Canonical has announced official Ubuntu support for the NVIDIA Rubin platform, NVIDIA’s next-generation GPU and platform architecture for AI and high-performance computing, including the rack-scale NVIDIA Vera Rubin NVL72 systems. The announcement was made at CES 2026.
The move positions Ubuntu as a supported operating system across NVIDIA’s next-generation AI platform, unifying the NVIDIA Vera CPU, Rubin GPU, and BlueField-4 DPU into a single, production-ready execution environment.
According to NVIDIA, the collaboration is designed to reduce friction for organizations deploying Rubin-based systems. Enterprises will be able to run the full NVIDIA platform stack on Ubuntu, from development through production, while maintaining enterprise-grade security and operational consistency across cloud, on-premises, and edge environments.
A central part of the announcement is first-class Arm support for Vera Rubin systems. Canonical is integrating upstream features such as Nested Virtualization and MPAM, allowing memory bandwidth and cache resources to be controlled at the hardware level.
Canonical is also extending Arm-native support across key infrastructure and data platforms. OpenStack Sunbeam and Apache Spark are being adapted for Arm-first deployments, allowing cloud operators and data engineers to run complete AI and analytics pipelines directly on Vera Rubin hardware without architectural compromises.
Moreover, to simplify AI inference deployments, Canonical introduced inference snaps, a packaging approach that eliminates dependency conflicts when running large language models. In collaboration with NVIDIA, Canonical plans to distribute the NVIDIA Nemotron-3 family of open models using inference snaps, starting with Nano variants.
For those not in the know, these snaps provide an immutable, containerized runtime that includes all required libraries and optimizations for NVIDIA platforms, enabling consistent deployment across Ubuntu systems with a single install command.
Networking, security, and storage performance are addressed through renewed support for the NVIDIA BlueField-4 DPU. With 64 NVIDIA Grace CPU cores and up to 800 Gb/s throughput, BlueField-4 offloads networking, storage, and security processing from the host CPU.
Finally, Canonical is also working with storage partners to optimize Ubuntu for GPUDirect Storage on BlueField-4. This allows high-speed data movement directly between NVMe storage and Rubin GPU memory, reducing latency and avoiding traditional I/O bottlenecks.
The approach targets both on-premises AI factories and sovereign cloud deployments where performance isolation and data locality are critical.
For more information, see Canonical’s announcement.
