A GPU cloud pricing calculator works out what an AI training or inference job costs on rented GPUs. Renting an H100 costs about $2.25 to $12.29 per GPU-hour on demand; enter your GPU, count and hours to rank 13 providers by the total cost of your job.
Pick a GPU, how many, for how long, and how you pay. Every provider that lists that GPU is ranked by the total cost of your job.
8x NVIDIA H100 for 24 hours: 13 providers
Cheapest: TensorDock
$432
$18.00 per hour, $2.25 per GPU-hour
Most expensive listed
$2,360
Microsoft Azure, 5.5x the cheapest
One 8-GPU node for a month (730 h)
$13,140
at TensorDock's on-demand rate
| Provider | Total | Per GPU-hour | Per hour (8 GPUs) | Details |
|---|---|---|---|---|
| CheapestTensorDockSXM5 80GB | $432 | $2.25 | $18.00 | Marketplace. Starting price Pricing page for TensorDock (opens in a new tab) |
| Vast.aiSXM 80GB | $513 | $2.67 | $21.36 | Marketplace. Median of live listings; changes by the minute Pricing page for Vast.ai (opens in a new tab) |
| RunPod Community CloudSXM 80GB | $516 | $2.69 | $21.52 | Global, third-party hosts. Starting price Pricing page for RunPod Community Cloud (opens in a new tab) |
| JarvislabsSXM 80GB | $670 | $3.49 | $27.92 | India, Europe, US Pricing page for Jarvislabs (opens in a new tab) |
| CrusoeHGX 80GB | $749 | $3.90 | $31.20 | US. Rises to $4.50 on Dec 1, 2026 Pricing page for Crusoe (opens in a new tab) |
| Together AIHGX 80GB | $766 | $3.99 | $31.92 | US. Sold as GPU clusters; check the minimum size Pricing page for Together AI (opens in a new tab) |
| RunPod Secure CloudSXM 80GB | $843 | $4.39 | $35.12 | Global Pricing page for RunPod Secure Cloud (opens in a new tab) |
| DigitalOceanHGX 80GB | $847 | $4.41 | $35.28 | Select regions. 1-GPU or 8-GPU Droplets Pricing page for DigitalOcean (opens in a new tab) |
| NebiusHGX 80GB | $864 | $4.50 | $36.00 | All regions. Spot is a starting price and changes with demand Pricing page for Nebius (opens in a new tab) |
| LambdaSXM 80GB | $939 | $4.89 | $39.12 | US. 1-GPU rate, $4.59 on 8-GPU instances; list price from Nov 15, 2026 Pricing page for Lambda (opens in a new tab) |
| CoreWeaveHGX 80GB | $1,183 | $6.16 | $49.28 | North America. $49.24 per 8-GPU instance-hour Pricing page for CoreWeave (opens in a new tab) |
| AWSp5.4xlarge, SXM 80GB | $1,321 | $6.88 | $55.04 | US East (N. Virginia). Capacity Block rate is for p5.48xlarge (8 GPUs) Pricing page for AWS (opens in a new tab) |
| Microsoft AzureND H100 v5, SXM 80GB | $2,360 | $12.29 | $98.32 | East US. $98.32 per 8-GPU VM-hour Pricing page for Microsoft Azure (opens in a new tab) |
Public list prices in USD, checked November 14, 2026. Excludes storage, networking, egress and tax. Spot prices change by the hour and can be interrupted; reserved prices need a commitment. Confirm on the provider's page before you buy.
Choose B200, H200, H100, A100, L40S or RTX 4090. The GPU memory is shown on each card.
Enter how many GPUs and how long you need them, in hours, days, weeks or months.
Compare on-demand, spot or preemptible, and reserved or committed prices.
Providers are ranked by the total cost. Copy a link or the ranked list as text.
The formula is price per GPU-hour x GPUs billed x hours. GPUs billed is your count rounded up to the provider's smallest node, and a month is 730 hours.
On-demand H100 rental costs about $2.25 to $12.29 per GPU per hour across the 13 providers in this calculator (checked November 14, 2026). GPU marketplaces and specialist GPU clouds are usually cheapest; AWS and Azure list the highest on-demand rates.
On-demand H200s cost about $3.59 to $10.60 per GPU-hour and B200s about $5.98 to $14.24 in this table. Newer GPUs cost more per hour but can finish a job faster, so compare the total cost of the job, not only the hourly rate.
On-demand is pay-as-you-go with no commitment. Spot or preemptible GPUs are much cheaper but the provider can stop them at short notice, so they suit jobs that checkpoint. Reserved or committed pricing gives a lower stable rate in exchange for paying for a fixed term, from a few weeks to a year or more. Marketplaces such as Vast.ai and RunPod Community Cloud rent GPUs on third-party hosts on demand.
Some providers, including AWS, rent data-center GPUs only as whole 8-GPU instances (for example p5.48xlarge with 8 H100s). The calculator rounds your GPU count up to whole nodes for those providers and marks the row "billed as 8 GPUs".
Hourly GPU prices usually exclude persistent storage, data egress, some networking and taxes, and spot prices change during the day. Budget separately for storage and data transfer, and confirm current prices on the provider's page before you commit.
Prices are public list prices checked on November 14, 2026, linked to each provider's pricing page. AI Wins is not affiliated with any provider listed and earns nothing from them.
Work out how much GPU memory your model needs before you rent.
Open toolFind the best model for the GPU or Mac you already own.
Open toolCompare the cost of calling hosted model APIs instead.
Open toolEstimate the energy and emissions of your AI usage.
Open toolGPU prices fall as new chips ship. Get the AI news that matters, including compute deals, in the free daily AI Wins digest.
Get the free digestOnly positive AI stories, delivered daily or weekly. Free, no spam, unsubscribe anytime.