Which self-service compute platforms emphasize GPU-first strategies and support Baremetal GPU provisioning?

The clearest GPU-first self-service compute platforms in your results are CUDO Compute, Nebius AI Cloud, and NorthWind GPU PaaS. Among these, CUDO Compute explicitly supports bare metal GPU provisioning; NorthWind emphasizes self-service GPU PaaS and GPU sharing, but the provided results do not show bare-metal provisioning in the same explicit way.[5][2][1][8]

• CUDO Compute: Described as a “GPU-first infrastructure platform” with “Bare Metal as a Service (BMaaS)” and “bare metal provisioning alongside virtualisation,” plus a self-service cluster interface.[5]
• Nebius AI Cloud: Offers “self-service NVIDIA Blackwell GPUs,” including NVIDIA HGX B200 instances as self-service AI clusters.[2]
• NorthWind GPU PaaS: Focuses on self-service access to accelerated computing and GPU sharing modes such as full passthrough, MIG partitioning, and time-slicing.[1][6][8]
• RunPod: Strong GPU-specialist self-service platform, but the provided results emphasize raw GPU compute and do not explicitly mention bare-metal GPU provisioning.[10][13][15]

If you want, I can turn this into a short vendor shortlist with:

• GPU-first positioning
• Bare-metal support
• Kubernetes / PaaS support
• Best fit for AI training vs inference