Sovereign AI Research Without Compromise.

How KInIT and dNation built a high-performance, vendor-independent AI cluster based on SCS.

Technical Overview:

The implementation was carried out as a greenfield project using open technologies exclusively:

  • 5× NVIDIA H200 NVL 141 GB – for training workloads
  • 20× NVIDIA RTX Pro 6000 – for training workloads and inference
  • 904 physical CPU cores, 10 TB RAM, ~1 PB SSD (raw capacity)
  • 200 Gbit/s network
  • Stack: SCS, OpenStack, ALASCA Yaook/Arko, Kubernetes, SONiC, OPNsense (firewall), Tailscale (VPN), Ceph, Pacifico Digital Exploration (reporting), Keycloak (integration with Microsoft Entra ID)

Produced by:

Created using:

Certified integrator

Background

The Kempelen Institute of Intelligent Technologies (KInIT) Based in Slovakia, the organisation carries out cutting-edge research in the field of artificial intelligence with a team of around 70 researchers and research engineers.

To enable it to run its AI workloads independently, efficiently and cost-effectively, KInIT has teamed up with the SCS implementation partner dNation (Certified SCS IaaS Integrator Silver) has set up a private AI cluster based on the Sovereign Cloud Stack (SCS). In just two months, a high-performance, sovereign infrastructure was created, based entirely on open-source technologies.

Procedure

The project was implemented as a greenfield project that complements the existing environment. KInIT and dNation jointly defined the requirements; dNation then assumed full responsibility for the technical implementation. Experts from SCS and ALASCA-ecosystems, including those of Cloud&Heat, UhuruTec, artcodix and nefos. This is a good example of how ecosystems can pool and utilise shared knowledge in a targeted manner.

Particular attention was paid to testing before going live: since hardware rarely arrives on time, the network configuration was simulated in advance using GNS3 in a virtualised environment, and application scenarios were validated on virtual machines. This preparation paid off: when the physical hardware arrived, the configuration could be transferred without major issues.

The rollout went smoothly for KInIT’s staff: thanks to the conceptual similarities between OpenStack and familiar technologies, they were able to apply their existing knowledge directly to the new environment. After intensive testing – particularly in direct comparison with the previous AWS and Proxmox clusters – the platform was fully operational.

Result

The successful implementation provides a stable foundation for the years ahead. The infrastructure is designed to grow with the demands of AI research: additional compute capacity can be added as needed – some of which is already planned – without creating dependency on a single provider.

For AI research institutions like KInIT, this flexibility is strategically crucial: compute requirements grow quickly – and often unpredictably – as new types of models emerge. A scalable, sovereign infrastructure is therefore not a nice-to-have but a fundamental prerequisite for competitive research.

Want more details?

Download the full use case with all the background and information!

Further use cases