The boom is financed by a machine that looks like the year two thousand and settles like two thousand and eight, the jobs it ends will be replaced by an income with a switch on it, and the only defence he offers an individual is a skill most people have never been taught.
DeepDives · 2026 · An AI–human collaboration that weighs every source, mainstream or alternative, under strict rules of truth and logic.
The three claims, stated at full strength#
Simon Dixon's appearance on the Peter McCormack Show — an episode the show called "They're Coming For Everything" — is one long conversation, and this is the second of three examinations of it. The first took the debt half: the forty trillion dollar stock, the nine point seven trillion dollar rollover, the identity of the buyers, and concluded that he describes a real machine with the wrong idea about who drives it. This one takes the other half: the technology the debt is financing, the jobs it is said to end, the money built to replace the wages, and the one skill he believes separates those who survive the transition from those who are processed by it.
Three claims, stated at full strength before any of them are tested. First: that the AI boom is financed by a debt machine that looks like the year two thousand and settles like two thousand and eight. Second: that the technology will end enough middle-class jobs that the state must step in with a programmable replacement income, and that the replacement will arrive with conditions. Third: that the only reliable defence an individual has is to hold their own keys.
That last claim is where this series is unusually well equipped: the hardware wallet attack he discusses at length is one it has already published on, four episodes deep, so for once a guest's claim can be tested against detailed prior reporting rather than general principles. The incentive is stated in his own account — Dixon's first job was selling initial public offerings, so he sat on the sell side of the structure he now anatomises.
The money machine: the float as an exit#
His account of how the boom is financed is the sharpest thing in the conversation, and it starts where most finance commentary skips: the initial public offering as an exit. A company sets an enormous valuation; the banks underwrite it; a slice of stock is floated to the public; and the early investors — venture funds, angels, employees with options — carry lock-in schedules that let them sell into the peaks. His own task, in the job he once held, was to sell that stock to the public, and the buyer that mattered was always the institution. The public is not the customer of a float. The public is the exit.
Then the step that matters for everything after it. The newly public company uses its inflated stock as currency: it buys other companies with it, because the stock is worth more than cash to a seller being paid in something that has been rising, and it borrows against the stock as collateral, because lenders accept it. Dixon's phrase is that the stock becomes a money printer at the stock level, and the phrase is exact.
His illustration is the float everybody watched this year: a company went public at a valuation near one and three quarter trillion dollars and immediately went on an acquisition spree, buying other companies with its own paper. Then the chart. The stock ran up toward a hundred and ninety dollars, fell to around a hundred and ten within six weeks — nearly halving — and recovered to somewhere in the one forties, below where it started. The people who bought at the top of the most anticipated float of the decade are down; the people with the lock-ins are out. That is not a glitch; it is the system performing as designed, and Dixon says it bit him too.
The eight-hundred-billion-dollar queue#
Follow the money one step further and the analysis connects to work this series has already published. The newly public giants are not only buying companies with stock. They are borrowing, on an enormous scale, to build the physical infrastructure of artificial intelligence, in the form governments use: corporate bonds, sold to the same investors who buy government debt.
His numbers are large enough to write down. The industry needs to raise something like eight hundred billion dollars this year and another one point three trillion next year, at six or seven percent for thirty or forty years. The chip maker at the centre of the build reports roughly a hundred billion dollars of revenue in a quarter and projects a hundred and eight billion in the next. The largest technology companies now account for between thirty and forty percent of the entire American stock index, and the concentration is still rising: he counts the seven giants, adds the rocket company, adds the two largest private AI laboratories, and calls the result the magnificent ten.
Then comes the sentence that ties this examination to the one before it: the AI build-out is competing for capital with a government that must roll over nine point seven trillion dollars this year. The government and the industry stand in the same queue, borrowing from the same pool of lenders, and the pool is not unlimited. When a company offers seven percent for forty years and a government offers four percent for ten, the lender chooses. The same marginal buyers — the funds, insurers and pensions examined in the first part of this series — are being asked to fund both, and there is a price at which they cannot do both.
Two thousand, or two thousand and eight#
The comparison he offers next is the one most people in his position get wrong, and his answer is honest. Asked whether this ends like the year two thousand or the year two thousand and eight, he says he does not know. The year two thousand was a stock-valuation event: the money was lost in the equity of companies that never earned a profit, and the damage stayed with the people who owned the paper. The year two thousand and eight was a liquidity event: the securities at the centre were mortgage bonds held inside the banking system, so when they fell the banks could not fund themselves, and the damage went through the plumbing of the whole system.
The honest answer is both channels at once, and he says so. The equity part is obvious. The debt part has migrated into private credit, the funds that raise money from investors and lend directly to companies — the same Cayman-domiciled vehicles and large asset managers this series examined when it looked at that market, lending to data centres, funded in turn by borrowing from Japanese institutions at low rates, or from the banks. The bank exposure is indirect but real, and who rolls it all over is the same question as before.
His conclusion is that the answer is finance itself, the bond market, and that what is being witnessed is a transfer of power to three groups: the technology companies building the machines, the financial companies lending them the money, and the military companies being paid to protect the arrangement. The winners of two thousand and eight were the largest asset manager and the largest bank, left standing when the government paid to resolve the crisis.
One refusal deserves credit before the testing begins. Asked whether the pump is finished, he says no — there is more of the pump to come, because the sequence is not complete. The two large AI laboratories have not gone public, and the debt has not been raised, and the offerings must finish before the market can discover what the assets are worth. That refusal to call the top is the correct posture: the sequence has to complete before a reversal becomes possible. This series reached a compatible conclusion through its risk dashboard: the private credit market shows three of the stages of stress at once, but the forced mark — the moment funds must admit what the loans are worth — has not happened, and that unfired stage is the next transition.
The price signal pointing both ways#
The most interesting tension in his argument is also where he handles a contradiction best. He has just described an inflation he expects to become incomprehensible, then introduces the force pushing the opposite way and does not blink: he explains it. Artificial intelligence, he says, is disinflationary by nature. The more productivity a business gains from it, the cheaper its output becomes. He gives a number from his own experience worth sitting with: his worst month of spending on AI services cost him nine thousand pounds, and in that month he built software he values at about a million pounds.
Then the qualification that turns a slogan into an analysis. The models themselves get cheaper over time, and so will the expensive frontier layer, because the frontier laboratories are always several generations ahead of what they sell and will keep the best for themselves. His example is an image-generation company that developed a body scanner, and what matters is the business model behind it. The laboratories will gatekeep the most valuable capabilities, build businesses on top of them, and monetise what the public never gets to use: the medical advances, the discoveries, the applications that would be worth the most if everyone could have them. And almost every technology he uses was originally developed with military budgets. He is not predicting a betrayal. He is describing a pattern so old that it is the default.
The synthesis he never states directly. The value of the boom is captured by the people who own the infrastructure — the data centres, the data, the stocks — and those owners are the same companies announcing the layoffs. The Chinese are catching up on efficiency rather than spending, building cheap open models for industry, which eventually pushes Western frontier revenue down while Western valuations stay where they are. Put the halves together and the picture is neither simple inflation nor simple deflation: the things you must buy get more expensive as the currency is debased and the population ages, the things the machines make get cheaper, and the wage sits between them, squeezed from both sides.
Fifty percent, and the shape of the claim#
Now the claim that matters most to anyone with a job, or with children who will need one. Dixon is asked what share of jobs are under threat in the next two years, and he gives a number and a verdict: fifty percent of middle-class jobs, and no chance that the economy makes up new opportunities at anything like the rate the old ones disappear.
Note the shape of the claim, because the shape is where the argument lives. It is not that some jobs go, which is the uncontroversial part. It is that the replacement mechanism fails, which is the specific and falsifiable part. The historical record is unkind to confident predictions in either direction: the automation panics of the last two centuries have mostly been wrong about the totals and right about the distribution — the total number of jobs did not collapse, while the specific people in the specific jobs were destroyed.
His own examples point at that more careful version. He is describing the destruction of the middle-class administrative job: the person who gets on a train, travels to a city, sits in an office, and moves information around on a screen. That work is pure information processing, which is exactly what these systems do — which is why the displacement is concentrated in one stratum of the labour market rather than spread evenly. He puts the economics in the form of a question the listener cannot answer comfortably: what happens to a job when the machine does it better than the human for twenty dollars a week, twenty-four hours a day, with no employment rights, no minimum wage, no sick days, and no contract?
He also concedes the boundary, which most doomsayers refuse. The in-person jobs survive — the coffee, the beer, the flight attendant — because people want a human being across the counter, and he says flatly that he does not want to go to a pub where a robot serves him. That concession is what keeps the rest of the argument respectable.
The verdict on the number: it is assertion, offered without a source, and it should be treated as a mood rather than a forecast. The mechanism he identifies — displacement concentrated in the information-processing middle — is the one the evidence supports: the gains accrue to whoever owns the machines, and the labour displaced does not automatically reappear. The distribution is more certain than the total, and the claim that half the middle class is finished within twenty-four months is a way of saying something true with a number that cannot be defended.
Three kinds of people, and the money with a switch#
He goes next where a financial analyst is not required to go: into what the state will do about it, which is the part of the conversation giving the episode its title. His organising idea is a division into three kinds of people. The first owns assets and borrows against them, in a world where asset prices rise and the currency they are denominated in is the one being debased, and they are fine. The second owns nothing and owes something — the person whose labour is being displaced and whose debts are fixed in nominal terms — and for this group he says there is a plan, and the plan is an income: the universal basic income, delivered through the programmable money being legislated right now. The third category is the one he cares about: people who build their own digital infrastructure, run their own models, hold their own identity, custody their own money, and live on the other side of the ledger from the asset holders and the income recipients alike. He is not saying he hopes this is how it works out; he is saying this is what the three positions look like, and the question is which one you occupy.
The programmable money is his most testable claim, and the most often exaggerated by others in his lane; he is more specific. He points at the legislation — the American stablecoin act of recent years and the equivalent British proposals — and makes a distinction worth holding onto: the stablecoin legislation by itself is not the control grid. It creates a legal category for digital dollars issued by private companies, and what matters is what the category is for. His claim is that these instruments are being built as the delivery vehicle for the next rescue. When the crisis comes — the AI build-out unwinding, private credit seizing, a government failing an auction — the state will need to put money directly into the hands of households, and it will use a digital token rather than a bank deposit, because a token can be delivered instantly, targeted precisely, and conditioned.
The conditions are the point. He says the money will come with data requirements and social credit scores, and describes the sequence he expects: the crisis arrives; helicopter money is issued as a temporary measure; prices fall because of the AI deflation he keeps returning to; life feels affordable for the first time in years; and then the recipient discovers the income has a switch on it, and the switch belongs to whoever owns the data. He is not arguing that people should refuse to participate, because participation is the price of functioning in the economy; he is arguing that the terms should be understood in advance.
There is a visible on-ramp in his telling, and it is the part easiest to verify: the recent wave of age-verification laws, which require a person to prove they are an adult before using parts of the internet, are in his account the first mass deployment of the identity layer that conditional money will need. No conspiracy is required to see the shape of that architecture. A system which must know who you are before it pays you is a system that can stop paying you, and every piece of it is being built in public, with legislation, in the name of protecting children from harmful content. That is the strongest part of his case, and it requires no one to be evil: only that a state facing a genuine crisis reach for the most efficient tool available — and the most efficient tool is programmable.
The defence: keys are a discipline, not an identity#
The defence he proposes is what he calls digital resistance. The technology coming to displace and monitor people, he argues, is also available to individuals, and can build a parallel infrastructure: models running on your own machine, identity that you hold, money that you custody. Those skills used to belong to a small tribe, he says; now the tribe is everybody, because technical illiteracy is no longer an option. The skill that matters more than any other is key management, and this is where he stops talking about Bitcoin as an investment and starts treating it as a teaching device. Bitcoin, in his words, is a brutal mechanism for teaching you how to secure a key, because the money is on the line and there is no appeal process: to own it you must hold a secret, and whoever holds the secret owns the coins. He generalises that to everything in the coming world — your identity is keys, your data is keys, your access to services is keys — and the three kinds of people are separated by what they do with their keys. The asset holders give their keys to power and benefit from the system. The income recipients have no keys at all. The free hold their own.
This series has published four episodes on exactly this ground: the episode under examination discusses a hardware wallet attack at length, the same attack those episodes examined. A vulnerability in a device beloved by the most security-conscious bitcoin holders allowed attackers to derive private keys and drain wallets, and he tells anyone holding an affected device to move their funds now. His numbers: about two thousand bitcoin taken, across roughly five hundred people, in a sector whose historical total of coins lost to hacks and failures is around one point three million. It was an attack on the people who did everything right — not an exchange, not a custodian, but the self-custody purists, and the vulnerability went to the heart of that practice.
Here is where the prior reporting adds the correction his telling leaves out. The attackers did not break the cryptography, and they did not need to. The failure was in how the seeds and the passphrases at the centre of the practice were generated, which means the defence was always available and simply went unused by most of the affected holders. Dixon says this, but he buries it: those who had set up multi-signature wallets, requiring several devices from different makers, were not taken, and neither were those who generated their own passphrases instead of accepting the device's defaults. Two thousand coins went to people who bought the right device and skipped the procedures. Self-custody is not a device you buy. It is a set of procedures, and the procedures are the thing that fails. He frames it as an identity; the evidence says it is a discipline, and a discipline can be done badly by good people. The lesson applied honestly is not that the free hold their own keys, but that the free hold their own keys correctly.
The verdict on three claims, and the markers#
On the funding of the AI build, he agrees with what this series has published and he sharpens it. What he adds is the mechanism explaining why the build must keep borrowing: the assets being financed are not durable. A data centre, in his account, is a dead investment in three years, because the chips inside it are obsolete and must be replaced — so the industry cannot pay down its debt out of the cash flow of the first generation of machines; it must borrow again to replace the machines it already bought. It is a treadmill, and a treadmill needs the credit market to keep running. The private credit examination found funds lending against assets whose marks are determined by no market, and identified the forced mark as the stage that has not yet fired. Put the two together and the synthesis is this: the AI build-out is a private credit event waiting for a mark, and the question is not whether the technology works but whether the financing can survive a repricing of what it has bought.
On jobs, he is directionally right and numerically unsupported. The disappearance is concentrated exactly where he says it is — the administrative middle, the information-processing work, the entry-level rungs of the professional ladder. But the fifty percent figure in two years has no basis in the conversation, and the distribution is more certain than the total.
On programmable money, his analysis is correct in its structure and unproven in its motive. Everything he says is being built: the stablecoin legislation exists, the digital identity infrastructure is being deployed in the name of child safety, the technology for conditional payments exists, and the state's incentive to reach for it in a crisis is obvious. The published work on the machinery of financial repression described the same instrument without requiring any sinister intent — a state that must fund itself and cannot raise taxes will reach for the payment rails, because the rails are the point of failure it can control. Where he goes further is in asserting that this is the plan, and there the verdict of the first examination applies unchanged: the mechanism is sufficient, the conspiracy is not proven. If it is a plan, resistance means defeating the planners; if it is a mechanism, the only resistance available is personal — which is the conclusion he reaches either way.
The markers are what turn any of this from commentary into something that can be checked. First, the credit channel, the cleanest test of the AI thesis: watch the cost of borrowing for the data-centre build relative to government debt. If the spread between what the AI companies must pay and what the government must pay widens sharply, the competition for capital is live and the treadmill is slowing. Second, the sequence: he says the pump continues until the two large AI floats are complete, so watch for them and the credit spreads around them. If the offerings complete and the market holds, the reckoning is later than his critics say; if the sequence breaks before completion, it is earlier than he allows. Third, the private credit markers this series already publishes: the discounts on listed vehicles against the value of their holdings, the queues of investors trying to withdraw more than the funds will allow, and the gap between the default rate the funds admit and the one the borrowers experience. The forced mark has not happened. Fourth, the jobs claim: fifty percent of middle-class jobs within two years is specific enough to check against the employment data for administrative and entry-level professional work. Look at the hiring rates for young graduates and the job openings in the information-processing occupations; if the claim is right, those break down visibly and soon. Fifth, the programmable money claim, where the right marker is not the existence of a digital currency but the conditions: watch for the first mass payment program that arrives with requirements attached, and watch whether the identity system built for age verification becomes the gateway for anything else. Sixth, his own practice: multi-signature setups and self-generated passphrases are measurable, and if his warning was absorbed, the wallets that survive the next attack will be the ones that followed the procedures.
The close#
Dixon's advice to a young person is not political: understand the game, then learn the rules, then win, because the old rules no longer produce a win and complaining about it changes nothing. He tells the story of his own father, who worked his whole life, lost everything as a pensioner in the crash of the year two thousand, and asked his son the question that became his career: where did my money go, and who has it? His answer, which is really the thesis of the whole conversation, is that the money did not go anywhere. It was repriced, by a system that no one person controls and everyone in a position of advantage maintains. The advice he draws from it is to keep a balance sheet, to pay yourself an income, to buy assets even in tiny amounts, and to spend less than you earn.
Then, at the very end, the conversation turns away from money entirely: disconnecting from the screens, nine days in Italy with the people he loves, a stray cat they adopted. This one ends with the markers and the verdict, but it would be dishonest to pretend the last note is a market call. Dixon's real argument is that the machine is beyond the reach of any individual to stop, and that the only things an individual controls are their skills, their assets, their keys, and the people they love. The debt machine is real, the queue for capital is real, the displacement in the administrative middle is real, and the programmable rails are being laid. What is not established is the plan behind them — and the difference between a plan and a mechanism is the difference between an argument you can win and a discipline you have to keep.
Sources: this article names its sources inline — the episode under examination, Simon Dixon's appearance on the Peter McCormack Show, with his prior career selling initial public offerings and his present position as a bitcoin advocate stated as incentives; the market figures he cites, including the one and three quarter trillion dollar listing; the industry's stated funding needs and the concentration of the American index; the private credit structure this series has examined, its Cayman-domiciled vehicles and the unfired forced mark; the stablecoin legislation, the age-verification laws and the conditional-payment architecture; and four episodes of this series' own reporting on the hardware wallet attack he describes — the two thousand bitcoin, the roughly five hundred holders, the historical one point three million coins lost, and the correction his account buries, that multi-signature wallets and self-generated passphrases were not compromised. Every source is weighed with the same skepticism regardless of politics or business model. The audio version of this article, read by DeepDives, is available on Wavlake.
Comments
No comments yet. Start the conversation.