← Back to the blog

AI has a cost. Who pays for it?

We ask an AI system for an answer, an image or some code. A few seconds later, we carry on with our day. The experience is so straightforward that it is easy to overlook what was needed to produce that result.

AI has a cost. Who pays for it?

I use AI in my work and see its value. That is precisely why I also care about the infrastructure supporting it. Beyond a user’s subscription, there is a bill for equipment, energy, cooling and land. The question is how that bill is shared and who gets a say in the decisions.

The cloud has an address

A data centre houses servers, storage, networking equipment, power and cooling infrastructure. As the International Energy Agency, IEA, explains, these facilities are where AI models are mainly trained and run.

Working with servers makes the physical requirements clear. Equipment needs electricity, generates heat and requires maintenance. A large facility scales up that basic reality. The cloud occupies a real place, alongside real communities.

How much electricity are we discussing?

In its 2026 analysis, the IEA estimates global data centre electricity consumption at approximately 485 TWh in 2025 and projects approximately 950 TWh in 2030, around 3% of global electricity demand.

These figures cover all data centres, not AI alone. The 2030 figure is a forecast rather than an observed outcome. That distinction matters: a large number can become an impressive headline while losing its original meaning.

A global average also cannot tell us everything about a particular location. For a community, the practical question is how much capacity its own grid has and what must change to accommodate a new facility.

Water needs site-specific evidence

No single litres-per-query figure can accurately describe every request to every AI model. A study by Lawrence Berkeley National Laboratory researchers and collaborators identifies cooling, climate, equipment efficiency and utilisation, and the water associated with electricity generation as important factors.

I would want evidence for the actual facility: where its water comes from, how much it consumes and what happens when local resources are under pressure. An annual total does not necessarily answer what happens on the hottest, most demanding days.

Those answers should be available before construction begins. Broad assurances leave residents unable to assess basic aspects of an investment they are being asked to accept.

Who pays for new infrastructure?

The IEA’s Powering Artificial Intelligence report, published on 14 October 2026, makes an important distinction. New network or generation costs recovered from all users can raise other customers’ bills. A facility that uses existing spare capacity and pays its share can instead help spread fixed costs.

I therefore would not assume that every data centre makes everyone’s electricity more expensive. I would ask who funds the infrastructure, what commitments the operator makes and who remains responsible if the original plans change.

An investment can be useful. Assessing it means understanding its obligations alongside the benefits presented in its announcement.

Why Europe’s protests matter

The European demonstrations on 6–7 November 2026 brought these questions further into public discussion. The issue also concerns people living near proposed facilities.

The Stop Dirty Data Centres campaign calls for a temporary pause on large new developments while their social and environmental implications are examined. That is the organisers’ position; each project still requires assessment on its own evidence.

What I take from this is the importance of hearing residents while decisions can still change. Consultation has value when it can affect the outcome. Presenting a finished decision as consultation is likely to deepen distrust.

What I would ask before such an investment

If a large data centre were proposed in our area, I would begin with straightforward questions:

  • How much electricity and water will it need, and under what design?
  • Who pays for additional infrastructure, and what commitments are in place?
  • How many permanent jobs are expected beyond the construction phase?
  • Who verifies the measurements, and where can residents read them?
  • What happens if the facility scales back or leaves?

I would want the same discussion in Greece. An investment should come with understandable information, allowing a community to judge both its benefits and the responsibilities it creates.

Progress needs an account of its costs

The IEA notes improving efficiency per AI task alongside rising use and more demanding applications. I see room here for better software and more careful use of resources.

I want us to keep using technology for things that help us. I also want its success measured by how it operates within the society hosting it.

AI has a cost. Seeing that cost clearly helps us decide where investment is worthwhile. Progress becomes more convincing when it can also explain its bill.