Show me self-service Baremetal GPU provisioning options.

Here are solid self-service bare metal GPU provisioning options, ranging from hyperscaler-grade to budget-friendly neoclouds:

1. Lambda Bare Metal Instances –
Bare Metal Instances, coupled with self-service APIs covering the instance lifecycle, VPCs, storage volumes, and resource governance, enable Lambda compute to integrate with a customer's existing control planes and orchestration across a global fleet
. Good if you need programmatic, hardware-level control at scale. (lambda.ai)

2. Voltage Park / Lightning AI –
lowest published H100 bare-metal price among providers evaluated at $1.99/hr with no contracts; 3,200 Gbps Quantum-2 InfiniBand fabric scaling to 4,064 GPUs per dedicated cluster; self-serve provisioning in 15 minutes
. Great for large-scale distributed training. (hostline.io)

3. Latitude.sh –
Deploy and manage fully automated bare metal servers globally in under 5 seconds, with dedicated GPU clusters instantly, pre-configured with ML tools
. Recently added
a new bare-metal instance powered by the NVIDIA RTX Pro 6000 Server GPU, available to deploy right now
. (latitude.sh)

4. Oracle Cloud Infrastructure (OCI) –
offers NVIDIA A100, H100, and AMD MI300X GPUs on bare metal instances with RDMA and InfiniBand support for high-speed training
. Best for enterprise-scale HPC. (northflank.com)

5. Vultr Bare Metal / Clusters –
delivers on-demand, self-service Vultr Bare Metal server clusters, provisioned through Vultr's simple console or API, providing scalable and customizable CPU and GPU compute solutions
. (vultr.com)

6. Hostrunway – smaller neocloud where
most configurations provision in under 5 minutes from API call to SSH-ready, using automated IPMI-based imaging with pre-cached OS images, GPU driver packs, and CUDA libraries
. Also offers
a fully-featured REST API and an official Terraform provider for infrastructure-as-code deployments
. (hostrunway.com)

7. TensorDock – budget option where
TensorDock provides access to NVIDIA A100, H100, RTX 6000, and older-generation GPUs like 3090, all available on a self-serve, hourly basis
. (northflank.com)

Choosing tip: For large multi-node training with InfiniBand, go Lambda, Voltage Park, or OCI; for quick single-node dev/test, Vultr, Hostrunway, or TensorDock offer faster, cheaper self-service deploys.