This week one of us sat in an investor forum on European data centres in Marbella. It ran under the Chatham House Rule, so this post uses what was said and names nobody who said it. The figures come from public sources, linked below.

The room agreed on one thing. Money is not what limits data centres in Europe. The European Data Centre Association forecasts €176 billion of investment between 2026 and 2031, and warns that grid readiness, not access to capital, now decides how much of it gets built (Capacity, September 2026). Its own report puts the growth of European IT power demand at 17 % a year through 2031 (EUDCA).

The grid is the limit. In Europe’s established hubs, connecting a new data centre takes an average of seven to ten years, and up to 13 (Ember, June 2025). Ember argues that flexibility, better siting and smarter connection agreements could cut that to about a year. Today the queue, not the money, sets the pace.


One of us asked a builder what would change if there were no environmental requirements at all. His answer was that he could build in 18 months. So even with every environmental condition waived, the fastest version of the building takes a year and a half, and the grid connection is still years behind it.

One talk split AI demand into two kinds. Training goes where the power is. Inference for companies, and anything that has to stay sovereign, stays close to the client, in or near the big cities, and pays a premium to be there. That second kind is the one that most needs a place in the hubs with the longest queues.


The question that stayed with us came from the panel. A panellist said it was less about infrastructure and more about intelligence. Asked to unpack that, he said it really raises another question: what product? His answer was to stay close to the client and understand what they need. And to ask the client what would happen if they did not use AI at all.

That question comes before any decision about where the computing runs. If the honest answer is “not much”, no data centre helps. If the answer is “we would lose a process that matters”, the next question is how much computing that process needs, and whether it has to wait in a queue at all.

Our own enquiry triage runs on a Mac mini on a desk, in a building that already had power on the day we started. We have not measured how much energy it uses per enquiry, so this post makes no claim about that. That number belongs in an experiment, with the hypothesis written down before the run.

Not every workload fits on a desk. Training a large model does not, and a service for thousands of users needs more than one machine. But much of what businesses want from AI is the close-to-the-client kind: reading their own documents, sorting their own enquiries, answering their own customers. For that kind, the grid queue is not the only route.

Forum attended in Marbella on 30 October 2026, under the Chatham House Rule. No speaker or participant is identified and no slides are reproduced. Figures come from the public sources linked above.