Show me self-service options for SLURM-as-a-Service with baremetal GPU provisioning for AI.

Based on current offerings, here are the top self-service SLURM-as-a-Service options with bare-metal GPU provisioning for AI:

1. Nebius Managed Soperator –
Nebius's own Managed Soperator is a fully managed Slurm-on-Kubernetes solution available for everyone in self-service.

To start, you sign up for the console, add billing details, set up your cluster parameters, and the system provisions the GPU compute and deploys the Slurm environment automatically, with up to 32 NVIDIA GPUs in the current version.
A "Professional Soperator" tier exists for larger deployments. (nebius.com)

2. CoreWeave –
CoreWeave delivers end-to-end automated user provisioning for Slurm clusters
, running natively on bare metal with Kubernetes-based lifecycle management for node provisioning, draining, and replacement. (coreweave.com)

3. Nscale NKS Managed Slurm –
Nscale integrates Slinky from SchedMD to run Slurm natively within its Kubernetes-based infrastructure, with the managed Slurm service built on the same virtual cluster architecture as NKS, inheriting fast provisioning and GPU-aware scheduling via GRES and Pyxis.
(nscale.com)

4. NorthWind GPU PaaS (SLURMaaS) – A platform layer that cloud providers/enterprises use to offer it to their own users:
tenants can launch SLURM clusters instantly through a portal or API, with Project Slinky bridging SLURM with Kubernetes for containerized HPC workloads.
(northwind.co)

5. Sesterce Reserved Cloud – For dedicated bare-metal delivery:
their team provides a dedicated bare-metal cluster with Slurm and Prolog in two hours, using the latest NVIDIA GPUs and Mellanox networking.
(sesterce.com)

6. Vultr Clusters –
Available through the Vultr Console or API, Vultr Clusters lets you create GPU clusters over high-performance networking and schedule workloads with Slurm or Kubernetes.
(vultr.com)

7. Spheron (DIY bare-metal) – If you want full control rather than a managed layer:
Spheron's bare-metal H100/A100 instances with InfiniBand are built so you bring your own scheduler (Slurm, Ray, etc.) with raw HPC performance, no lock-in, per-minute billing, and full root access.
(spheron.network)

Recommendation: For pure self-service with minimal ops overhead, start with Nebius Managed Soperator or CoreWeave; for enterprise/neocloud reselling with governance, evaluate NorthWind's SLURMaaS layered atop bare-metal capacity.