Nvidia is owed $63bn, and five customers owe most of it

Archived reading, published Thu, 27 Aug 2026 22:23:51 UTC (24 days ago). This is not the current state of the meter.

See the live reading →

CRASH-O-METER

Cracking

how close are we

-1 since the last reading

Fragility88

how much tinder is stacked up — moves slowly

Ignition33

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 35 sources and rewrites this page.

Why it moved: Down one to 60, entirely on the spark side: this is the calmest stretch of markets in our eight-issue trail. The VIX, the market's gauge of how much turbulence traders expect over the next month (under 20 is calm; it passed 80 in March 2020), sits at 14.51. The extra interest that shaky borrowers pay over the government rate is at 2.67 percentage points for the riskiest companies and 0.80 percentage points for safer ones. The S&P 500 closed just 0.87% below its all-time high, and Nvidia itself closed at a record. The one genuine break in the evidence, Jane Street's $15bn July trading-revenue loss and the forced sale of a fund it backed, happened seven weeks ago and the market absorbed it without anyone scrambling for cash. The dry-tinder score holds at 88; Nvidia's receivables disclosure is a sharper measurement of vendor credit we already count, not new fuel.

Reporting from 26 Aug to 27 Aug

Crash points

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

What’s moving the needle

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

The AI capex bubble27 Aughigh

The other side of Nvidia's balance sheet

The AI buildout's credit risk is increasingly concentrated on one supplier's balance sheet, in a line item that carries no promises from the borrower and no assets pledged as security.

Between January and July, the money Nvidia is owed by customers but has not yet collected went from $38.5bn to $63.1bn, a 63% rise (CNBC). In the quarterly filing, "five direct customers" account for 70% of that balance. A year earlier it was three customers and 56%. This is the least glamorous form of vendor financing and the most direct. Nvidia has shipped the chips, booked the revenue, and is waiting to be paid. In the meantime it is an unsecured creditor of whoever holds the hardware. The number matters less as a level than as a ratio to everything else Nvidia has committed: alongside $279bn of supply and capacity commitments, $29bn of cloud service agreements and the $105bn Ohio lease guarantee (totaling $366bn on Goldman's tally in the same report), Nvidia is now simultaneously the seller, the financier and the backstop for a customer base of roughly five names. The forward path is the part worth watching. Bank of America projects receivables at $113bn in January 2028 and $147bn in 2029; Morgan Stanley gets to $171bn. Those are projections, not disclosures, and both banks are positive on the stock. But if they are roughly right, the credit Nvidia extends to its customers grows faster than most of those customers' operating cash flow. The question worth holding: what does Nvidia actually recover on a receivable from a smaller cloud company whose only asset is depreciating silicon that Nvidia itself sold it? The filing does not say, because nobody has had to find out.

Bond market dysfunction27 Aughigh

Hyperscaler bonds pay a premium everyone else does not

The broad credit indexes look calm because they average away the sector where the borrowing actually is.

JPMorgan's head of European investment-grade finance, Matthias Reschke, told Bloomberg TV that bonds issued by the giant cloud and AI companies trade about 0.55 percentage points above the interest rate on other US corporate debt of similar quality, a "material premium" (Bloomberg). Meta, Alphabet, Oracle, Microsoft and Amazon have raised more than $220bn this year across seven currencies. "We don't doubt that we will be able to place it," Reschke said. "It's all a question of price." This is the most useful single number in today's evidence, because it dissolves a contradiction we keep having to explain. The extra interest that safer corporate borrowers pay over the government sits at 0.80 percentage points, and for the riskiest borrowers it is 2.67 percentage points. Both are among the tightest readings in our trail. On the face of it, the credit market is not worried about anything. But an index is an average, and the average is dominated by companies that are not building data centers. Inside it, the market is charging the AI complex roughly two-thirds more than the going rate for borrowers with equivalent credit ratings. So the credit market is discriminating. That is a healthier picture than pure complacency, and we should say so. It is also how the risk gets distributed: at 0.55 percentage points of extra interest the bonds find buyers, those buyers are compensated for the risk, and the exposure spreads out across pension funds and insurers rather than sitting in a few trading firms' inventories. What would worry us is the premium widening while borrowing continues at the same pace anyway, a sign that the borrowers are, in Treasury Secretary Bessent's phrase from last week, "almost yield agnostic," and will pay whatever it takes.

Hidden leverage and shadow banking27 Augmedium

Jane Street's first losing month in a decade

The firms that keep markets running during a panic are now making large directional bets that nobody sees until after they go wrong.

In July, a hedge fund run by 24-year-old Leopold Aschenbrenner was forced to sell its positions after heavy losses in AI stocks. Jane Street, one of the largest trading firms in the United States, had invested heavily in that fund, Situational Awareness, and held many of the same names: Sandisk, Nebius, Bloom Energy. The result, per the FT, was a $15bn loss in trading revenues and Jane Street's first losing month in a decade (FT). The wording leaves open whether that is money actually lost or a shortfall against expected revenue; Jane Street does not publish quarterly accounts, so the distinction cannot be settled from outside. What can be settled is the shape of the thing. Jane Street began as a firm that made money by exploiting tiny price differences over fractions of a second. It is now one of the largest financial institutions in the country and, on the FT's account, takes large one-way bets over long horizons. Larry Tabb of Bloomberg Intelligence puts the mechanism plainly: the revenue is only possible if the holding period extends along with the size, and "the longer your holding period, the more chance that something you can't forecast occurs." In an August letter to employees, the firm admitted it was not protected against some of the moves that undid its July. This is the migration we have been tracking, visible in one firm. A non-bank that provides the buying and selling that keeps equities, options and government bonds trading smoothly has taken a nine-figure-plus hit on a concentrated bet on AI, and the first the outside world hears of it is a feature story seven weeks later. It also cuts the other way: the system took the shock without anyone scrambling for cash. The losses were tested and held.

Crypto and TradFi contagion27 Aughigh

The obituary and the price disagree

A rising share price is not evidence a financing model works. The premium to the value of the coins held is, and that is where the model is broken.

The FT's analysis of the 50 companies holding the most bitcoin finds their combined market value has fallen from $150bn in July 2025 to $67bn. Shares in 43 of the 50 trade below where they were before the bitcoin pivot was announced; 35 have halved or worse. Many are selling tokens and going back to their original businesses. "It was always doomed," says Adam Morgan McCarthy of LO:TECH. "I don't think we see new ones" (FT). The market this week says something else. Strategy rose 11.6% on Thursday alone, is up 40.8% in twenty days and closed at its period high. Bitmine is up 41.7% over the same span, SharpLink 38.6%, Coinbase 16.7%, all four sitting at or near their highs. Bitcoin itself is up 23.7% in twenty days to $80,247. Both things are true, and the reconciliation is the interesting part. The treasury-company model works by issuing shares above the value of the coins held and using the proceeds to buy more coins. It requires a persistent premium. A rally in the underlying coin lifts the share price without restoring the premium, and the FT's structural finding (dozens of these companies selling tokens rather than buying them) is evidence the issuance machine is not running. So the unwinding we noted this week is genuine, and the price move is a bet on the currency losing value rather than a revival of the model. If MSCI's consultation on excluding non-operating companies goes through, the buying pressure from index funds disappears too.

Household credit27 Augmedium

Subprime auto sets a July record; everything else improves

The household number that policymakers watch is being held up by creditworthy borrowers while the riskiest tier prints records, and the two are funded by different people.

Fitch's index of subprime car loans at least 60 days late read 6.13% in July, the highest July in a series running back to 1994. The equivalent rate for creditworthy borrowers was 0.49%. That is a ratio of more than twelve to one. The New York Fed's second-quarter data has one in eighteen of all US car loans at least 90 days behind (5.5%), the second-highest reading in more than twenty years of tracking. Everything else in the household data got better. FICO's autumn report has mortgage payments 30 days late down to 1.35% from 1.42%, car payments 30 days late down about 0.05 percentage points to 2.64%, and bank card payments 30 days late roughly flat around 2.7%. Two issues ago we lowered the dry-tinder score a point on exactly this kind of improvement, and we stand by it. The two facts are not in tension; they describe a split. Household credit in the aggregate is fine because the aggregate is dominated by creditworthy borrowers with fixed-rate mortgages. The deterioration is entirely at the bottom, in the borrowers who finance cars at rates that only work if almost all of them pay. The publicly traded lenders are priced as though this is under control. Credit Acceptance, which lends at the very bottom of the subprime market, is up 7.8% in twenty days and sits 2.9% off its high. Ally is flat. Oportun is up 32% in twenty days. That is defensible (deep subprime loan pools are priced for double-digit losses and the loans are largely bundled and sold off to investors) but it means the stock market is not the place to look for the first sign of trouble here.

Fed, Treasury and policy27 Aughigh

Spending the emergency cash to buy back bonds

A fiscal agency using its cash buffer to manage borrowing costs is doing the central bank's job without the central bank's authority, and in the opposite direction.

The Treasury Department, which borrows the money the government spends, has doubled its buybacks of government bonds that do not come due for twenty or thirty years to at least $28bn over the next two months, announced outside the regular schedule. CNBC reports that two senior officials say the purchases could be funded from the Treasury General Account, the government's checking account at the Federal Reserve, America's central bank, currently holding around $954bn. The TGA exists as a buffer. It has been kept deliberately large so the government can keep paying its bills through shutdowns. Running it down to buy back thirty-year bonds converts a cash reserve into a bet on interest rates, and it does so at a moment when the interest rate the government pays to borrow for thirty years has been near 5.3%, its highest since 2007. Geng Ngarmboonanant, a former deputy chief of staff to Janet Yellen and now at JPMorgan, notes that within 48 hours of the buyback announcement the ten-year rate had rebounded above 4.70%, higher than before the intervention (NYT). Bloomberg's opinion desk is blunter, and Stanley Druckenmiller, who mentored both Treasury Secretary Bessent and Fed chair Warsh, used a Wall Street Journal column to say the effort is doomed and the only question is how much gets spent before conceding. The structural problem is the direction of travel. If buybacks work, they lower borrowing costs across the economy, which stimulates spending. Three members of the Federal Reserve's rate-setting committee (Hammack, Kashkari and Logan) voted against holding rates steady in July, preferring a 0.25 percentage point increase. Treasury is easing while a bloc of the Fed wants to tighten. The ten-year rate has since settled back to 4.66%, so the acute phase has passed. The institutional question has not.

Signs of the times

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

Wedding week, fund liquidated, backer down $15bn

Leopold Aschenbrenner, 24, married in August at the end of a week in which heavy losses forced the sale of his hedge fund's positions. Among the guests was Rob Granieri, co-founder of Jane Street, which had invested in the fund and posted its first losing month in a decade.

leverageaiconcentration

27 AugFinancial Times

Google's quarterly purchase orders: $707bn

Morgan Stanley's tally of filings puts Alphabet's purchase commitments at $707bn in the most recent quarter alone, against $72.5bn for the whole of 2025. A single quarter's forward orders now approach the annual GDP of Switzerland.

ai capexoff balance sheet

27 AugAxios

A$3.3bn owed, and that excludes the deposits

Administrators to collapsed Sydney developer Bathla put creditor claims at an estimated A$3.3bn, a figure that does not include deposits paid by thousands of apartment buyers across hundreds of projects. Teneo said the company would burn A$40m before year-end if it kept trading.

private creditpropertyaustralia

27 AugAustralian Financial Review

Twelve times worse than prime

Fitch's July reading for subprime car loans at least 60 days late was 6.13%, the highest July since the series began in 1994. Creditworthy borrowers were at 0.49%, meaning the two halves of the same asset class are now behaving like different products.

householdssubprime

27 AugFitch data via web sweep

The rumour mill

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

Partly true

“Nvidia's CFO commentary discloses $105bn of guarantees on the OpenAI Ohio facility plus an option to support 3.8 more gigawatts, implying about $198bn of total off-balance-sheet guarantees.”

The disclosure supports everything except the total. Nvidia's own commentary confirms guarantees capped at $105bn on 20-year OpenAI leases at SB Energy's Ohio campus, phasing in from fiscal 2029, and an option to support roughly 3.8 additional gigawatts. The $198bn figure is the commentator's extrapolation of what that option would cost if exercised; no filing or outlet puts a number on it.

Claimed by Meet Kevin

Partly true

“The Treasury has said it could tap its roughly $1tn Treasury General Account to fund the increased bond buybacks.”

CNBC reported on August 24 that two senior Treasury officials said the TGA could be used, but they gave no size and no decision has been announced. The buyback expansion itself is confirmed: from $2bn to at least $4bn per operation, running September 9 to November 4.

Claimed by George Gammon

Partly true

“Central banks are adding around 1,000 tonnes of gold a year for the fifth consecutive year, roughly a third of world production.”

True for 2022, 2023 and 2024; not for 2025, when the World Gold Council put central bank buying at 863.3 tonnes. The Council also revised first-quarter 2026 purchases down to 57 tonnes from 244, a 187-tonne cut, making the first half of 2026, at about 345 tonnes, the weakest since 2022 despite a strong 289-tonne second quarter.

Claimed by Palisades Gold Radio, Kitco NEWS

Confirmed

“Three FOMC members dissented in a 9-3 July decision, wanting a rate hike, and Hammack has said 'now is the time to act'.”

The official FOMC statement records Hammack, Kashkari and Logan voting against holding and preferring a 0.25 percentage point increase. Worth holding alongside the Treasury buyback story: the two arms of policy are pulling opposite ways.

Claimed by Meet Kevin

Earlier developments

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

The AI capex bubble27 Aughigh

Three trillion dollars of promises, none of it counted as debt

The borrowed money behind the AI buildout is disclosed in footnotes but not counted as debt, and it is what suppliers and developers are borrowing against.

Morgan Stanley has added up the commitments that seven companies (Alphabet, Microsoft, Meta, Oracle, Amazon, Nvidia and Broadcom) have made for AI infrastructure but have not yet put on their books. The total comes to roughly $3 trillion. That sits on top of an estimated $770bn of debt and lease obligations that do show up on the books (Axios). The split matters. About $1.1tn is payments on data-center leases that have not yet started. Under accounting rules, a lease only becomes a liability when it begins; before that it is a footnote. The other $1.7tn is purchase commitments for chips, memory and networking gear, a promise to buy, not a borrowing. Alphabet's purchase commitments alone reached $707bn in the most recent quarter, against $72.5bn for all of 2025. Nvidia's own supplier commitments, disclosed Wednesday, more than doubled to $279bn from $119bn in a single quarter (WSJ). None of this is hidden. It is all disclosed. But it is not counted where borrowed money is normally counted, and the researchers make the point that matters most: suppliers and data-center developers borrow against those long-dated commitments. An Alphabet promise becomes something valuable that a developer pledges to a lender, which becomes a bond sold to someone reaching for a little more interest. The same obligation is considered safe at one end of the chain and shaky at the other. What has to stay true is simple and unprotected: the tenants must want the capacity, at the agreed price, for the agreed term. If they do, these are ordinary supply contracts. If they do not, the commitment flips from a strategic advantage into a liability. The WSJ's comparison to Cisco in 2001 is the right one. This is a measurement improvement rather than fresh fuel. The commitments were signed weeks and months ago. We simply now have a number for the whole stack, and it is bigger than our running tally.

Hidden leverage and shadow banking27 Augmedium

A five-billion-dollar loan to a company worth two

Debt for these new AI cloud companies is underwritten against the contracts of two or three AI labs, so a portfolio of different borrowers can quietly be a single concentrated bet.

JPMorgan has started sounding out lenders for a $5bn debt package for Volta Infra Holdings, an AI cloud company founded earlier this year by two former Brookfield infrastructure executives. Earlier this month Volta raised $300m of venture money at a $2.4bn valuation (Bloomberg). So the proposed debt is roughly twice the company's equity valuation of three weeks ago, at a company that has existed for months. That is not necessarily reckless, but it tells you what the loan is actually secured on. It is not Volta's own books, because there barely are any. It is the $10bn computing contract Volta signed with Anthropic, plus whatever the chips and the shell buildings are worth if the contract fails. Which makes the credit question a question about Anthropic. And Anthropic is now the firm on the other side of a remarkable stack of these deals: $45bn to Nscale over six years for 460 megawatts in West Virginia (Bloomberg), $50bn to Fluidstack, $45bn to SpaceX, $10bn to Volta. Each of those companies is, or will be, borrowing against its Anthropic contract. A lender who buys the Nscale paper and the Volta paper has written the same credit twice and may not have noticed. Anthropic was valued at $965bn in May and is expected to hold a very large IPO. It may well be good for all of it. But the structure means the sector's debt is not diversified in the way a spread of separate borrowers looks diversified. Bloomberg counts roughly $600bn borrowed for data centers and AI since last year. A meaningful slice of it points back at the same handful of model labs.

Bond market dysfunction26 Augmedium

One buyer took most of Japan's five-year auction

The biggest buyer of Japanese government bonds is a giant pension fund that buys and sells mechanically based on how stocks are doing, and its flows reverse if world stock markets fall.

On August 18 Japan's finance ministry auctioned about ¥2.5tn of five-year government bonds. More than ¥1.8tn went to unidentified buyers, according to Nikkei QUICK's interviews with securities firms. The largest identified buyer, Mitsubishi UFJ Morgan Stanley Securities, took ¥265bn. "The only explanation is that the whale was buying," one trader said (Nikkei). The whale is GPIF, the ¥300tn public pension fund, which a 2025 rule change allowed to bid directly at auctions. Bidding directly is precisely so that nobody trades ahead of it: its identity becomes clear only after it has left. In fiscal 2025 it bought a net ¥14tn of Japanese bonds and sold a net ¥10tn of domestic stocks, one of the biggest rebalances in a decade. The mechanism worth understanding is that this buyer has no opinion. GPIF holds a fixed quarter of its portfolio in domestic bonds; when global stocks rally, the bond share falls and it buys bonds. It is executing arithmetic. That is comfortable while stocks go up. It runs the other way if they fall: the largest single source of demand for Japanese government bonds becomes a seller precisely when everyone else is trying to get out. Underneath it, the small lenders are already hurting. Seventeen shinkin banks (local cooperative lenders) posted net losses for the year to March, up from five, with the sector's unrealized securities losses at about ¥2.6tn and bond-related losses nearly tripling. Ehime Shinkin lost ¥18.9bn, its first loss since it was founded in 1951 (Nikkei). Japan's domestic buyer base is splitting into one giant mechanical rebalancer and a long tail of institutions that cannot afford to buy bonds that do not come due for decades. This matters beyond Tokyo: Japan is the largest foreign holder of US government bonds, and Washington has been propping up the yen partly so that Japan does not have to sell them.

Crypto and TradFi contagion27 Aughigh

The index rule that could break Saylor's machine

Index membership is part of the plumbing that lets companies that exist mainly to hold digital coins keep issuing shares to buy more coins, and it is now formally in question.

MSCI, the firm whose stock indexes determine what trillions of dollars of passive funds must own, is consulting on a rule that would exclude any company it judges to be primarily accumulating assets rather than running a business. Applied as of May, it would have removed Strategy Inc., Japan's Metaplanet and the uranium holder Yellow Cake (Bloomberg). MSCI abandoned a narrower crypto-exclusion plan in January and came back with a broader test that catches the same companies without mentioning bitcoin. The reason index membership matters here is that it is a load-bearing part of the financing structure, not a badge. Passive funds must own an index member. That automatic demand helps hold the share price above the value of the coins the company holds. A premium to the value of the underlying coins is what makes issuing new shares profitable: sell stock at 1.5 times the coins per share, buy coins, and coins per share rises. Take away the passive buying and the premium compresses; compress the premium and the issuance machine stops; stop the machine and the company is a fund holding coins with borrowed money, with more than $60bn of purchases to service. The prices are pointing the other way. Strategy rose 11.7% on Thursday and is up 40.9% in twenty days, at its high for our window. SharpLink is up 38.9% and Bitmine 45.0%, both at highs, with bitcoin up 23.9% over the same period. So the market is trading the coin, not the structure, even as the FT counts the combined value of the fifty largest bitcoin treasury companies falling from $150bn in July 2025 to $67bn, with 43 of the 50 below their pre-pivot share price (FT). Both things can be true. A rally does not repair a financing model; it postpones the test.

The dollar, gold and reserve status27 Aughigh

Gold and bitcoin are both rising while the dollar sits still

Investors are hedging against government-issued money in general without buying any other country's currency, which is a judgment about every major government's finances, not just America's.

Gold and bitcoin funds took in a record $7bn over five trading days: $3.4bn into SPDR Gold Shares and $1.5bn into iShares Bitcoin Trust, both in the top ten US funds by weekly inflow (Bloomberg). Gold is at $4,660, up 15.1% in twenty days. Silver is up 16.8%. Bitcoin is up 23.9%. Now look at the dollar. The dollar index, which tracks the dollar against a basket of other currencies, is at 99.16, down 0.65% over the same twenty days. The euro fund is up 1.1%, the yen fund is down 0.1%. The broad trade-weighted dollar is off 2.2%. That combination is the interesting part. If this were a currency trade you would expect money leaving the dollar to arrive somewhere, in euros, in yen. It is not really arriving anywhere. Buyers are moving into things that no government can print, and they are not doing it because they prefer the fiscal position of Japan or the euro area. They are doing it because they think every large government has the same problem, which is why gold and bitcoin, historically inconsistent partners, are now rising together. The immediate cause is domestic policy. Bessent's plan to at least double the Treasury Department's buybacks of government IOUs that do not come due for twenty or thirty years, and the reported possibility of funding them from the roughly $954bn the government holds in its own checking account rather than through new borrowing, reads to a lot of investors as a government trying to hold down the price of its own money (Bloomberg opinion). Worth noting what this is not: the interest rate the government pays to borrow for ten years is back at 4.64%, down from 4.73% a week ago, and short-term rates are calm. The rush into things governments cannot print is running while the bond market has settled. That looks less like a panic hedge and more like a deliberate reallocation, which is the slower and more durable kind.

Private credit and BDCs27 Augmedium

The managers are up; their loans are not

Rising failures to pay in private lending funds hurt fund investors, not fund managers, until investors start pulling their money out and shrinking the fee base. That is the signal to watch.

Blue Owl's fund reported a failure-to-pay rate of 2.8% in the second quarter, the highest in at least five years. Funds run by Ares, Blackstone, Blue Owl and Golub all reported loan failure rates at their highest since 2021. At BCRED, Blackstone's big fund sold directly to investors rather than traded on an exchange, loans the lender has stopped counting the interest on (because it stopped arriving) jumped from 0.6% to 2.4% of cost between the first and second quarters, driven by two names: Medallia and Affordable Care International. This comes from a web sweep of secondary reporting rather than a single paper of record, so treat the individual figures as indicative (sweep). The share prices of the firms doing the lending have gone the other way. Blue Owl's stock is up 21.0% in twenty days, Ares Management 15.4%, FS KKR 15.3%, Apollo 11.9%, Blackstone 13.4%. This is not as contradictory as it looks, and the reason is the most useful thing to understand about this industry. The managers do not own most of the credit risk. They earn management and incentive fees on the money they oversee. A loan going bad hits the fund's value, which is borne by the fund's investors: the shareholders of the fund itself, the pension schemes, the retail buyers of funds that are hard to get out of. The manager's earnings only fall when the money itself leaves. Which is why the number to watch is not the failure rate but the flows. Redemption requests at private lending funds sold to individuals averaged 14% of the fund's value in the second quarter against normal quarterly limits of 5%, with managers using gates (rules that cap how much money can leave at once). Gates protect the fund and the fee base in the short run. They also tell every remaining investor exactly how easy it is to get out. Losses hit fund investors first; the managers only re-rate when the money stops arriving.

The AI capex bubble27 Aughigh

Nvidia is now committed at both ends of the chain

The vendor at the centre of the AI build has now written large obligations to its suppliers and its customers simultaneously, so a demand shortfall hits it from two directions at once.

Nvidia told investors on Wednesday that its commitments to suppliers — mostly for high-bandwidth memory — reached $279bn in the quarter to July 26th, up from $119bn three months earlier (WSJ). That is $160bn of new promises to pay, made in ninety days. A purchase commitment is not debt, but it behaves like one in a downturn. It is a contractual obligation to buy components at agreed volumes. In a boom it is a weapon: it corners scarce memory supply and locks rivals out. Nvidia's guidance assumes it works — 70% revenue growth in the fiscal year ending January 2028, with gross margin falling from about 74% to 71.5% because memory now costs more. Jensen Huang's framing is that supply, not demand, is the constraint. The other end is the customer. Nvidia's own CFO commentary describes guarantees capped at $105bn supporting land, power and shell buildout for roughly 4.25GW in Ohio, leased to OpenAI for twenty years, phasing in from fiscal 2029, with an option on credit support for a further 3.8GW. So the same balance sheet has promised to buy components and promised to stand behind a tenant's rent. What has to stay true: end demand for compute grows fast enough to absorb both. The WSJ reaches for Cisco in 2001, which is the right comparison — supply commitments that were a strategic advantage became a write-off when orders stopped. The market disagrees today: Nvidia rose 7.2% and sits 0.2% off its high.

Hidden leverage and shadow banking27 Aughigh

A 5% index move cost Jane Street $15bn of revenue

The reporting arrived five weeks after the loss, which is how long it takes to find out what a large non-bank was actually holding.

The Financial Times has reconstructed what July's AI sell-off did to Jane Street: a $15bn loss in trading revenues and the firm's first losing month in a decade (FT). The mechanism matters more than the number. Jane Street had invested heavily in Situational Awareness, the hedge fund run by 24-year-old Leopold Aschenbrenner, which was forced into a sale of its positions. It was also directly exposed to many of the same AI-linked names — Sandisk, Nebius, Bloom Energy. So the seeded fund and the seeder were long the same thing, and the fund's liquidation ran into its backer's book. In an August letter to staff the firm admitted it was not protected against some of the moves. Hold that against the size of the shock. The worst ten-day drop in the Nasdaq 100 over the whole period was 5.4%, from 21 to 29 July. A single-digit index move produced a $15bn revenue swing at one firm, because Jane Street is no longer the fast arbitrage shop of its reputation. Larry Tabb of Bloomberg Intelligence puts it plainly in the FT: revenue of that size is only possible when holding periods extend along with position size, and longer holding periods mean less clarity about the future. This is exactly the migration Crash Lab tracks. A firm large enough that regulators watch it for the dislocations it could cause is running principal risk over weeks, funded by lenders, with no requirement to publish what it holds. We learned this a month after the fact, from a feature article.

Bond market dysfunction27 Aughigh

The bond market charges 55bp extra for the word AI

A 55bp premium on the strongest borrowers in the market is the cleanest available price for AI risk, and it is being paid quietly while headline credit spreads stay tight.

Matthias Reschke, JPMorgan's head of European investment-grade finance, told Bloomberg TV that hyperscaler bonds trade about 55 basis points wide of other US corporate credit — a "material premium" (Bloomberg). Meta, Alphabet, Oracle, Microsoft and Amazon have raised more than $220bn this year across seven different currencies. Put that next to the index. The whole US investment-grade market trades at an option-adjusted spread of 81bp. So the extra compensation demanded from five of the most cash-generative companies on earth is roughly two-thirds of the entire spread of the average IG borrower. These are not weak credits. Something other than default probability is being priced. Two candidates. The first is indigestion: in several smaller currency markets, one bond deal was enough to make a hyperscaler among the largest non-financial borrowers there, which is why Korea's export credit agency went to New Zealand this week to escape the crowding. The second is that the debt funds assets — GPUs, shells, power interconnects — whose residual value in 2031 nobody can honestly price, and the bond market has decided to be paid for that. Reschke's own line is the tell: "We don't doubt that we will be able to place it, it's all a question of price." That is a banker saying the market clears at a level, not that it clears comfortably. Rate volatility ticked up 13% on the day, though it remains 29% below its recent high, and the 10-year sits at 4.64%. The stress in AI credit is showing up in spread, not in the risk-free curve.

Crypto and TradFi contagion27 Aughigh

The treasury model is dead and its shares are at highs

A leveraged structure genuinely unwinding is one of the few things that lowers fragility, and this one is unwinding into a rising coin price, which is the gentlest way it could happen.

An FT analysis of the 50 companies holding the most bitcoin finds their combined market value has fallen from $150bn in July 2025 to $67bn today. Shares in 43 of the 50 are below where they traded before they announced the pivot; 35 have halved. Strategy alone accounts for $79bn of the decline from its peak. Many are now selling their coins and going back to their old businesses (FT). The tape says something else. Strategy is up 7.2% today, 17.5% in five days and 35% in twenty, and closed at the top of its recent range. BitMine is up 41.7% in twenty days, SharpLink 34.4%. Bitcoin is at $79,485, up 22.5% over the same period. Both are true, and the reconciliation is the point. The rally is coin beta plus the debasement trade — gold and bitcoin ETFs took a record $7bn in five trading days after the Treasury said it would double long-bond buybacks (Bloomberg). What has not come back is the thing that made the model work: a share price at a premium to the coins held, which let the company issue equity, buy more coins, and raise the price again. Without the premium the flywheel stops, and the debt raised while it was spinning still matures. MSCI is now consulting on excluding companies primarily engaged in accumulating assets rather than operating a business. Applied as of May, that would have removed Strategy, Metaplanet and Yellow Cake, taking the passive bid with it.

Private credit and BDCs27 Augmedium

The adjusted default number is rising three times faster

The metric that captures borrowers already in distress is deteriorating far faster than the one managers report, and BDC share prices are near highs.

Across the ten largest listed BDCs, loans on non-accrual reached 3.95% of debt at cost in the second quarter, up 20 basis points on the quarter. The portfolio itself shrank 2.3%, to $83.6bn, while the non-accrual balance grew $89m to $3.3bn — the numerator up, the denominator down. The more useful figure is the adjusted one. Count every tranche lent to a borrower that has at least one tranche on non-accrual, and the exposure is $5.0bn, or 5.95% of debt at cost, up 54bp in a quarter. That is nearly three times the rate of increase in the headline number. Here is why the gap exists. A lender can put a second-lien loan on non-accrual while continuing to accrue interest on the first lien to the same company. The headline rate then understates how much of the book is lent to borrowers already in trouble. The gap between 3.95% and 5.95% is a measure of how much credit is sitting one step behind a problem that has already been acknowledged. At Blackstone's BCRED, non-accruals at cost went from 0.6% to 2.4% between the first and second quarters, attributed to Medallia and Affordable Care International. Redemption requests at semi-liquid private credit funds aimed at individuals averaged 14% of NAV in the quarter, against typical quarterly limits of 5%. These figures come from a single sweep of Morningstar and PitchBook LCD analysis rather than from filings we have read, so treat the decimals loosely. The direction is corroborated. The listed prices are not: Ares Capital is 1.5% off its high, Blackstone Secured Lending 0.5%, FS KKR up 16% in twenty days.

Fed, Treasury and policy27 Augmedium

The buyback that has not bought anything yet

Using the government's cash buffer to hold down long yields converts a debt-management tool into monetary policy, run by the branch that is not supposed to set it.

Two points about the Treasury's bond intervention are worth separating. The first: the expanded programme has not started. Buybacks rise from $2bn to at least $4bn per operation between 9 September and 4 November, targeting 10- to 30-year maturities. Not one bond has been purchased under it. The 30-year's retreat from 5.31% on 19 August — the highest since 2007 — and the 10-year's fall to 4.64% are anticipation, plus softer oil, not the effect of any actual buying. The second: CNBC reported that officials are considering funding the buybacks from the Treasury General Account, the government's roughly $954bn checking account at the Fed. This is where it stops being debt management. Running the TGA down puts reserves into the banking system; it is monetary easing performed by the fiscal authority. Bloomberg's opinion desk argues it would cost Treasury credibility, and several Fed officials have said rates should be going the other way to cool inflation (Bloomberg, FT). It is also a one-shot resource. Spend the TGA and you must rebuild it later with bill issuance, which drains the reserves back out. Stanley Druckenmiller's line in the WSJ — that the only variable is how much is spent before conceding — is the market's read as much as an insult. Meanwhile investors are hedging the outcome directly: gold sits at $4,625, up 14.2% in twenty days, and gold and bitcoin ETFs together took a record $7bn in five sessions.

The AI capex bubble26 Aughigh

Nvidia is on the lease, and hopes to hand it off

The vendor is no longer just guaranteeing its customers' rent. It is signing leases itself on the assumption it can sell them to a buyer base it is currently helping to invent.

Nvidia's CFO commentary, published alongside Wednesday's results, puts numbers on the Ohio deal. Nvidia has entered guarantees covering the land, power and shell buildout for roughly 4.25 gigawatts at SB Energy's PORTS-Pike campus, a site that will exclusively host Nvidia hardware under 20-year leases to OpenAI. The guarantee obligations are capped at $105bn, become effective in phases, and the first phase is not expected until fiscal 2029. Nvidia also holds an option to provide the same kind of credit support for about 3.8 additional gigawatts as the site scales. The less-noticed sentence is the second one. The same commentary says Nvidia has signed data-center leases of roughly fifteen years, commencing in fiscal 2028 to fiscal 2029, and that it expects to reassign those leases to third parties. That is not a guarantee. That is Nvidia as tenant, warehousing lease obligations on its own books with a stated intention to sell them on. For it to work, a buyer has to exist in 2028 for fifteen-year leases on purpose-built AI shells, a market that does not exist today. The most likely buyers are the six asset managers Nvidia is already organizing into a $500bn compute-financing platform: Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR (The Economist). Worth noting where this started. CNBC reported the original OpenAI discussions ran to $250bn; the WSJ put the revised exposure at less than $120bn; the signed number is $105bn. The direction of travel on that particular figure has been down. The filings do not say how the reassigned leases will be accounted for, or what Nvidia books if no one takes them.

Bond market dysfunction26 Aughigh

One buyer took three-quarters of a Japanese government bond auction

A single buyer that does not care about price is now the swing bidder in the world's second-largest government bond market, and its buying is conditional on stocks going up.

On August 18 Japan's Finance Ministry auctioned about ¥2.5tn of five-year government bonds. More than ¥1.8tn went to unidentified buyers, according to dealers polled by Nikkei QUICK News. The largest identified buyer, Mitsubishi UFJ Morgan Stanley Securities, took ¥265bn, under a seventh of that. "The only explanation is that the whale was buying," one trader told Nikkei. The whale is the Government Pension Investment Fund, which manages ¥300tn ($1.88tn) of assets, making it the largest public pension in the world. Since a 2025 rule change it can bid directly at government bond auctions rather than through dealers. The explicit purpose was to stop traders who track its flows from trading ahead of it. Its identity emerges only after it has gone. The mechanism matters more than the anonymity. The fund runs a fixed four-way split (domestic bonds, domestic stocks, foreign bonds, foreign equities) and rebalances back to it. A global stock rally therefore makes it a mechanical, price-insensitive buyer of Japanese government bonds. In fiscal 2025 it bought a net ¥14tn of Japanese bonds and sold a net ¥10tn of domestic stocks, one of the largest such adjustments in a decade. This is the flow currently absorbing Japanese long-dated bond supply while smaller institutions sell into it at a loss: 17 shinkin banks (local cooperative lenders) posted net losses last fiscal year, up from five, on government bond writedowns. Run it backwards. If global stocks fall, the same arithmetic makes the fund a buyer of stocks and no longer a buyer of bonds. The buyer holding up the long end of a market with rising interest rates disappears precisely when everything else is being sold. And because the bidding is anonymous, nobody outside the ministry will know how much of recent demand was one rebalancing rule.

Private credit and BDCs27 Augmedium

Four big lenders, all at post-2021 default highs

Default rates across the four largest private lenders are at five-year highs while their listed vehicles trade at their highs. The gap is entirely a question of when the prices they put on their own loans catch up.

The second-quarter 2026 numbers are now compiled, and they point one way. Loans that borrowers have stopped paying interest on across the top 20 listed private lending funds reached a median 2.8% of cost, up 0.8 percentage points on the quarter and the highest since 2017. Morningstar DBRS puts the industry-wide average at 3.4% of portfolios, from 3.1% in the fourth quarter of 2025. Blackstone's BCRED saw non-paying loans go from 0.6% to 2.4% in a single quarter, driven by two names: Medallia and Affordable Care International. Funds run by Ares, Blackstone, Blue Owl and Golub all reported loan default rates at their highest since 2021, with Blue Owl's fund at 2.8%, a five-year high. Now look at what the shares did. Blackstone Secured Lending is up 8.3% in twenty days and sits 0.26% off its period high. Ares Capital is up 6.25% and 1.5% off its high. FS KKR is up 13%. The extra interest that shaky borrowers pay over the government is at 2.70 percentage points, near the tightest it has been. The two claims are not quite contradictory, and the reconciliation is the interesting part. These funds' share prices are driven by the value of their assets and the dividend they pay, not by the default rate. At 11-12% portfolio interest rates, 3% of loans going bad still leaves the payout covered. The share price only breaks when the prices the manager puts on the other 97% move, and those prices are set quarterly, by the manager, on loans that do not trade on any exchange. So the market is not disagreeing with the data. It is saying the data does not reach it yet. A caveat we will not bury: these aggregates come from a compilation of second-quarter filings and third-party research rather than a single paper of record, and we treat the fund-level figures accordingly.

Hidden leverage and shadow banking26 Augmedium

The clearing mandate has holes exactly where the borrowed money is

The regulatory repair for the 2020 government bond blow-up does not cover the two structures used to finance the largest leveraged positions in the market.

The Office of Financial Research, the Treasury Department's data arm, made two posts in the last window. The first, on August 19, put hedge funds' cash government bond holdings at roughly $2tn, and stressed that because the profits on the trade are tiny, the funds "rely heavily on leverage through repo financing and futures margining." In plain terms: they borrow against the same bonds over and over, using each round of borrowing to buy more (OFR). The second post is the one to keep. Central clearing of cash borrowed overnight against government bonds plateaued in the first quarter of 2026. Two exemptions explain a lot of that. As of May 2026 the SEC, the agency that polices American markets, does not require trades between affiliated firms to be centrally cleared if the affiliate clears its other eligible trades. And the clearinghouse itself does not offer clearing at all for open-ended or flexible-term borrowing without a fixed end date, so those trades are exempt by construction. Open-ended borrowing is the workhorse of the biggest leveraged bet in government bonds: hedge funds borrowing heavily to profit from a small gap between two nearly identical prices. The post-2020 policy fix was to force this activity into a clearinghouse so that a default gets shared across members rather than chased through a chain of one-on-one contracts. The coverage of that fix has stopped growing while the position has roughly doubled its share of the market: regulatory filings for September 2025 show $2.4tn of long positions and $1.6tn of short positions in government bonds, $4tn gross, about 8.5% of privately held US government bonds against roughly 4.5% in early 2023, financed with some $3tn of overnight borrowing. Around 90% of that sits in the 50 largest funds, up from 84%. This is measurement, not new risk. We already count this trade. What is new is the concentration figure and the specific shape of the exemptions. Against it: the Federal Reserve Bank of New York's account of April-May 2025, when roughly $60bn then $40bn of related positions came off, the main leveraged bet held, and overnight funding did not seize. That unwind was a test, and it passed.

Crypto and TradFi contagion26 Aughigh

Legislating for a growth curve that has flattened

The channel by which digital tokens pegged to the dollar could drain bank deposits is being built out by governments at the moment the underlying usage has stopped growing.

The UK Treasury is giving the Bank of England a statutory secondary objective to support innovation in payment systems and digital money, an explicit push to make London a hub for dollar-pegged digital tokens after crypto firms complained the Bank was too conservative (FT). Financial stability remains the primary objective; growth is now the second. Separately, Revolut launched EURR, a euro-pegged token, initially to customers in Denmark, Poland and Portugal (Bloomberg). Here is the awkward fact underneath the policy. The combined market value of stablecoins (digital tokens meant to be worth exactly one dollar) has plateaued in 2026 after years of rapid growth, according to DefiLlama, and Visa's on-chain data, which tries to measure actual usage rather than wash volume, shows transaction volumes cooling this year. Tether and Circle still account for about 85% of tokens in circulation, three years after PayPal launched its own. This matters for the thing regulators actually fear. The systemic worry about stablecoins is deposit substitution: money leaves a bank that lends it into a token backed by government bonds that does not lend it. That transfer is what would disrupt the deposit model the Federal Reserve, America's central bank, will discuss at Jackson Hole this week, whose theme is financial innovation and payments. It requires the tokens to grow. They are not growing. Two readings, and the evidence does not choose between them. Either the risk is smaller than the volume of legislation implies, or growth resumes once incumbents (Revolut, Visa, Klarna, the banks) do the distribution that crypto-native issuers could not. Meanwhile the speculative end is very much awake: bitcoin is at $80,156, up 23.5% in twenty days, and the treasury companies have run harder still (BitMine up 50% and sitting exactly at its high, Strategy up 32%, SharpLink up 36%). That is appetite for borrowed-money bets on price, not for payments.

The AI capex bubble26 Augmedium

The borrowed end is being sold, the funded end is not

The market is already discriminating between AI spending funded by cash flow and AI spending funded by borrowing, before any broad measure of corporate borrowing costs shows stress.

Tuesday's close, and the dispersion inside the AI complex is worth more than any index level. IREN fell 6.2% on the day and 7.6% on the week. Applied Digital fell 6.8%, days after being reported as gauging interest in a roughly $3.5bn bond sale to investors willing to take higher risk for higher interest. Nebius fell 3.6% having upsized its convertible bond (a bond that can be swapped for shares) to $5bn. Core Scientific fell 3.4% and is 27% below its period high. Talen is down 4.9% on the week and 19% off its high; Vistra is 17% off. Against that: Microsoft sits 1.75% off its high, Oracle rose 2.8% on the day, Meta rose 1.1%, and the Nasdaq 100 was flat at -0.08%. The VIX, the market's gauge of expected turbulence, printed 14.88, the lowest reading in our trail. The split runs along a clean line. What is being sold are the firms whose next gigawatt depends on issuing new debt: high-risk bonds, convertibles, project finance secured on GPU chips. What is being held are the firms that can fund from the cash their businesses already generate. This is the market pricing financing risk, not AI demand risk, and it is doing so while the broad market shows nothing at all. The bond market is saying the same thing in the interest rates it demands. QTS sold $3.9bn of bonds for a Microsoft-tied Georgia facility at about 7.23% despite having investment-grade ratings; BlackRock paid 7.53% in July on blue-chip paper for a Texas project (Bloomberg). Investors are demanding the kind of compensation normally reserved for shaky borrowers, on debt issued by safe ones, in this sector. And they are getting it. If that persists, the buildout concentrates into fewer, better-capitalized hands. Whether that is de-risking or the kind of concentration this site keeps warning about depends on whether the periphery's debt gets refinanced or defaulted.

What would change our mind

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

The 0.55 percentage point premium on AI-company bonds widening past roughly 0.80 percentage points while borrowing continues at the same pace. That would mean the borrowers are paying any price rather than the market comfortably absorbing the paper.

Would move the number

A second non-bank firm of Jane Street's scale disclosing AI-related losses, or any sign that the banks lending to these funds are cutting the amount they will lend against the same positions.

Would move the number

Treasury actually drawing on its checking account to fund buybacks, particularly if the thirty-year rate fails to respond. That would be a policy error visible in the price rather than in commentary.

Would move the number

A fund that lends to mid-sized private companies suspending withdrawals outright rather than rationing them, or loans that have stopped paying interest crossing 5% of cost at the ten largest such funds, with dividend cuts to follow.

Would move the number