H100 Cost Calculator
Indicative monthly and annual cost for an NVIDIA H100 SXM5 cluster across 10 GPU clouds. Adjust GPU count, hours, and term.
An NVIDIA H100 SXM5 in 2026 costs roughly $1.75/hr on Together AI at the low end and $4.10/hr on AWS p5 at the high end on-demand. Reserved 1-year terms cut 25–50% off list. A 64-GPU H100 cluster at 730 hours/month ranges from about $82K (Together AI) to $192K (AWS), with neoclouds like CoreWeave, Crusoe, Lambda, and Nebius clustering in the $90K–$120K band.
| Provider | $/GPU-hr | Monthly cost | Annual cost | Notes |
|---|---|---|---|---|
| ★ BestTogether AI | $1.75 | $81,760 | $981,120 | Open-model pricing |
| Voltage Park | $1.89 | $88,301 | $1,059,610 | Best reserved 1y/3y |
| Nebius | $2.00 | $93,440 | $1,121,280 | — |
| Crusoe | $2.30 | $107,456 | $1,289,472 | Vertical-integrated power |
| CoreWeave | $2.39 | $111,661 | $1,339,930 | Largest IB fleet |
| Lambda | $2.49 | $116,333 | $1,395,994 | 1-Click Clusters |
| RunPod | $2.79 | $130,349 | $1,564,186 | Per-second billing |
| Oracle (BM.GPU.H100.8) | $3.10 | $144,832 | $1,737,984 | — |
| Azure (ND H100 v5) | $3.67 | $171,462 | $2,057,549 | — |
| AWS (p5.48xlarge) | $4.10 | $191,552 | $2,298,624 | EFAv3 networking |
Pricing reflects public on-demand and 1-year reserved rates published by each provider as of Q1 2026. Excludes storage, egress, and support. Use as an indicative benchmark, not a quote.
Frequently asked questions
- What does an H100 cost in 2026?
- Indicative public on-demand pricing ranges from $1.75/hr (Together AI) to $4.10/hr (AWS p5). Reserved 1-year terms typically cut 25–50% off list. Neoclouds lead on price; hyperscalers charge a premium for integration.
- How much does a 64-GPU H100 cluster cost per month?
- At 730 hours/mo on-demand: ~$82K on Together AI, ~$112K on CoreWeave, ~$192K on AWS. On 1-year reserved: ~$52K to ~$96K. Adjust the calculator above for your exact configuration.
- Is reserved always cheaper than on-demand?
- Yes for utilization ≥40% over the term. For bursty workloads on-demand or spot can be cheaper. Neoclouds typically offer ~25–40% off for 1-year reserved; hyperscalers offer 45–55% off.
- What's not included in $/H100-hr?
- Egress bandwidth, persistent storage (typically $0.05–$0.10/GB-mo), object storage, and managed software. Add ~5–15% to raw GPU cost for a fully provisioned cluster.
Company → Datacenter → GPU → Customer → Industry
Every entity on this site is cross-linked. Follow the graph from operators down to specific facilities, GPU clusters, customers, and sectoral context.