Business and strategyEconomics and marketsArtificial intelligenceTechnology
How the AI Pie Will Be Divided
Artificial intelligence could grow the global economy. The uncomfortable question isn’t how much, but how it gets divided.
Translated from the Spanish original. Read the original
In my last article, I wrote about how AI is shifting the value of work from knowledge toward judgment. I was left with a question that kept nagging at me, and it’s bigger than which jobs will be valued.
If AI lets the economy produce more with less, who gets to keep the difference?
Because a technology can grow the pie and, at the same time, divide it less fairly. It’s happened before. And what’s interesting is that even though no one yet knows how much this pie is going to grow, the division has already started to take shape.
How much will the pie grow?
I’ll start with what we know least. Estimates of what AI could contribute to the global economy vary so widely they almost seem to be talking about different technologies:
| Who | Estimate | Time frame |
|---|---|---|
| Goldman Sachs (2023) | +7% of global GDP (about $7 trillion) | 10 years |
| McKinsey (2023) | Between $2.6 trillion and $4.4 trillion in value per year | Annual, at full adoption |
| Daron Acemoglu, MIT (2024) | At most, +0.66% in total factor productivity | 10 years |
The gap between the optimism of banks and consulting firms and the caution of Acemoglu (the 2024 Nobel Prize in Economics laureate) is enormous. The difference comes mainly down to how many tasks can actually be automated profitably and how much can be saved on each one.
My take is that, for today’s question, it doesn’t matter who’s right. Whether AI adds 1% or 7% to global GDP, the question of who gets it remains. And we have a lot more data on that.
How it’s divided among countries
In 2024, the IMF calculated what share of employment in each type of economy is exposed to AI—that is, how many jobs involve tasks AI can perform or transform:
At first glance, this seems like good news for poor countries: less exposure should, in theory, mean less risk. But the IMF turns that idea on its head. Being less exposed also means being less prepared to take advantage of the gains: less digital infrastructure, fewer skilled workers, and fewer companies able to adopt the technology.
And that’s exactly what the usage data show. At the end of 2025, according to Microsoft, one in four working-age people in the so-called Global North used generative AI tools, compared with one in seven in the South:
What’s most striking here isn’t the current picture, but the pattern taking shape over time. In the first half of 2025, the gap between North and South was 9.8 points; in the second half, it was 10.6. The North adopted AI almost twice as fast. So this gap between regions isn’t narrowing; quite the opposite—it’s growing.
Anthropic, the company that develops Claude, found the same pattern in its own data: every 1% increase in GDP per capita is associated with 0.7% more use of its AI, and so far there are no signs that lower-income countries are closing the gap.
Behind this is something even more concentrated: where investment goes.
In 2024, the United States invested almost twelve times more than China and twenty-four times more than the United Kingdom. Whoever builds the models, data centers, and patents also captures a good share of the gains.
How it’s divided between labor and capital
The second way to look at the division is the most traditional one in economics: what share of the wealth produced ends up in wages, and what share in returns to capital.
According to the ILO, the share of global income going to workers has fallen 1.6 points since 2004, to around 52%. That may seem small, but if it had remained at its 2004 level, workers around the world would have earned about $2.4 trillion more in 2024. The ILO points to automation and technological change as one of the causes, and warns that generative AI could maintain or accelerate that trend, although it stresses that this isn’t inevitable.
Behind all this, we could get fatalistic and think it would hurt less affluent social classes even more, but we find something very interesting.
Unlike factory automation, AI affects mainly tasks performed by well-paid workers, so it could reduce wage inequality. But at the same time, it could increase wealth inequality, because the gains are concentrated among those who own the companies and the technology. Less difference between paychecks, more difference between fortunes.
That’s why I think there’s growing fear of what could be called Corporatocracy or Technofeudalism.
Inside the company, a surprise
So far, everything points to more inequality. But there’s one piece of data that points in the opposite direction, and I find it one of the most interesting.
Brynjolfsson, Li, and Raymond studied more than 5,000 customer service agents at a large software company who were given an AI assistant. Productivity increased by around 15% on average, but not equally for everyone: the least experienced workers gained the most (up to 34% in the study’s preliminary version), while the best barely noticed a difference. The authors’ explanation is lovely: the AI had learned from how the best workers did their jobs and passed that knowledge on to newcomers.
This means that, basically, within the same job, AI can narrow the gap between someone just starting out and someone who’s been doing it for years. It’s the other side of what I wrote about judgment in the previous article: AI levels out execution, and what sets people apart becomes something else.
That said, not all groups are equally exposed. In 2025, the ILO updated its index: around 25% of global employment is in occupations exposed to generative AI, rising to 34% in high-income countries. And the highest exposure affects women much more, because they’re more likely to work in administrative tasks:
The ILO itself stresses that these jobs are more likely to be transformed than to disappear. But if that transformation doesn’t come with training, the division also has a gender dimension.
What about Spain?
Interestingly, Spain is well positioned when it comes to usage. It ranks as the sixth country in the world in Microsoft’s ranking, with 41.8% of the working-age population using AI, ahead of Germany and the United States.
Among companies, the jump over the past year has been enormous:
Source: Eurostat (isoc_eb_ai), accessed in October 2026
See the data
| Year | Spain | European Union |
|---|---|---|
| 2021 | 7.67% | 7.65% |
| 2023 | 9.18% | 8.06% |
| 2024 | 11.31% | 13.48% |
| 2025 | 21.52% | 19.95% |
In one year, the share of Spanish companies using AI has nearly doubled, and is already slightly above the European average. But the average hides what matters:
More than half of large companies use AI. Among small companies, fewer than one in five do. In a country of small and medium-sized businesses like Spain, that gap matters: if productivity gains are concentrated in large companies, so will profits, the best salaries, and the ability to keep investing.
So, what can be done?
I don’t think there’s a secret formula, but there are a few levers that come up again and again in everything that’s been discussed:
- Training, especially for people starting out. If AI helps the least experienced workers more, we need to make sure those workers have access to it and know how to use it with good judgment.
- Getting AI to small and medium-sized businesses. The gap between large and small companies is probably the easiest inequality to reduce: tools, support, and training designed for companies of ten people, not ten thousand.
- Competition. If a handful of companies control the models, infrastructure, and data, they’ll also control a large share of the gains.
- Infrastructure in countries that are falling behind. The IMF puts it clearly: without connectivity or human capital, AI widens the gap between countries instead of closing it.
- The tax debate. If gains shift from labor to capital, the question of how to fund governments that rely mainly on taxes on labor will come up sooner or later.
The pie and the knife
AI doesn’t decide on its own how what it produces gets divided. Steam engines didn’t decide either: the laws, education systems, and companies of each era did, which is why the same invention ended up being divided in very different ways in different countries.
The same will happen with AI. Technology puts the pie on the table. Who holds the knife—that’s still being decided.
Sources
- FMI, Kristalina Georgieva, «AI Will Transform the Global Economy. Let's Make Sure It Benefits Humanity» (enero de 2024)
- FMI, Rockall, Tavares y Pizzinelli, «AI Adoption and Inequality», Working Paper 25/68 (abril de 2025)
- OIT y NASK, «Generative AI and Jobs. A Refined Global Index of Occupational Exposure», Working Paper 140 (mayo de 2025)
- OIT, World Employment and Social Outlook, actualización de septiembre de 2024 (resumen de Noticias ONU)
- Microsoft AI Economy Institute, «Global AI Adoption in 2025. A Widening Digital Divide» (enero de 2026)
- Stanford HAI, AI Index Report 2025
- Anthropic, Economic Index, «Uneven geographic and enterprise AI adoption» (septiembre de 2025)
- Eurostat, uso de inteligencia artificial en las empresas (isoc_eb_ai)
- Daron Acemoglu, «The Simple Macroeconomics of AI», NBER Working Paper 32487 (2024)
- Goldman Sachs, «Generative AI could raise global GDP by 7%» (2023)
- McKinsey Global Institute, «The economic potential of generative AI» (2023)
- Brynjolfsson, Li y Raymond, «Generative AI at Work», The Quarterly Journal of Economics (2025)