The European Union has ordered another AI supercomputer, a €387.8 million machine called LUMI-AI, from the French state-owned maker Bull. On its face it is one more procurement in a long-running programme. But two details give it more weight than the price tag suggests: Europe is already turning away users because it does not have enough computing power, and this machine will run on AMD chips, not Nvidia’s.
Start with the demand. EuroHPC, the EU body that coordinates Europe’s publicly funded supercomputers, says it is having to reject some applications outright because capacity cannot keep up. That is a striking admission, and a healthy sign in one respect: it means European researchers, startups, and companies really want this compute. LUMI-AI, the sixth system under the bloc’s AI Factories programme, will sit beside the existing LUMI machine in Finland, offer ten times its AI capacity, and come online in the second half of 2027, one of several systems meant to ease the crunch.
Why the AMD choice matters
The hardware is arguably the more interesting story. LUMI-AI will be built around AMD’s Instinct accelerators and EPYC processors, with Bull’s own high-speed interconnect, IBM storage, and Nokia networking. In a market where Nvidia is so dominant that its financial entanglements with the whole AI industry have started to worry investors, a flagship European system deliberately choosing its main rival is a statement. Europe is signalling that it does not want to be wholly dependent on a single American chip supplier.
The catch is that AMD is American too, as is IBM. So the sovereignty on display here is real but partial. The machine is owned by the EU, built by a European champion, hosted on European soil, and shared across six countries, from Finland to Poland. The accelerators at its heart, though, still come from the United States. Europe owns the supercomputer; it does not yet make the silicon inside it, which is the harder, longer problem the continent is only beginning to tackle.
A real step, on a long road
It is worth keeping the scale honest. €388 million is Bull’s largest-ever order and a serious machine, and it validates France’s decision to rescue Bull from the wreckage of the collapsed Atos earlier this year. But set against the AI spending in the United States and China, it is small. American hyperscalers alone are pouring hundreds of billions into AI infrastructure every year; EuroHPC’s entire budget through 2027 is a fraction of that. Europe is closing the gap, but from a long way back, and slowly, a machine that will not switch on until late 2027 is not a quick fix.
There are signs the effort is bearing fruit. European developers are already using these systems to build real models, with France’s OVHcloud and Italy’s Domyn among those training on EuroHPC machines. Yet Europe’s grander plan, a network of giant AI “gigafactories,” has been stumbling on delays, a reminder that ambition and delivery are different things.
So is Europe finally getting serious about AI compute, or still bringing a garden hose to a wildfire? A bit of both. LUMI-AI is a genuine, well-designed step, pooled across nations, energy-efficient, and pointedly not beholden to Nvidia. It is also one modest machine in a race where the leaders are spending on a scale Europe has not begun to match. The appetite is clearly there. Whether the ambition can be built fast enough is the open question.
This website uses cookies.