PwC put a number on the AI build-out this week that most people would find pretty hard to wrap their heads around. Global spending on data centres, the firm reckons, will reach $31.6 trillion through 2050. It calls the boom one with no precedent, larger in real terms than the railways, electrification and the internet. If AI adoption runs hotter than the central case, the total could approach $50 trillion.
The entire United States produces around $30 trillion of output in a year. Europe’s share inside PwC’s figure is enormous and, by the standards of this race, modest: $5.6 trillion over twenty-five years, against $15.1 trillion for the United States and $8.2 trillion for Asia-Pacific.
If PwC is correct, $5.6 trillion will be spent, but the unresolved question is where the buildings will go.
Brussels has spent the past eighteen months describing data centres as the sovereign infrastructure under any plan to train models on European soil. What it has not done is decide where that infrastructure will be placed. The grid, the water and the local planning office now decide where a campus can go. All three are pushing the build away from the hubs you would expect.
The Plan Is to Triple Everything
The headline ambition is unambiguous. The Commission’s proposed Cloud and AI Development Act, tabled on 3 June 2026 and still going through Parliament and Council, sets out to at least triple the bloc’s data centre capacity within five to seven years.
Member states would designate acceleration zones with bundled permits and a twelve-month ceiling on the process. Favoured projects could get strategic-project status and a quicker route to funding. The diagnosis is that Europe’s constraint has been bureaucratic drag. Make building faster, and the buildings will follow.
Sitting next to that is the money. The Commission’s AI gigafactory tender, opened in July, invites consortia to build up to seven training campuses. Public money is capped near €10 billion and at 35% of any project. The €30 billion headline only appears if private capital puts up the rest, on chips Europe does not make.
The sovereignty case is the point of the programme, and the market share is why. European providers’ slice of the regional cloud market fell from 29% in 2017 to about 15% by 2022, according to Synergy Research, a figure the Commission now repeats. A continent that rents its compute from three American firms has a reason to want its own floor, and somewhere to run the home-grown models we mapped here. The plan is coherent, the problem Brussels faces is the execution.
The Grid that Has Other Ideas
The binding constraint is not necessarily the lack of capital. It is power, and the ability to hang a gigawatt-scale load on a grid that was never built for one.
The European Data Centre Association puts the investment needed to meet demand at €176 billion between 2026 and 2031. It is blunt that grid readiness, not money, decides whether that sum can be spent.
The evidence is already on the board. In Ireland, data centres used 23% of metered electricity in 2025. The old Dublin connection freeze has been replaced by a harder bargain: new sites must bring their own dispatchable generation and source 80% of annual demand from new Irish renewables within six years.
Amsterdam barred new data centres and expansions until at least 2030. That sits on top of a 2022 national rule that already confines the largest hyperscale sites to two designated locations. Germany wants the compute and has written some of the strictest efficiency and waste-heat rules in Europe. Every extra condition adds cost and time to the projects Brussels wants sped up.
Tie fast-track permits to strict green tests and you slow the build. Drop the tests and you undercut the Green Deal. That contradiction is not resolved in one simple proposal.
The IEA expects European capacity to grow by about 70% by 2030, but the political target is 200%.
The Water Nobody Costed
While the industry worries about electricity, it is water shortages that turn data centre neighbours into opponents.
Most large campuses still shed heat through evaporative cooling. Cold water chills hot air and then evaporates, so roughly four fifths of what is withdrawn does not come back. In the wet north, nobody much notices the water. In the dry south, it is already a fight.
In Aragón, the coalition Tu Nube Seca Mi Río, led by Aurora Gómez, has fought the regional government’s fast-track of an Amazon project. Amazon has already asked for 48% more water at its three sites than first planned. Those communities are not comforted by a national average, as we set out when the golf-course comparison did the rounds.
Where the Map Is Actually Moving
If connection times govern siting, the buildings go where the electrons are cheapest, cleanest and quickest to hook up. Energy think-tank Ember has spelled that out. Poor grid planning pushes developers toward the shortest queues. By 2035, more than half of Europe’s data centre capacity could sit outside the five cities that define the market today.
The direction of travel is already visible. The Nordics, with cold air, uncongested grids and deep hydro and wind, are absorbing load fast enough that regional demand could more than double by the end of the decade. That is the shift we traced when Iceland became an unlikely capital.
The Iberian peninsula is the other magnet, with campuses such as Start Campus’s Sines site in Portugal and the cluster around Aragón. A third band is forming in parts of Central and Eastern Europe, where queues are shorter even if the brand is not.
The pattern is close to the opposite of the sovereignty script. The strategy wants compute near the industries and public bodies meant to use it. The physics concentrates it wherever the grid says yes.
The Case That Europe Pulls This Off
Acceleration zones and a twelve-month permit cap go at the delay operators complain about most. The €10 billion of public money is there to pull in private capital, not to build the halls. Where the grid cooperates, as in the Nordics and on the Iberian coast, Europe can already build fast.
The backlash is not unique. In the US, Data Center Watch counted 75 projects worth about $130 billion blocked or delayed in the first quarter of 2026 alone.
The timetable is the big problem for Brussels. As mentioned, the IEA sees about 70% growth but the political target is 200%. A stop in Amsterdam or a fight in Aragón is not something an acceleration zone can wave away.
The likely result could be that Europe builds the centres, slowly, and in the wrong places.
Author: Grace Sharp
