Europe’s race to build more of its own artificial-intelligence and supercomputing infrastructure has reached a striking turning point: French manufacturer Bull says roughly 70% of the components in the machines it builds can now be sourced in Europe, up from only 20% to 30% five years ago. But one critical piece remains overwhelmingly dependent on suppliers outside the region — memory chips.
The gap matters because advanced AI systems and scientific supercomputers need enormous quantities of fast memory to keep processors supplied with data. Bull CEO Emmanuel Le Roux told Reuters that Europe has made progress in boards, interconnects, cooling and processors, while memory remains the major component for which there is no significant European supplier.
Europe’s supercomputer supply chain has changed rapidly
Five years ago, Bull sourced only around one-fifth to one-third of its machine components in Europe. Today that figure is around 70%, according to Le Roux. The shift reflects years of investment aimed at reducing dependence on foreign technology for strategically important computing systems.
European processor projects including SiPearl, Vsora and Openchip are part of that effort. Bull itself holds a stake in SiPearl. The company can also assemble systems around processors from Nvidia, AMD and Intel depending on customer requirements, showing that technological sovereignty does not necessarily mean eliminating global suppliers. Instead, Europe is trying to ensure that it has credible domestic options for more parts of the computing stack.
Why memory is becoming the key bottleneck
Processors receive most of the attention in the AI boom, but they cannot work efficiently without memory. Memory chips temporarily hold the data a processor is actively using. In AI training and high-performance scientific computing, enormous datasets must move rapidly between memory and processing units.
Le Roux said processors account for roughly 10% to 20% of a supercomputer’s value. Memory prices, meanwhile, have risen sharply and have become one of the industry’s largest concerns. Samsung Electronics, SK Hynix and Micron collectively manufacture more than 90% of global memory, according to Reuters.
That concentration creates a strategic challenge. Europe may be able to design processors, build cooling systems and assemble sophisticated machines domestically, but a disruption in the Asian or U.S.-linked memory supply chain could still affect its ability to expand computing capacity.
Bull doubles production at its Angers factory
The supply-chain debate comes as Bull rapidly expands production. The French state-owned company has reopened an enlarged factory in Angers after an €80 million investment. Output has doubled from six supercomputer racks per month to 12, executives told Reuters. The facility could increase capacity again to 24 racks per month next year if demand continues to grow.
That expansion is significant because the Angers plant is described by Reuters as Europe’s only factory dedicated to building these machines. Bull says it has received more orders this year than ever before and has won 15 of 18 tenders issued by EuroHPC, the European body that co-funds supercomputing infrastructure with participating states.
Europe is spending billions on computing power
The European Union has committed €7 billion between 2021 and 2027 to its supercomputer network. The objective is to narrow Europe’s computing gap with the United States and China while giving researchers, governments and companies access to infrastructure capable of handling advanced scientific simulation and AI workloads.
Bull built JUPITER in Germany, Europe’s first exascale-class system. Exascale computing refers to machines capable of performing operations at a scale measured in quintillions per second, opening new possibilities in climate modelling, medicine, engineering, physics and artificial intelligence.
The company is also assembling the Alice Recoque supercomputer for France. The system, named after the pioneering French computer scientist, accounts for 94 racks and is due to be delivered in stages beginning later this year. Reuters reported a combined French government and European Union cost of €554 million.
Finland’s LUMI-AI adds another major order
At the same time, Bull is working on Finland’s €388 million LUMI-AI system, which the company describes as its biggest contract to date. Before the Angers expansion, executives said, Bull could not have assembled the French and Finnish orders simultaneously.
This illustrates why manufacturing capacity is becoming almost as important as chip design. Governments can fund new AI systems, but those plans still require factories capable of integrating processors, memory, networking, cooling and power-management technologies at enormous scale.
AI demand is changing the economics of supercomputing
Traditional supercomputers were associated primarily with government laboratories, weather forecasting, nuclear research and universities. Generative AI has expanded the market. Training frontier AI models requires many of the same ingredients: large numbers of accelerators, extremely fast networking, huge memory capacity and sophisticated cooling.
Cloud operator OVHcloud and European Commission-backed Domyn are among organizations attempting to train frontier-class AI models using Bull’s JUPITER system. The overlap between AI infrastructure and conventional high-performance computing is therefore increasing.
NewsNationOnline has been following the enormous financing requirements created by this trend. Amazon, for example, is exploring new ways to finance expensive AI hardware, as explained in our report on its proposed $8 billion Nvidia chip financing structure. Broadcom’s relationship with Anthropic shows another side of the same infrastructure race, with potential financing linked directly to computing hardware.
Memory dependence will not disappear quickly
Europe does have initiatives aimed at strengthening semiconductor manufacturing, but creating a globally competitive memory industry would require enormous investment, specialized expertise and years of construction and qualification. Le Roux said European memory initiatives exist but would take a long time to become meaningful suppliers.
Memory manufacturing is particularly difficult because established producers operate at huge scale and continuously invest in new fabrication technology. AI has also intensified demand for advanced high-bandwidth memory, increasing the strategic value of a sector that previously attracted less public attention than CPUs and GPUs.
Europe’s strategy is broader than complete self-sufficiency
Even Bull’s increasingly European supply chain remains international. The Angers factory uses Nvidia, AMD and Intel processors when appropriate. Bull has also partnered with Taiwan’s Foxconn to manufacture Nvidia NVL systems at a Czech plant, with final assembly taking place in Angers.
That model suggests Europe’s practical goal is resilience rather than isolation: expanding local design and manufacturing while maintaining access to global technology. Having multiple sources for key components can reduce vulnerability without requiring every part of a machine to originate inside Europe.
Why this matters beyond Europe’s technology sector
Supercomputers increasingly underpin strategically important activities ranging from aerospace design and pharmaceutical research to climate modelling and national security. Airbus inaugurated Bull-built systems in Toulouse and Hamburg this year, tripling its simulation capacity, according to Reuters.
As AI spreads across industry, access to computing infrastructure may become an economic competitiveness issue as well. Regions that can build and operate large computing systems have greater control over where sensitive data is processed, which companies receive capacity and how quickly new research can move from laboratory to commercial deployment.
The next test: turning investment into a resilient ecosystem
Bull’s production expansion shows that Europe is adding real manufacturing capacity rather than relying only on policy targets. Moving from 20%-30% European component sourcing to around 70% in five years is also a substantial change.
But memory exposes the limits of that progress. A modern supercomputer is only as resilient as the supply chain behind its most difficult-to-replace components. For Europe, the next stage of the AI infrastructure race may therefore be determined not only by who designs the fastest processor, but by whether the continent can secure enough memory to keep those processors working at full speed.
Sources and image credit
This report is based on Reuters reporting from Bull’s Angers facility published October 1-2, 2026. Featured image is an illustrative supercomputer cluster image produced by the U.S. Department of Defense and made available through Wikimedia Commons as public domain.
आपके लिए सुझाव
