New Study: AI's Real Cost Is Power, Water, and Your Job

A 31-page paper published on 18 September 2026 by researcher Marcin Marciniak makes a simple but uncomfortable point: we keep talking about AI as a technological breakthrough, and that misses most of the story. AI is also moving money, energy, land, labour, and political power around the world — and it's doing so fast, largely without anyone voting on it.

The paper's biggest warning is about concentration. Building and running AI depends on a short list of things: advanced semiconductors (the chips that do the thinking), the machines that make those chips, cloud computing, reliable electricity, and scarce expert workers. Almost all of it sits in a few companies and a few countries. Whoever controls those pinch points gains real power over everyone else — power that doesn't come from armies, but from supply chains. That's why chip export rules have become a foreign policy tool.

The environment cuts both ways. Data centres (warehouse-sized buildings full of computers) consume huge amounts of electricity and water, and can overwhelm local power grids, drive up bills, or spark fights in the towns that host them. But the same AI can improve climate modelling, help integrate renewable energy onto the grid, and spot pollution or deforestation faster than people can. So it's not simply good or bad — it depends on how it's used and where it's built.

Finally, jobs. The paper says automation and smart robots will likely shrink employment in some occupations while boosting workers and creating entirely new tasks in others. Which way it tips isn't automatic. It depends on competition policy, infrastructure planning, environmental regulation, worker protections, collective bargaining, and how the gains from higher productivity get shared. In other words, the future of AI is a policy choice, not a forecast.