The aim is to build four to five mega facilities, each powered by 100,000 graphics processing units (GPUs), a type of chip used to train AI models. They would be four times the size of the AI Factories and rival some of the world’s leading projects, such as OpenAI’s Stargate data center in Norway.
The projects will be funded by combining public and private investment, and the EU has set up a €20 billion fund to back the plans from its side. A total of 76 bids to establish 60 sites across 16 countries were filed during an informal sense-check by the Commission last year, with established companies such as French Scaleway among them.
“This demonstrates an emerging substantial market demand,” the Commission’s Regnier said.
Efforts are ongoing to merge and finalize some of those bids and to iron out the practical details between EU countries that support the bids and the vehicle running the EU’s network of supercomputers, called EuroHPC. A formal call for proposals has been delayed twice, and is now expected to be launched this spring.
The idea is clear: to build very large facilities to train very large AI models, which requires vast amounts of computing power.
But critics say it is unclear which companies will use the power from the mega facilities compared to smaller facilities. “It’s way less clear what the audience for the gigafactories is,” said Nicoleta Kyosovska, a research assistant at the Brussels-based think tank Centre for European Policy Studies, in an interview.