Europe is building 35 AI supercomputers. The provisioning stack decides who controls them.

Research teams across Europe are about to bring more GPU capacity online in a single year than the continent has ever added at once, and the teams that provision it on infrastructure they govern will keep control of it. NVIDIA announced at ISC High Performance 2026 that 35 AI supercomputers are in development across 23 countries, serving more than three million researchers, including the MareNostrum5 AI upgrade at Barcelona Supercomputing Center, HammerHAI at HLRS, and BavariaAI's Blue Swan. The EuroHPC Joint Undertaking separately inaugurated the SOL quantum system and the LISA upgrade to Leonardo in Bologna.

The hardware is the headline. The stack that boots it, schedules it, and keeps its workloads reproducible is what determines whether a European institution runs its own AI infrastructure or depends on someone else to run it.

The teams that provision, containerize, and govern their AI clusters on open foundations keep sovereignty over the infrastructure, not just the data on it. That is the layer worth deciding deliberately while the systems are still being built.

Provision thousands of GPU nodes in minutes

A supercomputer is only useful once its nodes are running the right image. Warewulf Pro provisions nodes statelessly, bringing thousands of compute nodes online in minutes and keeping them consistent as the cluster grows. Warewulf is an open source provisioning project that CIQ contributes to upstream and supports commercially, and it runs at national labs and research institutions operating anywhere from 50 to more than 10,000 compute nodes.

Stateless provisioning matters most at the scale these European systems are targeting. A research team changes one image and redeploys the fleet, rather than configuring nodes one at a time as the cluster expands. The provisioning layer stays GPU-aware, so accelerator nodes come up ready for work.

Keep the operating system layer under your own governance

Sovereignty starts at the operating system. Rocky Linux from CIQ is open source and community-driven, governed through the OpenELA community rather than by any single vendor's roadmap. A European institution running Rocky Linux controls its own upgrade timeline and is not tied to one company's commercial strategy for the base of its stack.

RLC Pro AI extends that base to GPU workloads as a performance-tuned Enterprise Linux with CUDA, ROCm, and InfiniBand drivers pre-integrated. Research teams standing up accelerator clusters get a validated GPU stack on a foundation they can inspect, rebuild, and govern themselves.

A base governed through an open community, rather than one vendor's roadmap, lets a European institution set its own upgrade timeline and audit its own stack.

Planning infrastructure for a European AI or HPC program? See how Rocky Linux from CIQ supports sovereign AI infrastructure.

Carry reproducible workloads across sites and borders

Research does not stay on one machine. A workload validated on a national system often needs to run at a partner institution, a university cluster, or a EuroHPC allocation in another country. Apptainer is the open source container runtime built for that reality: rootless, daemonless, GPU-passthrough native, and MPI-aware, with the SIF format capturing a workload so it reproduces exactly on another site.

Apptainer is the de facto container standard across HPC clusters, so a workflow containerized at one European site already speaks the format the next one runs. For a research collaboration that spans borders, reproducibility is the difference between sharing science and porting it again. CIQ contributes to Apptainer upstream and supports it commercially, and it is one of several vendors that do.

Decide the stack while the systems are still being built

The 35 systems now under construction will define European research computing for years. The procurement conversations happening around them are the moment to decide the provisioning, container, and operating system layers, not after the hardware lands.

A team that chooses open foundations at this stage keeps its options open: it can inspect the source, support the stack through more than one vendor, and govern its own upgrade path. That is what sovereignty means in practice at the infrastructure layer.

Layer	What it does	CIQ option
Operating system	Governed base for the cluster	Rocky Linux from CIQ, RLC Pro AI for GPU nodes
Provisioning	Brings nodes online statelessly at scale	Warewulf Pro
Container runtime	Reproducible workloads across sites	Apptainer
Orchestration	Schedules HPC and AI jobs, bursts to cloud	Fuzzball

Each layer is open source at its foundation, with CIQ contributing upstream and offering commercial support. Adopt one layer or the full stack, and support them all.