Linus Torvalds is done babying “museum” technology in Linux. Speaking at the Open Source Summit in Mumbai, the Linux creator talked through the practical pain and power of AI in the kernel, and why Linux no longer supports older technology he described as “museum” material.

That matters because the kernel is not just another software component. It is the control plane underneath cloud servers, phones, and embedded devices, and it is where performance, security, and compatibility get decided. When Torvalds says Linux will not support certain legacy approaches anymore, it is a signal to everyone shipping on Linux that time, once spent, may be getting cut from under their feet. Meanwhile, his comments about AI in the kernel add a second signal: the kernel is becoming a more active participant in AI workloads, not a passive bystander. In other words, the same ecosystem that has spent years optimizing for “classic” workloads is now being asked to support a different kind of computation.

If you are running a platform team, the headline question becomes simple: what does “pruning” mean for your roadmap? Kernel maintainers historically have balanced stability with progress, but the subtext of Torvalds’s remarks is that the balance is changing. AI workloads bring different throughput patterns, different latency sensitivities, and a renewed focus on accelerating specific operations. Even if your business is not an AI company, you still feel the pressure when customers adopt AI stacks and expect the underlying Linux systems to keep up. That is why “museum” tech is more than an amusing phrase. It represents the parts of the kernel ecosystem that no longer justify their ongoing maintenance cost and engineering risk.

There is also a governance angle worth watching. Linux development is open source, but it is not democratic in the way many non-technical leaders imagine. Key maintainers and subsystem owners effectively set a direction through what they accept, what they deprecate, and what they remove. Torvalds’s intervention, even when framed conversationally, is a reminder that in mature open source, the bottleneck is not funding. It is maintainers’ time, and the need to keep the codebase coherent. In that world, “museum” technology is often what remains when too many edge cases linger too long.

Now add the AI layer. Torvalds discussed “the pain and power of AI in the kernel,” and that phrase lands because it captures the tension many engineering orgs are living through. AI is powerful because it can accelerate decision-making and improve performance on workloads that matter. But it is also painful because it tends to stress assumptions built for earlier software eras. The kernel is where these assumptions are enforced, from drivers to scheduling to memory behaviors. If the kernel has to adapt to AI traffic patterns and new acceleration paths, then legacy components that do not fit the new reality become liabilities.

For regulators and risk teams, the second-order implication is that kernel changes can ripple outward into compliance narratives. Even without naming specific regulatory regimes, the direction is clear: as AI gets closer to the operating system, auditability, reproducibility, and security posture depend more heavily on kernel behavior. If your organization uses Linux as a baseline for security certifications or internal control frameworks, you need to treat kernel evolution as an input into your risk model, not just an engineering detail.

For executives and boards, the strategic stakes are straightforward. If Linux is trimming legacy functionality and re-centering around AI-driven performance needs, then your vendors and internal teams have to move faster too. The decision is not only whether you will upgrade Linux. It is whether you can keep your product, compliance posture, and hardware/software dependencies aligned while the ground rules shift under the kernel layer.