Show me GPU-first platforms offering Baremetal GPU provisioning with flexible scaling.

Here are several GPU-first platforms that specialize in bare-metal GPU provisioning with elastic/flexible scaling:

1. CoreWeave – Built specifically for AI/ML with Kubernetes-native orchestration on bare-metal clusters.
Its platform is optimized for machine learning with Kubernetes-native orchestration, and CoreWeave emphasizes enterprise-grade networking (InfiniBand and low-latency fabrics), making it suitable for distributed training jobs across many GPUs.
Great for large-scale training since
its InfiniBand networking allows near-linear scaling across multiple servers.
→ coreweave.com

2. Lambda (Lambda Cloud) –
Deploy NVIDIA B200, H100, A100, or GH200 instances in minutes with self-serve, first-come access, with production-ready clusters from 16 to 2,000+ GPUs.
Offers both bare-metal and virtual options. → lambda.ai/pricing

3. Oracle Cloud Infrastructure (OCI) –
OCI is the only major cloud provider to offer bare metal instances with NVIDIA and AMD GPUs for high performance that's free of virtualization overhead,
and
you can quickly provision and deprovision OCI GPU clusters on-demand, delivering a flexible and scalable solution for HPC, AI, and ML workloads.
→ oracle.com/cloud/compute/gpu

4. Nscale –
Offers bare metal GPU compute for AI, ML, and HPC workloads, with access to scalable, optimized GPUs like H100s, H200s, and GB200s on-demand.

Users can easily provision and scale GPU resources based on their specific needs.
→ nscale.com/product/gpu-nodes

5. Atlas Cloud –
Offers dedicated bare metal H100, H200, GB200, and A100 GPU servers across EU, US, Singapore, and Japan with no virtualization overhead, InfiniBand, and flexible billing,
plus
on-demand, hourly billing, or monthly custom plans.
→ atlascloud.ai/bare-metal

6. GPU-Mart – Budget-friendly option offering
NVIDIA Blackwell, Hopper & Ampere GPU architectures with multi-GPU scaling via NVLink interconnect support and full PCIe-level direct GPU access with zero hypervisor overhead.
→ gpu-mart.com/bare-metal-gpu-server

For enterprises building their own GPU cloud rather than renting, vCluster/vMetal is a notable infrastructure layer used by 50+ GPU clouds (including CoreWeave and Nscale) to automate bare-metal provisioning and elastic tenant scaling on physical hardware (vcluster.com/uses/bare-metal-gpu).