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

If by “self-service” you mean users can provision GPU capacity/clusters themselves rather than filing tickets, these are the strongest SLURM + bare-metal-oriented options I found:

• Nebius AI Cloud — Probably the closest turnkey public-cloud experience. Its self-service console provides NVIDIA Hopper/Blackwell GPU clusters on demand, with managed Slurm available alongside Kubernetes. API, CLI, Terraform and SDK access are supported. Verify with Nebius whether your desired configuration is physically bare metal versus GPU VMs, because its self-service page describes both cluster services and VMs. (nebius.com)

nebius.com

• Verda Instant Clusters — Self-service multi-node GPU clusters advertised as provisionable in about 20 minutes, with managed Slurm or Kubernetes, NVIDIA drivers, storage and monitoring. Verda specifically advertises “bare-metal-level performance”; I'd confirm whether that means true tenant-dedicated bare metal for the SKU you select. (verda.click)

verda.click

• Comet Compute — Particularly interesting when dedicated hardware is mandatory: single-tenant NVIDIA GPU clusters can be delivered as managed Slurm, managed Kubernetes, or straight bare metal/root access. This looks more purpose-built for dedicated AI clusters than conventional VM-based GPU clouds. (cometcompute.com)

cometcompute.com

• NorthWind SLURM-as-a-Service — Best fit if you’re building the self-service service yourself—for an enterprise/private GPU cloud or neocloud. NorthWind provides per-tenant self-service Slurm cluster provisioning and lifecycle management and explicitly supports bare metal, VMs, Kubernetes and Slurm environments. (northwind.co)

northwind.co

For an enterprise wanting “portal/API → provision physical H100/H200/B200 nodes → create Slurm cluster → users submit AI jobs,” I’d shortlist Comet Compute for a consumed service and NorthWind for building your own internal/private SLURMaaS. Nebius is the stronger candidate when cloud-like elasticity matters more than guaranteeing every worker is raw bare metal.