French supercomputer maker Bull now sources about 70% of the components in its systems from Europe, up from just 20%-30% five years ago, CEO Emmanuel Le Roux told. Boards, interconnects, cooling equipment and an increasing share of processors can now be sourced domestically.
Memory is the major exception.
Europe does not have a large-scale supplier able to compete with Samsung, SK Hynix or Micron, leaving a critical piece of the AI computing stack dependent on foreign production.
Europe Has Rebuilt Much of the Supply Chain
The shift comes as Europe pushes to build more sovereign computing infrastructure.
Bull, which is now owned by the French state, recently doubled production capacity at its Angers factory from six to 12 racks per month and could reach 24 by 2027. The company built JUPITER, Europe’s first exascale supercomputer, while the EU plans about €7 billion of investment through 2027 to strengthen domestic AI computing capacity.
Processors are also becoming less dependent on overseas suppliers. European developers are beginning to produce their own chips, reducing another longstanding weakness in the region’s computing stack.
But the importance of processors alone may be overstated. Le Roux said processors account for roughly 10%–20% of a supercomputer’s value, while memory has become one of the industry’s biggest cost pressures.
Memory Is Becoming an AI Bottleneck
That matters because modern AI systems require increasingly large amounts of DRAM and high-bandwidth memory.
Rising memory costs have already pushed some Nvidia server prices more than 15% higher. The broader AI infrastructure buildout now depends not only on GPUs, but also on memory, networking, cooling and power.
The weakness could become more important as AI spending accelerates. The global buildout already requires trillions of dollars of AI financing, and shortages in any one component can raise the cost of entire systems.
