OpenNebula 7.4 will bring the first integration with NVIDIA Infra Controller (NICo), bringing bare-metal AI infrastructure into OpenNebula’s native provisioning and lifecycle management workflow. This initial integration demonstrates how operators can deploy and manage physical AI servers through the same interface and operational model already used for virtual machines.

Through two native drivers, OpenNebula incorporates NVIDIA NICo-managed bare-metal resources into its cloud management model. Physical GPU servers can therefore be represented, scheduled, provisioned, monitored, and managed using OpenNebula templates, lifecycle operations, scheduling policies, and user interfaces.

First Native Integration with NVIDIA NICo

NVIDIA NICo provides site-local, zero-trust bare-metal lifecycle management, automating hardware discovery, firmware validation, DPU provisioning, network isolation, and tenant sanitization for AI infrastructure.

To integrate these capabilities into its cloud management platform, OpenNebula introduces two native drivers. The Information Manager monitors NICo resources, while the Virtual Machine Manager provisions and controls bare-metal instances through the NICo API. Together, these drivers abstract the underlying NICo operations, allowing physical servers to be managed through the same OpenNebula data model used for virtual machines.

Mapping Bare-Metal Resources into OpenNebula

Once the required NVIDIA NICo resources have been mapped and configured, users can provision a bare-metal instance directly from OpenNebula.

OpenNebula submits the deployment request through the NVIDIA NICo API, follows the provisioning process, and exposes NICo-specific status and resource information through its standard interface.

After provisioning is complete, the physical server becomes available with the hardware configuration defined by the selected Instance Type. In the environment shown in the demonstration, the provisioned server provides direct access to four NVIDIA GPUs.

The server can then be operated through the standard OpenNebula lifecycle, giving infrastructure teams a consistent way to manage both virtualized and bare-metal AI resources.

Provisioning Bare-Metal AI Infrastructure

Once the required resources have been configured, users can provision bare-metal instances using the same self-service workflow employed for virtual machines. OpenNebula coordinates the deployment through NVIDIA NICo and monitors the provisioning process, exposing NICo-specific information through its standard interface.

After provisioning is complete, the physical server becomes available with the hardware configuration defined by the selected Instance Type. In the demonstrated environment, the deployed system provides direct access to four NVIDIA GPUs. The bare-metal server is then fully operational and managed through the standard OpenNebula workflow.

End-to-End Demonstration

The screencast walks through the complete workflow using an NVIDIA GB200 NVL72 system, from mapping NVIDIA NICo resources to OpenNebula objects and provisioning a bare-metal instance to validating the deployed GPU infrastructure.

Unified Bare-Metal and Virtual Infrastructure Management

The integration of NVIDIA NICo is an important step toward OpenNebula’s vision of a unified control plane for AI Factories. Rather than managing virtual machines, Kubernetes clusters, HPC environments, and bare-metal GPU systems through separate infrastructure silos, operators can bring them together under a common provisioning, scheduling, governance, and lifecycle management framework.

Discover more ways OpenNebula simplifies AI infrastructure management. Explore our AI Factory resources and watch additional screencasts covering advanced deployment and orchestration capabilities.