GPU cloud pricing

Cloud GPU prices per hour

25 GPUs from 22 clouds, per GPU-hour. An H100 costs from $1.99 an hour at Voltage Park to $10.98 at Google Cloud: same chip, 5.5× the price.

Cheapest cloud for each GPU

GPUMemoryOn-demand fromSpot fromClouds
NVIDIA H10094 GB$1.99 /h Voltage Park$1.30 /h Google Cloud20
NVIDIA A10080 GB$0.89 /h JarvisLabs$0.6788 /h Microsoft Azure15
NVIDIA H200143 GB$3.20 /h GMI Cloud$3.38 /h Verda (DataCrunch)14
NVIDIA L40S48 GB$0.97 /h Massed Compute$1.02 /h Verda (DataCrunch)12
NVIDIA B300288 GB$7.40 /h Hyperstack$6.27 /h Verda (DataCrunch)9
NVIDIA RTX PRO 600096 GB$1.80 /h Nebius$1.43 /h Verda (DataCrunch)9
NVIDIA B200192 GB$5.00 /h GMI Cloud$3.20 /h Vultr8
NVIDIA L424 GB$0.44 /h JarvisLabs$0.2907 /h Google Cloud6
NVIDIA RTX A600048 GB$0.35 /h Thunder Compute$0.4485 /h Verda (DataCrunch)6
NVIDIA L4048 GB$0.79 /h Thunder Computenot listed5
AMD Instinct MI300X192 GB$2.39 /h RunPod (Secure Cloud)not listed5
NVIDIA A1024 GB$1.00 /h OVHcloud$0.5914 /h Microsoft Azure4
NVIDIA RTX 6000 Ada48 GB$0.79 /h Massed Compute$0.811 /h Verda (DataCrunch)4
NVIDIA V10016 GB$0.246 /h Verda (DataCrunch)$0.123 /h Verda (DataCrunch)4
NVIDIA T416 GB$0.526 /h AWS$0.1641 /h Microsoft Azure3
NVIDIA A3024 GB$0.35 /h Massed Computenot listed2
NVIDIA A4048 GB$0.59 /h RunPod (Secure Cloud)not listed2
NVIDIA GB200192 GB$8.00 /h GMI Cloudnot listed2
NVIDIA GB300288 GB$14.41 /h Verda (DataCrunch)$7.21 /h Verda (DataCrunch)2
NVIDIA GH20096 GB$2.59 /h Lambdanot listed2
AMD Instinct MI350X294 GB$5.49 /h RunPod (Secure Cloud)not listed2
NVIDIA RTX 4000 Ada20 GB$0.28 /h RunPod (Secure Cloud)not listed2
NVIDIA RTX A400016 GB$0.15 /h Hyperstacknot listed2
NVIDIA RTX A500024 GB$0.27 /h RunPod (Secure Cloud)not listed2
NVIDIA RTX PRO 450032 GB$0.72 /h RunPod (Secure Cloud)not listed2

On-demand and spot list prices per GPU-hour in USD, checked daily (last Nov 17, 2026). Sources: GPU Rental Prices and Cloud GPU List Prices (AWS, Google Cloud and Azure price APIs), both CC BY 4.0; every row links the cloud's own pricing page. Order depends on price only. How we compare GPU prices.

Before you rent

Prices are per GPU-hour. Hyperscalers sell most data-center GPUs only in 8-GPU instances, so the smallest bill can be eight times the hourly rate shown. SXM, PCIe and NVL versions of a chip differ in memory bandwidth and interconnect; each row names the version.

Spot capacity costs less but can be reclaimed at short notice. Reserved and committed-use discounts, storage, network egress and taxes are not included.

Paying per token instead? Compare LLM API prices before you rent hardware to serve an open model.

Listed by one tracked cloud only: A10G, MI325X, MI355X, QUADRO RTX 5000, QUADRO RTX 6000, RTX 2000 Ada, RTX 3090, RTX 4090, RTX 5090, RTX A4500, RTX PRO 4000, RTX PRO 5000.