Before asking how to get more GPUs, we should ask why we're leaving so much of the GPUs we already have unused.

Kubernetes has traditionally treated GPUs as whole, indivisible resources, even when a workload needs only a fraction of one. In this lightning talk, we'll look at why GPU utilization is also a sustainability problem, how Kubernetes is evolving to manage accelerators differently, and why the greenest GPU might be the one we never have to add.

CNCF Cloud Native Sustainability Month 2026 - Virtual Meetup Romania