At the Linux Plumbers Conference 2026 in Prague, NVIDIA's John Hubbard and Red Hat's Danilo Krummrich jointly presented the state of Nova, the open-source NVIDIA driver written in Rust that aims to enter the upstream Linux kernel. The joint presentation is not a minor detail: it signals a concrete trajectory toward the stated goal of becoming the official NVIDIA driver on Linux.

For those managing GPU servers in on-premise or air-gapped environments, this carries a different weight than a routine driver update. An open-source, upstream driver allows distributions to integrate NVIDIA support without depending on external proprietary modules, reducing signing, update, and compatibility issues with newer kernels. The use of Rust, a language gaining ground in the Linux kernel for memory safety, adds a reduction in attack surface: a relevant factor when inference workloads process sensitive data that must remain under local control.

Red Hat's involvement in the presentation is not coincidental. Enterprise distributions need supportable and reviewable components to offer Linux systems with NVIDIA GPUs without chunks of proprietary code that are hard to govern. If Nova matures, incentives shift: NVIDIA can move part of its engineering from the closed binary to a shared path, while on-premise platform vendors can build more transparent and certifiable images. It is not an automatic win for anyone: performance and feature coverage will need to reach parity with the proprietary driver, and timing remains an open variable.

There is also a structural signal. The Linux kernel is accepting Rust code in critical areas, and a GPU driver like Nova is one of the most ambitious proving grounds. For those focused on data sovereignty and component auditing, an open-source driver reduces the need for blind trust in the vendor and enables independent inspection. It does not eliminate hardware dependency, but it shifts control of the base software toward the community. For organizations evaluating on-premise GPU deployments for LLMs and inference, the driver is an often-overlooked foundation. AI-RADAR offers analytical frameworks at /llm-onpremise to compare these trade-offs.