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

The clearest matches are CUDO Compute and NorthWind, with Nebius also showing a strong GPU-first posture. CUDO explicitly describes itself as a GPU-first infrastructure platform and says it offers bare metal as a service (BMaaS) with self-service cluster provisioning[4]. NorthWind emphasizes GPU PaaS / GPUaaS with self-service GPU consumption, including full passthrough, MIG partitioning, and time-slicing, though its published materials focus more on orchestration than baremetal provisioning[6][11].

• CUDO Compute: GPU-first infrastructure, on-demand VMs, dedicated bare metal, and self-service clusters; explicitly mentions Bare Metal as a Service[4].
• NorthWind: self-service GPU PaaS with enterprise controls; supports full passthrough GPUs, NVIDIA MIG, and time-slicing for tenants[6][11].
• Nebius AI Cloud: publicly available self-service NVIDIA Blackwell GPU clusters, indicating a GPU-first product strategy[1].
• GMI Cloud: provides instant GPU instances and also advertises bare metal servers with cloud integration, but the result is less explicit about baremetal GPU provisioning as a platform feature[15].

If you want, I can turn this into a short comparison table of “GPU-first” vs. “baremetal GPU provisioning” across providers.