executive of Guggenheim Partners' parent, over insurers he controlled lending to other parts of his own empire; after subpoenas, more than $20bn of loans were disclosed that had not been properly labelled as affiliated (FT). Guggenheim's financing entity, GIH Borrower, disclosed a 77% earnings fall on Friday. Its first-lien term loan traded below 80 cents on Monday, from 100 a month ago (Bloomberg). Walter pioneered tying insurance balance sheets to alternative assets. The whole industry copied it. That is why a twenty-point move in one loan is being read across the sector.

Hidden leverage and shadow bankingmedium

Jane Street lost a hedge fund's worth of money in a month

The firms that provide liquidity in a stress event are now taking proprietary risk on the scale of a large hedge fund, with none of the disclosure.

Jane Street took a $15bn hit in July, its first monthly loss in a decade, on directional bets that Bloomberg says go well beyond anything a market-maker does (Bloomberg). The FT separately puts a $35bn July trading loss at the door of Leopold Aschenbrenner's leveraged AI stock bets, and ranks it top of its all-time leaderboard of trading losses (FT). We cannot reconcile the two figures from what has been published, and we are not going to pretend we can. Either number is extraordinary. The structural point survives the arithmetic. Jane Street has described itself as a liquidity provider — the firm standing between buyers and sellers, earning a spread, holding inventory briefly. In its latest debt offering it described its strategy as evolving to include longer-term bets, in the manner of a hedge fund. It has still generated more than $40bn of net trading revenue this year, ahead of 2025. Why this matters for everyone else: market-makers are load-bearing. When an ETF trades, when a Treasury basis position is unwound, when options expire, firms like this are on the other side. A bank running this kind of directional risk would face capital charges and supervisory attention. A private trading firm faces neither. It funds itself in the bond market. The loss was absorbed. That is the reassuring part, and it is real. The unreassuring part is that nobody outside the firm knew the position existed until after it lost.

Bond market dysfunctionhigh

The long end is buckling and AI is part of why

When investment-grade paper has to pay junk yields to clear, the market is telling you the rating is not the risk.

The 30-year Treasury yield hit 5.31% on Monday, the highest since 2007 (Bloomberg). Last week's $25bn 30-year auction cleared at 5.22%, the highest since 2001 (FT). Wednesday brings $16bn of 20-years indicated near 5.27%, which would be the highest since the tenor was reintroduced in 2020. The usual explanation is fiscal: a near-$2tn annual deficit, debt approaching $40tn, inflation above target for five years. All true. The newer part is that corporate America is now competing with the Treasury for the same buyers. Investment-grade issuance hit $145.2bn in August, a record for the month, and $1.46tn year to date (Bloomberg). Nomura estimates the roughly $200bn borrowed by the biggest tech companies alone equals about 25% of Treasury's net issuance of notes and bonds to private investors — five times the 2025 share. "Whoever's issuing, be it a government or a hyperscaler," Nuveen's Tony Rodriguez told Bloomberg, "is now competing with more borrowers. And therefore yields have to be higher." Buyers are getting choosier, and the pricing shows it. A Blackstone-backed QTS bond funding a Microsoft-linked Georgia venture — expected to be rated investment grade — is being marketed at a yield of low-to-mid 7%, where single-B junk trades (Bloomberg). Demand topped $8bn on a $3.9bn deal, so the money is there. It just costs what junk costs.

The dollar, gold and reserve statusmedium

Gold up 11% in twenty days while the dollar sags

Gold rising while long yields rise is not a rate trade; it is a credibility trade, and it is the cleanest signal in the current market.

Gold is at $4,450, up 2.0% in five days and 11.0% in twenty. The dollar index sits at 99.6, a three-month low, after traders cut the odds of a September Fed hike to one-in-three from about 75% in late July on soft employment, inflation and retail sales data (Bloomberg). A falling dollar and rising gold on dovish repricing is ordinary. What is less ordinary is that it is happening while 30-year Treasury yields make 19-year highs. Normally weaker growth data pulls the whole curve down. Here the front end fell and the long end rose. That is the market saying the problem is not the cycle, it is the term premium — the extra compensation demanded for lending to the United States for thirty years. Two other data points from the window point the same way, and both are softer evidence. An OMFIF survey of reserve managers found 74% expect the dollar's share of global reserves to fall over five years and 89% expect central bank gold holdings to rise over the next year; it is described as the first such survey where more managers plan to cut dollar exposure than add. Separately, Japan is now able to borrow dollars from the Fed against its $1tn-plus Treasury holdings under a repurposed COVID-era repo facility, after the first joint US–Japan intervention in 28 years on 31 July (Nikkei). Japan's top currency diplomat called it "a US–Japan currency union." That is a phrase worth sitting with.

What would change our mind

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

Median BDC non-accruals falling back below 2.2% next quarter, or conversely a large non-traded private credit fund gating redemptions or cutting its NAV by more than 5%.

Would move the number

A data-centre SPV bond being pulled or repriced sharply wider — the QTS 'Project Odyssey' deal drew $8bn of orders at a single-B yield, so the bid is still there; the first failed deal is the signal.

Would move the number

High-yield OAS moving above 400bp from today's 267bp, which would mean credit is finally pricing what the default data already says.

Would move the number

The 30-year Treasury sustaining above 5.5%, or a tailed auction with dealers taking an unusually large share — evidence that duration demand has actually broken rather than merely repriced.

Would move the number