Provider index · hourly snapshot
GPU cloud providers compared: the cheapest you can actually rent
23 clouds rent GPUs by the hour. They carry different cards at different rates, and the cheapest price on the list is often sold out: on 10 of 12 popular cards it has no stock behind it. Below is the cheapest cloud you can actually rent each card from, then every provider side by side.
LIVE Prices and stock checked 1 minute ago
Cheapest with stock
Cheapest GPU cloud for each card
Cheapest provider reporting stock, per GPU hour on demand. Where a lower price exists with no stock behind it, it is shown struck through.
| GPU | Per GPU/hr | Grab |
|---|---|---|
| B300 Lium | $8.00 | Grab |
| B200 Lium | $5.60 | Grab |
| H200 Runpod | $3.59 | Grab |
| H100 Lium | $1.39 | Grab |
| RTX PRO 6000 Lium | $1.19 | Grab |
| RTX 5090 Vast.ai | $0.59 | Grab |
| A100 Runpod | $1.19 | Grab |
| L40S Vast.ai | $0.67 | Grab |
| RTX A6000 Lium | $0.42 | Grab |
| RTX 6000 Ada Vast.ai | $0.64 | Grab |
| A100 40GB Vast.ai | $0.35 | Grab |
| L40 Runpod | $0.69 | Grab |
fig — one row per popular card. The struck-through price is a listing nobody can fulfil.
The league table
Every GPU cloud provider
| Provider | GPU models | In stock | From |
|---|---|---|---|
| Runpod | 38 | 23 | $0.12 Nvidia RTX A2000 6GB |
| Vast.ai | 27 | 27 | $0.050 Nvidia GTX 1070 |
| Lium | 20 | 9 | $0.060 Nvidia RTX 3060 |
| Verda | 11 | 4 | $0.39 Nvidia V100 16GB |
| Massed Compute | 10 | 7 | $0.35 Nvidia A30 |
| Lambda Labs | 9 | 5 | $0.69 Nvidia Quadro RTX 6000 |
| Excess Supply | 8 | 2 | $0.40 Nvidia RTX 4090 |
| Paperspace | 8 | 8 | $0.80 Nvidia A4000 |
| QuantaCloud | 7 | 7 | $0.48 Nvidia RTX A6000 |
| BoostRun | 6 | 0 | $0.49 Nvidia RTX 4090 |
| Hyperstack | 6 | 4 | $0.15 Nvidia A4000 |
| DigitalOcean | 5 | 0 | $0.79 Nvidia RTX 4000 Ada |
| Crusoe | 4 | 2 | $1.10 Nvidia A40 |
| Denvr Dataworks | 4 | 3 | $1.10 Intel Gaudi2 |
| Latitude.sh | 4 | 0 | $0.74 Nvidia L40S |
| Nebius | 4 | 3 | $3.87 Nvidia H100 |
| Scaleway | 4 | 2 | $0.95 Nvidia L4 |
| IMWT | 3 | 0 | $0.49 Nvidia RTX A6000 |
| Vultr | 3 | 1 | $0.51 Nvidia A16 |
| Avian | 2 | 2 | On request |
| Horizon Compute | 2 | 0 | $1.95 Nvidia H100 |
| Phyntec | 2 | 0 | $0.40 Nvidia RTX 4090 |
| Voltage Park | 1 | 0 | $1.99 Nvidia H100 |
fig — ranked by catalogue size. In stock = models the provider confirmed available at the last check.
Field guide
Which kind of GPU cloud fits your job
Marketplaces such as Vast.ai and Lium rent machines that independent hosts own. They are usually the cheapest per GPU hour, and stock changes by the hour. Use them for experiments, fine-tuning and jobs that can restart.
GPU-first clouds such as Lambda Labs, Runpod, Crusoe and Nebius run their own data centres and sell GPU virtual machines by the hour. Prices sit in the middle, and they are the usual choice for multi-GPU training.
General clouds such as DigitalOcean, Vultr and Scaleway add GPUs to a full cloud with storage, networks and managed databases. You pay more per GPU hour and get everything else in one account.
AWS, Google Cloud and Azure are not in this index: the public APIs we collect from do not carry their GPU prices.
Questions
Frequently asked questions
Which is the cheapest GPU cloud?
It depends on the card, and it changes every hour. At the last check: Nvidia B300 is cheapest at Lium, $8.00 per GPU hour; Nvidia B200 is cheapest at Lium, $5.60 per GPU hour; Nvidia H200 is cheapest at Runpod, $3.59 per GPU hour. The table above has every popular card.
How many GPU cloud providers are there?
We track 23 that publish hourly GPU prices, from marketplaces like Vast.ai and Lium to clouds like Lambda, Runpod and Hyperstack. Each one carries different cards at different rates.
Why is the cheapest listed price often not the one I pay?
Because the cheapest listing is often sold out. On 10 of these 12 popular cards the lowest price on the market has no stock behind it, so a price-only comparison sends you to a card you cannot rent.
How is GPU cloud pricing compared here?
Per GPU per hour, on-demand, from each provider's public API, read every hour. Spot prices are left out of the headline because a preemptible hour is not the same product.
Residency
By country
- GPU clouds in United States of America 13 providers