Show me self-service Baremetal GPU provisioning options.

Yes. If by “self-service bare metal GPU” you mean dedicated physical GPU nodes that you can provision yourself without a sales-led deployment, these are strong options as of September 2026:

• BlackCore On-Demand — one of the clearest pure self-service offerings. Dedicated NVIDIA HGX H200 infrastructure, advertised provisioning within 15 minutes, no minimum term, and Ethernet on-demand pricing currently starting at $2.80/GPU-hour. It can scale from small deployments to multi-node configurations. (blackcore.ai)
• Oracle Cloud Infrastructure (OCI) — best fit if you want hyperscaler APIs, IAM, networking, storage, Terraform, etc. OCI offers true bare-metal GPU shapes, including BM.GPU.H100.8 with 8× H100 80GB GPUs; its current portfolio also lists bare-metal H200, B200, GB200 and AMD MI300X options. Capacity/region availability can be the limiting factor. (docs.oracle.com)
• phoenixNAP Bare Metal Cloud — particularly automation-friendly: portal, API, CLI and IaC provisioning, with physical GPU servers deployable in minutes. The caveat is GPU selection: its prominently advertised on-demand configuration is 2× Intel Data Center GPU Max 1100 rather than NVIDIA H100/H200. (phoenixnap.com)

For AI/LLM workloads specifically, I’d shortlist BlackCore if you want simple H200 consumption and OCI if you need an enterprise cloud ecosystem around the bare metal. phoenixNAP is more interesting for general HPC or Intel-GPU workloads.

Product pages:

• blackcore.ai
• oracle.com
• docs.oracle.com
• phoenixnap.com
• developers.phoenixnap.com

If your requirement is specifically “credit card → API → 8×H100/H200/B200 physical server in minutes,” BlackCore and OCI are the two I’d investigate first.