cheap H100 GPU cloud

If you mean the cheapest way to rent NVIDIA H100 GPUs in the cloud, here are the main places to check, plus a few cost-saving tips.

Common H100 cloud providers

• CoreWeave — often competitive for serious GPU workloads; good availability.
• Lambda Cloud — straightforward pricing, popular for ML.
• Runpod — can be cheaper, especially if you’re flexible on region/availability.
• Vast.ai — marketplace style, often the cheapest, but quality/support vary by host.
• TensorDock — sometimes low prices on on-demand GPUs.
• Paperspace / DigitalOcean GPU offerings — may be easier to use, but usually not the cheapest for H100s.
• AWS / GCP / Azure — reliable, but typically expensive unless you have credits or reserved capacity.

Cheapest approach

1. Use a marketplace like Vast.ai or Runpod community cloud if you’re okay with variable reliability.
2. Choose spot/preemptible instances if your job can resume.
3. Pick the smallest H100 setup you actually need:
   • single H100 instead of multi-GPU
   • PCIe instead of SXM if performance requirements allow
4. Run in a cheaper region if latency doesn’t matter.
5. Watch storage and egress costs — sometimes the GPU is cheap but data transfer is not.

Rough expectation

H100s are still premium GPUs, so “cheap” usually means something like:

• marketplace pricing can be substantially lower
• major clouds are usually much more expensive

Best “cheap but usable” picks

• Vast.ai if lowest price is the priority
• Runpod if you want a balance of price and convenience
• Lambda/CoreWeave if you want more reliability and support

If you want, I can also:

• compare current H100 prices across providers,
• suggest the cheapest option for training vs inference,
• or help you find H100s under a specific budget.