From an 11-Student Seminar to a 145-Student Lecture

CASE STUDY ·  UNIVERSITY OF ZURICH · 2025–2026

How educators at the University of Zurich scaled teaching computational subjects across 6 courses with a governed workspace spanning bare-metal and cloud compute

University of Zurich

Language models · RAG · Speech tech · Eye tracking & NLP · Accessibility · Advanced AI

Nuvolos

Governed Workspace

Infrastructure

Sovereign EU bare metal + elastic cloud capacity

Managed standby GPUs

Guaranteed capacity at scheduled times

DEPLOYMENT OVERVIEW

During the 2025–2026 deployment, UZH courses ran on a combination of the Nuvolos sovereign EU bare-metal fabric, dedicated EU-hosted compute and storage, and Nuvolos-managed standby GPU capacity (currently hosted on Microsoft Azure). Nuvolos provided the governed workspace across these resources, giving teaching teams consistent, reproducible environments while compute could be matched to the requirements of each course.

For additional capacity, Nuvolos-managed standby accelerators provide guaranteed capacity at scheduled times. The same governed workspace can extend to SWITCH Cloud Compute, or to commercial hyperscalers in an institution’s own tenancy via the OCRE (Open Clouds for Research Environments) framework.

The result was not one successful deployment. It was repeated use across substantially different teaching contexts, using the same underlying governed workspace for environments, compute and resource control.

WHAT CHANGED FOR TEACHING

One course environment, not 145 individual setups

The teaching team defines the course environment once: libraries, models, configuration and applications. Students enter the same browser-based environment, so support shifts from troubleshooting individual machines to maintaining one reference setup.

Compute tailored to each course

Courses can move between lighter compute, T4, A10-class and H100 resources without rebuilding the teaching environment. Instructors can design exercises around the computational task rather than around the hardware available on student devices.

Controlled and predictable resource use, as cohorts grow

Per-student credit quotas, Resource Pool wallets and automatic shutdown of inactive sessions allow GPU-backed teaching to remain governed at cohort level. Intensive workloads can be enabled without leaving resource use open-ended.

Reproducibility without platform dependence

Course environments are captured as Snapshots, immutable, versioned records of code, dependencies and configuration together, and can be exported as Docker images for full portability. The environment remains portable instead of becoming inseparable from the infrastructure on which the course ran.

FROM PILOT TO WIDER ADOPTION

Planning a computational course or lab?

Students work with the same research-grade tools and compute used for research. Just share your curriculum, compute requirements, and cohort size. We’ll prepare a ready-to-use course environment for your class.

Evaluating Nuvolos for Research IT?

Learn how Nuvolos integrates with institutional infrastructure, customer cloud tenancies and elastic compute while preserving governance and portability. See the Technical Overview below.

TECHNICAL DEPLOYMENT OVERVIEW

One governed workspace across institution-controlled and elastic compute

The University of Zurich deployment demonstrates the separation between the governed workspace and the underlying infrastructure. During 2025–2026, teaching workloads ran on the Nuvolos sovereign EU bare-metal fabric and Nuvolos-managed standby GPU capacity (currently Azure-hosted). Nuvolos kept the user workspace and course environments consistent across those resources.

DEPLOYMENT ARCHITECTURE

ONE WORKSPACE, FLEXIBLE INFRASTRUCTURE

Nuvolos separates the computational workspace from the underlying infrastructure. Institutions retain infrastructure choice and control over where workloads run, while course environments remain reproducible and portable across infrastructure.

ELASTIC ACCELERATORS AND SUPPLEMENTARY CAPACITY

Nuvolos-managed standby accelerators provide guaranteed GPU capacity at scheduled times, for example for weekly course blocks, a level of capacity assurance an individual tenancy cannot match. Beyond the sovereign core, workloads can burst to SWITCH Cloud Compute, the OpenStack-managed fabric of the Swiss national research and education network, or to Azure, AWS, GCP and Alibaba Cloud in an institution’s own tenancy via the OCRE framework. Institutional agreements include a pooled Burst Allowance covering orchestration of CPU workloads on bring-your-own capacity. These options integrate without changing the Nuvolos workspace model.

RESOURCE GOVERNANCE

Compute can be allocated through course- and user-level controls. Per-student credit quotas, Resource Pool wallets, fractional GPU sizes and automatic shutdown of inactive sessions provide mechanisms for governing expensive accelerators at cohort scale. GPU slicing, up to eight isolated workloads per physical device with a provider-independent control plane.

REPRODUCIBILITY AND VERSIONING

Course environments are captured as Snapshots, immutable records of the full software stack used for teaching, and can be exported as Docker images. Data egress from the Nuvolos sovereign core is free of charge. For academics, this supports reproducibility. For IT, it provides portability and an explicit exit path.

HOW NUVOLOS WORKS WITH EXISTING INFRASTRUCTURE

Nuvolos is a governed workspace that can complement existing institutional compute rather than requiring every course to create a separate infrastructure service. The same workspace model can span institution-controlled resources, customer cloud tenancies and managed standby capacity as requirements change.

PLANNING YOUR NEXT COURSE?

Get in touch contact@nuvolos.com ↗ or schedule a demo