Universities and research institutions in 2026 need cloud-native infrastructure for three overlapping missions: serious research computing (especially AI/ML), reproducible research environments for publication, and curriculum delivery for cloud-native courses. Cozystack provides a single open-source foundation that serves all three — multi-tenant for departments, labs, and student cohorts; KubeVirt for legacy and modern workloads; GPU-as-a-service for AI research; air-gapped support where research data sovereignty matters.
Ænix builds Cozystack-based platforms for universities, research institutes, and R&D organizations across the EU, DACH, and Central Asia.
Pairs with: Ænix Private Cloud Platform for sovereign student-data + multi-tenant research-group isolation; AI Platform for AI/ML research labs with GPU pools.
Three university missions Cozystack serves
1. Research computing infrastructure
Modern research increasingly demands GPU clusters, large-scale data processing, and HPC-adjacent workloads. AI/ML research, computational biology, climate modelling, materials science — all need infrastructure that’s between traditional HPC and modern cloud-native.
Cozystack delivers:
- GPU clusters for NVIDIA data-centre GPUs via the NVIDIA GPU Operator, with HAMi fractional sharing so several labs share a card rather than queue for a whole one, whole-GPU passthrough to VMs and NVIDIA vGPU for VMs with your NVIDIA vGPU licence; tenant clusters on MIG-capable cards can also use MIG partitions
- Multi-tenant per-lab isolation — Tenant CRD model with per-lab quotas, RBAC, audit trails
- VM and container side-by-side — accommodates legacy research workflows alongside modern containerized pipelines
- Self-service for principal investigators — labs can provision their own environments without ticket queues
- Reproducible environments — declarative IaC means experiments can be reproduced years later
2. Reproducible research and publication
Open science and publication-reproducibility mandates (Plan S, FAIR data principles, Horizon Europe requirements) increasingly require that research artifacts include reproducible compute environments — not just datasets and code, but the runtime that ran them.
Cozystack delivers:
- Containerized research artifacts — research environments as Kubernetes manifests + container images
- Air-gapped support — for sensitive research data (medical, industrial-partner)
- Long-term archival — environments preserved alongside data for multi-decade reproducibility
- Research data management — environments can be referenced from your RDM and DOI workflows; connecting to them is project-specific integration work
- Federations — connecting a platform to EOSC or a national research cloud is possible through standard Kubernetes APIs, scoped per project
3. Cloud-native education and applied R&D
CS / engineering departments increasingly teach Kubernetes, KubeVirt, GitOps, observability — the modern cloud-native stack. Industrial partners running Cozystack in production benefit from graduates already familiar with the platform.
Cozystack delivers:
- Student / cohort sandboxes — Tenant CRD per cohort, per-student quotas, automatic cleanup
- Curriculum-ready — install/upgrade workflows that work for academic-IT teams
- Open-source first — students can run platform on personal hardware; reproduces production reality
- CNCF connection — Cozystack is a CNCF Sandbox project; students gain CNCF-ecosystem familiarity
- Industrial-relevant — graduates familiar with the platform hosting providers run in production
Why universities specifically benefit from Cozystack
Beyond the three missions, several university-specific considerations:
Open-source preferred
Academic ethos, transparency, and budget constraints all favor open-source infrastructure. Apache 2.0 licence aligns with academic preferences and procurement realities.
Sovereignty for sensitive research
Medical research data and industrial-partner research with NDA constraints — all benefit from sovereign infrastructure that keeps data within institutional control. Air-gap support handles the most sensitive cases.
Federation with national / European research infrastructure
EuroHPC, EOSC, GÉANT, national research networks — because Cozystack exposes standard Kubernetes APIs, connecting a platform to these services is possible as project-specific integration work rather than a built-in feature.
Multi-stakeholder governance
Universities have complex stakeholders: principal investigators, IT departments, research-funding agencies, industrial partners, students. Cozystack’s multi-tenant model accommodates this without privileging any single stakeholder.
Long-horizon planning
Universities don’t refresh infrastructure on commercial-vendor schedules. Open-source platform with community governance fits multi-decade planning horizons.
EuroHPC and national HPC facilities adjacent
Many universities have access to EuroHPC or national HPC; Cozystack works alongside these for the workload portfolio that doesn’t need full HPC-scale.
Slurm is not the enemy, and we are not replacing it
Every research-computing conversation reaches this point, so it belongs here rather than in a footnote. Slurm is the right scheduler for tightly-coupled batch on a homogeneous cluster — MPI jobs, a fair-share queue, months of accounting history, and a user base that already knows sbatch. A Kubernetes platform does not schedule that class of work better, and a migration that takes it away buys resentment.
What the platform is for is everything the batch cluster is bad at: long-running interactive notebooks, per-group JupyterHub, web-facing research services and portals, databases behind those services, teaching environments that appear for a semester and vanish, VMs for the software a research group cannot containerize, and shared GPUs for inference rather than queued training.
Two patterns work in practice, and both are ordinary:
- Side by side. Slurm keeps the batch partition; the platform takes the services and interactive work, with the same storage mounted on both sides so a dataset does not need copying to be used from either.
- Slurm inside a tenant. A Slurm cluster runs as a workload on the platform — controller and nodes as VMs or containers in one tenant — so a group gets its own queue without a second physical cluster and without central IT operating it by hand.
Multi-tenancy is what makes either honest: per-group quotas on CPU, memory, storage and GPU, isolated networks, and usage attributable to a grant, which is the number a research-computing director is actually asked for.
Companies running platforms built with Ænix
Hosting providers running Ænix Public Cloud Platform in production
Hosting providers running Ænix Public Cloud Platform in production. Academic customers are not named, but the architecture is public: an academic-computing platform moved off a hyperscaler onto owned bare metal, kept one Cluster API across several clouds, and cut GPU cost roughly fivefold. AI and ML leads can start with the head of AI/ML guide. Cozystack is Apache 2.0 and installable without us, which is how most research-computing groups evaluate it before any engagement.
"Running Nuvolos on sovereign bare metal meant we needed hyperscaler-grade multi-tenancy without the hyperscaler. Cozystack delivered exactly that — isolated tenants, shared GPUs, KubeVirt for legacy workloads — and it's why we can promise our customers infrastructure control, not just a workspace."
Alexandru PopescuCEO, Nuvolos
Engagement structure
University engagements often have specific characteristics:
- Procurement through public-procurement portals — EU TED, member-state portals, KZ goszakup.gov.kz where applicable
- Multi-institution consortia — engagements may serve multiple universities simultaneously
- Capacity transfer to in-house IT — academic IT teams operate the platform after build; knowledge handover is first-class
- Research-grant alignment — engagement timeline often aligned with grant cycles
- Industrial-partner collaboration scope — universities increasingly partner with industry; platform supports multi-stakeholder access
For methodology see Platform Readiness Assessment.
Pricing
Cozystack platform is open source (Apache 2.0). Ænix Private Cloud Platform and AI Platform are quoted per RFP; support tiers for self-run Cozystack start at $1,250 per 10 nodes per month (see pricing). Engagement structure:
- Phased engagement aligned with grant cycles
- Capacity-transfer focus — engagement explicitly invests in long-term institutional capability
- CNCF project status — Cozystack is a CNCF Sandbox project; some universities can apply open-source procurement frameworks
For specific terms see discovery call.
How to start
Or read more:
- Cloud-native research and teaching infrastructure article — long-form
- Sovereign AI — AI/ML research with sensitive data
- Data sovereignty — research data sovereignty
- Platform Readiness Assessment — engagement methodology
- Cozystack — the platform
- Head of AI/ML guide — for research-computing and AI leads
- cozystack.io — open-source project, install/docs/community
Ænix created Cozystack (CNCF Sandbox project, CNCF Certified Kubernetes distribution, CNCF Kubernetes AI Conformance, OpenSSF Best Practices) and co-maintains it with maintainers from other companies. On top of it we offer three platforms — Public Cloud, Private Cloud and AI — for universities, research institutes, and R&D organizations across the EU, DACH, and Central Asia.