Kubernetes the Hard Way was treated as a rite of passage - something a brave soul would do to deeply understand Kubernetes and then never think about again. But what if I told you the "hard way" isn't just an exercise - it's just... the way.

Cloud GPU availability, egress costs, data sovereignty are harsh challenges in today's reality. The answer is a hybrid model - cloud orchestration, on-prem compute.

But how do you build it without falling into proprietary vendor lock-in?

We'll bypass cloud-packaged hybrid solutions and walk through an open-source blueprint. We'll explore how to connect a lightweight k3s control plane in the cloud to on-premises GPU over a secure, encrypted overlay network allowing standard Kubernetes primitives manage local AI inference workloads seamlessly. All built on CNCF projects that underpin planet-scale AI workloads at frontier labs and hyperscalers.