Europe shouldn’t bet it’s AI future on big data centres

OpenAI and Anthropic have started a process that could lead to public offerings to raise much-needed capital. Europe should be paying close attention. 

Both companies have confidentially submitted draft S-1 registration statements to the US Securities and Exchange Commission. An S-1 asks questions which have been obscured by AI industry hype. How does the company make money? How much capital does it require? Who does it depend upon?  

Europe is preparing to bet heavily on how AI compute will be produced. The European Commission’s proposed Cloud and AI Development Act aims to at least triple EU data-centre capacity within five to seven years. The Union is also moving ahead with up to seven AI gigafactories, backed by as much as €10 billion in public funding to unlock at least €20 billion in private funds. 

Europe absolutely needs computing capacity, but the harder question is what kind of compute ecosystem it should build. 

Silicon Valley’s answer has been large concentrations of processors demanding vast amounts of energy, water, and capital to support ever-larger models operated by a handful of companies. This is seen as AI’s natural physical requirements but it’s a business model more than a technological one. 

Before Europe spends tens of billions building the infrastructure that Silicon Valley says the AI future requires – not to mention new power plants to support them – the EU must examine whether the same approach is wise or necessary. 

Technological systems can seem inevitable after governments reorganise infrastructure around them, as Paris Marx describes in Road to Nowhere. For instance, he shows how automobile dependency seemed natural when cities had been redesigned for cars. Europe should be careful about doing the same with AI. 

Every data centre requires electricity, processors and enormous capital investment. Once governments reorganise energy systems and industrial policy around the hyperscale approach to AI, it becomes seen as a “natural” outcome.  

The risk is mistaking the capital requirements of today’s dominant AI companies for the computational requirements of Europe’s future economy. 

Some applications require enormous concentrations of compute, but much of the AI economy may not. Smaller specialised models, dedicated processors and powerful local devices are changing where computation happens. They also fit better into green energy grids. 

Moreover, there is a security case for distributing computing power. Iranian drone strikes this year damaged three Amazon data centres in the UAE and Bahrain, resulting in customer data which could not be restored. As compute becomes critical infrastructure, concentrating it also concentrates the physical risk.  

Quantum computing is another reason for Europe to keep its options open. Today's quantum computers can't do the work of the chips that train AI models, but they can handle certain narrow tasks remarkably well.  

In 2025, the company D-Wave reported in Science that its machine had solved a complex materials problem in minutes, a job it said would take one of the world's most powerful supercomputers a million years. The machine drew about 12 kilowatts. Frontier – the supercomputer it was measured against – draws around 20 megawatts, well over a thousand times more.  

The lesson is that more computing power does not always require more conventional computing infrastructure and the vast energy demands that come with it.   

Data centres, transmission lines and generating capacity are investments measured in decades, while computing technologies can change in months or years. Europe should not make generational infrastructure investments around technologies that could be superseded by cheaper, more efficient alternatives within years or months. 

A mixed approach can ensure Europe doesn’t waste its money by hyperscaling only where necessary, using distributed and edge computing where possible, and specialised processors where they offer an efficiency advantage. 

European technological sovereignty should mean more than building enormous data centres inside the EU. Real sovereignty means being able to change suppliers, protect data and continue operating when circumstances change. 

This is why the OpenAI and Anthropic filings bear watching once they are made public. Both companies face the same structural risks, from compute and energy costs to model commoditisation and uncertain infrastructure returns. Their filings may help Europe distinguish technological necessity from Big Tech's corporate choices. 

Europe arrived later to the AI boom. But this provides an opportunity to learn from, and avoid, costly pitfalls. It needs robust compute to compete in AI. But it does not need to spend a generation building yesterday’s answer only to find it obsolete before it’s complete. 

This Op-ed was originally published in The Rapporteur.

Chris Kremidas-Courtney is a Senior Adviser at the European Policy Centre.

The support the European Policy Centre receives for its ongoing operations, or specifically for its publications, does not constitute an endorsement of their contents, which reflect the views of the authors only. Supporters and partners cannot be held responsible for any use that may be made of the information contained therein.