Cloud GPU Rental Pricing
Hourly, on-demand, per GPU. USD figures kept as published; EUR given only where the provider quotes EUR directly.
| Provider | GPU | Price/hour |
|---|---|---|
| OVHcloud | H100 80 GB | €2.80 ($2.99) |
| OVHcloud | A100 80 GB | $3.07 |
| OVHcloud | L40S 48 GB | $1.80 |
| RunPod (Secure Cloud) | L40S 48 GB | $0.99 |
| RunPod (Community Cloud) | L40S 48 GB | $0.79 |
| Lambda | H100 80 GB SXM | $3.99-4.29 depending on node size |
| Lambda | A6000 48 GB (previous-generation workstation GPU) | $1.09 |
| GMI Cloud | H100 80 GB | $2.00 |
| GMI Cloud | H200 | $2.60 |
Hetzner (GEX131, dedicated bare-metal, not elastic hourly cloud) | RTX PRO 6000 Blackwell Max-Q 96 GB | €1.4247 (€889/month flat, excl. VAT) |
| Decentralized (Vast.ai-class marketplace) | H100 80 GB | $2.50-3.89 |
| AWS | H100 80 GB (no direct AWS tariff found; market range per AltStreet) | $8.00-12.29, mid estimate $10 |
| AWS | H200, p5en.48xlarge, on-demand (8-GPU node, per GPU) | $7.91 |
| AWS | H200, p5en.48xlarge, spot (per GPU) | $3.37 |
AWS does not sell a single-GPU H200 instance: the smallest P5en node is already 8 GPUs at $63.30/hour total. For a "one GPU as a remote workstation" use case, OVHcloud, Lambda, GMI Cloud, or RunPod fit the shape of the need better than AWS. GMI Cloud in particular undercuts OVHcloud on H100 ($2.00/h vs OVH's $2.99/h). Hetzner's GEX131 is a dedicated server, not a per-second elastic cloud instance: it bills a flat monthly rate regardless of how many hours you actually use that month, which is why it doesn't appear in the hourly-prorated table below. Sources: OVHcloud price list, Cloud Mercato H100-380, Lambda pricing, RunPod L40S, Vantage EC2 P5en, GMI Cloud pricing, Hetzner GEX131 press release.