Letters to the Telegraph

About Letters to the Telegraph

A letter to the Telegraph, and the essay it provokes. Defence, politics, markets and the British state, examined with a soldier's eye, a City head and a long memory. davidmccreadie.substack.com (https://davidmccreadie.substack.com?utm_medium=podcast)

Publisher

David McCreadie

Category

history · news

Language

en

Explicit

No

First episode

26 May 2026

Latest episode

14 Jul 2026

Latest episodes

6 episodes in the feed.

1. 

   14 Jul 2026

   The AI Illusion: The Fire Exit

   A year ago I published a series of fifty articles called The AI Illusion, covering artificial intelligence across finance, work, warfare, governance and truth, and I closed it with a short story. This is the second series, and I am beginning it the same way, with a story. A great deal has changed in twelve months. Not the pace of the technology, though that is real enough, but the widening gap between what the technology can now do and the near-total absence of anyone in a position to govern it. Most governments and regulators are not slow off the mark. They are not yet at the mark. We should ask whether we have already built something that sits beyond democratic reach. The story is in four parts. It is called The Fire Exit, and it describes a danger that is in plain sight and that almost no one is discussing, which is the usual condition of such dangers until the moment they arrive. You might conclude from it that the sensible course is to sell everything and run for the hills. Very few people ever do. There is a particular comfort in staying with the crowd, and at present the crowd is dancing. The trouble is that it is all dancing on one side of the boat. Prologue MERIDIAN ASSET MANAGEMENT LLC RISK OVERSIGHT COMMITTEE MINUTES OF EXTRAORDINARY MEETING 14 April 2025, 09:00 Eastern Standard Time Board Room C, 47th Floor, 383 Madison Avenue, New York Present: Chairman, Risk Oversight Committee; Chief Risk Officer; Chief Investment Officer; Head of Equities; External Risk Adviser (non-voting); General Counsel (observer) Apologies: None The Chairman opened the meeting at 09:04 and noted that a quorum was present. 1. PURPOSE OF MEETING The Committee had been convened to consider a formal proposal from the Investment Committee to waive the Fund’s standard single-sector concentration limit of 15 percent for the purposes of artificial intelligence infrastructure exposure within the domestic equity portfolio. 2. PRESENTATION OF PROPOSAL The Chief Investment Officer presented the proposal. He noted that the Fund’s current AI infrastructure exposure stood at 14.7 percent of the domestic equity portfolio, approaching the limit established in the 2019 risk framework. He further noted that peer analysis indicated median competitor exposure in the sector at 18.3 percent, with the upper quartile at 24.1 percent. The proposal was to raise the permitted limit to 25 percent for a period of eighteen months, subject to quarterly review. The External Risk Adviser noted that the Fund’s largest individual position within the sector, held across multiple vehicles, represented 11.2 percent of the domestic equity portfolio. She asked whether a single-name limit should also be reviewed. The Chief Investment Officer noted that the existing single-name limit of 5 percent applied at the vehicle level rather than the consolidated fund level and that the current structure was therefore compliant. The General Counsel observed that the minutes should reflect that the waiver was being sought in response to competitive positioning analysis rather than a change in the underlying risk assessment of the sector. 3. DISCUSSION One member raised the observation that Berkshire Hathaway’s reported cash position of approximately 347 billion dollars might be considered a relevant external data point. The Chief Investment Officer noted that Berkshire operated under structural constraints not applicable to the Fund, and that the comparison was not directly useful. No further discussion was recorded on this point. The Committee considered the proposal. The risk-adjusted return profile of the sector over the preceding twenty-four months was noted. The competitive implications of underexposure relative to peers were discussed at length. The word structural appeared six times in the course of the discussion. The word risk appeared eleven times. 4. RESOLUTION The Committee resolved unanimously to approve the waiver as proposed, subject to quarterly review. The next review was scheduled for 14 July 2025. There being no further business, the Chairman closed the meeting at 09:47. These minutes were approved as a true and accurate record. [Signed] Chairman, Risk Oversight Committee, Meridian Asset Management LLC Document reference: ROM/2025/04/14/EOC RESTRICTED: For internal distribution only. Not for external release. I. The Paper Beijing, 10 September 2026 Dr Fang Wei arrived at the Ministry of Industry and Information Technology building on West Chang’an Avenue at ten minutes past two. He had arrived at two o’clock, as instructed, but the security desk had no record of his appointment and the responsible officer had needed to make three telephone calls before the matter was resolved. This was not unusual. He was shown to a conference room on the seventh floor and given tea in a glass. The room contained a projector screen, a table that could seat twenty, and four other people, none of whom Fang knew well. A deputy director whose name he did not immediately recall was arranging papers at the head of the table. On the screen was a logo: NovaMind, with the characters below it that Fang had always thought were slightly too formal for a commercial product. He had written three of the four foundational papers underlying NovaMind’s architecture between 2021 and 2024. The fourth had been written by a team of twelve and was, in his private assessment, considerably worse for it. He had not complained. This was not how things were done. He had not been told what this meeting was about. The deputy director began at two-fifteen. The North American commercial licensing programme, he explained, had been reviewed and approved at the relevant level. The pricing reflected NovaMind’s commitment to accessible, high-quality artificial intelligence services. The programme would be valid for an initial period of eighteen months. Fang noted the word introductory, turned it over once, and put it down. He thought about what eighteen months of introductory meant when it was backed by state subsidy and applied to a market whose entire pricing architecture assumed you could not compete. The deputy director paused in the way of people who expect acknowledgement rather than questions. Two of the four others nodded. One wrote something down. He had written a paper in 2023 on cascade failures in complex interconnected systems. It had been accepted by a journal, presented at a conference in Shenzhen, and subsequently ignored by everyone who mattered. The paper argued, using examples drawn from the 1987 equity market crash and the 2010 Flash Crash, that complex systems did not fail gradually: they absorbed pressure until a threshold was crossed, and then reorganised catastrophically. The threshold event, the paper noted, could not be identified by those inside it. It became visible only after the reorganisation was complete. The paper had included an appendix on external validation signals: observable data that suggested a system was under stress before that stress became visible in prices. The most recent example he had used, added in the final revision, was the cash position of Berkshire Hathaway. At the time of writing, Berkshire held approximately 280 billion dollars in cash and short-term Treasuries. It had been selling equities, including a significant position in Apple, for six consecutive quarters. Its chairman had declined, publicly and repeatedly, to find assets at prices he considered rational. Fang had used it as an illustration of how signals could be simultaneously visible and ignored. He had noted, with some care, that Berkshire’s cash position was not a secret. It appeared in every quarterly filing, was discussed in every financial publication of consequence, and had been attributed by several commentators to the chairman’s age or the fund’s unwieldy size. Nobody at the Ministry had read the paper, as far as Fang knew. The appendix on Berkshire had not therefore been discussed. He raised his hand slightly, not quite a full gesture. “The American AI infrastructure companies,” he said. “Their current valuations are substantially dependent on their pricing models holding.” The deputy director looked at him in the way of senior officials presented with an inconvenient fact. “The programme reflects NovaMind’s commercial strategy,” he said. Fang nodded. He had asked the question. That was the limit of what was available to him. He knew, without having done the precise calculation, that eighteen months of subsidised pricing in the North American enterprise market would expose a valuation gap in American AI infrastructure companies of a magnitude the market had not priced. The figure that mattered was not the twenty dollars a month a business paid to put a chat assistant on an employee’s screen. That was the retail layer, the shop window, and its price had been falling for two years without anyone important seeming to mind. The figure that mattered was the one beneath it: the enterprise infrastructure price, the cost of the model capability that companies built their actual operations on, the layer where the margins lived and where the valuations had been raised. That was where the American firms made their money, and that was the number NovaMind had come for. He knew what the American companies charged for comparable or inferior capability at that level, having spent three years writing the papers that made NovaMind possible: 340 dollars per enterprise seat per month. The introductory NovaMind rate would be 38. The retail price war was the foothills. This was the mountain. The pricing had been set, he now understood, with precision. Not low enough to invite immediate anti-dumping action. Not high enough to leave room for the argument that there was no material competitive threat. Thirty-eight dollars was calibrated. He did not know by whom. He understood, broadly, what would follow. The American AI companies would respond with lobbying, anti-dumping complaints filed at the WTO, calls for trade sanctions. These processes would take months and produce uncertain outcomes. In the interim, enterprise buyers would do what enterprise buyers always did when presented with comparable quality at eleven percent of the prevailing price. They would switch. Once switched, they would not switch back. The cost of rebuilding workflows around a different system was prohibitive; the adoption curves in his own papers were clear on this. The 38 dollar rate was not a permanent commercial position. It was an 18-month investment in market capture. After which, when the American providers had been destroyed or crippled, the price could be whatever the market would bear. Which, by that point, would be whatever NovaMind decided it needed to be. He also understood, in a way the deputy director’s framing had carefully avoided articulating, that a market crash in the United States triggered by AI pricing was not only an economic event. An economy weakened by recession, burdened by debt, already stretched by a war it could not afford to lose, and now facing a crisis of confidence in the technology sector that had been the primary argument for its equity premium over the rest of the world: that was a geopolitical event. The dollar’s reserve status was not a law of nature. It had been sustained for eighty years by military power, economic dominance, and the absence of a credible alternative. Two of those three were now genuinely in question. He thought about Berkshire’s cash pile. He thought about the appendix that nobody had read. The meeting concluded at two-fifty. He took the lift to the ground floor. His car was waiting. He sat in the back and opened Bloomberg on his phone. The Nasdaq had reached 28,041. He had been watching it for three months with the attention he gave things he could not stop. The number had acquired, over those three months, the quality of a thing that everyone has agreed not to look at directly. The analysts whose quarterly notes he read described the multiple on forward earnings as reflecting AI’s structural transformation of the productive economy. Fang thought the revenue was real. He thought the question of what happened to that revenue when a state-backed competitor entered at eleven percent of the prevailing price had not been modelled by anyone whose notes he had read. His 2023 paper had contained a line he had been quietly pleased with at the time: the threshold event cannot be identified by those inside it. He put the phone face down on the seat beside him. At home that evening his wife was cooking. He sat at the kitchen table while she worked and they talked about their daughter’s school, about a trip they were considering in the spring, about a film a colleague had recommended. He contributed to this conversation without difficulty. His wife asked once whether everything was all right at work. He said it was, which was true in the sense that nothing had yet happened. She returned to what she was doing. After dinner he went to his study and looked at the shelf where his published papers sat in bound copies. The 2023 cascade failure paper was the fourth from the left. Its spine was uncracked. He wrote a short email to his former doctoral supervisor, now retired in Chengdu, whom he spoke to three or four times a year. The email said that he hoped the older man was well, that the weather in Beijing had turned, and that he had been thinking about some of their old arguments about system complexity. He did not mention the meeting, or NovaMind, or the number on his Bloomberg terminal. He sent the email and went to make tea. Four days later, on a Monday morning, the Nasdaq would open at 27,640 and close at 21,883. Fang Wei would be in a different meeting when it happened. He would not check his phone until lunch. II. The Spreadsheet Brussels, the week before Elise Vandermeer had owned a copy of Kindleberger’s Manias, Panics and Crashes since 2019. She had read it twice and had a system: passages she found merely interesting were underlined in pencil; passages she found important were underlined in ink; passages she found frightening she left unmarked. On the Wednesday evening she took it from the shelf and opened it without quite intending to. It fell open where the spine had softened from repeated reading: the chapter on the dynamics of the mania phase. One sentence underlined in ink: “There is nothing as disturbing to one’s well-being and judgement as to see a friend get rich.” She read it and closed the book. Her fund ran a little over a billion euros. The AI infrastructure position, 180 million euros across three American names, was 17.4 percent of the European equity book and had returned 31 percent in eleven months. In a normal year a 31 percent gain on a single sector bet would have prompted a quiet conversation about taking some off the table. This had not been a normal year. Her CIO had mentioned the position at the last quarterly review in the tone of someone filing a piece of good news for later use. The waiver that had made the position possible had been approved the previous year. Her fund’s standard single-sector concentration limit was 12 percent. The AI infrastructure position now stood at 17.4 percent of the European equity book. The risk committee had approved the waiver on the basis of peer analysis. The minutes recorded that the word structural had been used a great many times. The word risk had been used more. She had counted, afterward, and had not enjoyed the exercise. The public equity book was not, in any case, the part that frightened her most. That distinction belonged to the private markets allocation, which had grown over six years from a tidy eight percent of the fund to a little over a quarter of it, because private equity and private credit had returned more than public markets every year and had done so, conveniently, without the indignity of a daily price. You could not see a private holding fall. It was marked once a quarter, by the managers who ran it, using models that were their own. The marks were always smooth and they were always late. Elise had sat through enough presentations to know that a private portfolio reporting a steady twelve percent while the public market it invested alongside fell twenty was not reporting a different reality. It was reporting the same reality, slowly, and with a flattering lag. A good deal of that private allocation was, when you traced it through, exposed to the same theme as the public book: data centres, the power and construction to build them, the AI infrastructure companies that were still private and the credit extended to the ones that were not. Some of it sat in continuation funds, vehicles into which a manager who could not sell a holding moved it instead, selling it from one of his own funds to another of his own funds at a price he set himself, so that the investors in the first fund could be shown an exit and the investors in the second could be shown an opportunity, and the underlying business, which had not in fact gone anywhere or done anything new, could be carried for another few years without anyone having to write down what it was actually worth. Elise had declined two of these and been talked into a third. She was not proud of the third. What she also knew, because the requests had begun to arrive in the spring, was that her own end-investors, the corporate pension schemes whose money the fund managed, had started asking for liquidity. Quietly at first, a larger scheme rebalancing, a smaller one with a funding gap to plug. But the requests were arriving more often, and a fund that has promised its investors they can have their money back, and that holds a quarter of that money in things it cannot sell, is running a mismatch that works perfectly until the day it does not. She had raised this at a meeting in June. She had been told the redemption pipeline was being monitored. She had come to understand that monitored, in the language of her own institution, meant much what closer monitoring would shortly mean in Richard Ashworth’s bank, although she had never heard of Richard Ashworth and never would. The spreadsheet she had been looking at all week was not complicated. On one side: what her own fund paid for the AI services it used internally, taken straight from its contracts. On the other: NovaMind’s published pricing for the same enterprise infrastructure, the model capability her fund’s systems actually ran on rather than the chat seats her staff logged into, which was no secret, and which she had supplemented with what she had gathered second-hand. A former colleague, now at a fund in Amsterdam, had mentioned over dinner that they had switched providers in the spring and what they were now paying. A vendor representative had been indiscreet at a conference. None of it was documentation she could have put in front of a board, but the numbers all pointed the same way and the gap was not subtle. The per-seat differential was around 300 euros, the Chinese product at a small fraction of the American price. She had checked her arithmetic three times because she did not want it to be right. The numbers were not ambiguous. She had spent a week trying to find the ambiguity and had failed. She had raised a version of this concern at a board presentation in the spring. Not the spreadsheet, which she had not yet built, but a broader point about competitive risk from state-backed Chinese AI providers. The CIO had listened carefully and had then mentioned that Berkshire Hathaway’s cash position had also been raised as a concern at the previous quarter’s review. He had smiled in a way that indicated he found both observations interesting and neither decisive. “Berkshire,” he had said, “has a different problem from ours. Their problem is size.” The room had agreed with this. Elise had written it in her notebook and had not raised the subject again. On the Tuesday of the previous week she had drafted an email to her CIO. She described the pricing discrepancy and the exposure it created, attached the spreadsheet, and suggested a 30 percent reduction in their three largest AI positions as a prudent risk management step. In a footnote she had included the Berkshire cash position as corroborating evidence of broader market overvaluation. She had been at her desk at eleven-fifteen in the evening when she wrote it, and she had looked at it for four minutes and then closed the window. She told herself it was because she was not certain about the quality differential between NovaMind and the American products. This was true, as far as it went. The whole truth was that a 30 percent reduction in three performing positions, on the basis of a pricing comparison she had assembled herself without external validation, would require a conversation she did not know how to have with a CIO who had just praised the position at a board review. She thought about her nephew Jakob, who had been made redundant from his position as a junior credit analyst at a Frankfurt bank in February. They had had dinner in March at a Turkish place near her flat that they both liked. He had been careful not to seem sorry for himself, which made her more aware of what the months of searching had cost him. He had good analytical instincts and she had always thought he would do well. The bank had cited restructuring. Fifteen people in his department in three months, replaced by a single AI system. He was twenty-six. He had applied for forty positions, had three interviews and no offers. Over dessert he had asked her, without particular bitterness, whether she thought the market for his kind of work was coming back. She had said she thought so. She had not been certain when she said it and was less certain now. She had not connected Jakob’s situation to her portfolio at the time. She connected them now, and was not comfortable doing so. On the Thursday she drafted the email again. She changed the recommended reduction to 20 percent, which was easier to defend, and removed the sentence that used the word urgent, and removed the Berkshire footnote, which she had decided might seem alarmist. She read it back. It sounded reasonable. It sounded like something a cautious person would send before a problem became undeniable. She saved it as a draft and went to make coffee. She did not send it. She told herself she would wait for NovaMind’s North American announcement, which she had heard through a contact was coming the following week, and send it with the announcement as supporting evidence. This was rational. It was also a way of waiting until the decision was easier to justify. The announcement came on the Monday, at twelve forty-three Brussels time, while she was at lunch. III. The Floor New York, the morning of Marcus Reeves read the piece on the subway. He had not meant to; he had opened the Journal app looking for the CoreLight IPO update and found instead, at the bottom of the Markets section, a piece headlined: Forty Years On: What Black Monday Still Has to Teach Us. The fortieth anniversary fell the following month; the markets desk had run its retrospective early. He read it standing, holding the rail with one hand. He had left his coffee on the platform at 47th Street, where he stopped each morning to finish it before boarding, a habit his assistant called eccentric and he called efficient. The piece was competent without being particularly good. It summarised what happened: in October 1987 the Dow fell 22.6 percent in a single session, the largest single-day percentage decline in the index’s history. The proximate cause was described as portfolio insurance: algorithmic hedging strategies which, when equity prices began to fall, sold futures contracts automatically. The selling drove prices lower. Lower prices triggered more selling. The machines had done exactly what they were told. What they were told turned out to be wrong. Marcus had heard this story before. His father had been a junior trader at Kidder Peabody in 1987, on the floor for the actual Monday, and had told the story once in the flat way of someone who has processed a thing until it stops producing sensation. The floor had been quiet by midday. Not because trading had slowed. Because everyone had understood, at roughly the same moment, that the mathematics had turned against them and that the exit was smaller than the position. Marcus had asked, when he was perhaps fifteen, what his father had done. “I sold,” his father said. “Not enough and not fast enough. But I sold.” He thought about a dinner in the spring, at a client’s apartment on Park Avenue. Twelve people, a long table, a conversation about whether the Nasdaq could sustain its trajectory into the following year. He had made a joke about Berkshire Hathaway’s cash pile. He had said that Buffett had been wrong before and that being wrong for long enough was indistinguishable from wisdom. The table had laughed. It was the kind of joke that lands because everyone in the room has made the same private calculation: that the most credible long-term capital allocator in history had been sitting on more than 340 billion dollars in cash for two years while the market went up 40 percent, and that if you were going to be wrong, it was better to be wrong with everyone else than right too early and alone. Being wrong with the market was survivable. Being right too early was not. He had not repeated the joke since. He was aware it had a different quality depending on how things turned out. There was an old man, a famous one, who had been saying all of this in public for two years. He ran a large firm and had called the great bubbles of the past before they broke, the Japanese madness of the late eighties, the internet in two thousand, the housing collapse before two thousand and eight, and he had come on the television that spring and said in plain words that this was the largest bubble in American history, that the leading names could fall seventy percent, and that a transforming technology and a ruinous stock could sit perfectly well in the same company, as the railroads had, as the internet had. He had been right about every previous bubble and early about all of them, and being early, in this business, is indistinguishable from being wrong until suddenly it is not. He had lost clients for it the last time, good clients, who left because he would not buy what everyone else was buying and who did not come back even after he was proved right. Marcus had watched a younger analyst on a panel in June roll his eyes while the old man spoke, and had said nothing, and had been ashamed of saying nothing, and had gone on holding the position anyway. That was the trap, and the old man had named that too. The market can stay irrational longer than the client can stay patient, and a manager who is right too early is fired by clients who are wrong on time. He folded the Journal as the train pulled in and thought about his position. One point one billion dollars long across seven AI infrastructure names, representing 23 percent of the domestic equity book: eight points above the concentration limit the risk committee had waived in the spring at the Investment Committee’s request. He had believed in five of the seven names a year ago. He believed in one of them now, approximately. He had tried to address this in August. He had prepared a twelve-page note arguing that the sector’s forward revenue projections did not adequately model competitive risk from state-backed Chinese providers, and that the concentration limit waiver should not be renewed at the next quarterly review. He had attached one chart he was particularly attached to: the price of call options on the Nasdaq had risen so far that they now cost almost as much as the equivalent puts. In a normal market puts cost more, because large institutions buy them as insurance against a fall and there are always more sellers than buyers of the underlying optimism. For calls to cost nearly as much meant the speculative crowd betting on further gains had grown so large it was outweighing the insurers. It was, Marcus had written, the clearest single sign he knew that the market had entered its manic phase. He had made two further points the committee had liked even less. Two of the seven names were, when you stripped away the language, bets on a single human being each, founders whose departure or disgrace or simple loss of interest would erase a third of the company’s value between one morning and the next, a risk no model priced because it could not be made into a number. And the entire American AI complex, the chips, the labs, the data centres and most of the founders, sat within a few hundred miles of one another in a state built across an active fault line, so that the market had carefully priced a hundredfold growth in revenue while pricing at essentially zero the single most predictable geological event in California. The committee’s most senior member, who ran twice Marcus’s AUM on substantially the same thesis, had described the note as thoughtful and premature. The waiver had been renewed. Marcus had kept the note. He had not distributed it. There was a thing he had not put in the note, because it could not be put into a note, which was that he had come to despise the men at the top of these companies. It was not a feeling he was proud of and it was not, he knew, entirely clean, because he had made a great deal of money from them and his disdain had not once stopped him taking it. But it was real, and it was more widely shared on the floor than anyone said aloud. A handful of them, the most visible, had stopped behaving like the founders of companies and started behaving like a separate order of being, men who had decided that they personally were the ones who would carry the species forward and that the rest of humanity was a kind of weather to be managed. Several of them were not, quite, American, which a certain sort of patriot on the floor noted with a relish Marcus found distasteful even where he shared the sentiment beneath it. His own objection was simpler and he thought more defensible. They had built things, some of them remarkable, and had concluded from this that they were owed the future; and the market had agreed with them, and the agreement had been priced into everything Marcus held. There was a part of him, and he admitted it only to himself and only that morning, that wanted to see them brought down so badly that he was prepared to be brought down a little himself in order to watch it. He suspected he was about to get his wish. He would find, when it came, that the satisfaction and the ruin arrived in the same parcel and could not be prised apart. There was one more figure he had put in the note, a plain one that needed no interpretation. Margin debt, the money investors had borrowed against their portfolios to buy yet more stock, had reached a record one point four trillion dollars, up more than half in a single year, a level of borrowing against holdings never seen before in the history of the American market. What he had written beside it was the part that mattered. That figure was only the leverage that had agreed to be counted. It was the reported, regulated, retail-and-broker slice, the borrowing that filed a return. It said nothing of the total return swaps that had quietly destroyed a family office called Archegos a few years earlier, nor of the money the private banks lent their wealthy clients against their positions, nor of the layering by which one real dollar at the bottom could support many times its value in exposure higher up. The visible leverage was a record. The invisible leverage, by its nature, could not be measured, only discovered, and it was always discovered in the same way and at the same time, which was on the way down, when the borrowing that had magnified every gain on the way up reversed and magnified the losses, and the people who had to sell could not choose what they sold but only what still had a buyer. Every previous record in margin debt, he had noted at the foot of the page, had been set within months of a market peak. Nineteen ninety-nine. Two thousand and seven. Twenty twenty-one. Nobody on the committee had asked him about the footnote. Marcus arrived at his desk at ten past seven. He poured a coffee from the pot Dana kept on the counter beside the window. The first alert about NovaMind’s North American announcement appeared at nine forty-four. He read it twice. Then he looked at the CoreLight float. CoreLight Systems was to have been the listing of the year: a company that had never made a profit, that had lost 5 billion dollars in its most recent accounts, priced at ninety times its annual revenue on the strength of a banker’s projection that its sales would grow a hundredfold by 2031. The book had been covered eleven times over. The roadshow had been a sensation. The float itself had been engineered with some care: only a small slice of the company was being released, which kept the supply of stock tight and the price easy to move, and the real event was not the public sale at all but the near-automatic inclusion in the indices a fortnight later, when every passive fund on earth would be obliged to buy the thing at whatever price it had reached, regardless of what it was worth, because the index said so. Nobody in Congress had said a word about any of it. Pricing was due that evening and trading was to begin on Wednesday. At nine forty-nine the lead underwriter issued a statement: the offering was suspended pending a further review of market conditions. The most wanted piece of paper on Wall Street had become, in five minutes, a piece of paper nobody would price. He thought about something he had read years ago, in a book a mentor had pressed on him: A speculative boom requires not just optimism but a mechanism that converts optimism into purchasing power. He had thought it was about 2008. He now thought it might be about most things. He called Dana Marsh, his head of trading. She picked up on the second ring. “Have you seen the NovaMind announcement,” he said. “Yes.” “And the CoreLight delay.” “Yes.” “I want to reduce. Significant reduction. Start with the three names I flagged in August. Quietly, and now.” A pause. Not a long one. Dana had been his head of trading for six years and had learned that when he said now in that particular register, the time for questions was afterward. “How much?” she asked. “As much as you can move without moving the price,” he said. “Then move more.” He put the phone down and looked at the terminal. The Nasdaq was already down four percent and it was not yet ten o’clock. He thought about his father on the floor of Kidder Peabody. He thought about the line about not enough and not fast enough. IV. The Day Monday, 14 September 2026 New York · 09:51 Marcus reached the options desk while Dana was starting the unwind. He wanted puts on the seven names, at the money, whatever was available. The trader paused before answering, which told Marcus most of what he needed to know. There was barely a two-sided market left. The makers who would normally quote a price to buy and a price to sell had widened their spreads so far apart that they were, in effect, no longer quoting, and on three of the names had pulled their bids entirely. What he could buy, he could buy only at a price set by people who could see the building was alight and were charging accordingly. The insurance was not so much expensive as unavailable. You could not buy a fire exit once the fire had started. Marcus told him to take whatever he could get. Brussels · 15:51 Elise Vandermeer had sent the email at seven-eleven that morning, the moment the NovaMind announcement crossed. By the time the American market opened, at half past three her time, her CIO had already agreed in principle to a 30 percent reduction and asked her to prepare a paper for the Thursday board. The paper would not now be needed. She watched the American open at the end of her own working day. The Nasdaq fell four percent in nine minutes. She picked up the phone and called her CIO and told him she was not going to wait for the board. She would begin reducing the European book’s AI exposure immediately, on her own authority, and they could discuss the governance of that decision afterward. He did not argue. She heard, in the way he did not argue, that he had understood the same thing she had. She gave the instructions to her dealing desk. Then she sat for a moment and did the arithmetic she had been avoiding. They would get a fraction of the position away today. The rest would have to be sold into whatever Tuesday brought, and Tuesday would have read about Monday. At twenty to five her head of client relations put her head round the door, which she did not normally do, and said that two of the pension schemes had been in touch within the hour asking about their ability to redeem at the next window, and that a third, the largest, had used the word immediately. Elise understood the shape of what was coming before the sentence was finished. The public book she could sell, badly and at a loss, but she could sell it. The schemes pulling their money would have to be paid, and the easy money to pay them with was the money that could be sold, which was the public book she was already selling to reduce risk, which meant she would shortly be selling not because she judged it wise but because she had no choice, the worst reason there is and the one that moves the most stock. And behind the public book sat the quarter of the fund she could not sell at all, the private holdings with their smooth and lying marks, which would not save her today and would, in their own slower time, present a bill of their own. She had spent the morning trying to get ahead of one fire. She saw now that she had been standing in a building with several, and that the one she had been fighting was not the one that was going to bring the structure down. New York · 10:31 At 10:31 the first circuit breaker tripped. A seven percent fall in the S&P 500 triggered a fifteen-minute halt across all American equity markets, the mechanism built after 1987 precisely to interrupt the cascade the retrospective in Marcus’s coat pocket had described. For fifteen minutes nothing traded. Marcus used the time to call the three largest holders in his fund and tell them, before they heard it elsewhere, that the fund was reducing and that he would have numbers for them by the close. Two were grateful. One was already shouting. When the market reopened at 10:46 it fell straight through the level it had paused at, the way water resumes when you take your hand away. Guildford, Surrey · 15:00 Richard Ashworth signed for a delivery of tool steel at three o’clock and stood in the yard for a minute in the last of the good September light. It had been a reasonable day. An enquiry had come in that morning from a firm in Yeovil that he thought would turn into an order, and the new optimisation system had flagged a tolerance drift on the number four lathe before it became a scrappage problem, which was the kind of thing it had been bought to do. He made a note to mention it to Phil. He thought, not for the first time, that the loan had probably been the right decision. He went back inside. He had heard, on the radio at lunchtime, that there was chaos on the American markets, some business with Chinese artificial intelligence, the Dow and the Nasdaq down hard. He had taken it in the way you take in news of a storm at sea: serious, no doubt, for the people out in it, and no concern of a man on dry land. On the far side of the Atlantic, where it was ten o’clock in the morning, the value of the assumptions underneath his loan was changing by the second. Richard knew the markets were falling. He had no reason whatever to connect that to the number four lathe, or to the firm in Yeovil, or to himself, and the afternoon in Guildford was quiet and bright and entirely ordinary. Beijing · 22:47 Fang Wei could not sleep. His wife had gone to bed at eleven and he had told her he would follow shortly, and had then sat in his study with the lamp off and the screen the only light in the room. He had been in meetings all day and had kept his phone face down through most of them, a discipline he had set himself and was finding harder to maintain than he expected. Now, at a quarter to eleven, he let himself look. The American market had been open for an hour. The Nasdaq was down eleven percent. He read the figure twice, the way he had read the word introductory twice in the meeting four days before, and felt the same sensation: not surprise, which would have implied he had not known, but the particular vertigo of watching a thing you have predicted arrive on schedule. He opened the appendix of his 2023 paper on his screen, the section nobody at the Ministry had read. The external validation signals. Berkshire’s cash position, which he had updated in a footnote to more than 340 billion dollars. The call-put parity on the Nasdaq, which he had flagged as the signature of a market that had stopped pricing risk and started pricing hope. The widening gap between American AI valuations and the cost at which the same capability could be delivered, a gap he had not, when he wrote the paper, known he would one day be instructed to close with a single number. He thought about the chain that ran from the conference room on West Chang’an Avenue to the screen in front of him. A deputy director reading approved language. A price set with precision by people he would never identify. An announcement crossing the wires at nine forty-four New York time. And now an hour of trading that would be in every newspaper on earth by morning, and a number, eleven percent and falling, that represented the savings and pensions and livelihoods of people who had never heard of NovaMind and would shortly be told that what had happened to them had been a normal commercial decision reflecting market conditions. He could see where it went from here, because he had drawn the diagram in the paper nobody read. A fall of this size, into an economy already carrying that much debt and the cost of a war, would not stay in the markets. It would become a recession, and not a shallow one, because the usual instruments for softening it had been spent or compromised in advance. An American recession would not stay American either. The export economies would feel it within a quarter: Germany through its machine tools, the Gulf through oil demand, half of Asia through the orders that would simply stop arriving. A factory owner somewhere in England who had never heard the word NovaMind would find his order book thinning and would not, for some time, understand why. The dollar would come under a pressure it had not faced in his lifetime, and the question that had sat unasked beneath the whole arrangement for eighty years, why the world should continue to settle its business in the currency of a country that could do this, or have this done to it, would begin, very quietly, to be asked aloud. He had set none of this in motion. He had merely understood it, in advance, and been ignored. He did not feel triumph. He had been an instrument, the same as the algorithms in New York that were at that moment selling because they had been built to sell. The difference was that he had understood, and understanding without the power to act was, he was beginning to think, its own kind of complicity. He closed the appendix. He did not send it to anyone. There was no one to send it to who had not already decided not to read it. He went to bed at midnight and lay awake until two. San Francisco · 08:20 Three time zones west of the trading floor it was twenty past eight in the morning, and the founder of one of the seven names was awake because he was usually awake, not because anything was wrong. His chief operating officer had been trying to reach him since before six. When the call finally connected she was talking quickly, the way people talk when they have been holding bad news for two hours and need to set it down. The stock was down a third. The NovaMind announcement. The CoreLight suspension. The analysts who had loved them on Friday were revising their models. Two board members wanted an emergency call. The communications team needed a statement. She had a list and she was working through it and her voice had the particular tightness of someone watching a structure she had helped build come apart in real time. The founder let her finish. Then he said that she should tell the board he would join the call at ten, after his run, and that there was no need for a statement today because anything they said now would only sound frightened. She started to say something about the share price. He stopped her, not unkindly. “Look,” he said. “We closed the last raise in March. We have seventy billion in the bank. Seventy. We are not going to miss a payroll, we are not going to miss a quarter, we do not have to think about money for a decade if we choose not to.” He paused. “The market’s having a moment. Let it. We got on the bus before it left town. Everybody scrambling out there this morning,” and she understood that by everybody he meant the funds, the pensions, the people, the entire apparatus downstream of him, “they’re the ones who were still standing at the stop. That was always going to be their problem. It was never going to be ours.” He went on for a moment, not heated, almost gentle, in the register he used when he was explaining something he believed self-evident. The people panicking today, he said, had never built anything. They allocated. They moved other people’s money from one thing to another and took a clip, and they had spent a decade taking a clip of what he and a handful of others had actually made, and now that the thing they did not understand had turned against them they were astonished, as though the weather had betrayed them. He did not say it cruelly. That was the part the COO would remember. He genuinely saw it this way: a small number of people who made the future, and a very large number who lived in it and were, in the final accounting, its passengers. The crash was simply the passengers discovering they were passengers. It had nothing to teach him, because he had always known which seat he was in. He told her to enjoy her morning and that he would see her on the ten o’clock. Then he went for his run. The COO sat at her desk with her list, on which nothing had been resolved, and understood for the first time that the calm in his voice had not been reassurance. It had been accuracy. He was going to be completely fine. That was the whole of what he had told her, and it was true. What unsettled her, watching him jog out across the empty campus in the early light, was not that he was unkind, because he was not. It was that within the closed logic of how he saw the world he was entirely correct, and that she had worked three years inside that logic without once standing outside it to see its shape, and that she could see it now only because it was, this morning, costing other people everything and costing him nothing at all. New York · 11:20 Dana called across from her desk. Through the glass of the meeting room, where Marcus had been summoned by a chairman who had reserved his appearances on the floor for exactly this kind of morning, Marcus could see her holding the phone against her shoulder while watching two screens. She was getting the same report from two brokers independently. A European bank’s structured products desk was short gamma on one of the seven names, a position built over six months writing put options to institutional clients who wanted protection against a fall. At current prices their hedge required them to sell roughly 400 million dollars of the underlying stock for every five percent the name dropped. Marcus did the arithmetic without writing it down. The bank’s desk was not bearish. They did not want to sell. They had sold insurance against a fire and were now obliged, by the mathematics of their own hedge, to pour petrol on it: every sale they made drove the price down, and every fall in the price obliged them to sell again. They could not stop. The position would sell itself down to nothing or until the name reached zero, whichever came first, and neither of those was going to be orderly. He had described exactly this mechanism, in the abstract, in the August note nobody had distributed. It was less satisfying to watch than it had been to write. In the meeting room the chairman asked what it meant for the fund’s remaining position. Marcus said it meant the last 280 million was not going to move at anything close to the price on the screen. The head of compliance wrote that down. The chairman looked at the floor through the glass and said, quietly, that they should have had this conversation earlier. Marcus did not disagree. He thought about the note. He decided not to mention it. The silence afterward was the silence of people who have understood that the time for the conversation has passed. New York · 12:15 At twelve-fifteen, three algorithmic trading systems, two American and one European, identified the same anomaly in AI sector options pricing within the same instant and executed coordinated sell orders across a window of ninety milliseconds. No human authorised the trades. The systems had been built to act on exactly this pattern and they acted correctly. The Nasdaq fell another four hundred points in the time it took Marcus to walk from the meeting room to his desk. He saw Dana Marsh stand up as he crossed the floor. In six years he had never seen her stand up during a session. New York · 12:47 A Bloomberg alert noted that the yen had strengthened 3.2 percent against the dollar in ninety minutes. The carry trade was unwinding. For years, money had been borrowed in yen, where it cost almost nothing, and invested in dollar assets that paid more. Now the borrowers were buying yen back to repay the loans, which drove the yen up, which meant anyone still holding the trade was losing on both legs at once: their dollar assets falling and the currency they owed rising against them. The most recent survey had put the trade above four trillion dollars. Marcus had read the survey when it came out. He had filed it under things that were true but not yet urgent. It had become urgent at the rate such things always did, which was slowly and then all at once. New York · 13:15 At a quarter past one the bond market did the thing that stopped him. Ten-year Treasury yields, which had fallen at the open as money fled into the safety of government debt, were now rising. Twelve basis points in forty minutes, and the direction was wrong. In every crash since 1987 the reflex had held: equities fall, Treasuries rally, the loss in one cushioned by the gain in the other. The reflex had lasted two hours this time and then reversed, which meant the bond market had begun, for the first time in Marcus’s career, to ask out loud a question it had spent years declining to ask. Whether a government already carrying that much debt, already paying for a war it could not afford to lose, could be relied upon to absorb a crisis of this size as well. The deficit was not new. The war was not new. What was new was that the equities were crashing and the usual shock absorber had stopped working, and a market that had always treated American government debt as the safe place to stand was beginning, very slightly, to step back from it. New York · 14:04 At four minutes past two a single line appeared on the sovereign credit screen. US 5Y CDS, plus fourteen basis points. Marcus looked at it for a while. Fourteen basis points on a five-year credit default swap on United States government debt was not, by itself, a catastrophic number. What it was, was movement on an instrument that was not supposed to move in that direction. A credit default swap is insurance against a borrower failing to pay. For the premium on insuring the United States to rise meant that somewhere, someone with real money was paying to protect themselves against the possibility, however remote, that the American government would not meet its obligations. The premium was small. That it existed at all, and was rising, was the largest thing that had happened all day, and almost no one outside the room he had just left would understand why. New York · 14:31 At half past two the market turned and went up. It went up hard, the way these things always do in the afternoon, four percent off the lows inside twenty minutes, and the floor came alive with it. The rumours started at once and raced the length of the room. The New York Fed was on the wires buying. The President was about to sign an order banning Chinese AI from American enterprise altogether, which would put NovaMind back in its box and the seven names back where they belonged. There was a G7 call tonight and it would produce a coordinated facility by morning. None of it was confirmed. All of it was exactly what a falling market wants to hear, and a market that wants to hear something badly enough will, for a while, behave as though it were true. Marcus watched the younger traders around him lean into it. He could see the reflex firing in them, the one thing their entire careers had taught them, which was that the dip is the opportunity, that every fall since they had started was a thing you bought, that the people who held their nerve and added were the people who won. They had never once been wrong about this, because the market they had grown up in had never once made them wrong. They were buying the rally now with the particular conviction of people who have only ever seen it work. He did not move. He knew the desk at the European bank was still there, still short gamma, still mechanically obliged to sell into any strength because every tick higher in the underlying changed its hedge and the hedge did not have opinions about the Fed. He knew the margin calls that had gone out at noon had not been met and that the forced sellers behind them would use this rally for exactly what it was, which was the first decent bid they had seen all day, a better price to sell into. A rally in a market like this one was not hope arriving. It was liquidity arriving, briefly, for the people who had to get out and could not. The believers were providing the exit the forced sellers needed. What made it worse than any fall he had studied was the sheer quantity of borrowed money the market had carried into it. The record margin debt he had flagged in August was now doing exactly what he had said it would. Every forced seller was forced because of leverage, and there had never been this much of it, so the selling that begets selling had more fuel than in any previous break, and it burned faster. It lasted forty minutes. At ten past three the buying ran out of conviction and the selling, which had never actually stopped, came through it and over it, and the market made a new low, lower than the morning, and this time it took with it everyone who had bought the bounce. The morning’s losses had belonged to the people who came in long. The afternoon’s belonged to the people who had been brave. By the close there was no one left who had not been hit, which is the precise condition a market has to reach before it can stop falling, and it had not reached it yet. He went down to the lobby a little after three, because the floor had become a place he needed five minutes away from. By the lifts two of the building’s receptionists were comparing notes on their weekends, one of them describing a man her sister had set her up with, the date a qualified disaster, the restaurant at least very good. The other was laughing. Neither of them knew, and why would they, that eleven floors above their heads the savings of a generation were being marked down by the minute. The lift arrived. The world that did not trade was going to be fine, or at least no worse than usual, and there was something steadying in the ordinariness of a man who had ordered the wrong thing at a good restaurant. Back at his desk he watched the last of the position fail to sell, and thought about his father on the floor at Kidder Peabody, the quiet by midday, the exit smaller than the position. Not enough and not fast enough. He had got more of his own book out than his father had, and earlier, and it had still not been enough, and he understood now that it never could have been, because the whole point of the thing was that everyone was trying to leave through the same door at the same time. The Nasdaq closed down a little over nineteen percent on the day. With the ground it had already given up in the fortnight before, that put the index more than twenty percent below the high it had set in August, which is the line the textbooks and the broadcasters use to mark the end of one thing and the start of another. The bull market that had run, with interruptions, for the better part of seventeen years, and that everyone trading that afternoon under the age of forty had taken for the natural condition of the world, was now formally over. They would call it a bear market in the morning, the broadcasters reaching for the word with a kind of awed relish, and most of the people on the floor would be hearing it used about their own working lives for the first time. The machines had not failed today. The algorithms had done precisely what they were built to do. The hedgers had hedged exactly as their models required. The founder in San Francisco had been entirely right about his own position. Nothing had malfunctioned. Everything had worked. That was the part his father had never quite said, and that Marcus said now, under his breath, to no one in particular. The technology did exactly what it was told. Thanks, Dad. V. The Credit Line Guildford, Surrey, October 2026 I want to be clear about one thing from the start: I am not someone who complains. I built Ashworth Precision Components from a redundancy payment and a leased lathe in 2001. We make components for aerospace subcontractors and medical device manufacturers. Twenty-three people. I know all of them by name and most of them by the particular sounds their machines make at full load. I know which of them is buying a first house and which has a daughter starting at Surrey in September and which is quietly carrying a wife through a long illness. I did not build this company by sitting around wondering why things were difficult, and I am not in the habit of explaining myself. I am explaining myself now because I have spent three months trying to understand how something assembled over twenty-five years can be put at risk in a fortnight by people I will never meet, for reasons that have nothing to do with anything I or my customers or my workforce ever did, and I have not managed it, and writing it down is the nearest I can get. In the spring I took on an additional facility of three hundred and forty thousand pounds. This was not reckless. The business case was straightforward: an AI-assisted optimisation system for the production line, projected to reduce material waste by fourteen percent and to let us take on higher volumes without more headcount. Our accountant ran the numbers four ways. The bank approved the facility without much discussion. Three of our six major clients had signalled larger orders for the coming year. We were, by any reasonable assessment, well positioned. I want that word to sit there for a moment, because every decision I made was sound on the information any sensible person had. That is the part I keep returning to. What happened afterwards happened in stages, and the stages arrived far enough apart that for a while I did not connect them, which I now think was its own small mercy. The first was a call in late September from a client in Bristol who makes components for the automotive sector. He wanted to discuss order phasing. He was careful about it, mentioning a review of capital expenditure, but I had known him twelve years and I understood what he was telling me. His own biggest customer, an American firm, had frozen a programme. I said I would send a revised schedule. I put the phone down and stood in my office for a minute and told myself it was one client having one difficult quarter. The second was an email a fortnight later from one of our three largest aerospace clients, pausing a planned expansion and asking to review the supply arrangements accordingly. Politely worded. The kind of politeness that tells you more than the words do. I read it twice and felt, for the first time, something I can only describe as the floor moving very slightly under a building you had believed was on rock. The third arrived in the first week of October: a letter from the bank, raising the rate on our facility by 1.4 percentage points with effect from the first of November, citing prevailing market conditions. I put it on the desk and looked at it for a long time. I am not a man who shouts. I went out to the yard instead and stood in the cold for a while, and what I felt was not panic, not yet, but a sort of disbelieving anger that I could not aim anywhere, because there was no one in the building to aim it at and the people it belonged to were five thousand miles away and did not know my company existed. I knew about the crash, of course. Everyone did. You could not turn on the radio or pick up a paper without it. The American markets had fallen further in a day than at any time since 1987, something to do with Chinese artificial intelligence undercutting the Americans, and the broadcasters were using the word recession, and there was a good deal of talk about whether it would reach us. I followed it the way the whole country followed it, as a spectacular calamity happening to other people, to bankers and Americans, with, I will admit, a small and not very creditable satisfaction that the people who played with money for a living had finally been caught out. I assumed it would sort itself out. These things always seemed to. What I did not do, not once, was connect any of it to Ashworth Precision Components. Why would I. I make parts for aeroplanes and surgical instruments. I had never owned a share in my life. The notion that a number on a screen in New York had anything to say about my order book would have struck me, had it occurred to me, as faintly ridiculous. It did not occur to me. That, I have come to understand, was the whole trick of the thing. On a Friday in early November I had a call from Andrew at the bank. I have dealt with Andrew for nine years and trust him, which in banking means he tells you what is coming before it arrives rather than after. He said he wanted to discuss the facility. He used the word covenant twice, which Andrew does not do lightly. He asked if I could come in on the Monday. I said I could, and then I did not sleep especially well for the three nights in between, which I would not admit to him or to my wife or to anyone except a sheet of paper. On the Sunday I read a long piece in the paper about the crash. Graphics, a timeline, the whole apparatus. I read it at the kitchen table with my second cup of tea while my wife made lunch. She asked what I was reading and I said something about the stock market and she said it all sounded very complicated, and I agreed, and turned the page, and did not tell her that I had begun to understand it was about us. The piece laid out the sequence. A Chinese company had announced it would sell artificial intelligence to American business at a fraction of the going rate. The American AI companies, whose enormous stock-market values rested on charging the old rate, fell. Funds that owned them and had borrowed to own more were forced to sell. The selling fed on itself through machinery I did not fully follow, options and hedges and something it called short gamma, until the fall had spread far beyond technology and tipped the United States, the piece said plainly, into recession. There was the word. A recession in America, it explained, pulls down everyone who sells to America, and everyone who sells to them. Somewhere in the middle of that sentence, although the piece did not mention Guildford and had never heard of me, was a precision engineering firm in Surrey whose order book had begun to empty in late September for reasons that had been set in motion, I worked out, on a Monday morning by a pricing announcement made on the far side of the world. I am a careful man. I do not believe in blaming distant forces for failures closer to home. But I have turned this over every way I know how, and I cannot make it my fault, or my customers’ fault, or the fault of anyone who actually makes or builds or repairs a single thing. A handful of enormous companies in California, run by people of colossal self-regard, were valued by the market at sums that assumed a future that was never going to arrive, and when it did not arrive the bill was sent, not to them, but outward and downward, to people like me and the twenty-three people who depend on me, who were doing nothing more reckless than competent work. That is the part I cannot get past. The competence did not protect us. It was never going to. I drove to the bank on the Monday. Andrew’s office is on the second floor with a view of the car park. He offered coffee and I took it. He is a compact, careful man in his forties who handles a hard conversation the way a good doctor does, directly and without drama and with the particular kindness of someone who knows the news will land however gently he sets it down. He said the bank was reviewing its small-business lending in light of conditions and that our facility had been flagged for closer monitoring. He used that phrase, closer monitoring, three times. He asked about the order book for the first quarter. I told him about Bristol and the aerospace client. He wrote it down. He said he would need monthly management accounts and a quarterly review, and that if things deteriorated further there might have to be a conversation about the covenant, and that the bank remained supportive. I know what supportive means. It means supportive until it is not. I thanked him for his time, because none of it was Andrew’s doing either, and drove back. I walked the shop floor when I got in. We were running at about seventy percent, because two orders that should have been on the machines were not. Phil, my shop manager, caught my eye from across the floor. He has been with me nineteen years and can read me better than my wife can. I gave him the nod that means we are fine. It is the nod I give when I need us to go on being fine, and he knows that too, and gave it back, and neither of us said anything, and that wordless exchange between two men who have built something together and can feel it being threatened from somewhere they cannot see was, I think, the hardest moment of the whole business. On the wall of my office is a photograph of my grandfather’s machine shop, taken in 1962. Two men in overalls beside a capstan lathe, looking at the camera the way working men did then, as a brief interruption to be got through. My grandfather is on the left, with the expression of a man who intends to get back to work the moment the shutter closes. I have looked at that photograph for fifteen years and only lately noticed that expression. He ran his business through three recessions and two upheavals in manufacturing that people at the time called the end of everything. I do not know whether he ever felt what I felt that Monday. I suspect the names were different and the experience was the same: a practical man, who deals all day in steel and tolerances and things you can hold, discovering that abstractions have mass after all, and that they can come through a wall he did not know was thin. I learned, from that newspaper piece, that one of the great American investors had been sitting on more than three hundred billion dollars in cash for the better part of two years before the crash, having said openly and repeatedly that he could not find anything worth buying at the prices on offer. I had never heard of this. I have since asked a number of people whether they had, and most of them had, and not one of them had thought it had anything to do with them. I am not certain I would have thought so either. But a man whose entire trade is judging the worth of things had been quietly refusing to buy, in plain sight, for two years, and the rest of us, who do not have the time or the training to read the signs, were never told in any language we would have heard. I did not know I was in this story. But then, as far as I can tell, that was the whole design of it. The people who were in it knew. The people it was going to happen to did not. And I have come to think that the distance between those two facts is the truest measure of what was done. Epilogue April 2027 They called it the AI Crash, in the end, which was the name the newspapers settled on within the week and the rest of the world adopted because there was no competing version worth the trouble. It was not a clever name. It did not need to be. The industry had spent a decade making those two letters a promise, and now they were the name of the thing that had taken the savings of a generation, and the flatness of it was its own small verdict. On the floor they said worse, of course. One of Marcus’s traders, walking out at the close of the second day with twenty years of his own money gone, had stopped at the lifts and said, to no one and everyone, that he had spent his whole career being told the machines were intelligent, and that he would like to meet the genius who had named them, because artificial was the only honest word in it. Nobody printed that. Everyone who was there remembered it. The second day was worse than the first, which almost no one had expected. Overnight the coordination had come, the calming statements, the finance ministers on their call, the central banks promising whatever it took, and the market had opened sharply higher on the strength of it. For two hours it held, and the same generation that had bought the dip on the Monday afternoon bought it again on the Tuesday morning, more convinced this time, because now the authorities were visibly behind them and the system was doing what the system was supposed to do. And then, slowly across the Tuesday and faster across the Wednesday and the days after, the forced sellers ground it back down, because leverage does not leave a system in a morning, it leaves over weeks, and every rally was simply another window through which the people who had to sell could sell. By the end of the second week the people who had bought the second-day rally had learned the same lesson as the people who had bought the first, which was the oldest lesson there is and the one the long bull market had quietly persuaded an entire generation to forget: that sometimes the dip keeps dipping, and that the name for what they were now living through was the word their whole careers had taught them to treat as a museum piece. A bear market. The one they never thought they would see. In March, the Ministry of Industry and Information Technology announced a review of NovaMind’s commercial licensing structure. Dr Fang Wei was not involved in the review. He had been transferred in November to a research coordination role with fewer external responsibilities and more time, as his director had put it, for long-form work. His 2023 paper on cascade failures had been cited in three post-crash academic papers in the first quarter of 2027. He had not been told this. He found out by running a citation search on a Tuesday afternoon, alone in his office, eating a sandwich. He had not shown anyone. His doctoral supervisor’s reply to the September email had arrived eleven days after the crash, full of questions Fang could not then answer, and they had spoken once on the phone in January. His wife had asked him twice since November whether he was sleeping. He had said yes, which was true four nights out of seven. Elise Vandermeer’s fund closed the year down 22 percent. She kept her job. The CIO who had said that Berkshire had a different problem from theirs was moved to a strategic advisory role in February, which is the financial industry’s form of a departure that has not yet become one. The risk committee waiver was the subject of an internal review whose conclusions had not been shared with Elise. Jakob found work in March with a regulatory consultancy in Frankfurt. It was not what he had trained for, but it paid and he had sounded, when she spoke to him, like himself again. She had taken the Kindleberger from the shelf once in February, stood holding it for a moment, and put it back. She had not opened it. Marcus Reeves no longer ran a portfolio. He described himself as an independent adviser. The phrase was accurate and meant several things at once. He agreed to speak at a risk management conference in March. His talk was titled “Concentration Risk and the Limits of Consensus Positioning” and it was well received. On his way out a younger fund manager stopped him and asked whether he had really known in August that it was going to break. Marcus thought about the note he had not distributed, the Berkshire joke at the Park Avenue dinner, his father’s voice saying not enough and not fast enough. He said he had been concerned, which was true, and that concern and action were not the same thing, which was also true. He made the Berkshire joke again at a dinner in February, differently inflected. It did not get a laugh. Richard Ashworth’s facility had been restructured in January under a three-year covenant monitoring agreement. He had not taken on new debt. He had not replaced the two members of staff whose positions had been effectively eliminated when the Bristol and aerospace clients reduced their orders. The AI optimisation subscription had been cancelled in November; the system was removed by the supplier’s engineer on a Wednesday morning while the shop floor ran at 60 percent. Phil had watched without comment. Richard was running at 78 percent capacity in April and had two new enquiries he thought might come to something, though he had stopped letting himself count on things before they arrived. The photograph of his grandfather was still on the wall. The founder in San Francisco was, as he had predicted, completely fine. His company’s share price had fallen 71 percent from its peak and stabilised at a level that still made him one of the wealthiest men alive. He gave a widely shared interview in February in which he described the AI Crash as a necessary clearing of unserious capital and said the firms that survived it, his own foremost among them, would define the coming decade. He was not wrong about his own survival. He was asked, once, about the pension funds and the smaller businesses that had not survived, and he said that creative destruction had always been the price of progress and that history would judge the technology by what it built, not by what it disrupted on the way. He believed this. That was the part that stayed with the few people who were paying attention: not that he had said it, but that he believed it, and that nothing in his experience of the previous six months had given him any reason not to. The larger consequences were slower and are still unfolding. The American recession that the markets had forecast within hours arrived within the quarter, and was not shallow. Europe followed, as Fang Wei had drawn it on a diagram three years before: Germany through its machine tools, the Gulf through oil, and a great many firms like Richard Ashworth’s through orders that simply stopped. The dollar fell and did not fully recover, and in the capitals that had spent two decades looking for an alternative to settling the world’s business in it, a number of quiet conversations that had previously gone nowhere began, for the first time, to go somewhere. No one called the pricing of an artificial intelligence product an act of war, because it did not look like one and there was no provision in any treaty for what to call it. But a great deal had been demonstrated, on a single Monday in September, about how a modern economy could be reached and damaged without a shot, and the demonstration had been watched very carefully by people whose business it was to watch such things. What had been done once was now known to be possible. That, in the end, may prove the most expensive thing about the whole affair, and the bill for it has not yet been presented. This Substack is reader-supported. To receive new posts and support my work, consider becoming a free or paid subscriber. Get full access to David McCreadie at davidmccreadie.substack.com/subscribe (https://davidmccreadie.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4)

2. 

   16 Jun 2026

   Hi Vis

   The Telegraph letters page is a competitive place. This week it had the Iran ceasefire, two ministerial resignations and an American AI story to contend with. My letter about a West Sussex car park did not make the cut. It finds its home here instead, as the centrepiece of Episode 5, which turns out to be about rather more than parking. It is about the summer jobs of the early seventies, the travel they funded, the lessons they left behind, and the unexpected satisfaction of a hi-vis jacket on a fine June afternoon. There is a Victorian fraud, a Hollywood cowboy, a hotel general manager called Mr Smart, and a brief encounter with a relatively unknown comedian in a hotel ballroom in Inverness. The comedian was Billy Connolly. We were sixteen. Nobody can hold that against us. This Substack is reader-supported. To receive new posts and support my work, consider becoming a free or paid subscriber. Get full access to David McCreadie at davidmccreadie.substack.com/subscribe (https://davidmccreadie.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4)

3. 

   12 Jun 2026

   Letters to the Telegraph: The Temperament Question

   This week I sent a single quotation to the Telegraph. No comment, no gloss. Just Churchill, House of Commons, November 1936, on a government that could not bring itself to rearm. The letters editor will decide whether to run it. But the resignation letters of John Healey and Al Carns, landing within hours of each other on Thursday, made the argument for me. This episode takes those letters as its starting point. Not the politics of the reshuffle, which will be analysed to exhaustion elsewhere, but the question the letters raise about the man who received them. What the discipline of leadership profiling, a tool built by the CIA and used on foreign heads of government before every major summit, reveals when applied to a Prime Minister closer to home. It is the companion piece to an essay I published here on the seventeenth of April, The Hostile Hypothesis, which tested Starmer’s record against an adversarial template. This episode borrows a different instrument from the same tool-box. It asks not what he intends but how he decides, what he reaches for under pressure, and where the gap sits between what he declares and what he allocates. It also invokes the Goldwater rule, the reason psychiatrists are forbidden from diagnosing public figures they have never examined, and explains why the profiling question is both more rigorous and more troubling than any clinical label would be. The Asquith parallel is not decorative. It is, I think, exact. Running time: approximately twenty minutes. This Substack is reader-supported. To receive new posts and support my work, consider becoming a free or paid subscriber. Get full access to David McCreadie at davidmccreadie.substack.com/subscribe (https://davidmccreadie.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4)

4. 

   10 Jun 2026

   The Road That Ends at Broadwater Farm

   Last week I sent a short letter to the Telegraph about the disorder in Southampton and the death of Henry Nowak. Its argument was a simple one. When a government meets legitimate public anxiety with a charge of bigotry, it does not make the anxiety disappear. It abandons the people who hold it, and leaves the ground to those who will put it to worse use. In this third episode I read that letter, and then the essay it grew into, “The road that ends at Broadwater Farm”. A letter can only name a parallel; the essay argues it in full. It follows a pattern that runs through the Northern Ireland civil rights marches of the late 1960s, the Brixton riots and Lord Scarman’s warning, Michael Heseltine’s very different answer to Toxteth, the burning of Tottenham in 1985 and again in 2011, and a national scandal vindicated in plain sight only this past year. The thread is the same in every case. A grievance dismissed is not a grievance resolved, and the cost of ignoring it is always paid, usually late, and usually in full. Get full access to David McCreadie at davidmccreadie.substack.com/subscribe (https://davidmccreadie.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4)

5. 

   7 Jun 2026

   Letters to the Telegraph; The Phoenix

   Welcome back to Letters to the Telegraph, the weekly series in the spirit of Alistair Cooke’s Letter from America. Each episode takes one of my letters to the editor, sometimes published, sometimes not, and walks around the story behind it. This episode is built around a letter on a subject that is hard to escape at the moment, the state of British defence, and one document in particular that was promised a year ago and has still not appeared. A letter can only assert. It has room to say that the war in Ukraine has changed the character of land warfare, and to reach back to the Spanish Civil War for a warning, but it has no room to show why any of that is true, or what it costs. So this episode does what the letter could not. It begins in Washington, with a story a Ukrainian general told against his own allies, and it ends with a young man who told me why he would not join the Army. In between, the Phoenix drone, nineteen years in the making and nine in service, Northern Ireland, and Munich. The letter ran as follows.Sir,A year on from the Strategic Defence Review, the Defence Investment Plan that was supposed to give it teeth has yet to appear. The conventional reading is procrastination.There is a more generous one. The war in Ukraine has demonstrated a transformation in the character of land warfare. Persistent drone surveillance has made the battlefield transparent. One-way attack drones at a few hundred pounds apiece are doing the work of guided missiles costing many times as much. The innovation cycle now runs in weeks, not years; a Russian electronic warfare adaptation is met by a Ukrainian software counter within days. Authority over drones now extends from purpose-built command centres running coordinated strikes down to platoon operators acting in minutes. These are not features of one theatre. They are how land warfare now works. If the Plan has been delayed to absorb that, the delay is well spent.The historical parallel is instructive. The Spanish Civil War of 1936 to 1939 was the Wehrmacht's laboratory, and the Germans entered the Second World War three years ahead doctrinally. Britain and France watched the same war and drew few practical conclusions about combined arms. By June 1940, France had fallen.The Defence Investment Plan, whenever it arrives, will tell us which lessons we have drawn. The harder question is not the equipment list but whether the procurement cycle, the command authorities and the relationship with industry have been redesigned for an environment in which capability is updated in weeks. If the answer is yes, the optimist will be vindicated. If the Plan is a shopping list issued on the customary timetable, the historical parallel will be the other one.Yours faithfully,David McCreadieIn the episode I take its claims one at a time and put the evidence underneath them. Why the innovation cycle now runs in weeks rather than years, and what that has done to the battlefield. Why a country that has understood battlefield drones since 1964, and flew its first that year, still cannot field them on a sensible timescale. What the Phoenix, nineteen years in the making and nine in service, tells us about how we buy our equipment, and why Spain in the 1930’s and Munich in 1938 are the parallels that matter now. It runs to a little over twenty minutes, and it ends, as these things should, not with a document but with a person, a young man who told me why he would not join the Army. I hope you will listen. This Substack is reader-supported. To receive new posts and support my work, consider becoming a free or paid subscriber. Get full access to David McCreadie at davidmccreadie.substack.com/subscribe (https://davidmccreadie.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4)

6. 

   26 May 2026

   Letters to the Telegraph; The Cap Badge

   Welcome to Letters to the Telegraph, a small weekly series in the spirit of Alistair Cooke’s Letter from America. Each episode will take one of my letters, sometimes published in the paper, sometimes not, and walk around the story behind it. This first episode is built around a letter prompted by two stories from the same week, both involving Glasgow and the same regiment. The letter ran as follows. Sir, Two stories from the same week, both involving Glasgow and the same regiment. The family of Fusilier Dennis Donnini VC, the youngest Victoria Cross winner of the Second World War, have donated his medal to the Royal Highland Fusiliers Museum in Glasgow. Donnini was 4ft 10in tall, barely above the Army’s minimum height, and nineteen. Shot in the head at Selfkant in January 1945, he charged the enemy alone, was hit again, and kept firing his Bren until killed. His regiment, the 4/5 Royal Scots Fusiliers, lives on as 2 SCOTS of the Royal Regiment of Scotland. In the same week, by 39 votes to 28, SNP and Green councillors blocked a Labour motion granting the regiment the Freedom of Glasgow on its twentieth anniversary. The SNP called the honour “inherently problematic”. More than twenty Scottish councils, including Edinburgh and Aberdeen, have conferred it. Glasgow itself has rarely granted this honour to a regiment. The Highland Light Infantry received it in 1948, the Royal Highland Fusiliers in 1959, and the Royal Marines in 2014. The first two are the antecedents of today’s 2 SCOTS. One Durham family gave the regiment something priceless. One Scottish council refused it something that costs nothing. A Labour councillor called the vote shameful. The Prime Minister agreed. They are both right. Yours faithfully, David McCreadie The episode talks around the letter for twenty minutes or so. It is about three Fusiliers across eighty years, all of whom served in the same regiment. One won the Victoria Cross at nineteen in 1945. One joined the Army as an orphan and listed his Commanding Officer as next of kin. The third was my friend Dick Davidson, who served in Northern Ireland, at Lockerbie and in the Gulf and who was made compulsorily redundant under Options for Change. Dick died last month. Running time: about twenty-two minutes. Best heard with a cup of tea or, if the hour is right, something stronger. Get full access to David McCreadie at davidmccreadie.substack.com/subscribe (https://davidmccreadie.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4)