OpenNebula's AI Infrastructure Capabilities
This session explores how OpenNebula's cloud orchestration platform supports the computational demands of modern AI workloads, particularly Large Language Models. João Pita Costa, Senior Technologist in AI at OpenNebula, introduces the platform's Enhanced Platform Awareness (EPA) features, which enable fine-grained matching of processor capabilities to VMs and Kubernetes workflows. The platform supports GPU passthrough for direct hardware access, NVIDIA vGPU for resource distribution across multiple VMs, and PCI device management for specialized hardware allocation. OpenNebula's architecture allows organizations to deploy AI processing clusters with optimized scheduling, placing execution closer to data sources to reduce transfer times and bandwidth usage. The platform facilitates a 'Cloud for AI' paradigm where users can provision LLM virtual servers with pre-trained models as a service.
Production AI Platforms: Iguane Solutions Case Study
Jean-Philippe Foures, VP of Products at Iguane Solutions, details how his company leverages OpenNebula to build AI platforms for customers who cannot use public cloud services. Their architecture spans multiple layers: infrastructure (GPU servers with NVIDIA H100/H200, StorePool storage, VXLAN networking), cloud orchestration (OpenNebula with GPU passthrough and CPU pinning), LLM core services (model servers, API proxies, observability stack), and applications (private copilots, web search, no-code development tools). Iguane Solutions has used OpenNebula in production for over six years, managing platforms that provide OpenAI-compatible APIs while ensuring data privacy and intellectual property protection. Their internal use cases include a private GitHub Copilot alternative that scans codebases for context-aware suggestions, automated documentation maintenance, and AI-generated monthly customer reports.
Multi-Site AI Testbed: AI Sweden Implementation
Kim Henriksson, SVP for Technology, Innovation and Ecosystems at AI Sweden, describes how the Swedish National Center for Applied AI uses OpenNebula to manage on-premises infrastructure across three physical locations. The organization serves as a neutral nonprofit facilitating collaboration between private sector, public sector, and academia, providing hands-on access to AI infrastructure for experimentation and learning. AI Sweden's testbed handles heterogeneous hardware from multiple vendors, various GPU types (NVIDIA, AMD, Intel), and diverse edge devices, requiring extreme flexibility in orchestration. The platform supports multiple concurrent projects with different frameworks and workloads, using OpenNebula as an abstraction layer between hardware and software. Key capabilities that enable this complexity include easy host addition/removal, template management, SSH contextualization, live migration, VM resizing, and GPU passthrough. The organization operates with administrators from both AI Sweden and partner organizations, emphasizing the importance of intuitive configuration and management.
Strategic Considerations and Future Directions
Both speakers emphasize data sovereignty as a critical driver for private AI infrastructure in Europe, with organizations seeking to maintain control over their data and models rather than relying on public cloud providers. Iguane Solutions is working with OpenNebula to develop native integrations including AI-specific appliances, NVIDIA K8s device plugin equivalents, and unified metrics dashboards. The company positions its offering as a plug-and-play private AI stack with infrastructure management, analytics, and user-facing features included. AI Sweden highlights OpenNebula's price competitiveness, noting they alternate between the open-source community edition and enterprise version depending on requirements. The session concludes with discussion of key metrics for AI platforms: model usage patterns, query success rates, and system resource utilization for capacity planning.
