H100 rental prices per GPU-hour
Hopper, 80 GB HBM. Forms: SXM (HGX boards, NVLink fabric), PCIe, NVL (paired PCIe cards, 94 GB each on some hosts). Collected 2026-11-10 from the vendors' own pages. 38 price points, 21 providers.
| Provider | Secure | Community | Market | Serverless | Spot | Reserved |
|---|---|---|---|---|---|---|
| RunPod | $3.99 | $2.69 | · | · | · | · |
| Vast.ai | · | · | from $2.13 | · | · | · |
| Lambda | $4.59▲ | · | · | · | · | · |
| Nebius | $4.50 | · | · | · | from $0.79 | · |
| Hyperstack | $3.20 | · | · | · | · | from $2.72 |
| Verda (ex-DataCrunch) | $4.99 | · | · | · | $2.49 | · |
| Jarvis Labs | $3.49 | · | · | · | $1.54 | · |
| Together AI | $3.99 | · | · | · | $1.99 | from $3.69 |
| DigitalOcean GPU Droplets | $4.41 | · | · | · | · | from $3.26 |
| CoreWeave | $6.16 | · | · | · | · | · |
| Voltage Park | from $1.99 | · | · | · | · | · |
| Modal | · | · | · | $3.95 | · | · |
| Crusoe | $3.90 | · | · | · | · | · |
| Vultr | from $2.30 | · | · | · | · | · |
| AWS | $6.88 | · | · | · | · | · |
| Google Cloud | $11.06 | · | · | · | · | · |
| Microsoft Azure | $12.29 | · | · | · | · | · |
| Oracle Cloud (OCI) | $10.00 | · | · | · | · | · |
cooler = cheaper warmer = pricier (scaled within this map, reserved excluded). Purple dashed = "from" price, committed term or announced change. Not every provider has every tier or form.
| Provider | Secure | Community | Market | Spot |
|---|---|---|---|---|
| RunPod | $2.89 | $1.99 | · | · |
| Vast.ai | · | · | from $2.47 | · |
| Hyperstack | $2.50 | · | · | $2.00 |
| Thunder Compute | $3.20 | · | · | · |
cooler = cheaper warmer = pricier (scaled within this map, reserved excluded). Purple dashed = "from" price, committed term or announced change. Not every provider has every tier or form.
All H100 offers
Rows ordered by price within each tier. The form column is mandatory: compare only matching forms.
| Tier | Provider | Form | $ / GPU-hr | Notes | Source |
|---|---|---|---|---|---|
| Secure | Latitude.sh | form not stated | $1.68 | bare-metal server g3.h100.small with 1 x H100 80GB | price page |
| Secure | Voltage Park | SXM (HGX-class) | from $1.99 | FAQ: "starting at $1.99 an hour without a contract" | price page |
| Secure | Vultr | SXM (HGX-class) | from $2.30 | HGX H100, starting from; 8-GPU node | price page |
| Secure | Hyperstack | PCIe | $2.50 | price page | |
| Secure | Hyperstack | NVL / NVLink-bridged | $2.60 | listed as "NVIDIA H100 NVLink" | price page |
| Secure | RunPod | PCIe | $2.89 | price page | |
| Secure | RunPod | NVL / NVLink-bridged | $3.19 | price page | |
| Secure | Hyperstack | SXM (HGX-class) | $3.20 | price page | |
| Secure | Thunder Compute | PCIe | $3.20 | H100 PCIe, per-minute billing | price page |
| Secure | Jarvis Labs | SXM (HGX-class) | $3.49 | price page | |
| Secure | Crusoe | SXM (HGX-class) | $3.90 | HGX H100; new price $4.50 from Dec 1, 2026 | price page |
| Secure | RunPod | SXM (HGX-class) | $3.99 | price page | |
| Secure | Together AI | SXM (HGX-class) | $3.99 | GPU Clusters, on-demand | price page |
| Secure | DigitalOcean GPU Droplets | SXM (HGX-class) | $4.41 | price page | |
| Secure | Nebius | SXM (HGX-class) | $4.50 | price page | |
| Secure | Lambda | SXM (HGX-class) | $4.59 | 8x HGX H100 instance; announced price effective 2026-11-16 | price page |
| Secure | Verda (ex-DataCrunch) | SXM (HGX-class) | $4.987 | price page | |
| Secure | CoreWeave | SXM (HGX-class) | $6.16 | per GPU-hour from the 8-GPU node price ($49.24) | price page |
| Secure | AWS | SXM (HGX-class) | $6.88 | p5.48xlarge/8, us-east-1 | price page |
| Secure | Oracle Cloud (OCI) | SXM (HGX-class) | $10.00 | BM.GPU.H100.8, flat per-GPU rate | price page |
| Secure | Google Cloud | SXM (HGX-class) | $11.06 | a3-highgpu-8g/8, us-central1 | price page |
| Secure | Microsoft Azure | SXM (HGX-class) | $12.29 | ND96isr H100 v5/8, East US | price page |
| Community | RunPod | PCIe | $1.99 | price page | |
| Community | RunPod | NVL / NVLink-bridged | $2.59 | price page | |
| Community | RunPod | SXM (HGX-class) | $2.69 | price page | |
| Market | Vast.ai | SXM (HGX-class) | from $2.13 median $2.67 | price page | |
| Market | Vast.ai | NVL / NVLink-bridged | from $2.33 median $2.54 | price page | |
| Market | Vast.ai | PCIe | from $2.47 median $2.47 | price page | |
| Serverless | Modal | SXM (HGX-class) | $3.949 | $0.001097 per second × 3,600 | price page |
| Serverless | Replicate | form not stated | $5.49 | gpu-h100 hardware, $0.001525 per second | price page |
| Spot | Nebius | SXM (HGX-class) | from $0.79 | preemptible, "from" | price page |
| Spot | Jarvis Labs | SXM (HGX-class) | $1.54 | price page | |
| Spot | Together AI | SXM (HGX-class) | $1.99 | preemptible compute | price page |
| Spot | Hyperstack | PCIe | $2.00 | price page | |
| Spot | Verda (ex-DataCrunch) | SXM (HGX-class) | $2.494 | price page | |
| Reserved | Hyperstack | SXM (HGX-class) | from $2.72 | reservation "starting from"; term not stated on page | price page |
| Reserved | DigitalOcean GPU Droplets | SXM (HGX-class) | from $3.26 | 12-month reserved price | price page |
| Reserved | Together AI | SXM (HGX-class) | from $3.69 | 3–30 day reservation column | price page |
Questions
How much does it cost to rent an H100?
On 2026-11-10: secure on-demand SXM-class prices ranged $1.99 to $12.29 per GPU-hour across 16 providers; community and marketplace prices were lower and spot lower still. Form and tier matter, see the table.
What is the difference between SXM and PCIe H100?
SXM cards sit on HGX boards with a fast GPU-to-GPU fabric and are usually priced higher; PCIe cards are standard add-in cards. We never rank them against each other without a label.
Is spot cheaper than on-demand?
Per hour, yes, but spot can be interrupted. The estimator adds your own rerun rate so you can compare the whole job.
Sponsored row: none yet. Advertise on this page (waitlist). Sponsored rows would sit above the table and never change the order. Total a job.