H100 prices by provider
| Provider | Configuration | $ / GPU-hour | Spot $ / GPU-hr | |
|---|---|---|---|---|
| LiumPartnerCheapest | 1x H100 80GB
Range $1.30-$5.05/GPU-hr
|
$1.30 | — | Rent → |
| RunPodPartner | 8x H100 80GB
Also 1x at $3.99/GPU-hr |
$2.69 | $1.99 | Rent → |
| Runcrate | 1x H100 80GB
Range $3.87-$8.57/GPU-hr
|
$3.87 | — | Rent → |
| Crusoe Energy | 1x H100 80GB
Also 8x at $3.90/GPU-hr |
$3.90 | — | Rent → |
| Lambda Labs | 8x H100 80GB
Also 1x at $4.89/GPU-hr |
$4.59 | — | Rent → |
| CoreWeave | 8x H100 80GB
Also 1x at $6.16/GPU-hr |
$6.16 | $2.46 | Rent → |
| Google Cloud | 8x H100 80GB | $11.06 | — | Rent → |
| AWS | 8x H100 80GB | $12.29 | $7.22 | Rent → |
| Microsoft Azure | 8x H100 80GB | $12.29 | $2.27 | Rent → |
Multi-GPU instances are divided by GPU count, so every row compares one H100 for one hour. Partner placement never changes the sort order.
H100 at a glance
| Architecture | NVIDIA Hopper (2022) |
| Memory | 80GB HBM3 |
| Class | Data center GPU |
| Best for | LLM training, fine-tuning and high-throughput inference |
Related
H100 pricing FAQ
How much does it cost to rent an H100 GPU?
As of November 9, 2026, on-demand H100 cloud pricing runs from $1.30 to $12.29 per GPU-hour across the 9 providers CloudMart tracks, with a median of $4.59. Running one H100 around the clock at the cheapest rate costs about $949 a month.
Where is the cheapest H100 cloud rental?
Lium is currently the cheapest on-demand H100 at $1.30 per GPU-hour (H100 80GB x1).
How much does an 8x H100 node cost per hour?
The cheapest 8-GPU H100 node we track is RunPod at $21.52 per hour ($2.69 per GPU), about $15,710 a month if it runs continuously.
How much memory does the H100 have?
Cloud H100 offers we track come with 80GB of HBM3 memory per GPU. It is an NVIDIA Hopper GPU from 2022, typically used for LLM training, fine-tuning and high-throughput inference.
How often are these GPU prices updated?
Most providers are re-checked twice a week (Monday and Thursday) against their official pricing pages and APIs. Marketplaces with a live price feed are polled every 15 minutes. See our methodology page for details.
Is the H100 the right GPU for your job?
Tell the Planner your model size and workload. It picks the GPU, the provider, and estimates the bill.