The index reads, at every provider in its sample, the on-demand price per GPU-hour of each data-centre GPU it lists, for the smallest instance it prices. Reserved, spot and community-hosted prices are left out: they need a commitment or can be taken away.
Every price is the one on the provider's own pricing page, excluding taxes, on the day in its row. A row older than 90 days is marked stale.
The rows stay in dataset order. The index does not rank providers or name a best one, and ANAROQ is not in it: it rents no GPU.
The index
Filter by GPU, provider or a minimum of GPU memory. The GPU column keeps each provider's own name for the GPU.
20 of 20 rows shown.
| Provider | GPU | GPU memory (GB) | GPUs in the instance | Region | Listed price per GPU-hour | Per GPU-hour (USD) | Per GB of GPU memory an hour (USD) |
|---|---|---|---|---|---|---|---|
| Lambda | NVIDIA H100 SXM | 80 | 1 | United States | $4.29 | $4.29 | $0.0536 |
| Lambda | NVIDIA H100 SXM | 80 | 8 | United States | $3.99 | $3.99 | $0.0499 |
| Lambda | NVIDIA A100 PCIe 40 GB | 40 | 1 | United States | $1.99 | $1.99 | $0.0498 |
| Lambda | NVIDIA A100 SXM 80 GB | 80 | 8 | United States | $2.79 | $2.79 | $0.0349 |
| Lambda | NVIDIA B200 SXM6 | 180 | 1 | United States | $6.99 | $6.99 | $0.0388 |
| DigitalOcean | NVIDIA HGX H100 | 80 | 1 | One price everywhere | $4.41 | $4.41 | $0.0551 |
| DigitalOcean | NVIDIA HGX H200 | 141 | 1 | One price everywhere | $4.47 | $4.47 | $0.0317 |
| DigitalOcean | NVIDIA L40S | 48 | 1 | One price everywhere | $1.57 | $1.57 | $0.0327 |
| DigitalOcean | NVIDIA RTX 4000 Ada Generation | 20 | 1 | One price everywhere | $0.76 | $0.76 | $0.038 |
| Runpod | H100 SXM | 80 | 1 | One price everywhere | $3.99 | $3.99 | $0.0499 |
| Runpod | H100 PCIe | 80 | 1 | One price everywhere | $2.89 | $2.89 | $0.0361 |
| Runpod | A100 SXM | 80 | 1 | One price everywhere | $1.79 | $1.79 | $0.0224 |
| Runpod | L40S | 48 | 1 | One price everywhere | $1.09 | $1.09 | $0.0227 |
| Runpod | L4 | 24 | 1 | One price everywhere | $0.59 | $0.59 | $0.0246 |
| Hyperstack | NVIDIA H200 SXM | 141 | 1 | One price everywhere | $3.99 | $3.99 | $0.0283 |
| Hyperstack | NVIDIA H100 SXM | 80 | 1 | One price everywhere | $3.2 | $3.2 | $0.04 |
| Hyperstack | NVIDIA H100 | 80 | 1 | One price everywhere | $2.5 | $2.5 | $0.0313 |
| Hyperstack | NVIDIA A100 SXM | 80 | 1 | One price everywhere | $1.6 | $1.6 | $0.02 |
| Hyperstack | NVIDIA B200 | 192 | 1 | One price everywhere | $6 | $6 | $0.0313 |
| UpCloud | NVIDIA L4 | 24 | 1 | One price everywhere | €0.58 | $0.65 | $0.027 |
Medians
The median price per GPU-hour of each GPU in each region, over every row. Different GPUs are never pooled: their prices say nothing about each other.
| GPU | Region | Rows | Providers | Per GPU-hour (USD) |
|---|---|---|---|---|
| NVIDIA H100 SXM | United States | 2 | 1 | $4.14 |
| NVIDIA A100 PCIe 40 GB | United States | 1 | 1 | $1.99 |
| NVIDIA A100 80GB SXM | United States | 1 | 1 | $2.79 |
| NVIDIA B200 | United States | 1 | 1 | $6.99 |
| NVIDIA H100 SXM | One price everywhere | 3 | 3 | $3.99 |
| NVIDIA H200 SXM | One price everywhere | 2 | 2 | $4.23 |
| NVIDIA L40S | One price everywhere | 2 | 2 | $1.33 |
| NVIDIA RTX 4000 Ada Generation | One price everywhere | 1 | 1 | $0.76 |
| H100 PCIe | One price everywhere | 1 | 1 | $2.89 |
| NVIDIA A100 80GB SXM | One price everywhere | 2 | 2 | $1.7 |
| NVIDIA L4 | One price everywhere | 2 | 2 | $0.62 |
| NVIDIA H100 | One price everywhere | 1 | 1 | $2.5 |
| NVIDIA B200 | One price everywhere | 1 | 1 | $6 |
Currencies
Each price is listed in the provider's currency, with its source and the day it was read. The other columns convert it with the European Central Bank's euro reference rates, the only rates the site converts with:
- 1 EUR = 1.1188 USD, published for 13 Oct 2026
- 1 EUR = 3.4404 ILS, published for 13 Oct 2026
Download
The whole index as CSV: every row with its source page, read date and stale mark, the listed price and the normalized values in USD.
Download the CSV gpu-price-index.csv
Example
One NVIDIA H100 SXM for an hour costs from $3.2 to $4.41 across 5 on-demand offers from 4 providers in the sample, with a median of $3.99. Running it all month (730 hours) at the median comes to about $2,912.7, before storage and traffic.
How the index is built
Sample: providers whose own pricing page is public (no login, no sales contact) and lists a fixed on-demand price per GPU-hour that can be bought self-serve, from GPUs in data centres they operate. Per provider, each listed data-centre GPU at the smallest instance priced; older rows may name a larger instance, and the GPU count column says which.
Normalization: the price per GPU-hour, and the same divided by the GPU's memory in GB. A GPU that @anaroq/gpu describes is grouped under its name; any other keeps the provider's name.
Left out of this snapshot: Akamai GPU (access on request), and every reserved, spot or community-hosted price.
Refresh: by hand, about once a month, in a reviewed change to the dataset; a row is re-read from its source or marked stale. Nothing is scraped.
Limits
- A sample of a few providers, not the whole market: medians describe the sample.
- The instance around the GPU differs (vCPUs, RAM, disk, interconnect) and is not priced separately; per-hour prices of different instance sizes are not interchangeable.
- On-demand capacity may be unavailable when you need it, and the price says nothing about availability.
- Taxes, storage and traffic are not in the price; converted prices use one day's ECB rate.