Ericsson and Jülich put JUPITER to work on AI for 5G and 6G

Ericsson's Forschungszentrum Jülich collaboration is useful because it moves 6G work closer to Europe's high-performance-computing layer. The public event is a March 2026 agreement to develop advanced AI for 5G and 6G networks, with Jülich's supercomputing environment, including the JUPITER exascale system, as the computational backdrop. The signal is not that a commercial 6G network is near. It is that radio research, AI model training, neuromorphic computing and energy-efficiency work are being tied to compute resources that can test network ideas before standards and procurement settle.

• Ericsson and Forschungszentrum Jülich signed a memorandum of understanding on 25 March 2026 to research advanced AI for 5G and future 6G networks.
• The partners say they will use JUPITER to design and test AI solutions, with work spanning high-performance computing, neuromorphic approaches and energy-efficient network intelligence.

A research programme, not a 6G launch

Ericsson and Forschungszentrum Jülich are linking telecom research to Europe’s largest class of scientific computing. A European Commission 5G Observatory note says the organisations launched a research partnership on 25 March 2026 to explore how high-performance computing and artificial intelligence can support continued 5G development and future 6G systems.

Under the memorandum, the partners intend to use the JUPITER supercomputer to design and test advanced AI solutions. The stated areas are high-performance computing, neuromorphic approaches and energy-efficient intelligence intended to improve network performance while limiting energy use. Those are research objectives, not reported results: no named telecom workload, benchmark, energy saving or proof of concept has yet been disclosed publicly.

The distinction matters because 6G remains pre-commercial. The Commission note refers to commercial 6G services being expected around 2030, but the agreement does not set a launch date, announce operator adoption or settle a technical standard. It creates an environment in which Ericsson and Jülich can test ideas before those decisions are made.

What JUPITER contributes

EuroHPC describes JUPITER as Europe’s first exascale supercomputer. It was inaugurated at Jülich in September 2025 and is capable of one exaflop. Its accelerator partition contains about 24,000 NVIDIA GH200 Grace Hopper superchips and is designed for demanding simulations and AI-model training.

That scale makes JUPITER a plausible platform for testing computationally intensive network-AI methods. It does not, however, give Ericsson ownership or exclusive control of the machine. JUPITER is owned and co-funded by EuroHPC and operated in the Jülich environment. Public material does not specify Ericsson’s allocation, scheduling priority or duration of access.

SDxCentral’s independent report places the collaboration in a wider vendor race to apply AI and high-performance computing to future radio systems. The practical value will depend on whether the partners turn compute access into reproducible findings rather than broad research claims.

Why it matters

For Ericsson, the programme offers a way to examine AI methods at a scale that is difficult to reproduce in an ordinary vendor laboratory. For Forschungszentrum Jülich, it brings a demanding industrial network problem into an exascale research setting. For European policy, it is an example of publicly backed computing infrastructure being used by industry before a new mobile generation reaches the market.

The collaboration could eventually influence algorithms, network-management techniques or evidence presented in standards discussions, but none of those outcomes is guaranteed by the MoU. The next credible milestones are a named JUPITER workload, a published methodology, performance and energy measurements, peer-reviewed work, or a disclosed contribution to a standards process.

What to watch

Watch for the first technical deliverable, the amount and terms of JUPITER access, the datasets or network models used, and whether results separate gains from the AI method from gains attributable simply to more compute. Until such evidence appears, this is best understood as a research-capability agreement rather than proof of 6G readiness.