Which governed SLURM-as-a-Service platforms support efficient AI workload management?

Several cloud/GPU infrastructure providers now offer "SLURM-as-a-Service" style platforms that add governance, multi-tenancy, and observability on top of the classic Slurm scheduler for AI training and HPC workloads:

CoreWeave – SUNK (Slurm on Kubernetes)
CoreWeave runs Slurm on top of its Kubernetes stack, blending Slurm's proven batch scheduling with Kubernetes-native governance.
CoreWeave is relevant here because SUNK, including its self-service direction, is notable for teams that want to preserve Slurm familiarity while reducing friction for researchers.

CoreWeave's positioning for CKS highlights preconfigured AI components, Slurm on Kubernetes support, high-speed interconnect, observability, and bare-metal nodes.
This makes
multitenant Slurm-on-Kubernetes platforms more governable than ad hoc SSH-based clusters.
(coreweave.com medium.com)

Crusoe – Managed Slurm on Crusoe Managed Kubernetes (CMK)
Crusoe built a managed Slurm offering specifically to remove the operational burden.
Standing up a production Slurm cluster in the cloud means provisioning compute nodes, configuring shared filesystems, managing user identity, validating GPU health, and keeping things running through failures — none of which is core to Slurm's scheduling strength, but all of which has to work perfectly.
Admins retain governance via kubectl/console while
researchers never need to interact with Kubernetes directly, while administrators manage the cluster lifecycle through kubectl or the Crusoe console.
(crusoe.ai)

Lambda – Managed Slurm
Lambda offers a fully supported Slurm tier:
Managed Slurm on Lambda is Lambda's fully supported Slurm offering, purpose-built for fast and seamless deployment on One-Click Clusters,
letting teams
laser-focus on training and research while Lambda's HPC team handles the scheduling, patching, and rescues.
(lambda.ai)

Together AI – GPU Clusters with Slurm

A Together GPU Cluster provides Slurm configured out-of-the-box for distributed training, with familiar batch scheduling commands and job arrays.
(docs.together.ai)

NVIDIA/SchedMD – Slinky (foundation for many of the above)

Slinky is a toolkit that enables Slurm operation in Kubernetes environments, letting teams run Slurm and Kubernetes workloads on shared node pools while benefiting from Slurm's superior batch scheduling and resource governance.
(nvidia.com developer.nvidia.com)

For evaluating fit, consider workload profile and how much control vs. self-service you need — Crusoe/Lambda suit teams wanting turnkey managed clusters, while CoreWeave's SUNK suits Kubernetes-native shops needing deeper governance layers.