A parable, first
Someone invents a machine that makes anything. Raw material in, finished goods out. No workers, no supply chain, no wages. It belongs to one person.
Everything gets cheap immediately, which sounds wonderful. It also means every factory closes, because nobody can compete with a machine that has no labour cost. Within a few years almost nobody has a job, which means almost nobody has money, which means nobody can buy the cheap goods.
The government does the obvious thing: tax the machine, send everyone a cheque.
Follow the money round one cycle. The state hands out $100. People spend it on goods from the machine. The state taxes the machine's owner at a punishing 90 per cent and gets $90 back. The owner keeps $10.
Next cycle the government has $90 to hand out. Then $81. Then $73.
Nothing was stolen and nobody cheated. The owner is simply entitled to a share, and a share taken every cycle from a closed loop drains it. To keep the payment at $100 the government must find the missing $10 somewhere — borrowing, printing, or taxing something else — every cycle, forever, at a rate that compounds.
Try it at a 50 per cent tax rate and the loop empties in a handful of cycles. The only rate that closes the circle is 100 per cent, at which point the owner is not an owner in any meaningful sense and you have nationalised the machine while maintaining the paperwork of private property.
This is the whole argument. Everything below is the same problem with more institutions attached.
The spiral, in an economy that is not a parable
Money in a real economy moves in a circle. Firms pay wages to households. Households spend those wages at firms. That circulation is what employment is.
Break a piece of it and the break propagates. Workers who lose income stop spending. The businesses they spent at lose revenue and cut staff. Those workers stop spending. Round and round, each turn smaller than the last.
This was the ordinary condition of capitalism before the Second World War — the repeated booms and busts of the 1800s and early 1900s were largely this mechanism, running unchecked. What changed is that governments learned to interrupt it, by replacing lost income until confidence returned. Unemployment insurance, stimulus, and in 2020 direct payments at extraordinary speed. It works well enough that most people have never seen the spiral run.
AI-driven displacement breaks the interruption, for three reasons.
It is not cyclical. Every tool in the kit assumes the jobs come back when confidence does. Here they do not, because the thing that replaced them is cheaper, does not sleep and does not quit. There is no recovery to bridge to.
The scale is different in kind. In 2020 the government replaced the wages of a few per cent of the workforce for a few months. Displacement on the scale we are describing is a quarter of it, permanently.
And you cannot do it by halves. To stop a you have to replace enough income to keep spending roughly where it was. Replacing a fraction of lost wages slows the spiral; it does not stop it. The full bill is the real bill.
The bill, with real numbers
The original version of this argument used figures we have since checked and found wrong, so here it is rebuilt on sourced ones. The conclusion survives. It did not need the exaggeration.
- Total US wages and salaries were $12.98 trillion in 2025 (BEA).
- Total federal receipts were $5.24 trillion in FY2025, of which individual income tax was $2.60 trillion — 49.6 per cent of all federal revenue — and corporate income tax was $497 billion, or 9.5 per cent.
- About 162 million people were employed in July 2026.
Displace a quarter of the workforce and, if the displaced earned the average wage, roughly $3.2 trillion of annual income disappears.
But they will not be average earners. The occupations most exposed to a machine that works through a screen are among the better paid: management work averages $144,860 a year, legal occupations $129,880, computer and mathematical occupations $108,620, business and financial operations $84,510 — against an all-occupation mean of $69,770. Weight the displacement towards those and the lost income is not $3.2 trillion but something closer to $4 trillion.
Set that against total federal revenue of $5.24 trillion. Replacing the lost income alone would cost somewhere between 60 and 80 per cent of everything the federal government currently collects — before a single road, school, aircraft carrier or Medicare payment.
And the obvious place to get it is not there. Corporate income tax raises $497 billion. Confiscate all of it — a 100 per cent rate, no allowances, no avoidance — and you have covered about an eighth of the bill.
Four reasons the money is not there
The arithmetic above assumes the tax base sits still. It will not, and it fails in four independent ways. Any one of them is enough.
1. You are taxing the people you just made unemployed
Half of federal revenue is individual income tax, and it is paid overwhelmingly by the upper half of earners — the accountants, engineers, analysts, marketers and lawyers first in line for displacement. The event that creates the need for the money simultaneously destroys the source of it. Every worker replaced is both a new claimant and a lost taxpayer.
2. The profits you planned to tax get competed away
This is the deepest problem and the least intuitive, because it is caused by AI succeeding.
Large profits require barriers to entry — capital, expertise, scale, distribution. AI dissolves them. When four people can build what took five hundred, industries flood with competitors, prices fall towards the cost of production, and the cost of production falls towards the cost of electricity. Profit is what survives competition, and this technology is a machine for manufacturing competitors.
It is not hypothetical. In early February 2026 software stocks fell hard — roughly $285 to $300 billion of market value in about 48 hours, with the broader early-2026 selloff running to something like a trillion — as investors worked out that AI made the products easier to build and therefore harder to charge for. The market was pricing exactly this: the same technology that promised enormous profits is the thing that competes them to nothing.
You cannot fund a permanent transfer from a profit pool that the technology is actively eliminating.
3. The escape routes were built years ago
Companies whose product is software and whose workforce is a data centre can be domiciled anywhere. The infrastructure for moving profit to the lowest-tax jurisdiction is mature, legal and already in use by every large technology company. A tax steep enough to fund income replacement for a quarter of the workforce is a very large incentive to leave, applied to the most mobile businesses that have ever existed.
4. Some of the value never becomes a taxable transaction at all
A model running on a company's own hardware, or on a laptop, produces real output and generates no revenue anywhere. Nobody invoices anybody. The value is genuine and there is no entity to tax. The more capability moves in-house and onto personal devices, the more of the economy becomes invisible to the tax system — not through evasion, but because no transaction ever occurred.
And even if you win all four, the leak is still there
Suppose you solve every problem above. Assume a co-operative world government, no offshoring, no competition, profits intact and taxable.
still fails, for the reason the replicator showed. Every cycle, the owner keeps a share. The loop shrinks. The state must inject the difference forever, borrowing or printing to do it.
Today's economy has the same leak — capitalists take profits from every cycle — and it survives because of one recycling mechanism: profits get reinvested in ways that create jobs. Build a factory, hire workers, and the wages flow back into the circle as demand. The loop closes, imperfectly, and the gaps are patched with , government deficits and asset bubbles.
AI severs that mechanism. Reinvested profit buys compute, not workers. Investment continues, output grows, and no wages come back. Money flows up and stays there.
So UBI under private ownership has exactly two destinations. Either the state taxes at nearly 100 per cent — in which case it has nationalised the economy and left the ownership paperwork as a courtesy — or it taxes less, the loop drains, and the payment falls year after year: $2,000 a month becomes $1,200, becomes $800, each cut its own small demand crisis, against a background of permanent deficits.
There is also a problem no spreadsheet solves. The displaced accountant receives an income for not working. Her neighbour, a construction worker whose job is not yet automatable, does hard physical work in the heat for less. Ask how long that arrangement survives a political system, and what happens when the construction worker's children ask why they should bother.
What this actually points at
Every failure above has the same shape. The money leaves the circle because someone owns the machines and is entitled to a share of everything they produce. Taxation tries to claw that share back after the fact, every year, forever, from an owner with every incentive and every means to avoid it — and from a base that is shrinking while the bill grows.
Change who owns the machines and the problem does not get solved. It stops existing. There is no leak to patch, because the surplus never leaves the public in the first place. Producer and distributor are the same entity.
That is not a tweak to the tax code. It is a different arrangement of who owns the productive capacity of a society, and there is no polite word for changing that.
The three paths out are laid out on the next page: three futures.