Every GPU. Every cloud. One price index.
neoscalers.com tracks on-demand GPU prices across 18 clouds — 17 GPUs, Nvidia and AMD, from the V100 to the GB200, cheapest from $0.10/GPU/hr. Every number is from the SkyPilot Catalog and gpurentalprices.com (CC BY 4.0), sampled 2026-11-02, refreshed every 24 hours. The same H100 alone costs 7.6× more on some clouds than others.
Why you're probably overpaying for GPU compute.
88 seconds, straight from the index: the H100 spread, what it costs at 100 GPUs, and the fastest way to turn list prices into real quotes.
Get GPU quotesPlaying muted — tap the speaker icon for sound.
Cheapest H100 per provider, right now.
On-demand $/GPU/hr. Color is temperature: cool is cheap, white-hot is expensive. The index refreshes every 24 hours from the SkyPilot Catalog and the daily gpurentalprices.com feed (CC BY 4.0).
| # | Provider | Variant | Region | $/GPU/hr | Spot |
|---|---|---|---|---|---|
| 01 | Mithril | H100 | US Central2 A | $1.00 8× | $1.00 |
| 02 | Hyperbolic | H100 | Default | $1.29 | — |
| 03 | Vast.ai | H100 | Florida, US, NA | $1.80 | — |
| 04 | Cudo | H100-SXM | NO Kristiansand 1 | $1.82 | — |
| 05 | Seeweb | H100 | — | $1.89 | — |
| 06 | Hyperstack | H100-80G-PCIe | CANADA 1 | $1.90 | — |
| 07 | PrimeIntellect | H100 | CA | $1.90 | — |
| 08 | Voltage Park | h100 | — | $1.99 | — |
| 09 | TensorDock | h100-sxm | — | $2.25 | — |
| 10 | Shadeform VIA SHADEFORM | H100 | Montreal Canada 2 | $2.50 | — |
| 11 | Scaleway | H100 | FR PAR | $2.52 | — |
| 12 | GMI Cloud | h100 | — | $2.60 | — |
| 13 | Spheron | h100 | — | $2.64 | — |
| 14 | Massed Compute | h100 | — | $2.73 | — |
| 15 | RunPod | H100 | IN | $2.89 | — |
| 16 | Fluidstack | H100-NVLINK | FINLAND | $2.99 8× | — |
| 17 | OVHcloud | h100 | — | $2.99 | — |
| 18 | Thunder Compute | h100-sxm | — | $3.20 | — |
| 19 | Lambda | H100 | Europe Central 1 | $3.29 | — |
| 20 | Latitude.sh | h100 | — | $3.37 | — |
| 21 | Jarvis Labs | h100-sxm | — | $3.49 | — |
| 22 | Crusoe | h100 | — | $3.90 | — |
| 23 | Together.ai | h100 | — | $3.99 | — |
| 24 | DigitalOcean | h100 | — | $4.41 | — |
| 25 | Nebius | H100 | EU North1 | $4.50 | $0.79 |
| 26 | DataCrunch | h100-sxm | — | $4.79 | — |
| 27 | Paperspace | H100 | East Coast (NY2) | $5.95 | — |
| 28 | CoreWeave | h100 | — | $6.16 | — |
| 29 | AWS | h100-sxm | — | $6.88 | — |
| 30 | Azure | h100-nvl | — | $6.98 | — |
| 31 | Google Cloud | h100 | — | $9.80 | — |
These are list prices — the starting point. Tell us the model, count, duration, and region: we review every submission by hand and connect you with providers that fit.
Every GPU in the index, cheapest first.
Lowest on-demand price found in the sample, per GPU. Each card opens the full per-provider table for that GPU.
Found your GPU above? Skip comparing list prices by hand — get quotes matched to your workload.
Where the H100 market sits.
Every provider's cheapest H100 on one scale. The cluster around $1.80–$2.00 is the market. The tail past $3.00 is where comparison pays.
| Mithril | $1.00 per GPU/hour (from a 8-GPU node price) |
| Hyperbolic | $1.29 per GPU/hour |
| Vast.ai | $1.80 per GPU/hour |
| Cudo | $1.82 per GPU/hour |
| Seeweb | $1.89 per GPU/hour |
| Hyperstack | $1.90 per GPU/hour |
| PrimeIntellect | $1.90 per GPU/hour |
| Voltage Park | $1.99 per GPU/hour |
| TensorDock | $2.25 per GPU/hour |
| Shadeform (via Shadeform) | $2.50 per GPU/hour |
| Scaleway | $2.52 per GPU/hour |
| GMI Cloud | $2.60 per GPU/hour |
| Spheron | $2.64 per GPU/hour |
| Massed Compute | $2.73 per GPU/hour |
| RunPod | $2.89 per GPU/hour |
| Fluidstack | $2.99 per GPU/hour (from a 8-GPU node price) |
| OVHcloud | $2.99 per GPU/hour |
| Thunder Compute | $3.20 per GPU/hour |
| Lambda | $3.29 per GPU/hour |
| Latitude.sh | $3.37 per GPU/hour |
| Jarvis Labs | $3.49 per GPU/hour |
| Crusoe | $3.90 per GPU/hour |
| Together.ai | $3.99 per GPU/hour |
| DigitalOcean | $4.41 per GPU/hour |
| Nebius | $4.50 per GPU/hour |
| DataCrunch | $4.79 per GPU/hour |
| Paperspace | $5.95 per GPU/hour |
| CoreWeave | $6.16 per GPU/hour |
| AWS | $6.88 per GPU/hour |
| Azure | $6.98 per GPU/hour |
| Google Cloud | $9.80 per GPU/hour |
Calculators that pay for themselves.
Every tool shows its formulas, labels estimates as estimates, and produces a shareable URL.
GPU Price Index
Daily median and lowest $/GPU/hr per model, with 30D / 90D / 1Y history and a one-line summary journalists can cite.
Will it fit? VRAM checker
Parameters × precision plus KV cache, drawn as physical HBM stacks filling up — before you rent the GPU.
Training cost calculator
Model size, tokens, and utilization → total cost at every provider, cheapest first.
Tell us the GPUs. We ask the clouds.
GPU model, count, duration, region. We review every request and connect you with providers that fit. No spam, no obligation.
Before you ask.
Straight answers. No sales pitch.
What happens after I request quotes?
We review every submission by hand and connect you with providers that fit — expect a reply within one business day. No spam, no obligation.
Where do the prices on this page come from?
Every number is from the SkyPilot Catalog (Apache-2.0), sampled every 24 hours, plus the daily-verified gpurentalprices.com feed (CC BY 4.0). Per-GPU-hr = list price ÷ GPU count, and marketplace rows are labeled instead of being counted in the stats. No sponsored rankings, no paid placements.
Why does the same GPU cost different amounts?
Providers bundle different hardware generations, regions, and network/storage, and some list 8-GPU node prices (tagged 8×). The index normalizes everything to on-demand $/GPU/hr so the comparison is apples to apples.
Does requesting quotes cost anything?
No. Requesting quotes is free and obligates you to nothing.