GPU price comparison: the cheapest cloud GPU rental
Cheapest on-demand H100: $1.80/hr at HyperAI. Spot from $1.42 (Microsoft Azure). Marketplace hosts from $2.50 (Lium, Helsinki, Finland). The typical provider charges $3.83 an hour (median of 26).
Cheapest rental price for each GPU
Showing 53 GPU models across all tiers. Prices shown per GPU-hour.
| GPU | Memory | On-demand from, $/hr | Provider median, $/hr | Spot or marketplace from, $/hr | Providers | 30-day low | 7 days | Rent |
|---|---|---|---|---|---|---|---|---|
| H100 | 80–94 GB | $1.80HyperAI · 80 GB | $3.8326 providers | $1.42spot · Microsoft Azure | 33 | $1.80 | unchanged | Rent → |
| A100 | 40–80 GB | $0.89Jarvislabs · 40 GB | $1.8918 providers | $0.68spot · Microsoft Azure | 22 | $0.89 | unchanged | Rent → |
| RTX 4090 | 24 GB | $0.35TensorDock | $0.595 providers | $0.16community · SaladCloud | 7 | $0.35 | unchanged | Rent → |
| B200 | 180 GB | $5.00GMI Cloud | $8.3412 providers | $3.59spot · Daytona | 18 | $3.49 | unchanged | Rent → |
| H200 | 141–143 GB | $3.20GMI Cloud · 141 GB | $5.3318 providers | $2.61spot · Daytona | 24 | $3.14 | unchanged | Rent → |
| L4 | 24 GB | $0.44Jarvislabs | $0.807 providers | $0.10spot · AWS | 10 | $0.44 | unchanged | Rent → |
| L40S | 48 GB | $0.55Runcrate | $1.6116 providers | $0.38marketplace host · Lium | 21 | $0.54 | +1.9% | Rent → |
| A10 | 24 GB | $1.00OVHcloud | $1.574 providers | – | 6 | $1.00 | unchanged | Rent → |
| RTX 5090 | 32 GB | $0.35HyperAI | $0.836 providers | $0.25community · SaladCloud | 8 | $0.35 | unchanged | Rent → |
| Tesla T4 | 16 GB | $0.35Google Cloud | – | $0.21spot · Google Cloud | 5 | $0.35 | unchanged | Rent → |
| RTX 3090 | 24 GB | $0.20TensorDock | – | $0.09community · SaladCloud | 4 | $0.20 | unchanged | Rent → |
| A40 | 48 GB | $0.59RunPod | $1.713 providers | – | 3 | $0.59 | unchanged | Rent → |
| RTX A6000 | 48 GB | $0.35Thunder Compute | $0.578 providers | $0.40marketplace host · Lium | 10 | $0.35 | unchanged | Rent → |
| L40 | 48 GB | $0.79Thunder Compute | $0.957 providers | $0.69community · RunPod | 7 | $0.47 | +68.1% | Rent → |
| RTX 6000 Ada | 48 GB | $0.75TensorDock | $0.967 providers | $0.75marketplace host · Lium | 8 | $0.75 | unchanged | Rent → |
| RTX PRO 6000 | 96 GB | $0.80HyperAI | $2.3814 providers | $0.45spot · Google Cloud | 21 | $0.80 | unchanged | Rent → |
| RTX A5000 | 24 GB | $0.27RunPod | $0.443 providers | $0.16community · RunPod | 3 | $0.27 | unchanged | Rent → |
| MI300X | 192 GB | $2.59DigitalOcean | $3.226 providers | – | 6 | $2.39 | unchanged | Rent → |
| B300 | 288 GB | $7.10Daytona | $9.9913 providers | $4.08spot · Daytona | 17 | $4.99 | unchanged | Rent → |
| Tesla V100 | 16 GB | $0.17TensorDock | $0.786 providers | $0.12spot · Verda (DataCrunch) | 7 | $0.17 | unchanged | Rent → |
| MI355X | 288 GB | $5.99Daytona | – | $3.44spot · Daytona | 3 | $5.99 | unchanged | Rent → |
| GB200 | 186 GB | $8.00GMI Cloud | – | – | 3 | $8.00 | unchanged | Rent → |
| GH200 | 96 GB | $2.59Lambda | $3.023 providers | – | 3 | $2.29 | +13.1% | Rent → |
| RTX PRO 5000 | 48 GB | $0.96RunPod | – | $0.80marketplace host · Lium | 3 | $0.96 | unchanged | Rent → |
| RTX PRO 4000 | 24 GB | $0.57RunPod | – | – | 3 | $0.57 | unchanged | Rent → |
| RTX 4000 Ada | 20 GB | $0.76DigitalOcean | – | $0.20community · RunPod | 3 | $0.28 | unchanged | Rent → |
| RTX A4000 | 16 GB | $0.10TensorDock | $0.153 providers | $0.17community · RunPod | 3 | $0.10 | unchanged | Rent → |
| GB300 | 288 GB | $13.86Verda (DataCrunch) | – | $6.93spot · Verda (DataCrunch) | 2 | $11.17 | +4.1% | Rent → |
| MI325X | 256 GB | $3.80DigitalOcean | – | $2.00spot · Vultr | 2 | $3.80 | unchanged | Rent → |
| RTX PRO 4500 | 32 GB | $0.72RunPod | – | – | 2 | $0.72 | unchanged | Rent → |
| A30 | 24 GB | $0.35Massed Compute | – | – | 2 | $0.35 | unchanged | Rent → |
| RTX 5070 Ti | 16 GB | –no on-demand rate | – | $0.10community · SaladCloud | 2 | – | Rent → | |
| RTX 4070 Ti Super | 16 GB | –no on-demand rate | – | $0.09community · SaladCloud | 2 | – | Rent → | |
| RTX 4080 | 16 GB | –no on-demand rate | – | $0.11community · SaladCloud | 2 | – | Rent → | |
| RTX 5080 | 16 GB | –no on-demand rate | – | $0.15community · SaladCloud | 2 | – | Rent → | |
| RTX 3080 Ti | 12 GB | –no on-demand rate | – | $0.08community · SaladCloud | 2 | – | Rent → | |
| RTX 4070 Ti | 12 GB | –no on-demand rate | – | $0.08community · SaladCloud | 2 | – | Rent → | |
| RTX 3080 | 10 GB | –no on-demand rate | – | $0.06community · SaladCloud | 2 | – | Rent → | |
| MI350X | 288 GB | $6.16DigitalOcean | – | – | 1 | $5.49 | unchanged | Rent → |
| RTX 5000 Ada | 32 GB | –no on-demand rate | – | $0.49community · RunPod | 1 | – | Rent → | |
| Quadro RTX 6000 | 24 GB | $0.69Lambda | – | – | 1 | $0.69 | unchanged | Rent → |
| RTX 5090 Laptop | 24 GB | –no on-demand rate | – | $0.10community · SaladCloud | 1 | – | Rent → | |
| RTX A4500 | 20 GB | $0.25RunPod | – | $0.19community · RunPod | 1 | $0.25 | unchanged | Rent → |
| RTX 2000 Ada | 16 GB | $0.24RunPod | – | – | 1 | $0.24 | unchanged | Rent → |
| Quadro RTX 5000 | 16 GB | $0.60OVHcloud | – | – | 1 | $0.60 | unchanged | Rent → |
| RTX 4080 SUPER | 16 GB | –no on-demand rate | – | $0.28community · RunPod | 1 | – | Rent → | |
| RTX 5060 Ti 16GB | 16 GB | –no on-demand rate | – | $0.07community · SaladCloud | 1 | – | Rent → | |
| RTX 4060 Ti 16GB | 16 GB | –no on-demand rate | – | $0.08community · SaladCloud | 1 | – | Rent → | |
| RTX 5070 | 12 GB | –no on-demand rate | – | $0.08community · SaladCloud | 1 | – | Rent → | |
| RTX 4070 | 12 GB | –no on-demand rate | – | $0.07community · SaladCloud | 1 | – | Rent → | |
| RTX 3070 | 8 GB | –no on-demand rate | – | $0.13community · RunPod | 1 | – | Rent → | |
| RTX 5060 | 8 GB | –no on-demand rate | – | $0.065community · SaladCloud | 1 | – | Rent → | |
| RTX A2000 | 6 GB | –no on-demand rate | – | $0.12community · RunPod | 1 | – | Rent → |
One row per GPU: the cheapest current offer in each billing class. On-demand is capacity the provider runs itself, at a fixed price and without interruption; the provider median takes each provider's cheapest on-demand rate once. Spot and community capacity is cheaper but can be taken back; a marketplace host is one independent machine owner's server. Popular first puts the GPUs people search for most when renting at the top. The 30-day line is the cheapest on-demand rate each day, with its lowest point beside it; 7 days is the change in the cheapest on-demand rate since 4 November 2026, among the providers that listed the GPU on both days, the same measure as This week below. A month is 720 hours, the convention used across the site. See all 594 offers with filters, shortlist and CSV export. Sources include AWS on-demand pricing.
Cite this data
GPU Rental Prices, "Rent a GPU: compare cloud GPU rental prices per hour", prices checked 11 November 2026, https://gpurentalprices.com/
Quote it with a link to this page. Downloads of the whole dataset are on Data & API.
Fresh prices come first. Within each billing type, prices run lowest first, with unavailable offers last. Commercial relationships never affect ordering. How we collect and rank prices.
This week
The cheapest on-demand price of each GPU, 4 November 2026 to 11 November 2026, among the providers that listed it on both days: the same figures as the table's 7 days column.
Biggest cuts
All price moversNo GPU's cheapest on-demand price fell since 4 November 2026.
Start from the job
- LLM training80 GB+ cards, nodes of 4 to 8 GPUs$2.50/GPU-hour39 offers →
- InferenceOne GPU with 24 GB or more$0.35/GPU-hour62 offers →
- ServerlessBilled per second, scales to zero$0.50/GPU-hour51 offers →
- Consumer cardsRTX 3090, 4090, 5090 and more$0.20/GPU-hour13 offers →
- EnterpriseContracts, regions, support, compliance18 providersProcurement matrix →
Plan your spend
- Rent or buy →17,384 GPU-hoursbefore one H100 NVL 94GB at $29,900 (NetworkOutlet) costs less than renting at $1.80, power included
- Price a job →$284.0610 hours on 8× H100 (Scaleway), before storage and egress
- Training cost →$35,50710,000 GPU-hours on 8× H100 nodes (Scaleway, $3.55 per GPU-hour)
- API or rent →Token bill vs GPU billthe same model through an API or on a rented card
- Model memory →44 GBDeepSeek R1 Distill Llama 70B at 4-bit, 4,096 tokens of context
- Hosting income →Gross, payout, powerwhat renting out your own GPU earns
The market
How GPU rental prices are compared
Source and date on every price
Every price links to the provider's own page and shows when we checked it. A price we could not re-check keeps the date of its last check and is never shown as today's.
History kept every day
Every daily snapshot since 5 July 2026 is kept in a public ledger, so you can see whether today's price is low for that GPU.
Math with the formula shown
The calculators show how each result is worked out. Ranking is by price; commercial relationships never change the order.
Browse by what you need
By VRAM
By provider
- TensorDock: cheapest on-demand for 5 GPUs
- GMI Cloud: cheapest on-demand for 3 GPUs
- HyperAI: cheapest on-demand for 3 GPUs
- All 38 providers