8.1 C
London
Sunday, November 8, 2026

Europe Is Building Its Own AI. But Can It Afford To?

Date:

Related stories

Child Online Safety Rules Raise a Bigger Question: Who Controls the Internet?

Governments across the continent are moving to keep children...

Europe Commits €525 Million to Build a New Generation of Chip Design Start-ups

The European Chip Design Platform, launched under the EU...

AI Ambitions Are Running Into Europe’s Infrastructure Reality

Brussels wants far more AI computing power, and it...

Constantinople to Istanbul: One City, Two Empires Opens at the V&A

The V&A opens its biggest new exhibition of the...

Beyond Children’s Theatre: How Contemporary Puppetry Is Finding a Wider Stage

At Finland’s TIP-Fest, puppets, everyday objects, shadows and physical...

Christine Lagarde warned this month that Europe risks being cut off from AI entirely, dependent on infrastructure it doesn’t own and can’t fully control. Days later, the IMF told EU finance ministers something else. The technology could strain power grids and widen inequality, even if it works. Both warnings point to the same uncomfortable truth. Wanting AI sovereignty and being able to fund the infrastructure it requires are two very different problems.

Lagarde’s Warning

Speaking in Vienna on 14 September, the ECB president laid out the stakes bluntly. Within a few years, she said, AI “will be screening goods at the border, deciding which tax returns are audited, dispatching trains.” Losing access to that technology would hit every sector at once. So would a foreign supplier simply changing the terms.

The infrastructure gap behind that warning is stark. Europe already lacks enough data centre capacity to meet its own demand. Lagarde said that gap is projected to grow more than sixfold within a decade on current trends. Done right, she argued, AI could lift European productivity by up to 4% over ten years. The catch is that reaching those gains requires computing infrastructure Europe doesn’t currently have. Building it isn’t simply a matter of writing a cheque.

What “Afford” Actually Means Here

The EU’s answer so far is the AI Gigafactories programme, opened for tender on 30 July. Up to seven sites will be built, backed by as much as €10 billion in combined EU and national funding. That public money is designed to attract at least €20 billion in private investment on top. Each facility bundles AI processors, cloud software, high-speed networking, and energy-efficient data centre design into a single integrated site.

That’s a serious commitment on paper. It’s also worth sizing against the competition. A single American hyperscaler is guiding toward $175 billion in capital expenditure this year alone. That means the EU’s public commitment amounts to roughly three weeks of one company’s annual spending. The original plan called for only five Gigafactory sites; member state interest pushed the number to seven before the tender even opened. The Commission’s own framing acknowledges this isn’t really about matching that scale. It’s about something narrower and more achievable. European institutions and startups need to access serious compute at all, on terms decided in Brussels rather than Washington or Beijing.

The Physical Cost Nobody Budgeted For

Here’s where “afford” stops being purely financial. An IMF background note was prepared for EU finance ministers, who met in Dublin on 18-19 September. It found that Europe’s data centres already consume roughly 3% of the continent’s electricity. That demand will climb sharply as AI use expands.

Frankfurt, London, Amsterdam, Paris, and Dublin are Europe’s main data centre hubs. Local power networks in each are already straining under existing clusters. Building Gigafactories on top of that pressure means budgeting for grid capacity and cross-border energy integration alongside chips and software. Money alone doesn’t solve this kind of bottleneck. It’s rooted in physical infrastructure and permitting timelines that move far slower than computing demand does.

The IMF’s Reality Check

Beyond electricity, the IMF’s paper flagged a second cost rarely discussed in AI sovereignty debates: who actually benefits. The Fund estimated AI could lift European productivity by roughly 1% over five years, a more conservative figure than Lagarde’s. It warned those gains would land unevenly across countries, regions, and workers. Around 60% of jobs in advanced European economies sit in occupations highly exposed to AI. Some workers will gain productivity; others face displacement instead.

The paper echoed a concern also raised by Mario Draghi. Europe’s fragmented capital, labour, and energy markets actively hold back the investment and innovation AI sovereignty depends on. Building Gigafactories inside 27 separate national markets adds real friction. Each has its own grid rules, permitting processes, and labour regulations that a single unified market wouldn’t face.

Put together, these two reports reframe what “affording” European AI actually means. It’s not simply €30 billion versus not having €30 billion. Electricity infrastructure, cross-border grid coordination, and semiconductor access all matter just as much. So does permitting speed, and a labour market ready to absorb genuinely uneven gains, all layered on top of the headline investment figure. Lagarde is right that Europe cannot risk permanent dependence on foreign AI infrastructure. The IMF is equally right about the risk underneath. Simply building Gigafactories, without fixing the market fragmentation and energy bottlenecks beneath them, could deliver sovereignty on paper while leaving the real vulnerability intact.


Key Sources


Subscribe to EuroLuminant for independent European journalism.

Kay
Kay
The reporter/editor based in London

Subscribe

- Never miss a story with notifications

- Gain full access to our premium content

- Browse free from up to 5 devices at once

Latest stories

LEAVE A REPLY

Please enter your comment!
Please enter your name here