Distributed AI training workloads impose complex scheduling requirements that Kubernetes’s built-in scheduler can’t meet. Kubernetes schedules pods individually and independently, but distributed training introduces two requirements that break this model: Pods must land on hardware with…