Crashopedia
Everything we keep referring to, explained. This exists so that nothing in our writing has to go over your head: click any underlined term in a dispatch and the explanation comes to you. 57 entries so far, and it grows as the coverage does.
The crash points
The eight places we watch for stress. Everything else hangs off these.
Bond market dysfunctionBond market dysfunction is what happens when the market for government debt stops absorbing new supply smoothly: yields climb, auctions clear cheaper than the market had marked them, and the financing plumbing underneath starts to strain. At the moment only the first half is happening. The US Treasury market is repricing the cost of lending to the government for thirty years, relentlessly but in good order, with gross federal debt past $40tn in August 2026 and annualised net interest payments of about $1.21tn against a defence budget of about $1.17tn. Sitting underneath all of it is a cash-futures basis trade the Federal Reserve sized at roughly $830bn in September 2025, which is where an orderly repricing would turn into a disorderly one.Crypto and TradFi contagionContagion here is the set of mechanisms by which a fall in crypto prices now lands on bank balance sheets, inside ordinary mutual funds and in the market for US Treasury bills — connections that American regulators deliberately kept closed until 2025. The repeal of SAB 121 that January and the GENIUS Act signed in July opened them: banks now lend against bitcoin, and Tether alone reports $114.96bn of Treasury bills in its most recent 2026 attestation, which would rank a stablecoin issuer among the world's largest holders of US government debt. The practical consequence is that the next crypto drawdown no longer stays inside crypto.Hidden leverage and shadow bankingShadow banking is the business of doing bank things — lending, financing other people's positions, borrowing short against long assets — inside institutions that are not banks and therefore hold no bank capital against any of it. Hidden leverage is the borrowing inside those institutions that never appears as debt, because it lives in repo haircuts, derivative margin and bilateral agreements that no single counterparty can see in aggregate. The Financial Stability Board put the global non-bank financial sector at roughly $250 trillion at end-2023, about 49% of all financial assets, up around 130% since 2009. It stays invisible right up to the day everybody asks for their collateral back at once.Household creditHousehold credit is everything American consumers owe — mortgages, car loans, credit cards, student debt — counted up each quarter by the New York Fed, which put the total at $18.771 trillion in Q2 2026. In aggregate it is fine, and slightly better than the quarter before: the share of balances in some stage of delinquency ticked down to 4.7%. Underneath that average the bottom of the income distribution has already broken. Sixty-day delinquency on subprime auto loans reached 6.9% on Fitch and Equifax data, above the 5.0% peak of 2008, and about one credit-card dollar in eight is 90 days or more past due; nobody has panicked because the lenders no longer own most of the risk, having sold it into asset-backed securities to investors who wanted the yield and did not always check what was behind it.Policy, the Fed and fiscal dominanceFiscal dominance is what happens when a government's debt grows large enough that the central bank's rate decisions become the government's funding problem, and price stability starts losing arguments to the borrowing calendar. The United States is currently running a live experiment: three Fed officials voted for a rate hike at the July 2026 meeting, while the Treasury has doubled its purchases of its own long-dated bonds to push long yields down. Federal debt passed $40tn on 18 August 2026, and net interest ran to $963bn in the first ten months of FY2026 against $804bn of national defence spending. Nobody at the Treasury uses the phrase. It is what the arrangement looks like from outside.Private creditPrivate credit is lending to companies by investment funds instead of banks: mostly floating-rate senior loans to mid-sized, often private-equity-owned borrowers, held to maturity by the fund that made them. Its defining feature is that these loans have no secondary market and therefore no market price, so what a loan is worth is whatever the manager who owns it estimates. Estimates of the market's size run from $1.97tn to $3.5tn globally on 2024-2025 data, depending on what you count, which tells you something about how well observed it is. As of the second quarter of 2026 median non-accruals at the twenty largest listed vehicles sit at a decade high of 2.8% of cost, retail investors have asked for a record share of their money back, and the loans themselves are barely marked down.The AI capex bubbleThe AI capex bubble is the hundreds of billions of dollars a year now being spent on data centres, chips and power for artificial intelligence, funded increasingly by debt that sits in special purpose vehicles, footnotes and vendor guarantees rather than on the balance sheets of the firms that have promised to pay. The commitments run well ahead of the revenue: OpenAI carries roughly $1tn of compute commitments against annualised revenue above $40bn as of August 2026. Five US tech giants disclose $1.65tn of off-balance-sheet obligations, more than the $1.35tn of debt they actually report. The market has begun sorting the complex into companies that pay with cash and companies that pay with leases, and it is doing it violently.The dollar, gold and reserve statusThe dollar's reserve status means foreign central banks, companies and investors park their savings in it, which lets the United States borrow cheaply in a currency it prints itself. That status is not vanishing, but it is thinning: the dollar's share of allocated global foreign exchange reserves was 57.13% in Q1 2026, against roughly 71% in 1999-2000, while central banks bought around a thousand tonnes of gold a year from 2022 to 2024. What has been happening through August 2026 is not a run but a repricing, with the cost of managing a difficult bond market being paid in the currency instead of in yields.
Mechanisms
How the machine is actually wired.
10y10y forward yieldsThe 10y10y forward yield is the ten-year bond yield that today's curve implies will exist ten years from now: a rate nobody quotes directly, backed out by arithmetic from the ten- and twenty-year yields that people do trade. Because everything a central bank might plausibly do in the next decade already sits inside the first ten years, the measure strips out the business cycle and leaves the long-run residue — the inflation regime, fiscal credibility, and the premium investors demand for holding very long paper. It is the closest thing to a clean read on what lending to a government for thirty years actually costs. When it rises in a week of soft inflation data, something other than the business cycle is being repriced.2s10s spreadThe 2s10s spread is the ten-year Treasury yield minus the two-year Treasury yield, quoted in basis points. It is the shape of the entire US government borrowing curve squeezed into one number. Positive means long money costs more than short money, which is the normal and boring state of the world. Negative means investors expect the Fed to be cutting hard before long, which historically means they expect something to break.Auction tailAn auction tail is when a Treasury auction clears at a higher yield than the market was quoting seconds before bidding closed, meaning the government had to pay up to find enough buyers. It is measured in basis points, and it is the closest thing the bond market has to a live demand reading: a tail says the bidders who showed up wanted a discount, and the dealers who are obliged to bid ended up owning the rest. A run of tailing auctions is how a slow repricing announces itself as indigestion.Cash-futures basis tradeThe cash-futures basis trade is the business of buying a Treasury bond, selling the futures contract that promises to deliver that bond, and collecting the small gap between the two prices — a gap so thin it only pays a salary if you do it with fifty or a hundred times more money than you have. The borrowed money comes from the overnight repo market, refinanced every morning. Federal Reserve research puts the position at roughly $830bn as of September 2025, about double its early-2020 peak and around 35% of hedge funds' total long Treasury exposure.Circular flow of incomeThe circular flow of income is the observation that one person's spending is another person's income: firms pay wages to households, households buy goods from firms, and the same money goes round again. It is the plainest idea in economics and the one every argument about demand quietly rests on, because income lost in one place does not stay lost in one place. Break the loop anywhere and the damage turns up somewhere else, usually attached to a person who had nothing to do with it.CovenantA covenant is a promise written into a loan or lease document and given legal teeth: keep leverage below this multiple, do not move that asset, pay the rent for fifteen years. Break one and the lender is entitled to demand its money back immediately, which is why covenants are rarely enforced and almost always sold — the borrower buys a waiver with a higher coupon, faster repayment or a tighter set of promises next time. Confusingly, credit people also use "the covenant" to mean the party doing the promising, as in "Microsoft's covenant", meaning Microsoft's creditworthiness.Demand doom spiralA demand doom spiral is the self-feeding loop in which lost wages become lost sales: workers who stop earning stop spending, the businesses they spent at lose revenue and cut staff, and those newly unemployed workers stop spending in turn. Each round is smaller than the last, which is the only good news on offer. It is the ordinary failure mode of a market economy, and interrupting it is most of what post-war macroeconomic policy is for.Directional betA directional bet is a position that only makes money if a price moves the way you predicted, and loses money if it moves the other way. It is the opposite of a market maker's normal business, which is to quote both sides, buy at the bid, sell at the offer, hedge whatever is left over, and earn a spread thousands of times a day without caring where anything goes. The distinction matters because the two activities need completely different amounts of capital, and because a firm doing the second can quietly start doing the first.DurationDuration measures how much a bond or a portfolio of bonds falls in price when interest rates rise. It is quoted in years, which is the most misleading convention in fixed income, because what it behaves like is a multiplier: a duration of ten means roughly a ten per cent price fall for every percentage point that yields go up. It is the number that converts a rate move into a solvency problem, which is why institutions holding long assets against short money keep dying of it.HaircutA haircut is the discount a lender applies to the value of collateral before deciding how much cash to lend against it: post a bond worth 100, get 98, and the haircut is 2 per cent. It is the lender's cushion against the collateral falling in price before it can be sold, and it is the single dial that sets how much leverage a borrower can run. Raise haircuts across a market and every borrower has to shrink at once, which is the polite name for a fire sale.High-yield spreadsA high-yield spread is the extra yield, measured in basis points, that investors demand for holding junk-rated corporate bonds instead of US Treasuries of the same maturity. It is the public bond market's running price on the risk that companies stop paying, quoted to the basis point and updated every day. When the number is low, the market is saying it expects almost nothing to go wrong. The interesting question is always whether the things going wrong are inside the index.Interval fundAn interval fund is a registered fund that holds assets it cannot sell quickly — private loans, property, credit nobody quotes daily — and therefore lets investors out only on a schedule it sets, typically once a quarter and typically capped at 5% of the fund. If more shareholders ask to leave than the cap allows, everyone is cut back pro rata and the rest wait for the next window. The queue is not a malfunction. It is the design.Liability-management exerciseA liability-management exercise is a debt restructuring done outside bankruptcy, usually with only some of the lenders, in which a struggling borrower moves collateral beyond their reach or reorders the repayment queue so the participating creditors come out ahead and everyone else is quietly subordinated. Because the loan documents permit it, no default is declared and no court is involved. The losses are real anyway. They just do not have a label.MarkA mark is the value an asset is recorded at on someone's books. If the thing trades on an exchange, the mark is the last price it changed hands at; if it does not — a private loan, a data-centre bond, a pool of lawsuit advances — the mark is an estimate produced by a model built by the people who own the thing. The gap between those two situations is where a great deal of this crash lives.MOVE indexThe MOVE index is the bond market's fear gauge: a measure built from options prices on US Treasuries of how much traders expect yields to swing over the coming month. It is the VIX for interest rates, and because Treasury yields are the price of everything else, a spike in it tends to matter more than a spike in equity volatility. It measures expected movement, not direction — it can sit at the floor while yields grind relentlessly higher.NAV loanA NAV loan is money lent to a private equity or private credit fund secured against the fund's entire portfolio of holdings, valued at whatever the fund's own manager says they are worth. It lets a fund that cannot sell anything raise cash anyway, usually to pay distributions to investors who were promised them years ago. The collateral is not a building or a bond you can price on a screen. It is a spreadsheet of marks, cross-collateralised, and senior to everyone who put equity in.Non-accrualA loan goes on non-accrual when the lender formally stops booking the interest it is owed as income, because the borrower has stopped paying or is expected to stop. It is one of the few near-binary facts in a business where almost every other number is an estimate produced by the person paid to manage the asset. Across the twenty largest listed business development companies, the median reached 2.8% of cost in the second quarter of 2026, the highest since 2017.Option-adjusted spread (OAS)Option-adjusted spread is the extra yield a bond pays over the risk-free Treasury curve once you strip out the value of any options buried in the contract, such as the issuer's right to call the bond early and refinance the moment it suits them. It is the standard like-for-like measure of how much a lender is being paid to take credit risk, and the average OAS across a bond index is the market's running blood-pressure reading on corporate credit. When a single deal prices at several times the index OAS, the market is saying it disagrees with the rating.Payment-in-kind (PIK) incomePayment-in-kind income is interest a lender books as revenue even though no money arrived, because the borrower settled it by adding the amount to what it owes instead of paying it. The loan gets bigger, the income statement looks exactly as it would if a cheque had cleared, and the cash question is postponed until maturity. It is entirely legal, sometimes entirely sensible, and it is also the most convenient place a struggling loan can hide.Repurchase agreement (repo)A repurchase agreement is a cash loan secured by a bond, structured as a sale plus a promise to buy it back, so that if the borrower fails the lender already owns the collateral instead of queuing for it in bankruptcy. It is the plumbing under almost all modern financial leverage: buy a Treasury, repo it for cash, buy another, repeat until the arithmetic gets frightening. Two things decide how much leverage the system can carry and how fast it comes off — the size of the haircut, and whether the trade is centrally cleared or done privately between two parties who tell nobody.Residual valueResidual value is what an asset is worth once the contract paying for it runs out: the building after the tenant's lease expires, the hardware inside it after two more chip generations. In data-centre finance it is the whole argument, because a lender is underwriting Microsoft's rent for a fixed term and then a purpose-built shed in Georgia for everything after that. Nobody knows what the second part is worth. No AI data centre has yet reached the end of its first lease.Residual-value backstopA residual-value backstop is a promise by a third party to pay the difference if an asset is worth less than an agreed floor at some future date. It is a put option written on the second-hand value of a building, an aircraft or a data centre, and it exists so that lenders can underwrite the guarantor's credit instead of the tenant's. In the AI buildout it is the structure holding up projects nobody would otherwise finance.Scalable Human EquivalentA scalable human equivalent is an AI system that has become good enough at a specific job, at a low enough price, to do that job instead of the person — and which, being software, can then be copied without limit and without hiring anyone. It is our term for why AI displacement should arrive as a step rather than a slope: capability accumulates quietly, but the decision to switch is made once and applies to the whole department on the same morning.Secured Overnight Financing Rate (SOFR)SOFR is the interest rate at which the financial system borrows cash overnight against US Treasury collateral, published each morning by the New York Fed as the volume-weighted median of the previous day's actual repo trades. It is the benchmark under trillions of dollars of loans, floating-rate debt and interest-rate swaps, having replaced LIBOR, which was famously based on what banks said they would pay rather than what anyone paid. Because it is built from real overnight funding, it is also the cleanest single reading of whether the plumbing is working.Significant risk transferA significant risk transfer, or SRT, is a trade in which a bank pays outside investors to absorb the first losses on a pool of loans it keeps on its own balance sheet, which lets it hold less regulatory capital against those loans without selling them or telling the borrowers anything happened. The loans stay exactly where they are; only the risk moves, and it moves to private credit funds and alternative asset managers, some of whom are financed by banks. Banks globally issued €30 billion of new SRT tranches on €378 billion of loans in 2025, according to an IACPM survey. The structure dates to JPMorgan's 1997 BISTRO deal, died in the financial crisis, and came roaring back after Europe extended favourable treatment to synthetic deals in 2021 and the Federal Reserve clarified the US capital rules in September 2023.Special purpose vehicle (SPV)A special purpose vehicle is a company created to do exactly one thing, usually own a single asset and borrow against it, with its own accounts, its own creditors and a legal existence separate from whoever set it up. Because the debt belongs to the vehicle rather than the sponsor, it need not appear on the sponsor's balance sheet, even when the sponsor has promised the payments that will service it. This is how a company can commit to twenty years of fixed obligations and still report an unremarkable debt load.Sponsored repoSponsored repo is a repo trade in which a bank sponsors a client, usually a hedge fund, into the central clearing house for Treasury repo, so that the clearing house rather than the bank becomes the counterparty of record. The bank still guarantees the client's obligations and still collects a fee; what it gets rid of is the balance sheet, not the risk. The point is netting, and netting lets a dealer finance far more client leverage per unit of capital. It is the plumbing that makes very large, very cheap Treasury financing possible.SubordinationSubordination is the rule that decides who eats a loss first. In a securitisation, the deal's junior slices absorb every dollar of loan losses until they are wiped out, and only then does the slice above them lose anything — which is how a pool of loans to borrowers with bad credit gets turned into a bond rated AAA. It is also why a subprime lender's shareholders can watch delinquencies pass their 2008 peak without visible distress: the loss is landing on somebody else's tranche.Syndicated loan marketThe syndicated loan market is where a very large corporate loan gets arranged by a bank and then sold in slices to dozens or hundreds of institutional investors, instead of one lender holding the whole thing. It is the public, tradeable end of corporate borrowing: the loans get quoted, priced and marked daily by people who did not write them. That daily mark is why a borrower who shops a deal to private credit and then comes back here is telling you something.Tail riskTail risk is the risk of the rare, very large loss: the outcomes far out at the thin end of the probability distribution, the ones that almost never happen and ruin you when they do. The thing to understand about a tail is that it never actually disappears, it only moves. When an insurer declines to write the policy, the risk does not vanish; it relocates to whoever owns the asset, or the debt secured against it, unpriced and unreserved.Term premiumThe term premium is the extra yield investors demand for lending long rather than rolling over short-term paper: the slice of a thirty-year yield that is not a forecast of central bank policy but a fee for committing money for thirty years and not knowing what happens in them. Nobody can observe it directly — it is what is left after you subtract an estimate from a fact, which is why every model gives a slightly different answer. When it rises, every asset priced off a long discount rate reprices at once, and no central bank has a lever that reliably pushes it back down.UBI under private ownershipUniversal basic income under private ownership is the proposal to answer permanent, automation-driven job loss by taxing whoever owns the machines and mailing the proceeds to everyone else. It breaks as arithmetic before it breaks as politics: at any tax rate below 100 per cent the owner keeps a slice of every cycle, that slice leaves the spending loop, and the pot the state has to redistribute shrinks each round. The only rate that closes the circle is the one that abolishes the private ownership the scheme was designed to preserve.UnitrancheA unitranche is a single loan that collapses what used to be a stack of senior and junior debt into one facility at one blended interest rate, provided by one lender or a small club rather than a syndicate of banks and bond buyers. It is the product that let private credit funds take corporate lending away from the banks: the borrower gets a firm commitment in days from one counterparty instead of weeks of syndication risk. The price of that speed is that nobody outside the lender ever sees what the loan is worth.
Actors
The firms, funds and institutions holding the risk.
Business Development Company (BDC)A business development company is a closed-end fund that borrows money, lends it to mid-sized private companies, and hands nearly all the interest straight to its shareholders so it never pays corporate tax — a structure that works beautifully right up until the borrowers stop paying. Congress created the category in 1980 to get capital to small business; today it is the main wrapper through which ordinary investors own private credit. There are two flavours: listed BDCs, whose shares you can sell at whatever the market will pay, and non-traded "perpetual-life" BDCs, which you can exit at roughly 5% of the fund per quarter, if the queue lets you. As of mid-2026 both flavours are saying something uncomfortable, and they are not saying the same thing.EreborErebor is a US national bank, founded by Palmer Luckey and Joe Lonsdale and granted its final charter in February 2026, built to bank the technology, defence and crypto companies that lost their lenders when Silicon Valley Bank, Signature and Silvergate all failed in 2023. It is named after the mountain in Tolkien where a dragon sleeps on a hoard of gold, and it intends to run on blockchain rails so clients can move money at any hour of any day. As of the August 2026 reporting, deposits had reached $4.6bn and the bank was in talks to raise $1.5bn at a valuation reported between $8bn and $9.5bn. The interesting part is not the branding: it is a concentrated, tightly networked depositor base with the last mechanical brakes on a bank run deliberately removed.HyperscalerA hyperscaler is one of the handful of companies that runs computing infrastructure at a scale no ordinary business attempts — Microsoft, Alphabet, Amazon, Meta and now Oracle — by adding thousands of identical commodity machines rather than buying bigger ones. For about a decade this looked like the most capital-efficient business model ever invented, largely because the cost of owning the hardware had moved from the customer to the provider and nobody minded. It no longer looks that way: the four biggest have guided to something close to $700bn of capital spending in 2026, and Nomura estimates the biggest tech firms' roughly $200bn of bonds equals about a quarter of the Treasury's net note and bond issuance to private investors. The hyperscalers have stopped being an equity story and become a rates story.Jane StreetJane Street is a privately held trading firm that quotes buy and sell prices in ETFs, options and equities using its own capital, handling more than 20% of US ETF trading on the most recent estimate. It made $39.6bn of net trading revenue in 2025, more than JPMorgan's entire fixed income and equity trading operation, and then lost $15bn in July 2026 on directional bets that had nothing to do with quoting prices. It now tells bond investors that its strategy has evolved to include longer-term positions in the manner of a hedge fund. The balance sheet that makes prices for everyone else is also running risk for its own account, outside bank capital rules, with no supervisor and almost no disclosure.MicroStrategyStrategy, the software company formerly called MicroStrategy, is a business-intelligence firm that spent six years converting itself into a leveraged bitcoin holding: 840,447 coins at an average cost of $75,385 each, funded by convertible bonds and preferred stock. It invented the digital-asset treasury model, in which you issue shares above the value of the coins behind them, buy more coins, and repeat until the premium goes away. The premium went away in late June 2026, when the company's market value fell below the value of its bitcoin for the first time. Since then the machine has run in reverse, and Strategy has been selling stock and coins to pay fixed dollar obligations of about $1.76bn a year against a software business generating roughly $490m.NeocloudsNeoclouds are specialist cloud companies — CoreWeave, Lambda, Nebius, Crusoe — whose entire business is renting out enormous clusters of Nvidia GPUs by the hour to anyone training or serving AI models. They fund the buildout with debt secured on the chips themselves, which means the whole model depends on what a two-year-old GPU turns out to be worth. Several of them started life as cryptocurrency miners and kept the power contracts. Their most important benefactor is also their supplier: Nvidia sells them the chips, owns a slice of their equity, and under a September 2025 order form is contractually obliged to buy CoreWeave's unsold capacity through April 2032.Private equity sponsorA private equity sponsor is the firm — the general partner — that raises a fund from outside investors, buys companies with that money plus a great deal of borrowed money, charges fees the whole time, and is supposed to sell those companies at a profit inside about ten years. The selling part has stopped working: 13,509 PE-backed companies were sitting unsold in the US as of the second quarter of 2026, and continuation vehicles, in which a sponsor sells an asset to a fund it also manages, accounted for about 14% of sponsor-backed exits globally in 2025. What sponsors do instead of selling — dividend recaps, continuation funds, pledging unusual collateral — is now one of the main sources of demand for private credit. It is also where the strain is showing.Thoma BravoThoma Bravo is a private equity firm that buys software companies and only software companies, running more than $172 billion of assets across roughly 80 portfolio companies as of 31 March 2026. Its model is to take a subscription business private using debt sized against recurring revenue rather than profit, cut costs, bolt on rivals, and refinance when the loan comes due. That last step is the one that has started failing: creditors took control of Medallia in June 2026, a $5bn Proofpoint refinancing needed sweetening in July after a lender revolt, and in August a $2bn Sophos refinancing could not find private credit backing at all. If you want a single name to watch for whether the software debt boom unwinds quietly or badly, this is it.
Characters
People and episodes that tell you something.
Leopold AschenbrennerLeopold Aschenbrenner is the former OpenAI researcher who turned a 165-page essay about artificial general intelligence into a hedge fund, and then turned the hedge fund into one of the largest trading losses on record. Situational Awareness LP launched in 2024 with roughly $225 million, was reported at about $45 billion at its peak, and lost something in the region of $35 billion in July 2026 before its stock book was sold to Citadel. He matters here less for what he predicted — he was largely right about AI infrastructure — than for who was funding him: Jane Street and PIMCO, which is to say the plumbing.Mark WalterMark Walter is the billionaire chief executive of Guggenheim Partners and the founder of TWG Global, the holding company behind the Los Angeles Dodgers. He is under federal criminal investigation over roughly $20bn of loans his life insurers made to businesses he controls without disclosing the connection. His empire runs on the standard modern trick of pairing annuity money with private credit originated by an affiliate, which works right up until someone asks who set the prices. Since the grand jury subpoenas landed in February 2026 he has been raising cash hard — offering to pledge his Guggenheim stake, selling the Lakers nine months after taking control, talking about selling Chelsea — which makes him the first live test of what happens when a prosecutor opens the insurance-balance-sheet leg of private credit and looks inside.Plaza AccordThe Plaza Accord was the agreement of 22 September 1985 in which the finance ministers of the United States, Japan, West Germany, France and the United Kingdom committed to drive down an overvalued US dollar. It worked, and it worked cheaply: the dollar fell around 40% over the following two years on roughly $10 billion of actual dollar selling in the first six weeks, against a market then turning over about $200 billion a day. That ratio is why it is still the reference point for every currency intervention since. Forty-one years later it is the explicit model for the so-called Mar-a-Lago Accord and the joint US-Japan yen interventions of 2026.Radiant WorldRadiant World is a Singapore-headquartered iron ore trader, founded in 2003 by Pinkesh Nahar, that in the summer of 2026 was dropped by its banks and its largest counterparties over allegations that the invoices it showed lenders described cargoes which had never moved. Vitol and Cargill stopped trading with it at the end of July 2026, Deutsche Bank and KBC froze some of its Singapore accounts, and the US Justice Department and CFTC opened investigations. The firm denies wrongdoing and calls the claims "inaccurate and unsubstantiated". It is the clearest live test of a question nobody in commodity finance likes asking: what happens when the lenders finally phone the counterparty to check.Scott BessentScott Bessent is the 79th US Treasury Secretary, and before that a macro hedge fund manager who helped George Soros break the pound in 1992 and made Soros around $1.2bn shorting the Japanese yen in 2013. In office he has used the Treasury's own plumbing as a trading desk: on 19 August 2026 he at least doubled long-end bond buybacks to support a collapsing 30-year, and since 3 August he has been openly defending the yen he once sold. America's fiscal problem is now being managed by a man who knows exactly how such problems are traded against, which is reassuring right up to the moment he runs out of tools.Silicon Valley Bank (SVB)Silicon Valley Bank was the bank of the American venture industry — at its 2021 peak it held banking relationships with nearly half of all US venture-backed startups — until it failed in March 2023, when it discovered that a deposit base which all talks to each other can also all leave at once. It had parked its pandemic deposit flood in long-dated bonds and used held-to-maturity accounting to keep roughly $15bn of losses in a footnote; when it admitted the problem, depositors pulled $42bn in a single day. Regulators seized it on 10 March 2023 and invoked the systemic risk exception to make even uninsured depositors whole. It is the template for every fast, coordinated, screen-mediated bank run since.Situational Awareness blow-upThe Situational Awareness blow-up was the July 2026 collapse of Situational Awareness LP, an AI-focused hedge fund that lost about 67% of its value in a single month, roughly $30 billion, when its leveraged bets on AI infrastructure stocks went the wrong way and its prime brokers asked for money. It was run by Leopold Aschenbrenner, a former OpenAI researcher who was 24 at launch and had grown the fund from a $225 million seed in late 2024 to a reported peak of $45 billion, on the thesis that artificial general intelligence was arriving by 2027. The portfolio was sold to Citadel in a distressed transaction. It is the cleanest demonstration to date that the AI trade was being financed with borrowed money, not just conviction.Telecom buildout of 1999-2001Between 1996 and 2001 American telecoms companies raised roughly $1.9 trillion to bury fibre-optic cable across the country and under the oceans, on the strength of a widely repeated claim that internet traffic was doubling every hundred days. It was not; measured growth was about 100% a year, and the distance between those two numbers is the entire buildout. By 2000 less than 5% of the installed fibre was lit, bandwidth prices had fallen by nearly 90%, 47 telecom firms went bankrupt between 2000 and 2003, and something like $2 trillion of shareholder value went with them. It is the episode analysts including Apollo's Torsten Sløk now reach for when they want a historical yardstick for the AI datacentre boom.
Everything, A–Z
10y10y forward yields2s10s spreadAuction tailBond market dysfunctionBusiness Development Company (BDC)Cash-futures basis tradeCircular flow of incomeCovenantCrypto and TradFi contagionDemand doom spiralDirectional betDurationEreborHaircutHidden leverage and shadow bankingHigh-yield spreadsHousehold creditHyperscalerInterval fundJane StreetLeopold AschenbrennerLiability-management exerciseMarkMark WalterMicroStrategyMOVE indexNAV loanNeocloudsNon-accrualOption-adjusted spread (OAS)Payment-in-kind (PIK) incomePlaza AccordPolicy, the Fed and fiscal dominancePrivate creditPrivate equity sponsorRadiant WorldRepurchase agreement (repo)Residual valueResidual-value backstopScalable Human EquivalentScott BessentSecured Overnight Financing Rate (SOFR)Significant risk transferSilicon Valley Bank (SVB)Situational Awareness blow-upSpecial purpose vehicle (SPV)Sponsored repoSubordinationSyndicated loan marketTail riskTelecom buildout of 1999-2001Term premiumThe AI capex bubbleThe dollar, gold and reserve statusThoma BravoUBI under private ownershipUnitranche