As Artificial Intelligence moves to the forefront of digital transformation, organizations are facing increasing pressure to process data closer to its source—either because of privacy concerns, performance, or regulation. This is where decentralized AI comes in, enabling AI workloads to run directly at the edge or across multiple clouds, without having to transfer sensitive data to a single location.
With OpenNebula, deploying and orchestrating decentralized AI and distributed AI applications becomes not only possible, but also quite straightforward. For instance, you can build a Federated Learning cluster with E-Group’s FedX sovereign AI platform running across a real-world multi-cloud environment—such as the Virt8ra testbed, developed as part of the €3B IPCEI-CIS initiative.
Federated Learning with FedX on OpenNebula
FedX is a federated learning appliance developed by E-Group. It is designed for privacy-first AI collaboration and it enables multiple institutions to train AI models without ever transferring their data off-site. For example, with FedX you can build a federated data network across healthcare providers, allowing each of them to store anonymized data within their own isolated edge environments. This setup allows organizations to leverage its own sensitive data more effectively by sharing insights and analysis for AI in a secure way—without compromising data privacy.
Key features of OpenNebula to support federated learning include:
- Pre-packaged orchestration. Comes with Kubernetes via OneKE—OpenNebula’s native CNCF-certified Kubernetes distribution. It comes with routing capabilities based on Traefik, distributed storage through Longhorn, and observability with Prometheus and Grafana.
- Flexible network configuration. Supports public and VXLAN interfaces for secure multi-site communication.
- Federated client scalability. Spin up edge nodes in multiple regions—each with local data and full autonomy.
- Privacy-preserving training. Only model updates are exchanged—raw data never leaves the local environment.
Virt8ra: A Real-World Decentralized Testbed
To showcase the power of Federated Learning, we have deployed E-Group’s FedX across the Virt8ra testbed, a pan-European infrastructure built on OpenNebula. This testbed spans research institutions and cloud providers in different countries, each contributing compute and data while maintaining full sovereignty. The screencast showcases how the storage and networking infrastructure services provided by OpenNebula are used to run the geo-distributed federated learning application.
Watch the Full Demo: Deploying Federated AI with FedX
We’ve prepared a screencast that walks you through the entire process—from launching the appliance to configuring federated clients and monitoring training progress in real time. Each node runs localized training on datasets like Spanish blog posts, German legal texts, or Polish newsletters, all while contributing to a single, shared AI model.
In this screencast, you’ll see how to:
- Deploy the FedX appliance
- Configure multi-site networking with VXLAN and BGP EVPN using FRR
- Scale federated clients and assign provisioning roles
- Launch a distributed training job, monitor results and resource usage
- Fine-tune models across multilingual datasets
- Use JupyterLab inside the cluster for experimentation.
It’s ideal for teams interested in decentralized, privacy-preserving AI workflows—especially in multi-jurisdiction environments where data can’t be centralized.
What’s Next?
This demo is just one step toward building sovereign, decentralized AI and federated AI infrastructure. As always, OpenNebula continues to focus on open, flexible, and secure cloud infrastructure to support the next generation of distributed AI systems.
To learn more, click here.
0 Comments