LXD 6.4 and the Orchestration Divide

The November 2026 release of LXD 6.4 does not change the fact that Kubernetes dominates production environments. According to the CNCF 2025 survey, 82% of container users run Kubernetes in production, while Docker Swarm holds a niche share that is not separately tracked. This gap widens because major hyperscalers like AWS, Google, and Microsoft provide managed Kubernetes services such as EKS, GKE, and AKS, but they do not offer managed Swarm. Teams choosing between these platforms must weigh the steep learning curve of Kubernetes against the simplicity of Swarm, especially when considering the operational overhead of managing control planes versus using single-command initialization.

I find the cost structure decisive for small teams. A three-node cluster on Amazon EKS costs roughly $73 per month for the control plane alone, plus node fees like $30 per month for a t3.medium instance. In contrast, running Swarm on DIY virtual machines incurs zero orchestration markup; you pay only for the underlying VMs. This means a team with five developers can deploy a working stack in minutes using docker swarm init, whereas Kubernetes requires weeks of competency for new engineers. The trade-off is clear: Swarm offers negligible overhead and shallow setup, but it lacks native auto-scaling, service mesh support, and dynamic storage provisioning. If your application has fewer than ten nodes and no complex networking needs, Swarm remains the pragmatic choice. However, if you require GPU scheduling, Istio integration, or "scale-to-zero" capabilities introduced in Kubernetes 1.37, the ecosystem tools are simply unavailable in Swarm.

The decision often hinges on whether you need enterprise-grade features like RBAC, network policies, and audit logging. Kubernetes provides these natively through its controller-managed reconciliation loops, which ensure self-healing at the pod level. Swarm relies on simpler service-level restarts and built-in overlay networks without a native policy engine. For teams already using Docker Compose, the transition to Swarm via docker stack deploy is minimal, preserving workflow continuity. Yet, the job market favors Kubernetes skills; hiring engineers proficient in CRDs and operators is easier than finding those experienced with Swarm’s limited extensibility. I would skip Swarm if your roadmap includes AI/ML workloads, as Kubernetes supports NVIDIA NIM and KubeFlow out of the box, while Swarm does not support GPU scheduling.

Consider the staffing implications carefully. Kubernetes demands engineers who understand CNI plugins like Calico or Cilium, and the complexity of managing ingress controllers and gateway APIs. Swarm operates within the comfort zone of general backend engineers who know Docker basics. If your organization cannot dedicate specialized DevOps staff, the operational burden of Kubernetes may outweigh its feature advantages. Conversely, if you are building for production at scale with complex microservices, the lack of dynamic provisioning in Swarm becomes a bottleneck quickly. The 82% adoption rate reflects this reality: most organizations eventually outgrow simple orchestration.

One question remains unanswered: how long will Docker Inc. maintain Swarm’s viability given the absence of managed cloud offerings? The vendor support extends to 2030, but without hyperscaler backing, the ecosystem growth stagnates. For now, the split is functional rather than ideological. Small teams prioritize speed and low cost, choosing Swarm for its "five-minute" deployment time. Larger teams prioritize scalability and tooling depth, accepting the higher initial investment in Kubernetes. The data shows that while Swarm is still maintained and functional, it is no longer a contender for large-scale infrastructure. It serves as a bridge for simple applications, not a foundation for complex systems.

In my assessment, the choice is rarely about technical superiority but about organizational capacity. If you have under five developers and a simple app, Swarm is clearly the better fit. If you need managed services, GPU access, or advanced networking, Kubernetes is the only viable path. Do not let the marketing of "simplicity" obscure the fact that Swarm lacks the foundational components required for modern cloud-native architectures. The 2026 landscape confirms that Kubernetes won the orchestration war, and LXD 6.4 does not alter that outcome. It merely reinforces the bifurcation: use Swarm for quick, local experiments, and use Kubernetes for anything that matters in production.