Find self-service platforms for Baremetal GPU provisioning.

If by “platforms” you mean software you can deploy in your own data center to expose GPU servers as self-service bare-metal infrastructure, these are the strongest candidates I found:

• Canonical MAAS — Probably the cleanest starting point for an internal GPU BMaaS. It discovers CPU, memory, NICs, storage, and GPUs/accelerators, then lets users allocate and provision physical machines through a web UI or API. Supports PXE, BMC control, OS imaging, cloud-init, resource pools, and multi-user access. (canonical.com)

canonical.com

• Saturn Cloud Bare Metal Control Plane — Particularly interesting if you're building a commercial GPU cloud rather than just automating servers internally. Saturn positions this specifically as a tenant-facing/self-service layer for GPU-cloud operators, including multi-tenancy and integration with underlying BMaaS/Metal3 infrastructure. (saturncloud.io)

saturncloud.io

• RackN Digital Rebar — Enterprise bare-metal lifecycle automation with BMC discovery, BIOS/firmware/RAID configuration, OS deployment, APIs and reusable workflows. It explicitly supports GPU infrastructure, but is more of an infrastructure automation engine than an out-of-box GPU-cloud storefront; you'd typically integrate your own portal/IAM/billing layer. (rackn.com)

rackn.com

• OpenStack Ironic — Best open-source building block if you want to construct a full private/public BMaaS. Ironic provides API-first allocation, provisioning, cleaning and lifecycle management of physical servers and can operate standalone or behind OpenStack Nova. GPU-equipped machines can be represented through resource classes/traits. More engineering-intensive than MAAS. (docs.openstack.org)

ironicbaremetal.org

For a GPU-cloud business where customers should log in → select something like “8× H100” → provision → receive networking/credentials → later release the machine, I would shortlist Saturn Cloud, MAAS, and Ironic first. Digital Rebar is especially compelling underneath a custom commercial control plane when deep hardware lifecycle automation matters.

If instead you meant vendors where you can self-service rent bare-metal GPUs, I can produce a comparison of those providers (H100/H200/B200/GB200, hourly pricing, API, locations, minimum commitment, and provisioning time).