{"id":31878,"date":"2026-10-20T04:00:00","date_gmt":"2026-10-20T02:00:00","guid":{"rendered":"https:\/\/www.thebrokernews.ch\/?p=31878"},"modified":"2026-10-19T07:44:49","modified_gmt":"2026-10-19T05:44:49","slug":"how-i-learned-to-stop-worrying-leverage","status":"publish","type":"post","link":"https:\/\/www.thebrokernews.ch\/en\/how-i-learned-to-stop-worrying-leverage\/","title":{"rendered":"How I Learned to Stop Worrying and Love the Leverage"},"content":{"rendered":"
Movie recommendation from Eric Lefebvre: Stanley Kubrick's scathing satire, *Dr. Strangelove*.<\/span><\/div>\n\n\n\n

A week of quiet cracks, a rumour that moved billions on almost nothing, and the doomsday machine the market built out of game theory. Asia detonated. The fear gauge did not. If you have not watched Dr. Strangelove lately, watch it again next weekend. It is shorter than a market briefing and it is, at the moment, more accurate.<\/strong><\/p>\n\n\n\n

Dr Strangelove<\/a> never existed, which is a pity, because he was the most honest man in the room. The wheelchair and the rogue arm were Peter Sellers and Stanley Kubrick, but the mind behind the mask was stitched together from men who were entirely real. One of them was John von Neumann<\/a>, who invented game theory, the mathematics of how rational people behave when their fate depends on what everyone else does, and who then sat on the committees that wired American nuclear strategy. The other was Herman Kahn<\/a>, the strategist at RAND<\/a> who taught a generation to think the unthinkable and who gave the film its doomsday machine. Kubrick fused them into one character, and the joke of the film is the joke of this week. A machine assembled from nothing but rational steps can produce a perfectly insane result, and because every step was reasonable, no single person is able to switch it off. The film is sixty years old. It has not dated. If anything, the market has caught up with it.<\/p>\n\n\n\n

I raise the good doctor because this was a week to think about games rather than prices. A game, in von Neumann\u2019s sense, is any situation in which your best move depends on the moves of everyone else. His uncomfortable discovery was that a room full of individually rational players can walk together off a cliff, each one correct at every step. The market is the largest non-cooperative game ever run, and this week, for a few days, you could watch the equilibrium begin to wobble.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

The first lines of defence<\/strong><\/p>\n\n\n\n

The wobble began, as these things do, at the most fragile edge. CoreWeave is what the trade calls a neocloud, a company that builds data centres for artificial intelligence on borrowed money and rents out the computing power. Its debt is rated junk. This week its bonds fell below eighty cents on the dollar, the yield on one of them climbed past sixteen per cent, and the cost of insuring the debt against default rose above a thousand basis points, a record. The shares fell about sixteen per cent in a session. The chief executive called the reaction irrational. More tellingly, the founder of Nvidia felt obliged to go on the record to say that CoreWeave was improving and needed no rescue, while Elon Musk summarised the bond chart with a single word: Yikes.<\/p>\n\n\n\n

A word on that phrase, rated junk, because the industry would much rather you did not dwell on it. Junk is the honest term. The polite term, the one printed on the marketing, is high yield, which is to a junk bond what pre-owned is to a used car: the same thing in a clean shirt. A bond is rated junk when the agencies are no longer confident the borrower will pay it back, so the lenders demand a fatter coupon as danger money. Investment grade means they think you are good for it. Junk means they are not sure. And a yield of fifteen per cent, which is where CoreWeave now sits, means the market has stopped waiting for the agencies altogether and begun, quietly and with some embarrassment, to price the funeral.<\/p>\n\n\n\n

There is a joke buried in the name, and it is worth a moment, because it tells the whole story. CoreWeave did not begin life in artificial intelligence. It was founded in 2017 as Atlantic Crypto, a cryptocurrency-mining operation that ran banks of graphics chips to mint Ethereum. When the crypto price collapsed, the founders realised that their real asset was the pile of chips itself, much of it bought cheaply from other miners going bust, and that those chips were worth more renting out computing power than digging up coins. So they changed the sign over the door. The company that now sits at the foundation of the most celebrated boom in a generation is a crypto miner in a new suit, and its name, if you listen to it, describes the method exactly. It weaves. What it has woven, these past two years, is debt.<\/p>\n\n\n\n

The thing to watch with a company like this is not the share price but the price of its borrowing, because a bond yield is a thermometer. It measures, in a single number, how likely the lenders believe they are to be paid back. CoreWeave sold a tranche of bonds as recently as June at around nine per cent, which was already a steep, speculative rate for a firm selling the future. By this week the market would only hold that same paper at a yield close to sixteen, the bonds had slipped below eighty cents on the dollar, and the cost of insuring them against default had passed a thousand basis points, a level S&P itself files under distress. Every credit analyst carries a rough ladder in his head. Around six per cent a growing company copes. Past eight the ground turns slippery. Into the double digits you are no longer talking about refinancing but about restructuring. Fifteen is the market saying, with great courtesy, that it does not expect to be paid in full.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Approximate annual yields on company debt. Investment grade and junk thresholds are illustrative; a yield near 15 per cent signals the market is pricing default.<\/em><\/p>\n\n\n\n

And here is a tell within the tell. While the bond desks were quietly pricing CoreWeave to fail, the equity enthusiasts kept buying the shares, Cathie Wood chief among them, a manager so dependable in reverse that the trading floors treat her as the Jim Cramer of her side of the aisle: whatever she is buying, sell. The point beneath the cruelty is simple. The people lending CoreWeave money and the people buying its shares cannot both be right, and only one of the two groups holds a lawyer and a claim on the assets. The bond market is the adult in the room, and this week the adult was reaching for the door.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Pause on why Nvidia had to speak at all. Nvidia sells the chips, and Nvidia also invests in several of the companies that borrow money to buy those chips. The dealer is quietly staking some of the players at his own table. It is a wonderful arrangement while the game is being won, because every chip sold validates the next loan and every loan funds the next chip. It is a very different arrangement when the fragile players start to wobble, because then the dealer is implicated in their debts. Around CoreWeave sit the other outworks of the same fortress. The smaller neoclouds. The model-makers themselves, one of which told the world last month that its annualised revenue was fifty billion dollars, having implied seventy not long before, with nobody able to say exactly what annualised is meant to mean. These are the first lines of defence. When they fall, the wave reaches the keep.<\/p>\n\n\n\n

And at the centre of the keep stands Oracle, which is the more disquieting case, because Oracle is not a repurposed crypto miner but a company more than forty years old that genuinely earns money. That is precisely what makes its week so sobering. To pay for its share of the build-out, Oracle\u2019s on-the-books debt has roughly doubled, to some one hundred and twenty-five billion dollars. Its capital spending now swallows more than four-fifths of its revenue, its free cash flow has turned sharply negative, and S&P has cut it to the very last rung of investment grade, a single notch above junk, with the other agencies muttering about worse. But the figure that rewards a torch and a look under the carpet is the one Oracle keeps off the balance sheet altogether. Tucked into the footnotes of its accounts sits a further quarter of a trillion dollars, some two hundred and forty-eight billion, of data-centre leases that appear nowhere on the balance sheet itself, up from a mere twenty-three billion only eighteen months earlier. They run for fifteen to nineteen years. The customers renting that capacity have, for the most part, signed up for five, and roughly half of Oracle\u2019s celebrated backlog rests on a single one of them, OpenAI, a company that does not yet make a profit. So the true shape of the bet is this: Oracle is committing to pay rent for nineteen years against contracts that last five, to serve a customer that is itself a cash furnace, and it is keeping the heaviest part of that promise in the notes rather than on the page. One commentator this week reached for the word Enron. That is too strong, and it is fair to say so, because Oracle sells real software to real customers at a real profit, which Enron never did. But the narrower echo is the one that matters, and it is exactly this: the art of keeping the frightening number just off the edge of the page, where the eye is not trained to look. When the solid old name at the centre of the fortress is financing itself like this, the question stops being whether the fragile edge gives way, and becomes how much of the fortress was ever as solid as the brochure claimed.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Oracle. On-balance-sheet debt versus the ~$248bn of 15-19 year data-centre leases disclosed only in the notes (10-Q, Nov 2025), up from ~$23bn eighteen months earlier. Greece GDP shown for scale.<\/em><\/p>\n\n\n\n

The doomsday machine nobody mentions<\/strong><\/p>\n\n\n\n

Kahn\u2019s doomsday machine had one iron rule. It only works as a deterrent if everyone knows it exists and knows it is automatic. The tragedy of the film, the part that still makes an audience laugh and then go quiet, is that the Russians build one and forget to announce it, so it deters nobody and simply goes off. The market has built its own version out of leverage, vendor financing, off-balance-sheet vehicles and the carry trade, and it is automatic in exactly the way Kahn feared, because margin calls and forced selling require no human decision once the collateral starts to fall. The difference is that this machine is semi-secret. Its wiring runs through the quiet cracks that the cheerful financial channels do not dwell on. In the film the general is proud of the machine. In the market the general is still denying it has been switched on.<\/p>\n\n\n\n

Look where it actually went off. In Asia the damage was severe. Seoul\u2019s market tripped its circuit breaker, the automatic halt that stops trading when prices fall too fast, and closed down around seven and a half per cent, nearer nine on the day that mattered. The Indonesian market fell about eleven per cent, its worst day since the depths of 2020, and this one had a name: MSCI said the market no longer merited emerging-market status, which is the sort of bureaucratic sentence that empties a room. Tokyo lost more than four per cent, Taiwan was badly shaken, and China kept sliding. And yet, turn to Europe, and Paris rose about two per cent on the week while Wall Street finished close to flat. The machine detonated in one hemisphere and the other hemisphere threw a party. That is not calm. That is the lag before the sound reaches you.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Asia: worst single-day fall this week. West: change on the week. Figures approximate.<\/em><\/p>\n\n\n\n

The instrument that ought to have screamed did not. The VIX, which everyone calls the fear gauge, is a single number distilled from the prices investors pay for options, which are in essence insurance against a falling market. When people are calm and few are buying protection, it sits in the low teens. When they are frightened and scrambling to hedge, the cost of that insurance leaps and the VIX leaps with it. The rough scale is easy to carry: below twenty is complacency, around thirty is real fear, forty and above is outright panic. For a sense of proportion, it closed above eighty when Lehman fell in 2008 and again in the depths of the 2020 pandemic, touched the mid-sixties in the yen scare of August 2024, and reached the forties as recently as last spring, when tariffs briefly frightened everyone. This week it rose, but only to about twenty, from a becalmed sixteen the week before. A continent was repriced and the fear gauge barely cleared its throat. That is not reassurance. It is the signature of stress that has not yet found the instruments that measure it, because the stress is not in the banks, which reported this week with their credit quality still placidly benign, but in the private credit and the artificial-intelligence debt the banks do not hold, which is precisely why it is so hard to see.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

The VIX at selected moments. Levels are closing values except the August 2024 spike, which is an intraday peak (it closed nearer 38). Real panic begins in the forties; this week the gauge reached only about twenty.<\/em><\/p>\n\n\n\n

Two tells that give the game away<\/strong><\/p>\n\n\n\n

Underneath the calm, two things happened this week that do not fit the cheerful story, and both are worth more than any index close. The first is in the bond market. Long-term interest rates fell, by around ten basis points, in the same week that the worst inflation figures of the year landed. American producer prices rose one point three per cent in a single month and more than six per cent over the year, driven mostly by energy. Consumer prices rose six-tenths in the month and are up 3.7 per cent over the year. Import prices are up more than eight per cent on the year, export prices more than eleven. This with a possible interest-rate rise on the twenty-eighth of October in plain view. Falling yields on rising inflation is not a considered view about monetary policy. It is a stampede towards the only fire exit anyone still trusts. When money runs into government bonds while inflation burns and rate rises loom, it is not forecasting. It is frightened.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

United States, year-on-year. Producer and consumer prices per the BLS September releases; trade prices as cited.<\/em><\/p>\n\n\n\n

The second tell is stranger and more revealing. Gold and silver, the textbook hedges against inflation, were among the very worst performers of the week, during the worst inflation scare of the year. Silver was heading for a weekly loss of around six per cent. The reason is pure game theory of the most practical kind. A leveraged player who receives a margin call does not sell what he would like to sell. He sells what he can, which means the liquid, still-profitable thing, and this week that was the precious metals he had been sitting on at a handsome gain. You raise cash by selling your winners, because your losers will not fetch enough and you do not want to crystallise them. So when the safe haven falls hardest in the middle of an inflation panic, the market is not making a prediction. It is answering a margin clerk, and liquidity beats logic every single time the clerk picks up the telephone.<\/p>\n\n\n\n

A rumour about nine people in Princeton<\/strong><\/p>\n\n\n\n

The purest Strangelovian moment of the week, though, was a rumour. Midweek the tape lit up with talk that an artificial-intelligence hedge fund called VARA, nine people in an office in Princeton, New Jersey, had blown up. The details were lurid and precise in the way the best rumours always are. Sixteen months after it opened its doors, the fund was said to have been running some twenty-six billion dollars by the end of June, to have been up around two hundred per cent on the year at its peak, and to have poured that money into exactly the artificial-intelligence names now under the most pressure: Nebius, Bloom Energy, and, with a certain poetry, CoreWeave itself. The story was that a fall of forty-four per cent in a single summer month had roughly halved the book, from twenty-six billion to thirteen. The shares it was said to hold shivered. Then came the counter-rumour, that a person close to the fund insisted it was perfectly fine, that its filings still showed billions, and, most revealing of all, that the various official figures did not even agree with one another: five billion in one document, nine in another, somewhere between thirteen and seventeen in a third. At the time of writing nobody outside the fund actually knows which number is true, or whether any of them is.<\/p>\n\n\n\n

And here is the whole game in a single episode. Keynes once compared investing to a newspaper beauty contest in which you win not by choosing the prettiest face but by choosing the face you think everyone else will choose. In that contest the truth of the rumour is almost beside the point. What moves the price is what every player believes every other player will do about it. Now add the detail that makes this week\u2019s version perfect. The fund was said to hold CoreWeave, the very name whose bonds were already falling. A rumour that a large holder of CoreWeave is being forced to sell pushes CoreWeave down, and a falling CoreWeave lends the rumour precisely the credibility it lacked, which invites yet more selling. The fear of a fire lights the fire, and the smoke is then taken as proof. A story that may well be false moved real billions, because in a market wired together by leverage the doomsday machine does not require anyone to verify it before it goes off. That is not a flaw in the players. It is the structure of the game.<\/p>\n\n\n\n

The wiring is getting more expensive<\/strong><\/p>\n\n\n\n

All of this sits on top of a cable that this column has traced before, the carry trade, the great flow of cheap money borrowed in Japan to be invested everywhere else. That cable is fraying. Japan\u2019s ten-year interest rate has pushed above three per cent for the first time since 1996. The Ministry of Finance has just raised the coupon on new ten-year bonds to 3.1 per cent, the highest in three decades. The era of the costless yen that funded a generation of speculation is ending, and money is being pulled home. Japanese institutions have been quietly selling the foreign bonds they have held for years. One of them, Sumitomo Mitsui, has sold its entire holding of French government bonds and rotated into German and Japanese paper. Japan was, for more than thirty years, the world\u2019s largest foreign holder of French debt. It is no longer volunteering. The detail that matters is not whose bonds they are selling. It is that the cheapest fuel in the global game is becoming dearer, and when the fuel becomes dearer, the most leveraged players are always the first to feel it. The fragile edges wobble first for a reason.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Japan 10-year government bond yield. Path approximate; latest figure as cited.<\/em><\/p>\n\n\n\n

And since we have strayed to France, a question the polite press circles but will not ask outright: how far is France from junk? The official answer is a comfortable-sounding five or six notches. Fitch and Standard and Poor\u2019s both rate it A+, which the agencies cheerfully note puts France on a par with Belgium and Slovenia, a sentence no French minister reads aloud. That is the distance on paper. The market keeps its own books, and they are less flattering. It already charges France about a point and a half more than Germany to borrow for ten years, and this autumn Goldman Sachs began writing insurance against a French default, which is not a product anyone troubles to invent for a country nobody worries about. None of this makes France junk. It makes France the one large economy at whose expense the question is now being asked aloud. And if the word still sounds abstract, remember Greece, which was investment grade too, respectable and European and safely inside the euro, right up until 2012, when its private lenders were handed a haircut of 53.5 per cent, the largest sovereign default in recorded history, and the ratings agencies stamped the whole affair, however briefly, with the word default. France is not Greece. France is larger, and in a currency union that is not a comfort but the problem, because Greece was small enough to be rescued and France is not. There is not enough money in the building. The distance between an A and a default is not counted in notches. It is counted in political will, and France has spent a decade demonstrating how little of it she has left.<\/p>\n\n\n\n

What to watch, and how<\/strong><\/p>\n\n\n\n

You cannot un-invent leverage, any more than von Neumann\u2019s generation could un-invent the bomb. His real lesson was never that the game can be won, only that a rational player is one who respects what the machine is capable of doing. The market has spent three years learning to love the leverage, and this week it was handed a quiet reminder of the other half of the title. The cracks are still discreet. The fear gauge is still low, which is the most discreet thing of all. The build-out is still applauded nightly. But the bond market is already edging towards the exit, the safe havens are being sold to pay the margin clerk, a rumour about nine people in Princeton can move billions, the Japanese are selling French bonds because the yen is no longer free, and the first lines of defence in the artificial-intelligence fortress have begun to give. None of that is a prophecy of catastrophe, and I am not in the business of selling one.<\/p>\n\n\n\n

It is simply the hum of a device that was assembled, one rational step at a time, to be impossible for any single player to turn off. The sane move, as ever on Planet Finance, is the one poor Strangelove never managed. Stop worrying just long enough to look calmly at the wiring, and then start worrying about the right thing. Keep a little of your wealth in the one thing the margin clerk cannot force you to sell, the metal that answers to no issuer, and keep your leverage where you can see it. The machine is still running. The switch is still not labelled. And if you have an hour and a half next weekend, watch the film again. You will recognise the room.<\/p>\n\n\n\n

Eric Lefebvre<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

See also: The Frog Prince<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

A week of quiet cracks, a rumour that moved billions on almost nothing, and the doomsday machine the market built out of game theory. Asia detonated. The fear gauge did […]<\/p>\n","protected":false},"author":2,"featured_media":31585,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"inline_featured_image":false,"_price":"","_stock":"","_tribe_ticket_header":"","_tribe_default_ticket_provider":"","_tribe_ticket_capacity":"0","_ticket_start_date":"","_ticket_end_date":"","_tribe_ticket_show_description":"","_tribe_ticket_show_not_going":false,"_tribe_ticket_use_global_stock":"","_tribe_ticket_global_stock_level":"","_global_stock_mode":"","_global_stock_cap":"","_tribe_rsvp_for_event":"","_tribe_ticket_going_count":"","_tribe_ticket_not_going_count":"","_tribe_tickets_list":"[]","_tribe_ticket_has_attendee_info_fields":false,"footnotes":""},"categories":[5100,12135,5134],"tags":[5102,15828,11870,15831,15830,15832,15829,15833,15834],"class_list":["post-31878","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-current","category-columns","category-general","tag-ai-en","tag-capital-expenditures","tag-debts","tag-doomsday-machine","tag-dr-strangelove","tag-junk","tag-leverage","tag-losses","tag-silver","ownarticle"],"acf":[],"cc_featured_image_caption":{"caption_text":"Movie recommendation from Eric Lefebvre: Stanley Kubrick's scathing satire, *Dr. Strangelove*.","source_text":"","source_url":""},"_links":{"self":[{"href":"https:\/\/www.thebrokernews.ch\/en\/wp-json\/wp\/v2\/posts\/31878","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.thebrokernews.ch\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.thebrokernews.ch\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.thebrokernews.ch\/en\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.thebrokernews.ch\/en\/wp-json\/wp\/v2\/comments?post=31878"}],"version-history":[{"count":2,"href":"https:\/\/www.thebrokernews.ch\/en\/wp-json\/wp\/v2\/posts\/31878\/revisions"}],"predecessor-version":[{"id":31893,"href":"https:\/\/www.thebrokernews.ch\/en\/wp-json\/wp\/v2\/posts\/31878\/revisions\/31893"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.thebrokernews.ch\/en\/wp-json\/wp\/v2\/media\/31585"}],"wp:attachment":[{"href":"https:\/\/www.thebrokernews.ch\/en\/wp-json\/wp\/v2\/media?parent=31878"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.thebrokernews.ch\/en\/wp-json\/wp\/v2\/categories?post=31878"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.thebrokernews.ch\/en\/wp-json\/wp\/v2\/tags?post=31878"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}