South African Tech billionaire Mark Shuttleworth says Ubuntu 26.04 is built for the AI agentic era

Mark Shuttleworth has a habit of making bold declarations about where technology is going and then spending years quietly building toward them. At Computex 2026 in Taipei this week, the South African billionaire and founder of Canonical made his latest one: Ubuntu 26.04 LTS, released in April, is the operating system for the AI agentic era.

"The agentic revolution begins with software engineering but will touch almost every discipline," Shuttleworth said at the conference. "Sandboxing is the critical foundational layer of agentic workflows."

The statement was not just philosophical positioning. It came attached to a concrete product announcement. On June 1, Canonical revealed a new collaboration with NVIDIA to integrate the NVIDIA OpenShell runtime for agents directly into the Ubuntu ecosystem, packaging it as a snap so enterprises can deploy next-generation agentic workflows across local devices, hybrid environments and private clouds with rapid, reliable updates and built-in confinement.

"OpenShell provides a securely-designed sandbox for agent work of any sort," Shuttleworth said. "Snaps deliver rapid and reliable updates as well as enterprise control of this critical new infrastructure."

The NVIDIA partnership is the most visible piece of a broader AI strategy Canonical has been assembling across multiple fronts. Ubuntu 26.04 LTS, codenamed Resolute Raccoon, shipped on April 23 with day-one support for NVIDIA's Vera Rubin NVL72 rack-scale architecture, NVIDIA CUDA integration, support for NVIDIA Jetson Thor, and real-time kernel patching through Kernel Livepatch, which allows organisations to address critical security vulnerabilities without taking systems offline. As agentic AI workloads move into production environments and uptime becomes non-negotiable, the ability to patch live systems without rebooting is no longer a luxury but a necessity.

The release also added full support for RVA23, the baseline standard for RISC-V processors, and delivered both guest and host support for confidential computing on Intel Trust Domain Extensions and AMD SEV, enabling AI workloads to run with silicon-level encryption in shared cloud environments.

The AI roadmap Canonical's engineering team has laid out goes beyond a single release. Jon Seager, Canonical's vice president of engineering, has outlined a phased approach to integrating AI across the Ubuntu ecosystem through 2026 and beyond. The strategy divides AI capabilities into two categories. Implicit features run in the background using on-device models to improve existing functionality, the most immediate example being high-quality speech-to-text and text-to-speech tools that improve accessibility without requiring any user interaction. Explicit features are new, optional AI-powered workflows including document composition tools, automated file management agents and tools that can troubleshoot a broken connection autonomously.

Starting with Ubuntu 26.10, the next interim release, these explicit AI features will be made available as a strictly opt-in preview. None of them will be switched on by default. The company has been emphatic on this point: AI features delivered as snaps can be removed individually at any time, and users who skip them during the initial setup wizard will not have them installed at all.

The local-first philosophy running through the entire initiative reflects a deliberate policy decision. Canonical's position is that inference happens on your machine by default. Your logs and your data stay local. If you want to use a cloud-based model, you supply your own API token. The company is building out inference snaps that include optimised and quantised versions of open-weight models including Qwen and DeepSeek, models whose licence terms Seager has said must be compatible with Canonical's open-source values before they make the cut.

Seager has also been careful to temper the AI enthusiasm with operational pragmatism. "Using AI for its own sake is not a constructive goal for anything but increasing exposure, and it rarely yields good results in production code," he wrote in a community post. His stated aim is for Ubuntu to expose the primitives needed for agents to operate within existing boundaries, including read-only analysis modes, tightly scoped permissions and full auditability of decisions and outcomes.

That audit trail capability is being delivered through Ubuntu's existing system architecture rather than bolt-on tools. Strict access controls and system logging track every decision an AI agent makes, which matters considerably in enterprise environments where regulators increasingly require organisations to explain automated decisions.

Shuttleworth built Canonical in 2004 after selling his first company, the internet security firm Thawte, to VeriSign for approximately $575 million in 1999. He used a portion of that windfall to buy a seat on a Russian Soyuz spacecraft in 2002, becoming the first African in space at a personal cost of $20 million. He then funded Ubuntu's development out of his own pocket for years before Canonical reached profitability. The company reported revenue of $250 million in 2023. His net worth is estimated at approximately $500 million to $1 billion depending on the valuation methodology applied to his Canonical stake.

Ubuntu is now used by Google, Amazon, Netflix and thousands of other enterprises to run their server infrastructure. It is the dominant operating system in cloud computing environments. Shuttleworth's claim that version 26.04 is built for the agentic era is not a marketing line without substance behind it. The NVIDIA OpenShell partnership, the confidential computing support, the Kernel Livepatch capability and the inference snap infrastructure are all real, shipped components.

Whether the agentic era arrives on the timeline Shuttleworth envisions is a separate question. What is clear is that Canonical has spent years laying the groundwork for a moment when enterprises need a secure, auditable, locally-inferenced operating system to run AI agents at scale. Ubuntu 26.04 is Shuttleworth's argument that the moment is now.