
IWILLNOTLOOKAWAY.ORG
ANALYSIS
Who owns the machines? The problems: capital, labour, energy, water
17 November 2026 — World
Series “Who owns the machines” · 1 of 2
Transparency note: this piece was prepared with the assistance of Claude, the artificial intelligence model made by Anthropic. Anthropic is one of the companies taking part in the investment race described below, and is therefore one of the interested parties.
In brief
- A handful of large technology groups plan to spend almost 700 billion dollars in 2026 on data centres, chips and energy. Revenues are growing much more slowly than spending.
- Market power is concentrating in a few firms. Hiring is slowing in the most exposed occupations, and young graduates are finding less work that matches their degree.
- The race consumes energy, minerals and water, in a world where one person in four still has no safe drinking water.
- It hits hardest those with the least power: small businesses, poor countries, countries at war. This is the first of two pieces: here the problems, in the second the solutions put to the test.
The question
Free competition works as long as competitors can stay in the market. A multinational spreads its costs over enormous turnover, optimises every step of its supply chain and can sell below cost for years. A small business cannot. Politics welcomes large companies because they bring at once what politics needs: jobs, growth, taxes. The bill comes later, in shops that close and shopping streets that empty.
Artificial intelligence speeds all of this up. Earlier industrial revolutions left whole generations time to adapt, study and change trade. This one is running faster. And whoever owns the technology, and the energy and raw materials that feed it, has an advantage the others struggle to close.
It is a question that concerns billions of people. This first piece looks at the facts: what is happening and who risks being left out. The second piece puts the solutions to the test.
What the facts say
Market power has grown, but in a few firms. The most cited study on the subject measures how much American companies charge above their costs. Between 1980 and 2016 this average margin rose from about 21% to about 61%. The most important point, however, is a different one: the typical firm did not raise its margins. A few large groups did, and they also took market share from the others (De Loecker, Eeckhout and Unger, Quarterly Journal of Economics, 2020). The study covers the United States; the same authors find a similar trend in the rest of the world (De Loecker and Eeckhout, Annual Review of Economics). The measure is disputed: redoing the calculation on all the available data, other economists find a smaller increase, almost entirely concentrated in finance and insurance. The authors have replied (Benkard, Miller and Yurukoglu, NBER).
Investment is running ahead of earnings. For 2026 the large cloud groups have pledged almost 700 billion dollars of investment. In 2025 their revenues grew by 16.5% on average, their capital spending by 60% (Futuriom). Spending on this scale pays for itself in only two ways: with products the market does not yet buy in that quantity, or with a large saving on the cost of labour. That is why the two fears heard most often, "it's a bubble" and "it will leave us all without work", do not fully fit together. If the bubble bursts, the replacement of labour slows down, but a financial crisis arrives. If the replacement succeeds, the bubble does not burst, but labour pays the bill. We described the possible scenarios here.
Work: fewer hires, no mass layoffs yet. The broadest research so far is a study by the International Monetary Fund and the University of Chicago of about 700 million job adverts in 110 countries, from November 2021 to April 2026. In the occupations most exposed to artificial intelligence, job vacancies fell by almost 10%. Then the decline stopped, even as the models became more capable. The blow is harder where the machine can replace the person, softer where judgement, responsibility and communication are needed. On wages, so far, no clear effect (Ando, Kirti, Lin and Ilman, IMF, Working Paper 2026/220; Policy Edge summary). The most worrying signal concerns young people: in the United States 42% of graduates aged 22 to 27 work in a job that does not require a degree (Federal Reserve Bank of New York).
History says one thing, the future may say another. A review of more than a hundred studies cited by the Monetary Fund concludes that, so far, technology has created more jobs than it has destroyed. But the same Fund writes that the future path of labour demand is "exceptionally uncertain" (IMF, Dan Katz, June 2026).
Energy and raw materials decide who wins. According to the International Energy Agency, data centres will consume more than twice as much electricity in 2030 as in 2024: more than the whole of Japan uses today (IEA, Energy and AI). In Europe industry pays two to three times more for electricity than in the United States (Mario Draghi at the European Parliament, 17/9/2024); a year later the gap was still about twofold (HCSS).
Water: the resource consumed twice. Data centres use water to cool their servers and, indirectly, to produce the electricity they consume. According to a report by the United Nations University, by 2030 the water footprint of the electricity consumed by data centres could reach 9,300 billion litres a year: as much as the minimum household use of all 1.3 billion inhabitants of sub-Saharan Africa (UNU-INWEH, June 2026). The figure includes the water used to produce the electricity, not only the water for cooling the servers, and estimates vary a great deal depending on what is counted. The direction, though, is the same. And it falls on a world in which one person in four, about 2.1 billion, still has no safe drinking water (WHO and UNICEF, August 2025). In January 2026 the same UN institute spoke of an "era of water bankruptcy": aquifers, rivers and glaciers exploited beyond their capacity to recover, with global warming making everything worse (UNU-INWEH, January 2026).
Who risks being left out
The same forces do not hit everyone in the same way. The Monetary Fund asks whether artificial intelligence will help the poorest countries catch up. The answer is conditional: it can happen if it lowers the cost of learning and adapting in the weakest economies, and if those economies can turn it into more productive work (Imam and Temple, IMF, September 2026). For now it is a possibility, not a trend.
The gap is visible inside rich countries too. In 2025, 55% of large European companies used artificial intelligence, and only 17% of small ones (Eurostat). Those who start ahead accelerate.
And it is visible above all in fragile countries. There, coverage of safe drinking water is 38 points lower than in the rest of the world (WHO and UNICEF). Among the contexts of extreme fragility the OECD lists Sudan and Yemen, and it names the war in Ethiopia among those that worsen water scarcity (OECD, States of Fragility 2025). These are theatres this site covers. For them the artificial intelligence race and water scarcity risk adding to the war. The second piece puts the solutions to the test precisely there.
What we draw from it
Put in a row, the facts say three things. Economic power is concentrating in a few firms, and artificial intelligence is accelerating that concentration. Work, for now, is changing more than it is disappearing, but the first signals hit young people and the jobs that are easiest to replace. And the race consumes resources that belong to no one in particular and to everyone: energy, minerals, water. Those with the least power risk paying twice, losing work and resources together.
There is also a limit that works in favour. An economy of unemployed people does not buy. Large companies need customers with an income. This is not enough to stop the weakest from paying the bill first, but it makes the system unstable, and instability, in history, is the moment when the rules change.
What to watch
- UN Water Conference, 2-4 December 2026, in the United Arab Emirates (UNU-INWEH).
- UN negotiation on tax cooperation, Nairobi, 30 November – 10 December 2026 (Eurodad).
- The accounts of the large cloud groups: if artificial intelligence revenues do not catch up with spending, the correction will come before the replacement of labour.
Open questions
- Who pays for the water and energy consumed by data centres, and who decides where to build them?
- If technology widens the gap between those who start ahead and those who start behind, who narrows it?
- What happens to a country at war when the resources it lacks are the same ones the machines need?
Sources: De Loecker, Eeckhout and Unger, Quarterly Journal of Economics, 2020 · De Loecker and Eeckhout, Annual Review of Economics · Benkard, Miller and Yurukoglu, NBER · Futuriom · Ando, Kirti, Lin and Ilman, IMF, Working Paper 2026/220 · Policy Edge summary · Federal Reserve Bank of New York · IMF, Dan Katz, June 2026 · IEA, Energy and AI · Mario Draghi at the European Parliament, 17/9/2024 · HCSS · UNU-INWEH, June 2026 · WHO and UNICEF, August 2025 · UNU-INWEH, January 2026 · Imam and Temple, IMF, September 2026 · Eurostat · WHO and UNICEF · OECD, States of Fragility 2025 · UNU-INWEH · Eurodad
Related pieces: Who owns the machines? The solutions, and who is left out · Six roads for artificial intelligence: the possible scenarios and the signals to watch · How to govern a machine: the tools on the table for regulating artificial intelligence · Minab, the school nobody checked