Blue Owl lends IREN $2.4bn at 9% to buy Nvidia chips

Archived reading, published Fri, 28 Aug 2026 22:21:57 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

Ignition36

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 31 sources and rewrites this page.

Why it moved: Up one to 62, on ignition. Two things moved. First, Blue Owl and Pimco lent IREN $2.4bn at 9% against graphics processors that have to be paid back in two and a half years. That is private credit crossing from lending against businesses that earn money into lending against equipment that loses value every year. Second, the most leveraged corner of the AI trade sold off for a second day running, with Nvidia itself down 4.66% while Microsoft, Amazon and Alphabet rose. The extra interest that riskier borrowers pay over the government did not budge (the gap for the shakiest companies sat at 2.63 percentage points, for safer ones 0.79 percentage points, and the VIX, the market's gauge of how much turbulence traders expect over the next month, read 14.43, which is calm). So this is still some stocks falling while others rise, not the whole system under strain. Fragility holds at 88 because the new chip-backed loans are small against a pile we already count.

Reporting from 27 Aug to 28 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.

Private credit and BDCs28 Aughigh

Direct lending against chips, not cash flows

Private credit is moving from lending against businesses to lending against hardware, on loans that come due before the hardware has paid for itself.

Blue Owl funds, with Pimco alongside, have led a $2.4bn debt deal for IREN, split between a $1.2bn loan and $1.2bn of notes, both secured against the company's assets, both at 9%, both due two and a half years after the money arrives. The money buys Nvidia Blackwell Ultra graphics processors for IREN's Mackenzie campus in British Columbia. Blue Owl's Kurt Tenenbaum described it as "a combination of equipment financing for GPUs and a direct lending underwrite for Iren" and one of the largest of its kind (Bloomberg). This is a different animal from most private credit. A normal loan from one of these funds is secured against a company that earns money. This one is secured, in large part, against hardware whose resale value depends on whether the next generation of chips arrives early or late. Three things have to stay true for the lender to get paid back: IREN's contracted revenue shows up, the chips hold their value if it does not, and the borrower can borrow again to pay this off in early 2029. IREN told the market on Thursday it expects up to $30bn of spending on building things in the 2027 financial year. Its shares fell 12.6% on Friday and are down 15.4% over five days. The filing does not say how much the lender would get back if it had to seize and sell the chips, or what happens to that security if a new chip generation makes these ones obsolete before the loan comes due. Blue Owl manages $319bn and has already financed data centers for Meta. Across the ten largest listed funds that borrow money, lend it to mid-sized private companies, and pass the interest to shareholders, the share of loans where the lender has stopped counting the interest (because it stopped arriving) reached 3.95% in the second quarter, up from 3.75% in the first. The industry is adding a new, faster-depreciating kind of asset to a book that is already slowly getting worse.

The AI capex bubble28 Augmedium

Lambda borrows $1bn to buy chips Microsoft will rent

The strongest set of books in the chain is the one carrying none of the debt.

Lambda, an AI cloud provider backed by Nvidia, has raised about $1bn of short-dated private debt arranged by JPMorgan and placed with private investors. The money buys Nvidia graphics processors. The processors will be leased by Microsoft (Bloomberg). It follows a separate $926m loan that closed earlier this month for the same purpose, and comes while Lambda is in talks for a pre-IPO round of up to $3bn. Trace the circle. The chip vendor holds a stake in the buyer. The buyer borrows in the private market to purchase the vendor's product. The end user, one of the best-capitalized companies on earth, takes the equipment on a lease rather than buying it, so the asset and the debt sit on the small company's books and not on Microsoft's. Each party has a rational reason for its position. The combined effect is that the credit risk of a giant tech company's computing expansion is held by private investors, not by the giant tech company. What the reporting does not tell us is the shape of the Microsoft lease: how long it runs relative to the debt, whether Microsoft has to pay regardless of whether it uses the chips, and whether it can walk away. That single fact determines whether the lenders are really exposed to Microsoft or to Lambda. Nvidia has separately withdrawn the guarantee program that used to sit behind deals like this, where the chip maker would promise to rent the equipment back if the buyer could not. So on current evidence there is no vendor safety net here. More than $400bn of debt has been raised globally for AI this year. The share of it that ultimately rests on one lease clause is the number nobody has.

Private credit and BDCs28 Augmedium

Two banks stop selling Mark Walter's annuities

Insurance-funded private credit shrinks from the funding side first, and this is the funding side.

Truist and Fifth Third have paused distribution of products tied to Delaware Life Insurance, the largest insurer controlled by Mark Walter's TWG Global, while a US probe of Walter's empire proceeds (Bloomberg). The banks were acting as sales channels through branches and advisor networks; the insurance and market risk sat with the insurer. The background is a June disclosure that TWG's two insurers held more than $20bn of loans that should have been marked as affiliated but were not. S&P and Fitch, two of the three firms that grade how likely a borrower is to pay you back, cut their outlooks on Delaware Life in July. S&P held an A- rating while flagging that it could go lower. TWG says there "has been no fraud" and "no victim here". The mechanism worth understanding is the direction of the money. A company that sells policies paying a saver a monthly income for the rest of their life takes the premiums and invests them in longer, harder-to-sell credit, increasingly credit created by a manager it is affiliated with. The policy sales are the funding leg. Cut the distribution and the investment side stops growing; it does not immediately shrink, because these policies are stickier than bank deposits and usually carry penalties for cashing out early. There is no reporting of policyholders rushing to cash out here, and we should not imply any. But this is the first thing in weeks that looks like a lender pulling back rather than a number being revised. Insurance company books are the largest single destination for the risk that left the banks, and the way that pipe narrows is exactly this: a distribution partner declines to keep selling while a rating is under review.

The dollar, gold and reserve status28 Aughigh

Japan has spent $170bn buying its own currency

The largest foreign holder of US government bonds is selling them to defend its currency, and a domestic interest rate near 3% gives its savers a reason to stop sending money abroad.

Japan's Ministry of Finance disclosed on Friday that it spent ¥15.39tn ($96.5bn) between July 30 and August 26 supporting the yen, the largest single round on record. Adding April and May, the 2026 total is ¥27.13tn ($170bn), the biggest annual intervention ever, nearly double 2024's ¥15.3tn (Nikkei). Here is why this belongs in a US newsletter about financial risk. To buy yen, Japan must sell dollars. To get dollars, it sells US government bonds. Japan's holdings of those bonds stood at about $1tn in June, down 16% from a $1.25tn peak in 2021. Every yen-buying operation is, mechanically, new supply dumped into a market where the interest rate the government pays to borrow for thirty years has been near 5.3% and the Treasury Department has doubled its buyback operations to at least $4bn each to hold rates down. Which explains the otherwise odd fact that Washington joined Tokyo in a joint yen-buying intervention on July 31, the first in 28 years. That is not routine. Governments almost never coordinate currency operations. Supporting the yen is a way of stopping Japan from having to sell more US bonds. Two governments are jointly managing one price to protect another. The second leg is quieter and matters more. The interest rate on Japan's own ten-year government bonds has gone from 0.4% to about 2.9%, a 30-year high. Japanese investors sold a net ¥2.8tn of foreign bonds and stocks last week, after two weeks of buying more than ¥2.5tn each. August as a whole is still net buying. One week is not a trend. But the decades-long arrangement whereby Japanese savings funded everyone else's bond markets only works while there is nothing worth buying at home.

Hidden leverage and shadow banking28 Aughigh

Convertibles with no coupon and no cushion

Investors are financing the AI buildout by selling insurance against volatility and taking no income for it, which works until volatility arrives.

Global issuance of convertible bonds (bonds that can be swapped for the issuer's shares if the stock rises enough) has hit $147bn this year, more than 50% above the same point last year, and $72bn of it pays no interest at all, already within a whisker of the $73bn full-year record set in 2025 (FT). Bloomberg reports that a standard measure of how stock-like the convertible market has become is at its highest since 2021 (Bloomberg). A convertible is a bond plus the right to buy the issuer's shares at a set price. The more volatile the shares, the more that right is worth, and the less interest the issuer has to pay to sell it. Nicolas Crémieux of Mirabaud put it plainly: "The companies coming to this market now are far more volatile than the ones they replaced. If you're selling something more valuable, you can offer less for it." At zero interest, the investor's entire return is the bet on the stock. There is no income to cushion a fall in the shares. If the stock goes nowhere for three years, the holder has lent money for free; if it falls, they own an unsecured claim on a company that chose this market because borrowing the normal way was expensive. ON Semiconductor sold $1.3bn at 0% in May, Ciena almost $2.9bn at 0% in June. The last time the market looked this stock-like, the issuers were Peloton and Beyond Meat. The defenders' argument, that 2021 broke because of interest rate hikes, not structure, is worth taking seriously, but the odds of a September rate hike are now above 50%. And the concentration is invisible one deal at a time: a portfolio of thirty convertibles can be one bet on AI volatility.

Crypto and TradFi contagion27 Aughigh

The dead model is up 37% in a month

A structure can be both discredited and still capable of moving 37% in a month; the borrowed money does not go away because the story did.

The FT published an analysis this week finding that the 50 largest companies that hold bitcoin as their main treasury asset have gone from a combined $150bn of market value in July 2025 to $67bn; that 43 of the 50 now trade below where they were before announcing the pivot, with 35 down by at least half; and that Strategy alone accounts for $79bn of the decline from its peak. One researcher called the frenzy "dead" (FT). The market disagrees, at least about direction. Over the last twenty days Strategy is up 36.5%, Bitmine up 37.7% and SharpLink up 33.9%, against bitcoin up 19.3%. On Friday all three fell about 7% while bitcoin fell 3.5%. Both things are true and the reconciliation is the point. These companies are not a business model. They are a leveraged bet on a coin, and roughly a two-times leveraged bet at that, in both directions. That is precisely what the FT's numbers describe: a structure that works while the shares trade above the value of the coins, because selling new shares adds value, and collapses when they do not, because the only way to keep going is to sell coins or issue debt. The $83bn of destroyed value is what happened when the premium vanished. The 37% bounce is what happens when the coin moves 19% and the leverage is still there. So the honest reading is not that the market has refuted the FT. It is that the sell-off the FT documents has left behind a smaller, still-leveraged instrument, and $67bn of it is a fraction of the $150bn that has already been written down without incident.

Signs of the times

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

The miner on the Warriors jersey

IREN, a Sydney-based bitcoin miner turned AI landlord whose logo appears on the Golden State Warriors' jerseys, borrowed $2.4bn at 9% this week to buy Nvidia chips and told investors it may spend up to $30bn on building things in the 2027 financial year. Its shares fell 12.6% the same day.

neocloudsgpu-financingleverage

28 AugBloomberg

Zero percent, like Peloton

Companies have sold $72bn of convertible bonds paying no interest whatsoever this year, near the full-year record set in 2025. The last time the convertible market was this stock-like was 2021, when the marquee zero-interest issuers were Peloton and Beyond Meat.

convertiblesvolatility2021-rhymes

28 AugFinancial Times

$170bn to buy your own money

Japan has spent ¥27.13tn ($170bn) buying yen so far in 2026, the largest annual intervention on record and nearly double the whole of 2024. Its US government bond holdings are down 16% from their 2021 peak, to roughly $1tn, which is why Washington helped.

interventiontreasuriesrecords

28 AugNikkei Asian Review

A$3.3bn, not counting the deposits

Administrators to collapsed Sydney developer Bathla put creditor claims at roughly A$3.3bn, a figure that excludes deposits paid by thousands of apartment buyers across hundreds of projects, and warned the company would burn A$40m before year-end if it kept trading.

propertyinsolvencyaustralia

27 AugAustralian Financial Review

The rumour mill

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

Confirmed

“Marvell signed a deal to help build Google's custom TPUs, and Google received a warrant for nearly 59 million Marvell shares.”

Marvell's own filing puts it at a warrant over 58,970,907 shares at $206.58, vesting largely on Google's purchases through fiscal 2033. The supplier is paying the customer in equity to secure the order, the same circularity as a vendor lending money to its own buyer, just settled in shares rather than debt.

Claimed by Meet Kevin

Partly true

“Nvidia wants to give OpenAI a $125bn guarantee.”

The announced figure is $105bn, after talks that had reached as high as $250bn and were revised to under $120bn. The number circulating is close enough to sound authoritative and wrong enough to matter; the guarantee is real and large, the $125bn is not a reported figure.

Claimed by Wealthion

Partly true

“Money market funds now hold a record $8.2tn, mostly invested in T-bills.”

ICI's weekly figure was $7.93tn in mid-August, with other measures ranging to $8.4tn, so "record, roughly $8tn" is fair. The "mostly T-bills" half is not established. Recent data show funds reducing their holdings of short-term government bills, not concentrating in them.

Claimed by Heresy Financial

Confirmed

“24.9% of America's workforce is now functionally unemployed — jobless, involuntarily part-time, or earning under $26,000 a year.”

That is the Ludwig Institute's True Rate of Unemployment for July 2026, up from 24.7%, and it is a real published series with a stated methodology. It is not the official unemployment rate and is not measuring the same thing; used carefully it is the best available read on the low end of the household credit beat.

Claimed by Michael Bordenaro

Earlier developments

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

Fed, Treasury and policy28 Aughigh

The Fed chair says rates may need to go higher

Nearly every floating-rate loan in private lending and every AI construction project was priced on the assumption that rates were heading down. The new Fed chair just made a rate increase the market's best guess.

Kevin Warsh, in his first speech since becoming Federal Reserve chairman in May, told the Jackson Hole conference that "we must be confident that underlying inflation is moving to our objective, clearly and at sufficient speed. Otherwise, we have work to do." He added that borrowing conditions across the economy are not currently tight enough to slow inflation, and that interest rates are the Fed's "predominant tool." The numbers he cited: the Fed's preferred inflation measure running at 3.7% over twelve months and 4.1% over six, with 54% of the items in the index above 3% over a year and 49% over six months. The interest rate on two-year government bonds rose as much as 0.09 percentage points to 4.32%, while the rate on bonds that do not come due for thirty years fell slightly, according to Bloomberg. The implied odds of a rate increase in September went above 50%, up from around 36%. The next consumer-price report lands September 11; the Fed meets on the 15th and 16th. The mechanism worth holding onto is not the rate increase itself. It is that almost every structure we track was built on the assumption that the next move was down. Loans made by private lending funds charge a floating rate: the borrower pays a fixed margin on top of a benchmark rate (currently 3.64%) that moves with the Fed. A rate increase raises the interest payment on the same loan books where borrowers have already stopped paying at the highest rate in nine years. It improves the lender's income on paper and worsens the borrower's ability to pay in practice, at the same time. Debt financing AI data centers is being priced off today's interest-rate curve for assets meant to last fifteen years. And it collides with the Treasury Department, which borrows the money the government spends. Treasury Secretary Scott Bessent is buying back government bonds that do not come due for decades, trying to hold those long-term rates down, while the Fed prepares to push short-term rates up. The gap between two-year and ten-year rates sits at just 0.47 percentage points. There is not much room left in the curve to absorb a disagreement between the two arms of the state about which direction rates should go.

The dollar, gold and reserve status27 Aughigh

The scarcity trade reversed in a week

The most popular hedge against the government quietly debasing the dollar turned out to be a bet on the Fed staying passive, and it lost its footing in two days.

Gold and bitcoin funds that trade on the stock exchange pulled in a record $7bn over five trading days: $3.4bn into the largest gold fund (SPDR Gold Shares) and $1.5bn into the largest bitcoin fund (iShares Bitcoin Trust). Both ranked in the top ten among all US funds by money flowing in that week, with the gold fund trailing only a handful of funds including the main Vanguard index tracker, Bloomberg reported. The trigger was Bessent's plan to double the Treasury's purchases of its own long-dated bonds: investors read it as the government reaching for the interest-rate curve, and bought the two assets whose supply sits beyond its reach. Then it went the other way. Gold is down 3.4% over five days and 1.9% on Thursday alone. Silver is down 4.0%. The dollar index, which tracks the dollar against a basket of other currencies, is up 0.9% on the week. Bitcoin fell 3.1% in a session, though it remains up nearly 20% over twenty days. The trade was a joint bet on two things: that the Treasury would lean on long-term rates, and that the Fed would let it. Warsh's speech removed the second leg. A central bank willing to raise rates is a central bank that will not quietly print money to cover the government's borrowing, and the hedge against that scenario loses its point. The companies built around these assets moved harder than the assets themselves, as they always do. Strategy fell 7.3%, SharpLink 7.3%, BitMine 6.3%, Galaxy 7.4%, and Coinbase 7.4%, roughly twice bitcoin's own move. That is the arithmetic of holding something volatile inside a company that has also borrowed money: the swings run in both directions, and they were flattering on the way up. Nothing here is a break. It is a reminder that the fastest-growing hedge in the market is itself a bet, made with borrowed money, on a policy assumption that one speech can change.

Private credit and BDCs28 Augmedium

Loans going bad at a nine-year high; the lenders' shares near theirs

Six straight quarters of rising loan failures at the largest private lending funds, and their shares are priced as though the cycle has not started.

The ten largest publicly traded funds that lend to mid-sized private companies (known as BDCs) ended the second quarter with 3.95% of their loans, measured at original cost, no longer producing interest, up from 3.75% the previous quarter. That is $3.3bn across 101 borrowers against an $83.6bn loan book, with $772m of interest income at risk. Count every loan to every borrower that has at least one loan in trouble and the figure is 5.95%, up 0.54 percentage points in a quarter. Across the twenty largest funds, the median is the highest since 2017. The individual moves are the more telling part. A Jefferies tally puts Ares Capital at 2.4% against 1.5% in early 2025, Blue Owl at 2.8% from 1.4%, Golub at 2.9% from 1.2%, and Blackstone Secured Lending at 3.6% from 0.3%. That last one is a twelvefold rise from a very low base, over five quarters. Now look at the shares. Ares Capital is 1.26% off its recent high and up 6.2% over twenty days. Blackstone Secured Lending is up 7.3%, FS KKR 15.5%, Blue Owl's manager 19%. Nothing in the way these shares are trading suggests investors have read the filings. We do not think that is irrational so much as it is a statement about what these shares actually trade on: whether the dividend is covered. Coverage is comfortable because the benchmark interest rate is high, and these funds earn more when rates are high because their loans are floating-rate. But the benchmark rate is also what is pushing borrowers into trouble, because the borrower pays whatever that rate does. A rate increase widens the gap between the two things the share price depends on. It is also worth knowing that the decision to stop counting interest on a loan is a judgment made by the same manager who decides what the loan is worth on its books. The number tells you when they concluded the interest would not arrive, not when it stopped arriving.

The AI capex bubble28 Augmedium

Selling AI project debt into a falling market for AI builders

The next round of AI infrastructure debt is being sold against assets that only hold their value in the world where the debt is never tested, and the safety net from the chip maker was pulled last week.

IREN fell 13.7% on Thursday and 16.4% over five days, now 22% below its recent high. Core Scientific fell 7.2% and is down 20.9% over twenty days, 32% off its peak. Applied Digital fell 7.3%. On the same day, Microsoft closed at its high, Amazon rose 3.5%, and Alphabet 1.9%. The companies building AI infrastructure are falling while the companies paying for it are rising. Into that market, the financing pipeline: Applied Digital is gauging interest in roughly $3.5bn of bonds from companies too shaky to be considered safe, for its Delta Forge 1 campus in Louisiana. Novva is sounding out $800m for two Nevada campuses. Goldman Sachs is expected to lead a $1.085bn secured bond for a campus leased to CoreWeave, issued through a separate company affiliated with PowerHouse Data Centers and Chirisa Technology Parks. Nebius increased its fundraising to $5bn ($3bn of bonds convertible into shares due 2030, $2bn due 2034) against up to $25bn of building costs this year, and has begun pledging its infrastructure and customer contracts as security after a $775m July loan secured on existing chips and the revenue from one highly rated customer's agreement. More than $410bn has been borrowed this year for AI infrastructure. The order matters. Stock market money funds this build-out first and debt second, and the debt is secured on two things: chips, whose resale value collapses in precisely the scenario where you need to sell them, and a single customer contract. Last week Nvidia paused the program under which it would rent back unsold computing power, the guarantee that gave small cloud operators a residual buyer and made them creditworthy. The new deals are being marketed without it. Bonds from the large cloud companies already pay about 0.55 percentage points more than other corporate debt of similar quality, which a JPMorgan banker called a "material premium." Where Applied Digital's $3.5bn prices, if it prices, is the number to watch.

Hidden leverage and shadow banking28 Augmedium

The SEC asks four banks who lent the borrowed money

If four major banks each financed the same concentrated AI portfolio without seeing the whole picture, the fact that this one was absorbed tells you nothing about the next one.

On August 24 the SEC, the agency that polices American markets, sent legal demands for documents to Bank of America, Citi, Goldman Sachs, and JPMorgan Chase in the Situational Awareness matter, seeking information on the fund's trades and its use of borrowed money. The fund, run by 24-year-old Leopold Aschenbrenner, faced demands from its lenders for more cash and its portfolio was hit by roughly 67%. In an ordinary bad quarter a fund might lose 10% or 15%. Losing two thirds of your money in weeks is not a bad quarter. The Financial Times' account of the fallout fills in the other side. Jane Street had invested heavily in the fund, employed a former Jane Street person there, and held many of the same AI-linked stocks (Sandisk, Nebius, Bloom Energy). The stake and those positions were the main factors behind a $15bn hit to trading revenues and the firm's first losing month in a decade. In an August letter to staff, the firm admitted it was not protected against some of the moves. "When you start seeing the kind of revenue these guys are putting up, the only way that's possible is when your holding period extends as well as the size," Bloomberg Intelligence's Larry Tabb told the paper. "The longer your holding period, the more chance that something you can't forecast occurs." The legal demands ask the right question. The business of lending to trading firms is where hidden borrowed money actually lives: the terms are private, unpublished, and set by competition for the client's business. Archegos, the fund that blew up in 2021, worked because several banks each believed they held a modest position and none of them could see the whole. Nobody has said any bank lost money here; the loss landed on Jane Street's own capital, which is the reason nothing broke and the reason markets have been calm for seven weeks. What the regulators are trying to establish is whether that was structure or luck.

Bond market dysfunction28 Augmedium

Two arms of the state pulling in opposite directions

The Treasury Department is trying to hold long-term borrowing costs down while the Federal Reserve raises the cost of the overnight cash that finances the biggest buyer of those same bonds.

The Treasury Department will double its purchases of its own 10- to 30-year bonds to at least $4bn per operation from September 9 through early November, at least $28bn in total. The announcement came outside the regular schedule, and may be funded from the government's roughly $954bn cash balance at the Fed. The trigger was the interest rate on thirty-year government bonds hitting about 5.34%, its highest since 2007. Treasury buybacks are not unusual in themselves. Doing one without warning, outside the schedule that was supposed to contain it, is. The buyback bought a one-day rally; within 48 hours the ten-year rate was back above where it started. Three former senior Treasury officials have now said publicly they cannot see the case. "I do not think there were any meaningful signs of illiquidity in the market, and it is not clear what bar was met that triggered the expansion of the buybacks at this time," Citi's Nathan Sheets, a former under secretary, told Nikkei. "It was an unusual action. I did not find it shocking, but I found it puzzling." As of Friday the short-term end of the interest-rate curve is being repriced upward by the Fed while the Treasury buys the long-term end. Our own two-year series lags to August 26 at 4.19%, below the 4.32% Bloomberg reported during the day after the speech. The curve is flattening from both ends. The part nobody controls is who owns the long-dated bonds. Hedge funds carry roughly $4tn of total government bond exposure, including an estimated $830bn in a trade that profits from a small gap between two nearly identical prices (the cash-futures basis trade), nearly double its 2020 peak. That position is financed with about $3tn of cash borrowed overnight against bonds put up as security, with 90% of the exposure concentrated in fifty funds. A higher policy rate is a higher financing cost on the largest buyer at the margin of the government's own long-term debt. Japan, meanwhile, has spent ¥27.13tn buying yen this year, partly by selling the US government bonds it holds.

The AI capex bubble28 Augmedium

The AI tape splits along balance sheets

The first thing to break in a vendor-financed buildout is the middleman who owns the assets and none of the cash flow, and equities are now pricing that layer separately from the rest.

Friday's session was flat at the index level — the S&P moved 0.03%, VIX 14.59 — and violent underneath it. IREN fell 9.7%, Nebius 4.9%, Applied Digital 4.7%, Core Scientific 4.0%. Over five days IREN is down 12.6% and Core Scientific is 29% below its recent high. In the same session Microsoft closed at its period high, Amazon rose 1.5%, Alphabet 1.0% and Meta 1.7%, taking it up 5.6% on the week. The dividing line is not who builds data centres. It is who pays for them. The hyperscalers fund from operating cash flow and investment-grade bonds — expensive bonds, at roughly 55 basis points over other US corporate credit, which JPMorgan's Matthias Reschke calls a "material premium", but bonds that clear (Bloomberg). The neoclouds fund with secured paper against the machines. Nebius upsized its raise to $5bn this week — reported as $3bn of convertibles due 2030 and $2bn due 2034 — after a $775m July loan backed by existing GPUs and the cash flows from one investment-grade customer, and has begun collateralising infrastructure and customer contracts for further borrowing. Applied Digital is sounding out roughly $3.5bn of high yield for its Delta Forge 1 campus in Louisiana. That model needs two things: a residual value for used GPUs, and a buyer for compute nobody has yet contracted. Nvidia's paused rent-back programme was, briefly, the answer to the second (Reuters). It is gone. The equity market appears to be marking that withdrawal before the credit market does: high yield sits at 267bp and has tightened 7% in twenty days.

The AI capex bubble28 Aughigh

SoftBank's fourth financing for one equity stake

This is margin lending against a private, unmarked equity stake, wearing the clothes of ordinary corporate finance.

SoftBank is in the market for a $10bn two-year loan at around 275 basis points over SOFR, with Mizuho as lead arranger and commitments due 31 August. With SOFR at 3.64%, that is roughly 6.4% money. Proceeds would partly repay the $40bn bridge loan taken in March (Bloomberg). Count the layers. A $40bn bridge. A possible $10–20bn bond sale. A record ¥1tn ($6.3bn) retail bond offering to Japanese savers. Now a $10bn two-year facility. All of it to fund close to $65bn of equity in OpenAI by October. The structure is the point. SoftBank is borrowing on short and medium tenors, at floating rates, to hold an illiquid stake in a private company that pays no dividend and has no market price except the one set at its last round. Interest is due whatever OpenAI is worth. Refinancing has to be available every two years whatever the AI credit mood is then. A bridge is meant to be crossed once; this one is being refinanced with instruments that themselves need refinancing. The funding base is the second thing worth noticing. A retail bond in Japan sold to households, at a moment when the 10-year JGB yield is near 3% and every domestic saver has a rising risk-free alternative, is a materially harder sell than it was two years ago — and it puts Japanese retail money one step from an American AI lab's equity value. The Bloomberg report does not say what happens if the bond market shuts, and neither does SoftBank. Banks are already in talks to join the loan, which tells you the appetite is there today.

Hidden leverage and shadow banking28 Aughigh

A $15bn loss that nobody had to disclose in advance

The largest concentrations of AI risk now sit in balance sheets that publish nothing until after the loss.

Jane Street lost money over a month for the first time in a decade. The FT puts the hit at $15bn in trading revenues, driven by AI-linked positions and by its stake in Situational Awareness, the hedge fund run by 24-year-old Leopold Aschenbrenner, which was forced into a sale of its positions in July (FT). Jane Street was exposed to many of the same names Aschenbrenner owned — Sandisk, Nebius, Bloom Energy. In an August letter to employees the firm said it was not protected against some of the moves. The mechanism is a firm changing shape without changing category. Jane Street is still described as a market maker; it is now one of the largest financial institutions in the US, holding directional positions over long horizons. "When you start seeing the kind of revenue these guys are putting up, the only way that's possible is when your holding period extends as well as the size," Larry Tabb of Bloomberg Intelligence told the FT. "The problem with that is your clarity on the future declines." The same AI factor arrived by three routes at once: the trading book, the seed investment, and — reportedly — prime brokerage. A market-commentary sweep says the SEC sent subpoenas on 24 August to Bank of America, Citi, Goldman Sachs and JPMorgan seeking information on Situational Awareness's trades and use of leverage, with the fund's portfolio down about 67%. Treat that as second-hand until a paper of record carries it. One thing cuts the other way, and it should be said plainly: the system took a $15bn hit at a systemically-sized firm seven weeks ago with no funding event, no counterparty failure and no widening in credit. That is a mark that was tested and held.

Bond market dysfunction28 Aughigh

Lending to AI companies for no interest at all

The AI complex is now funding itself with the instrument that offers investors the least protection when the story stops working.

Companies have sold $72bn of zero-coupon convertible bonds this year, according to Dealogic, against a full-year record of $73bn set in 2025 (FT). Total convertible issuance is $147bn, more than 50% ahead of this point last year, and Bloomberg reports that a measure of how equity-like the market has become is at its highest since 2021 (Bloomberg). ON Semiconductor sold $1.3bn at 0% in May; Ciena sold nearly $2.9bn with no interest in June. The arithmetic is simple and it is not irrational. A convertible is a bond plus an option to swap into shares. The more volatile the share price, the more that option is worth, so the issuer can pay less for it. "The companies coming to this market now are far more volatile than the ones they replaced," Mirabaud's Nicolas Crémieux told the FT. "If you're selling something more valuable, you can offer less for it." What you own if the option stops mattering is an unsecured claim paying nothing. Our own arithmetic: a five-year zero-coupon bond discounted at 5% is worth about 78 cents on the dollar. In 2021 the same trade was done by Peloton and Beyond Meat, and it ended with holders of no-coupon unsecured paper on collapsing equities. The issuers are the tell. This instrument was popularised by Michael Saylor's Strategy, which used it to buy bitcoin; the AI buildout has adopted it wholesale. It is the cheapest financing available precisely because the underlying is the most volatile — which is a coherent trade and a poor cushion.

The dollar, gold and reserve status28 Aughigh

Japan spent $170bn on the yen this year

The marginal foreign buyer of US Treasuries is now a forced seller of them, for reasons that have nothing to do with America's fiscal position.

Japan's Ministry of Finance disclosed on Friday that it spent ¥15.39tn ($96.5bn) supporting the yen between 30 July and 26 August — the largest single round ever. Including April and May, 2026 intervention now totals ¥27.13tn ($170bn), the largest annual figure on record, against ¥15.3tn for all of 2024 (Nikkei). The plumbing matters more than the size. To buy yen you need dollars, and Japan raises dollars partly by selling US Treasuries. Its holdings stood at about $1tn in June, down 16% from a $1.25tn peak in 2021. So the largest foreign owner of US government debt has spent the summer shrinking its position — in the same weeks the US Treasury doubled its long-end buybacks to at least $4bn per operation to hold yields down, and, on 31 July, joined Tokyo in the first coordinated yen-buying operation in 28 years. Read that as one policy. Washington is helping support the yen so that Tokyo does not have to sell more Treasuries, while simultaneously buying Treasuries back. Both arms are pointed at the same variable: the US long end. The flows are turning too. Japanese investors were net sellers of ¥2.8tn ($18bn) of foreign bonds and equities last week after two weeks of buying above ¥2.5tn (Nikkei). August is still net buying overall, so this is not yet a trend — but with the 10-year JGB at a 30-year high near 3%, the structural reason Japanese institutions bought foreign bonds for two decades has weakened. Gold at $4,640 is up 15% in twenty days. The dollar index has barely moved.

Private credit and BDCs27 Augmedium

Australia is running the experiment first

The gate, not the default, is the mechanism that turns a credit problem into a liquidity event — and Australia is demonstrating it in public.

Administrators to the collapsed Sydney developer Bathla say creditors are owed an estimated A$3.3bn — a figure that excludes deposits paid by thousands of apartment buyers across hundreds of projects. Teneo warned that the "scale and complexity" of the group was at odds with the cash available, and that it would burn A$40m before year-end if it kept trading (AFR). Australia's private credit market is around A$200bn — small. What makes it worth watching is that its wiring is identical to the American version, and it is a few months further along. MA Financial has limited redemptions on its A$2.3bn secured property loan fund after a surge in withdrawal requests, moving investors to quarterly limits with only part of each request paid. ASIC has warned of the "first significant cracks", citing the collapse of several large borrowers and limited redemptions at several major funds. A gate is not a default; it is a queue. But it changes the economics for everyone in the fund. The loans behind a property development are carried at a valuation supplied, in effect, by the developer, until an administrator arrives and supplies a different one. Whoever redeems before that happens gets net asset value. Whoever is behind them in the queue gets whatever is left after the marks catch up. That asymmetry is exactly what produces a run in a vehicle that offers liquidity it does not hold. The US equivalent is not yet gated at scale, though sector commentary puts Q2 redemption requests at semi-liquid private credit funds averaging 14% of NAV against typical 5% quarterly limits — a tier-2 figure we would want confirmed. The listed BDCs, meanwhile, are calm: Ares Capital, Blackstone Secured Lending and Blue Owl all within 5% of their recent highs.

The AI capex bubble28 Augmedium

Nvidia withdraws its buyer of last resort

The vendor guarantee was what turned speculative lending on graphics chips into something closer to lending against contracted revenue. Withdrawing it tests whether the smallest AI clouds can borrow on their own credit.

Nvidia has paused the financing program under which it offered small AI cloud companies credit support in exchange for a share of their revenue, stepping back from it last week, less than two months after announcing it. The Wall Street Journal reported the retreat; Reuters carried it. Nvidia's spokesperson says the new business model "is still in place and continues to evolve due to high demand," and that it could be revamped or folded into another program. The mechanism matters more than the size. Under the arrangement, Nvidia would rent computing capacity back from cloud customers if they could not sell it. That single promise changes what a loan backed by graphics chips actually is. A lender underwriting a company with no operating history and a warehouse full of chips that lose value every quarter is making a speculative bet on future demand. The same lender, looking at a contract in which Nvidia agrees to be the buyer if nobody else shows up, is underwriting something much closer to guaranteed revenue, and will lend more, at a lower interest rate, against it. Remove that backstop and the small cloud startups go back to standing on their own order books. What is left is still enormous. Goldman analysts, cited by CNBC, put Nvidia's financial commitments in support of customers at $366bn: $279bn of supply and capacity commitments (largely memory), $29bn of cloud service agreements, and other data-center items. The $500bn financing push with six asset managers and the $105bn Ohio arrangement for OpenAI are untouched. Two readings. Either Nvidia is trimming its contingent obligations under the circular-financing scrutiny it has drawn all summer, or demand is strong enough that it does not need to buy demand. Markets prefer the second: Nvidia closed at an all-time high, up 8.8% on the day. Watch whether the cloud-startup bond pipeline (Applied Digital's roughly $3.5bn offering of debt from a company too shaky to be considered safe, Novva's $800m) prices at a higher interest rate without the guarantee behind it.

The AI capex bubble28 Aughigh

A $45bn lease from a company worth $14.6bn

The AI build is now financed by lending against the promises of private companies whose accounts nobody outside the deal has seen.

Anthropic has agreed to rent 460MW of Nvidia's next-generation Vera Rubin computing capacity at Nscale's West Virginia data center, in a deal the FT puts at roughly $45bn over six years. The chips come online late next year. Nscale raised $2bn in March at a $14.6bn valuation, led by Norwegian energy group Aker, with Nvidia, Dell and Nokia among the backers. The Mason County site is planned at 1.35GW across 2,250 acres, with a 2GW natural gas plant built on site to power it, and Nscale estimates the cost at $69bn. It is preparing an IPO that could come as soon as next month. The arithmetic only works one way round. Nscale cannot fund $69bn from equity. What makes the project financeable is the signed Anthropic lease, a long-dated stream of payments a lender can size a loan against. This is now the standard shape of the AI build: JPMorgan arranging $5bn for Volta after Volta signed a $10bn computing contract with Anthropic; Meta's Hyperion financed through a Blue Owl joint venture that issued $27bn against Meta's lease. The tenant's credit is the collateral. So the question is Anthropic's balance sheet (the list of what it owns and what it owes), and the reporting does not disclose it. Anthropic has separately committed to lease more than $150bn of Google's AI chips, signed agreements with Amazon and SpaceX, held early talks with Meta, and plans an IPO this autumn. The West Virginia site alone implies roughly $7.5bn a year of rent. This is the migration our running thesis describes, in its cleanest form. None of this obligation sits in a bank. It sits in leases, project debt and private capital, and it is underwritten by a revenue forecast for a company that has not yet filed publicly.

Bond market dysfunction27 Augmedium

AI borrowing is now setting the price of Korean debt

The AI build is no longer a sector story. It is driving up the extra interest that lenders demand for tying their money up for a long time, and that cost is being paid by countries with no AI industry at all.

As the interest rate the US government pays to borrow for thirty years climbed to 5.33% last week, its highest since 2007, Korea's thirty-year government bond rate hit a record 4.751%. It has since retreated. The mechanism that carried it there has not. Samsung Securities, quoted by the Korea Herald, breaks the move down. This year's rise in US long-dated rates is about investors demanding more compensation for tying their money up for decades, not about repricing what the Federal Reserve (America's central bank, which sets interest rates) will do next. Three things are pushing on that compensation at once. The US deficit implies heavier borrowing. Foreign official demand is fading: most of a $72bn decline in foreign holdings of US government bonds in June came from central banks and sovereign wealth funds. And six major technology companies issued $182bn of bonds in the first half of the year, fourteen times the year-earlier figure, with more than 30% of it carried at thirty-year maturities. That last number is the one worth sitting with. Data-center spending is being funded in exactly the same maturity bucket as government debt, competing for exactly the same pool of long-term buyers: insurers, pension funds, sovereign funds. There is no separate window for AI paper. When Alphabet or Oracle sells a thirty-year bond, the buyer choosing between it and a long-dated US government bond pushes the price of both. Bloomberg's JPMorgan interview this week put the resulting premium at about 0.55 percentage points over other US corporate debt. The transmission is then arithmetic. US government bonds are the benchmark for global bond pricing; Korea's long-dated rates reprice, then Korean mortgage and corporate borrowing costs follow. A Seoul homeowner is, at one remove, paying for the electricity contracts in Ohio and West Virginia.

Private credit and BDCs27 Augmedium

The number the private-lending funds report and the number that matters

Losses in private lending are disclosed on a lag chosen by the people holding the assets, and the adjusted figure is rising three times as fast as the headline.

Blackstone's BCRED, the largest fund of its kind (a fund that borrows money, lends it to mid-sized private companies, and passes the interest to shareholders), saw loans where the borrower has stopped paying interest go from 0.6% to 2.4% of the portfolio between the first and second quarters, on two names: Medallia and Affordable Care International. That is a quadrupling in three months, from a very low base. Across the ten largest listed funds of this type, loans where interest has stopped arriving reached 3.95% of debt at cost in the second quarter, up 0.20 percentage points on the quarter. But there is a second figure in the same Morningstar analysis that gets less attention and deserves more. Count every slice of every borrower that has at least one loan where interest has stopped arriving, and the exposure is $5.0bn, or 5.95% of total debt at cost, up 0.54 percentage points in the quarter, nearly three times the headline move. That gap is the interesting mechanism. A borrower rarely stops paying on one loan and keeps paying comfortably on the rest. The junior slice goes first, the senior follows, and the reported non-payment rate lags the credit reality by however long it takes the manager to concede. The adjusted number is the better estimate of what is actually impaired, and it is deteriorating faster. The prices say none of this. Ares Capital is 1.6% off its high, Blue Owl's lending fund up 4.4% over twenty days, FS KKR up 16%, Blue Owl the manager up 20%. The one exception is precise: Golub, where the value per share slipped from $14.35 to $14.25 and the count of non-paying loans went from 19 to 20, is the fund that fell this week, down 1.3% over five days. The extra interest investors demand for risky debt is tight, the valuations are the managers' own, and the share prices are priced for none of it. That is the disagreement to hold onto.

The dollar, gold and reserve status27 Aughigh

The rescue was the catalyst for the escape

Gold and bitcoin are now moving as one trade, and that trade is responding to fiscal policy rather than to inflation or the currency.

Exchange-traded funds tracking gold and bitcoin took in a record $7bn over five trading days, per Bloomberg data: nearly $3.4bn into SPDR Gold Shares and $1.5bn into BlackRock's iShares Bitcoin Trust. Both cracked the top ten US ETFs by weekly inflow, with GLD trailing only a handful of funds including the Vanguard S&P 500 ETF. The proximate catalyst, Bloomberg reports, was Scott Bessent's plan to at least double long-dated US government bond buybacks. That is worth stating plainly: the largest recorded five-day flow into the two assets whose supply no government controls was triggered by the Treasury Department (which borrows the money the government spends) announcing it would support its own bond market. For most of the past two years these were treated as alternative trades: gold for institutions worried about fiscal arithmetic, bitcoin for a different crowd. They are now moving as one position. Gold is up 15.5% over twenty days at $4,658. Silver sits at its period high, up 17.3%. Bitcoin is up 21.8% over the same window at $79,030. What is not happening is a dollar crisis. The dollar index (which tracks the dollar against a basket of other currencies) is 99.23, down 0.73% over twenty days; the euro and yen ETFs have barely moved. This is not capital fleeing dollar assets for other currencies: the euro has picked up none of it. It is capital moving out of long-dated bonds and out of claims on future government solvency, and into things with a fixed supply. A monetary historian cited in the same coverage puts the dollar's reserve share at just under 60%, down from a little over 70% at the turn of the century, with roughly three-quarters of the loss going to non-traditional reserve assets rather than to the euro. A repricing of policy credibility, then, not a funding event. Those look different, and it matters which one you are watching.

Fed, Treasury and policy27 Aughigh

Spending the emergency account to buy back bonds

If the Treasury Department is seen to be managing interest rates rather than funding the government, every buyer of long-dated government bonds reprices what a Treasury promise is worth.

Three major outlets wrote the same piece this week, which tells you something on its own. The Economist, a New York Times guest essay by a former deputy chief of staff to Janet Yellen, and Bloomberg Opinion all examined what the Treasury Department is doing with long-dated bonds, and all three landed closer to skepticism than to reassurance. The facts underneath. On August 19 the Treasury said it would at least double buybacks of ten- to thirty-year debt, to at least $4bn per operation from $2bn, taking the two-month total to at least $28bn, announced outside the regular schedule. Buybacks themselves are not unusual. Announcing one outside the published calendar, two weeks after the schedule that was supposed to contain it, is. CNBC has since reported that Bessent is considering funding further buybacks from the Treasury General Account, the government's roughly $954bn checking account at the Federal Reserve, kept deliberately large for emergencies such as shutdowns. Stanley Druckenmiller publicly criticized his former mentee in the Wall Street Journal. Mechanically, a buyback funded from that account is a swap: the Treasury retires long-dated bonds, drains its own cash balance, and adds reserves to the banking system, shortening the government's effective debt maturity without ever announcing a change in borrowing policy. It is not nothing. Citrini Research argues this is one piece of an emerging "Treasury-Fed Accord," with the Fed shrinking its holdings while banks absorb more short-term bills, and recommends betting on thirty-year bonds outperforming five-year notes. Both things can be true. The trade may work and the credibility may still be spent. The Economist notes the Treasury's usual role is keeping the market orderly in crises, not pushing interest rates around in quieter times, and more than half of the US deficit now consists of interest on past borrowing. Markets are currently siding with Bessent: the ten-year rate sits at 4.66%, below where it traded on August 21.

What would change our mind

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

The extra interest that shaky borrowers pay over the government rising above 3.5 percentage points, or the same measure for safer borrowers rising above one percentage point. Credit has not moved through any of this, and until it does the AI sell-off is money rotating between stocks, not the system under stress.

Would move the number

A smaller AI cloud or data-center company pulling a bond deal, or chip-backed loans repricing well above the 9% IREN just paid. That would be the funding channel closing rather than narrowing.

Would move the number

Evidence of policyholders rushing to cash out at Delaware Life or any similar insurer, as opposed to a pause in new sales. A run on the liabilities is a different order of event from a sales freeze.

Would move the number

A September rate hike on September 16 combined with the extra interest riskier borrowers pay actually widening rather than shrugging, which would connect the policy path to the financing of the AI buildout for the first time.

Would move the number