The guide
What AI does to wages, who ends up with the income, and what could turn the next downturn into a depression. Every number here comes from the published research and the public model behind this site.
1. A wage is tied to what could replace it
When a firm can do a task with a person or with a machine, the wage for that task cannot stay above what the machine would cost. That link is standard economics. The research behind this site asks what sets the machine's cost.
A machine is built from other machines, labour, energy, land, minerals and permits. Follow those costs back far enough and they end in things no one can produce more of: land, locations, grid connections, mineral deposits. While building machines still takes a great deal of labour, part of every machine's cost is wages, and machines stay relatively expensive. As machines take over building machines, that part shrinks. More of a machine's cost is then rent on the scarce inputs, and the wage it competes with falls.
The same scarce inputs price the alternative to working. Life without a wage still needs somewhere to live, food, energy and warmth. When manufactured goods get cheaper and housing does not, what people can afford without work falls, and so does the point at which they leave work altogether.
Deflate US wages by the price of durable goods and by the price of shelter, and the two measures of the same wage diverged by a factor of 4.8 between 1964 and 2024. An hour's pay buys ever more manufactured goods relative to housing.
Båge and Wilson, Pinning the Wage to Scarcity and Technology (2026).
2. Output is growing without work
The change has started to show in the US national accounts. Since Q2 2023, output in the business sector has grown 2.75% a year while hours worked grew 0.23%. Labour's share of business output is the lowest since records began in 1947, down 3.3 points in a year. Since 1990 only recessions and the recoveries after them have seen a fall that fast.
Nonfarm business sector. Shaded: US recessions.
Nonfarm business sector.
The lost work is not showing up as unemployment. Since Sep 2024, the share of Americans aged 25–54 in work has fallen by 0.6 points, and 94% of that fall is people who have left the workforce rather than people looking for work. In 2007–10 it was 12%. People who stop looking are not counted as unemployed, so the unemployment rate misses most of the change.
The investment boom is narrower than it looks. Since 2019, US output of semiconductors has risen 69%, while manufacturing as a whole has grown 0.1% and employs 1% fewer people. Much of the equipment is imported: imports of computers and parts have risen from 0.7% of GDP in 2022 to 1.8%. Net of them, investment in computing, software and power is 0.3 points of GDP above its 2015–22 trend. A collapse in AI investment would land largely on the countries that make the equipment.
3. Where the income goes
The model follows one economy as machines take over tasks, at three speeds. Choose a speed and a year.
At the medium speed, machines go from doing 80% of tasks to 94% over fifteen years. Labour's share of income falls from 47% to 6%. The owners of the machines do not take its place: interest on machines falls from 17% to 8% of income, because machines get cheaper as the technology improves. The income goes to the land under homes, whose share rises from 13% to 73%.
The real wage first rises, by about 3% by year 6, as goods get cheaper. Then it falls, to 51% of where it started, as housing takes over household budgets. At the height of the change, goods prices fall 17% a year while housing costs rise 29% a year, with the overall price index on its 2% target. The share of people in work falls from 60% to 39%.
4. Public finances lose their base
Most tax revenue is raised on wages. In the model, taxes on wages fall from 78% of revenue to 22% as labour's share shrinks, while more people need support. Public debt rises from 40% of income to 77% after fifteen years at the medium speed, and to 131% at the fast speed.
Interest rates rise, then fall. Building the machines raises the demand for savings; later, income moves to owners who save more of it. At the fast speed the neutral real interest rate goes from 0.8% to 1.8% and then to −0.2%.
A tax on the rent of land and other scarce inputs can fund a payment made equally to everyone, in work or not. Taxing land does not reduce the amount of land, and a payment that does not stop when someone takes a job does not discourage work. Fully captured, measured US land rents would pay about a third of a basic living income per person (2025).
5. Markets will see it late, then all at once
Markets learn about this from data on labour's share of income, which is noisy and revised. In the model a market that starts with 5% odds on the new economy moves from 10% to 90% between years 2.3 and 3.6, when about 3% of the eventual fall in labour's share has happened.
Government bond yields give little warning. Across that window the ten-year yield moves by 11 basis points. Whether long-term rates end up higher or lower depends on two things: how much of the new income its owners save, which pushes rates down, and how much governments borrow, which pushes them up. Across 4,000 variations of the model, the ten-year yield ten years on ranges from −322 to +147 basis points around its start (5th to 95th percentile).
The warning is in the real economy: wages, labour's share of income, tax receipts and the jobs data on the indicators page. Headlines move markets without data; when the data do not follow, the move fades.
6. From a bust to a depression
An end to the AI investment boom would start like the dot-com bust: share prices fall, investment stops and lenders take losses. What follows depends mostly on how governments respond.
The model runs four economies together, the United States, the euro area, Sweden and China, quarter by quarter. It carries the channels that turned 1929 into the Great Depression: lost wages cutting spending, bank losses raising the cost of credit, bank runs, collapsing trade, and falling prices making debts heavier. One set of parameters was tested on two episodes. For 1933 the model gives US output at 72% of its 1929 level against 74% in the data, and unemployment at 30% against 25%. For 2008–10 it gives peak unemployment of 8.8% against 9.9%.
The same bust, different policies
A dot-com-sized bust, AI taking jobs at the central pace, robots arriving in two years. Six years. Out of work counts the unemployed and the people who have left the workforce after losing their job. Output is the four economies together against their capacity.
| Out of work, US peak | Of which unemployed | Output, worst point | Outcome | |
|---|---|---|---|---|
| Today's policies | 24% | 15% | −8.5% | A labour depression, output held |
| An income guarantee | 17% | 10% | −5.7% | A labour depression, output held |
| More care jobs, paid by borrowing | 18% | 14% | −6.7% | A labour depression, output held |
| Policies erode | 50% (limit) | 34% | −34.3% | A depression |
| The rules of the 1930s | 50% (limit) | 39% | −48.2% | A depression |
Today's policies stop a bust from becoming a depression in output. They do not stop a depression in jobs: within six years 24% of the US workforce is out of work, while output across the four economies falls 8.5% below capacity at its worst. The unemployment rate shows only part of it. It peaks at 15%, because 9% of the workforce has left it altogether. An income guarantee holds up spending, not employment, at a cost of up to 5.1% of US GDP a year. Spending cuts, failing banks and tariffs turn the same bust into a depression on the scale of 1929–33.
Today's policies, different shocks
| Out of work, US peak | Of which unemployed | Output, worst point | Outcome | |
|---|---|---|---|---|
| No bust, no AI job losses | 5% | 5% | −0.6% | A slowdown |
| AI job losses, no bust | 16% | 10% | −4.7% | A labour depression, output held |
| A dot-com-sized bust, no steady AI job losses | 15% | 12% | −6.6% | A 2008-scale recession |
| The bust and AI job losses | 24% | 15% | −8.5% | A labour depression, output held |
| ... and firms automate three times as many of the jobs they cut | 42% | 20% | −11.8% | A labour depression |
The more of the jobs they cut that firms automate instead of refilling, the more of the loss goes uncounted. When firms automate three times as many, 42% of the workforce is out of work, but the unemployment rate peaks at 20%: 22% have left the workforce.
Support that waits for unemployment comes late
Unemployment benefits, stimulus and spending cuts are set off by the unemployment rate. When people who lose their job leave the workforce instead, they start late. With no bust, AI's steady job losses take US output 2.7% below capacity within four years when support waits for unemployment, against 1.0% when it watches everyone out of work. A two-year stimulus then runs out either way, and output is about 7% below capacity by year six; an income guarantee holds the loss to 3.6%.
The United States is hit hardest of the four: firing is quickest there, the safety net is thinnest, and its exposure to AI and to share prices is the largest.
7. Office work first, physical work later
AI can already do much office, professional and technical work. Those jobs are 61% of employment in the United States, 56% in the euro area, 65% in Sweden and 20% in China. Physical work waits for robots.
Recessions hit physical work hardest. From 2007 to 2010, US physical jobs fell 11% while office jobs fell 3%. Until robots arrive, most jobs lost in a recession come back: only 37% of a typical US recession's job losses are in work AI can do. Once robots arrive, those jobs stop coming back.
A later arrival of robots delays the crisis in jobs. It does not prevent it.
8. Care is carrying job growth
From Aug 2024 to Aug 2026, health care and social assistance added 1,342 thousand of the 1,318 thousand net new US jobs: 102%. The rest of the economy shrank. Women took 76% of the new jobs.
Care absorbs people, not pay. It pays about the average in the United States (0.98 times) and less in most of Europe (Sweden 0.84). It is mostly paid for by the state: Medicare and Medicaid cover 64% of US spending on health care. And the people AI displaces are mostly not the people care hires: men work in care at about a quarter of women's rate.
Care cannot grow without limit. No rich country employs more than 20.1% of its workers in care (Norway); the United States and Sweden are at 15%. Across 34 OECD countries, care's share of jobs follows public spending on long-term care (R² 0.66), not the share of people over 65 (R² 0.04). The limit is financing, not need. Norway's oil fund paid the equivalent of 7–13% of mainland GDP into its budget each year from 2019 to 2025.
In the model, funding new care jobs up to Norway's share lowers the share of the US workforce out of work at its peak from 24% to 18%. How the jobs are paid for matters more as the crisis deepens: taxes on wages lose room as the wage base shrinks, while a levy on the income AI moves to owners gains room. Borrowing costs do not limit an expansion of this size.
9. What to watch
The indicators follow the data that would show this early, each against a fixed trigger:
- Labour's share of income and people leaving the workforce: output growing without work, outside a recession.
- Jobs outside care: whether growth outside health care and social assistance has stopped.
- Office jobs and factory and warehouse jobs: AI and robots taking work outside a recession.
- Permanent job losses and hiring: whether jobs lost in a downturn come back.
- Technology stocks and corporate credit: a bust in AI investment.
- France's borrowing costs: a government's finances under strain.
- A market bust and job losses together: the combination behind the worst outcomes.
Policy decides the outcome, so budgets, benefits, bank rescues and tariffs matter as much as the data.
Limits and sources
- The model is a simplified description of four large economies. Many of its settings are approximations; each is stated in its code.
- It has no exchange rates, so it overstates Sweden's exposure. A falling krona cushioned Sweden in 1931 and in 2008–09.
- It recovers from 2009 faster than the United States did, so it is optimistic about how long a slump lasts.
- How owners spend their income and how governments borrow are set by stated assumptions, not estimated.
- It shows what can happen and why. It is not a forecast.
Båge, J. and Wilson, S. (2026). Pinning the Wage to Scarcity and Technology. SSRN 7226858. The model, its data and its checks are public at github.com/wilsoniumite/labor. Data: FRED (Federal Reserve Bank of St Louis), US Bureau of Labor Statistics, US Bureau of Economic Analysis, Sveriges Riksbank, US Treasury, Statistics Sweden, ILOSTAT, OECD, BIS, Statistics Norway and the World Bank.