Ubuntu 26.04 LTS 'Resolute Raccoon' landed today with a clear message: we're the OS for AI workloads. The biggest news? NVIDIA CUDA and AMD ROCm are now in Ubuntu's official repositories.
Developers can install the compute frameworks behind most ML training and inference with a single command. No more wrestling with driver installations or toolkit compatibility. Andrej Zdravkovic, AMD's SVP and Chief Software Officer, says the integration covers everything from data center servers to Ryzen-based laptops.
Canonical is clearly positioning against Red Hat's OpenShift AI stack. Where Red Hat bets on enterprise management, Canonical pushes speed and tighter silicon integration.
The release runs on Linux 7.0 and adds support for Intel Core Ultra Series 3 processors and their integrated NPUs, relevant for edge AI and on-device inference. Confidential computing gets first-class support too, with Intel TDX and AMD SEV-SNP enabling encrypted workloads at the hardware level. Jules Drean, co-founder of Tinfoil, says his team can now deploy a single secure image for private AI workloads across Intel, AMD, and NVIDIA hardware without platform-specific changes.
Livepatch comes to Arm64 servers, meaning critical kernel updates without reboots. Canonical and Arm frame this as important for agentic AI and always-on workloads. They're also replacing C-based system utilities with Rust implementations to cut memory safety vulnerabilities and address the ongoing Linux kernel security surge. Dr. Rebecca Rumbul, Executive Director of the Rust Foundation, said the move raises the security baseline for millions of enterprise users worldwide.