Summary
- SiPearl has delivered the first Rhea1 CPU samples to Bull.
- Bull has begun integrating Rhea1 into its BullSequana XH3000 platform for JUPITER.
- The milestone moves the European-designed processor from silicon bring-up into system-level integration.
SiPearl has delivered the first samples of its Rhea1 processor to Bull for integration into the BullSequana XH3000 platform used in JUPITER, Europe’s first exascale supercomputer.
Bull has begun integrating the processors into the server platform, moving Rhea1 from chip-level bring-up towards operation as part of a complete high-performance computing system.
JUPITER was procured by the EuroHPC Joint Undertaking and is hosted at the Jülich Supercomputing Centre in Germany. Rhea1 is intended to provide a European-designed CPU option within the system.
The processor has been developed through the European Processor Initiative, part of a wider effort to strengthen European control over strategically important high-performance computing technology.
SiPearl says Rhea1 contains 80 Arm Neoverse V1 cores and integrates high-bandwidth memory alongside DDR5 and PCIe connectivity. The architecture is designed for HPC, AI, and data-centre workloads rather than mainstream desktop computing.
Delivery of first samples is not the same as volume availability. Bull now has to integrate the chips with memory, firmware, networking, management systems, power delivery, and cooling inside the wider XH3000 platform.
That system-level work is significant because very large compute environments expose issues that may not appear during individual-chip validation. Reliability, thermal behaviour, firmware stability, interconnect performance, and workload scaling all have to hold across large numbers of nodes.
JUPITER also illustrates the practical limits of describing a supercomputer as entirely sovereign. Rhea1 is European-designed, but modern HPC systems still rely on international semiconductor fabrication, accelerator hardware, networking components, memory, and other parts of the supply chain.
The European policy objective is therefore better understood as increasing control over key layers rather than eliminating every external dependency.
For data centre engineers, the processor’s origin is only one part of the deployment. Exascale systems concentrate substantial electrical and thermal loads into specialised halls requiring high-capacity power distribution, cooling, controls, and operational resilience.
The same pressures are increasingly appearing in commercial AI facilities. As compute density rises, facility infrastructure has to evolve alongside processor and accelerator design rather than being treated as a static shell around the IT equipment.
EuroHPC’s expanding AI factory programme adds further demand for European computing infrastructure and increases the strategic value of suppliers capable of occupying more of the hardware stack.
Rhea1’s first delivery is therefore an integration milestone rather than the end of the programme. The next test is whether Bull can qualify the processor at system level and whether SiPearl can move successfully from samples into production deployment.

