Who gets the extra €100?

The premise

If we can track where extra income goes, we can see which channels are widening and which are stuck — even without knowing exactly what caused the gain.

That is an assumption, not a conclusion.

Two years on, the payslips hadn’t moved

Two economists, Anders Humlum and Emilie Vestergaard, worked with Statistics Denmark to follow 25,000 workers at 7,000 workplaces. All of them were in jobs heavily exposed to AI chatbots — 11 in total, including accountants, customer-support staff, journalists, lawyers, software developers and teachers. The researchers matched what people told them in surveys with official records of what they earned.

Two years after ChatGPT arrived, the chatbots had made no measurable difference to earnings or to recorded working hours. Any effect larger than 2% can be ruled out.

That does not mean nothing changed. The work itself did. Of the people using chatbots, 85% said they had spent the time they saved on other tasks. The time went back into the job rather than into pay.

What the study does not show is where that extra output ended up instead.

Say you produce €100 more

Put it simply. Suppose the same people, working the same hours, now produce €100 more than before — thanks to a new machine, or to AI.

That €100 can end up in one of five places. It can go to workers, as higher pay. It can go to customers, as lower prices. It can go to shareholders, as dividends or share buybacks. It can go to the state, as tax. Or it can stay inside the company, as investment.

Three of my earlier pieces stopped at exactly this question. I said it couldn’t be measured. This time I followed it through.

Productivity is going up. Pay isn’t following.

The productivity gains from AI are real. In a study of 5,179 customer-support agents in the United States, an AI assistant raised the number of problems solved per hour by 14% on average, and by 34% for new and less-skilled staff. That study didn’t look at pay. The Danish one did, and found no effect on earnings.

Workers are getting a smaller slice

The national figures show the same pattern over a longer period. Worldwide, the share of income that goes to workers fell from 53.0% in 2014 to 52.4% in 2024, according to the International Labour Organization. Had it stayed at its 2014 level, workers would have earned $1 trillion more in 2024 — about $290 each.

Belgium shows it more sharply. The National Bank of Belgium describes value added as “an economic pie shared between labour and capital”. Since 2013, the slice going to wages has been shrinking, while across the euro area it has stayed broadly stable since 2010.

The bank is careful to add that this does not mean wages have fallen. It means they have grown more slowly than the value workers produce. Belgian wages are automatically indexed to prices, so they keep up with inflation. But since 2013, the bank says, hourly productivity has risen without a matching rise in real hourly wages.

About two-thirds of the decline comes from the top 1% of firms by value added. These tend to be capital-intensive businesses, where wages make up a smaller share of what the company produces.

In Belgium, the law caps pay rises

How much Belgian wages can rise above inflation is set for two years at a time. Under a 1996 law, the Central Economic Council compares Belgian labour costs with those in Germany, France and the Netherlands, and unions and employers negotiate within the resulting margin. If they can’t agree, the government imposes one by royal decree.

The margin has been shrinking: 1.1% for 2019–20, 0.4% for 2021–22, zero for 2023–24, and zero again for 2025–26. The last one was imposed by royal decree on 12 September 2025, after the two sides failed to agree.

Indexation and seniority increases are guaranteed regardless, so a zero margin is not a pay freeze. It means there is no room for real pay rises on top of inflation. According to the National Bank, negotiated real wage growth has “effectively flatlined” since the zero margin was introduced in 2023.

The benchmark, note, is not Belgian productivity. It is labour costs in the three neighbouring countries.

Payouts to shareholders are at a record

Companies around the world paid a record $1.75 trillion in dividends in 2024. Between January and March 2026, they paid out $424.5 billion in dividends — and spent almost exactly the same amount, $425.7 billion, buying back their own shares.

How much goes this way is up to each company’s board.

Prices rose faster than costs

When prices fall, customers get part of the gain. When prices rise faster than costs, the gain stays with the seller.

In a 2023 analysis, the European Central Bank found that over the previous three quarters, rising profits per unit of output accounted for roughly 60% of the increase in euro-area prices, as measured by the GDP deflator. Profits and the income of the self-employed make up only about 42% of GDP. The ECB concluded that companies had, on aggregate, not absorbed their higher costs: their price rises “have exceeded the increases seen in costs”.

That is a finding about prices and costs, not about anyone’s motives.

Prices are also where the other channels meet. The price of a product is made up of the wages, profits and taxes that go into it, so looking at what is pushing prices up shows how the gain is being split.

The state has set a minimum tax

Since 2024, groups operating in the EU with annual revenue above €750 million have had to pay a top-up tax if their effective tax rate falls below 15%. That puts a floor under this channel. How much of the extra income it actually moves to governments is something I couldn’t measure.

Companies are investing less

When a company puts its profits back into equipment, research or AI systems, the gain stays inside the business. In the last three months of 2025, EU companies invested about 22 cents of every euro of value they created, according to Eurostat — the lowest rate since 2015.

The five, side by side

Where the €100 can go	Who gets it	What’s happening	Who decides
Pay (Belgium)	Workers	Keeps up with inflation, no room above it	The law, royal decree, unions and employers
Dividends and buybacks	Shareholders	At a record	Each company’s board
Prices	Customers	Recently driven up by profits	The market
Tax	The state	A 15% floor since 2024	EU directive and national law
Investment	The company	Lowest since 2015	Each company

What has grown is the money paid out to shareholders. Pay has not kept up with productivity, and investment has fallen.

But this is not AI’s doing

Belgium’s wage share started falling in 2013, well before generative AI. Looking at 24 OECD countries between 1995 and 2014, researchers found that in most of them productivity and typical wages had already drifted apart.

So the way gains are split was settled long before AI arrived — by wage law, by cuts to employers’ social contributions, and by company boards.

That gives some reason to think gains from AI will flow the same way: the Danish study and the national figures point in the same direction. But that has not been measured directly. I have set the two pieces of evidence side by side rather than joining them into a single claim.

Where I stand

In three earlier pieces, I stopped at the same sentence: this couldn’t be measured. Following it through this time, I found that wasn’t true. The Danish study was already out when I wrote the sixth piece. I simply hadn’t looked for it.

What I found instead is that where the €100 goes depends on arrangements that were in place long before AI. This piece doesn’t say what should be done about that.

What I got wrong

• I started out assuming the gain from AI couldn’t be separated out and measured. A study had done just that. My three earlier pieces may have stopped not because it couldn’t be measured, but because I didn’t look
• I first counted four places the €100 could go. I had left out investment
• One summary attributed the ILO figures to the wrong report. They weren’t in it; I found them in the May 2025 update
• One summary said the ECB report compared profits’ recent share with a historical “one-third”. The report says no such thing. Its comparison is with the 42% of GDP that profits and self-employed income make up
• One summary quoted a “3% time saving” and a pay pass-through rate from the Danish study. I couldn’t find either in the revised paper, so I left them out

What I checked, and what I couldn’t

Checked

• The Danish study (May 2025, revised March 2026): 11 highly exposed occupations; a late-2024 survey of 25,000 workers at 7,000 workplaces. No effect on earnings or recorded hours two years after ChatGPT’s launch; effects larger than 2% ruled out. 85% of users reallocated saved time to other tasks
• The US customer-support study (2025): 5,179 agents; problems solved per hour up 14% on average, 34% for new and less-skilled staff. Pay was not measured
• ILO (May 2025): labour income share 53.0% in 2014 and 52.4% in 2024. At the 2014 level, labour income in 2024 would have been $1 trillion higher (constant PPP), about $290 per worker
• National Bank of Belgium (2025 No 7): the wage share (compensation of employees as a share of gross value added) has fallen since 2013; broadly stable in the euro area since 2010. About two-thirds of the 2013–2023 decline is due to the top 1% of firms by value added. Negotiated real wage growth has effectively flatlined since the zero margin of 2023. The bank says the decline “could be due to” wage moderation and cuts in employers’ social contributions
• Belgian wage norm: 1.1% (2019–20), 0.4% (2021–22), 0% (2023–24), 0% (2025–26, royal decree of 12 September 2025). Indexation and seniority increases guaranteed
• Global dividends: a record $1.75 trillion in 2024. January–March 2026: $424.5 billion in dividends, $425.7 billion in buybacks
• ECB (Economic Bulletin 4/2023): unit profits accounted for roughly 60% of GDP deflator growth over the previous three quarters, above the 42% share of gross operating surplus and mixed income in nominal GDP
• EU minimum tax: groups with revenue above €750 million; 15% effective rate; fiscal years starting January 2024
• EU business investment rate: 21.8% in Q4 2025, the lowest since Q3 2015. “About 22 cents of every euro” in the text is rounded
• OECD study (2017): 1995–2014, 24 countries; productivity decoupled from real median compensation in most of them

Couldn’t check

• Where the gain went, if not into pay, when narrowed down to AI
• How much of the extra income the minimum tax actually moves to governments
• European companies’ total share buybacks in 2025 — I couldn’t open the original page
• The long-run trend in statutory corporate tax rates — I couldn’t open a primary source
• Exactly which three quarters the ECB’s analysis covers

What this piece does not claim

• That companies took the gain
• That AI has held down pay
• That wages have fallen
• What should be done