A RAAK.work project
Capex WatchCould growth pay for the AI bet, so that wages do not have to? The front page measures the bet against the “advanced economies”: 43 countries and territories that the International Monetary Fund groups together. This page shows how much they produce and who makes up the total, then how fast they would have to grow for the AI spending to pay for itself out of new output instead of out of wages.
Every figure on the front page that describes the economy, from the wage pool in §3 to the growth dial in §6, is a figure for this group as a whole. The chart below takes it apart. The coloured layers are the countries you pick, up to four at a time; the grey layer is everyone else; the line on top is the whole group.
The years after 2026 are not a forecast. They show two paths. On the dotted line, and in the layers under it, the 43 carry on at today’s pace of 2.1% a year. The dashed line is the growth that would pay for the AI spending with nothing taken from wages: an average of 3.4% a year from 2025 to the end of 2030, and today’s pace after that. The band between the lines is the extra output. By 2030 it is $5.2T a year, which is what AI must be worth to its customers for the builders to earn back their spending. The chart below works that figure out.
The band looks thin because the economy is large: it is about 7% of what the 43 would produce in 2030. As growth it is not small. The 43 grew by an average of 2.0% a year from 2016 to 2025, so the dashed line asks for 1.7 times their recent pace, kept up for five years. The first of those years is nearly over, and the IMF expects it to end at today’s pace. Reaching the same level by 2030 would then take about 3.8% a year from 2027.
After 2030 the two lines run side by side. The faster growth does not have to last. What pays for the spending is the higher level of output, $82.4T a year by 2030 against $77.2T at today’s pace, and the economy then has to stay at that higher level. Every country is drawn growing at today’s pace, so the shares stay as they are in 2026.
How this was built: each country’s output in 2025, converted to dollars at 2025’s exchange rates, then carried back to 2016 and forward to 2026 with the IMF’s figures for its growth after inflation. So every year is in 2025 dollars: a layer that rises means the country produced more, not that prices rose or its currency strengthened. The 2025 column is also the plain dollar figure for that year. 2026 is the IMF’s forecast, made in April. The 43 are the IMF’s own list; they add up exactly to its total for the group in every year. Each country’s figure is checked against the World Bank’s. Sources below.
| Economy | Group | 2025, $B | Share | Growth since 2016 |
|---|
Growth since 2016 is the total rise in output after inflation, 2016 to 2025.
The AI spending can be paid for out of wages or out of growth. Paying for it out of growth alone would take growth averaging about 3.4% a year, against 2.1% today, to reach output of about $82.4T a year by 2030. After that, growth could return to its old pace.
The front page works out what the spending must earn: about $1.3T a year by 2030 for the five builders (the floor), or about $1.82T paid by their final customers (the bill). No customer pays the full value of what it gets, or there would be no reason to buy. If AI sellers keep a quarter of that value, the front page’s capture rate, customers must gain four times the revenue: $5.2T to $7.3T a year.
That gain can be one of two things. It can be wages no longer paid, because AI does the work: the front page’s §4, which counts 65 to 91 million jobs’ worth of pay. Or it can be output that would not otherwise exist, because AI helps people produce more. In that case nobody’s wage has to go, and workers’ share of income need not fall. The BIS’s report has this case too, and calls it a productivity boost. The more of the gain growth supplies, the less has to come out of wages. The chart shows the trade.
What counts is the level the economy reaches, not the path it takes. Growth could start slowly and build, so long as the 43 economies produce about $82.4T a year by the end of 2030, where today’s pace would bring them to $77.2T. The 3.4% is the average rate that gets there. Once the economy is at the higher level, growth can fall back to today’s pace and the extra output stays. The date is less forgiving than the path, because that is when the spending’s bill falls due: the next section works out what a late arrival means.
At 2.1% a year, growth supplies none of it. All $5.2T a year would have to come out of wages: 65M jobs’ worth of pay, with workers’ share of income falling from 60% to 53% by 2030. Growth of 3.4% a year would supply it all.
| If the spending must earn | Customers must gain, a year | All from wages: jobs’ worth, workers’ share | All from growth: a year to 2030 | US must grow, if the rest grows 2.0% |
|---|---|---|---|---|
| The floor, $1.30T | $5.2T | 65M, | 3.4% (1.7× the 2016–25 average) | 5.9% |
| The bill, $1.83T | $7.3T | 91M, | 4.0% (2.0× the 2016–25 average) | 7.3% |
Each extra tenth of a point of growth a year supplies about $0.4T more by 2030, which spares about 4.7M jobs’ worth of wages.
How likely is the growth? It would be a clear break with the record. The 43 grew 2.0% a year on average from 2016 to 2025. The studies the BIS surveys put AI’s lift to economy-wide productivity growth at less than 1%, where the floor needs 1.3 points and the bill 1.9. The BIS also notes that a lasting rise in trend growth has not happened since the Industrial Revolution. A one-off rise in output by 2030 would be enough to pay for the spending, though, so the growth is possible. On the evidence so far it is not the likely case, and the front page’s jobs row watches for the other one.
How this was built: what customers must gain is the required revenue divided by the capture rate, the front page’s §4 arithmetic. “Today’s pace” is the IMF’s forecast for the group’s growth in 2026, 2.1%, carried to 2030. The extra output is whatever the chosen growth adds above that path by 2030, and all of it is counted towards the bet, which is generous, because growth has other causes too. What growth does not supply is taken to come from wages. Jobs’ worth divides those wages by $80,000 a job, as the front page does. Workers’ share starts at 60% and falls by the wages displaced, as a share of 2030’s output. The floor is due by the end of 2030, the end of the BIS’s spending window, so the paths run five years from 2025. All figures are in today’s prices. The US column uses the US weight of 37% and everyone else growing 2.0%, as the front page’s §6 does. The capture rate is the front page’s; the FAQ sets out the evidence for it. Arithmetic in the derivations ledger.
Growth that comes late can spare workers and still leave the investors unpaid.
The economy can take its time. The spending cannot. The floor falls due by the end of 2030 whatever growth does, because the chips bought in 2025 and 2026 last about five years and are worn out by then. So the question is not only whether the growth comes, but whether it comes on the spending’s clock. The table takes the same $5.2T a year of gain and moves the growth later, or spreads it over ten years.
| When the growth arrives | Growth needed, a year | Supplied by the end of 2030 | Still missing in 2030, a year |
|---|---|---|---|
| On time, 2026–2030 | 3.4% | all $5.2T | nothing |
| One year late, 2027–2031 | 3.4% | $4.0T, 78% | $1.2T: 14M jobs’ worth of wages, or $0.3T of the floor unpaid |
| Two years late, 2028–2032 | 3.4% | $3.0T, 57% | $2.2T: 28M jobs’ worth of wages, or $0.6T of the floor unpaid |
| Spread over ten years, 2026–2035 | 2.7% | $2.3T, 44% | $2.9T: 36M jobs’ worth of wages, or $0.7T of the floor unpaid |
Whatever growth has not supplied by 2030 has to be found some other way until it catches up. It can come out of wages, with AI replacing work in the meantime. It can come out of the investors, who then earn less than they needed: in the ten-year case about $0.7T a year of the floor goes unpaid in 2030, twice the $0.35T a year of return investors required, and about $2.2T in all by 2035. Or it can be borrowed, which only postpones the question.
The ten-year row is the one to look at. It needs growth only 0.6 points above today’s pace, which is inside the range of the studies the BIS surveys (less than 1 point), where the on-time row needs 1.3. So the growth that looks likely is too slow for the spending, and the growth that is fast enough looks unlikely. A boom that arrives late is good for workers’ share of income and bad for the first investors: the chips bought now would be replaced before the revenue came in. The BIS’s own comparisons point the same way. It sets this boom beside the canals, the railways and the dot-com years, and describes boom and bust as a regular feature of past major innovations.
How this was built: every row keeps the front page’s floor ($1.30T a year), its capture rate (25%) and today’s pace (2.1%). Output grows at today’s pace until the faster growth starts, at a steady faster rate until it has added $5.2T a year, and at today’s pace again afterwards. A later start needs a slightly lower rate, because it starts from a larger economy. “Supplied by the end of 2030” is the output above today’s path at that date. What is still missing is shown two ways: as wages, at $80,000 a job, and as the part of the floor it leaves unpaid, a quarter of it at the 25% capture rate. The total unpaid in the ten-year case adds the shortfall at the end of each year from 2030 to 2034. Arithmetic in the derivations ledger.
The front page starts from the 2025 total, $69.6T. Its wage pool in §3 and §4 is a share of that figure, and the growth dial in §6 starts from it. Until 3 October 2026 it used a round $60T, which is close to the IMF’s figure for 2022, $59.4T. Only part of the rise since then is extra output: after inflation, the group produced about 6% more in 2025 than in 2022. The rest is higher prices and changes in exchange rates.
The US share depends on how currencies are compared. At market exchange rates the US produced 44% of the group’s output in 2025, a little under half. Adjust each currency for what it buys at home, which is how the IMF and the BIS weight countries when they average growth rates, and the US share is 37%. Prices are higher in the US than in most of the group, so a dollar of American output buys less there than a dollar of output elsewhere.