Europe Orders a Giant AI Supercomputer for €390 Million

Europe is building a massive new supercomputer to boost its artificial intelligence power. The European Union's supercomputing organization, called EuroHPC, has chosen a French company named Bull to build the machine for about €388 million — roughly $449 million. The new system will be placed in Finland and is expected to be ready in 2027. This is part of a bigger effort to help Europe keep up with the United States and China in the race to develop AI.

The new supercomputer is called LUMI-AI. It will be placed next to an already-existing machine called LUMI at a supercomputing center in Finland. Six countries are helping to pay for it: Finland, the Czech Republic, Denmark, Estonia, Norway, and Poland. Each of those countries is sharing the cost equally with EuroHPC, the EU body in charge of supercomputing.

This is the biggest contract Bull has ever won. The French government owns Bull, and it recently separated the company from a larger business called Atos, which had been struggling with debt. Bull's win shows that European-owned tech companies can compete for large and important projects.

LUMI-AI will be built to do two big jobs. First, it will train and run AI programs at a very large scale. Second, it will carry out heavy-duty scientific experiments. The computer chips inside will come from AMD, one of Nvidia's biggest rivals. Storage will be provided by IBM, and networking — which connects everything together — will come from Nokia.

EuroHPC has been working for years to build a network of powerful computers across Europe. It has a budget of €8.2 billion covering 2021 to 2027. So far, it has set up 19 AI factory centers built around 12 supercomputers, including one in Germany called Jupiter, and it plans to add up to seven more sites.

Some of these European supercomputers are already being used for AI. A large cloud computing company called OVHcloud said it used the Jupiter machine to start training an AI model. An Italian startup named Domyn is also using EuroHPC systems to build its own AI model, aiming for 400 billion parameters — a measure of how complex an AI model is. Both companies want to become top AI developers in Europe, alongside the Paris-based company Mistral, but neither has released a finished model yet.