How much energy do data centers and artificial intelligence use?
Data centers in Ireland account for more than 20% of the country’s electricity consumption. So, this is not just a minor technical detail for energy companies (though it is for them), but a significant governance hurdle for anyone planning the next wave of public digital services.
In fact, the International Energy Agency estimates that data centers use about 1.5% of global electricity, which sounds like a manageable number on paper. But that aggregate figure hides a geographic reality that will matter for European procurement and infrastructure planning. If we want to deploy AI in public services, we cannot ignore the fact that demand is concentrated in a few places (like Virginia in the US or specific regions in Europe) where it can actually strain local grids and push up prices for everyone else.
Moreover, if the grid is also a bottleneck, then the speed of AI adoption in public services could be influenced by utility planning rather than software capability. I have seen digital projects fail because the underlying infrastructure was not considered enough in the initial requirements This happens because we the focus is now on the model or data privacy, which are necessary but not sufficient.
Now, people often feel guilty about the energy cost of a single LLM query, which usually consumes around 0.3 Wh, a figure that is negligible compared to average daily household use in the EU. But for a public administration trying to scale a service to millions of citizens, the math changes quickly. The problem is no longer about the individual query (though that matters for transparency), but about the physical capacity of the grid to host these systems without creating local bottlenecks.
Infrastructure matters as much as code. Public procurement should considering energy capacity as a primary technical requirement, not a footnote in a sustainability report.
#AI #PublicAdministration #DigitalTransformation #Energy #Europe