# GPU cloud prices per GPU-hour: H100, H200, B200, A100, L40S, RTX 4090 and more, with AWS, Google Cloud and Azure > On-demand list prices per GPU-hour from the public pricing pages of RunPod, Lambda, DigitalOcean, CoreWeave, Hyperstack, Nebius, Modal, Crusoe, JarvisLabs, Voltage Park and others, next to AWS, Google Cloud and Azure GPU instances. Sort and filter them, and follow each row to its source. Prices that are not stated stay blank. By GPUCostLab. Updated 7 October 2026. Canonical URL: https://gpucostlab.com/gpu-cloud-prices/ No paid links on this page: vendor links go straight to the vendor. > **Interactive table.** Filter by GPU, provider, type (including hyperscalers), price and memory, sort any column and download CSV on the HTML version of this page (https://gpucostlab.com/gpu-cloud-prices/). The full data is below. ### On-demand GPU prices, USD per GPU-hour (read 2026-10-07) | Provider | GPU (as listed) | GPU family | Tier / instance | Memory (GB) | $ per GPU-hour | Source | |---|---|---|---|---|---|---| | RunPod | B300 | B300 | Community Cloud | 288 | 6.94 | https://www.runpod.io/pricing | | RunPod | B300 | B300 | Secure Cloud | 288 | 7.89 | https://www.runpod.io/pricing | | RunPod | H200 | H200 | Community Cloud | 141 | 3.59 | https://www.runpod.io/pricing | | RunPod | H200 | H200 | Secure Cloud | 141 | 4.59 | https://www.runpod.io/pricing | | RunPod | B200 | B200 | Community Cloud | 180 | 5.98 | https://www.runpod.io/pricing | | RunPod | B200 | B200 | Secure Cloud | 180 | 6.79 | https://www.runpod.io/pricing | | RunPod | RTX Pro 6000 | RTX PRO 6000 Blackwell | Community Cloud | 96 | 1.69 | https://www.runpod.io/pricing | | RunPod | RTX Pro 6000 | RTX PRO 6000 Blackwell | Secure Cloud | 96 | 2.09 | https://www.runpod.io/pricing | | RunPod | H100 NVL | H100 | Community Cloud | 94 | 2.59 | https://www.runpod.io/pricing | | RunPod | H100 NVL | H100 | Secure Cloud | 94 | 3.19 | https://www.runpod.io/pricing | | RunPod | H100 PCIe | H100 | Community Cloud | 80 | 1.99 | https://www.runpod.io/pricing | | RunPod | H100 PCIe | H100 | Secure Cloud | 80 | 2.89 | https://www.runpod.io/pricing | | RunPod | H100 SXM | H100 | Community Cloud | 80 | 2.69 | https://www.runpod.io/pricing | | RunPod | H100 SXM | H100 | Secure Cloud | 80 | 3.49 | https://www.runpod.io/pricing | | RunPod | A100 PCIe | A100 80GB | Community Cloud | 80 | 1.19 | https://www.runpod.io/pricing | | RunPod | A100 PCIe | A100 80GB | Secure Cloud | 80 | 1.59 | https://www.runpod.io/pricing | | RunPod | A100 SXM | A100 80GB | Community Cloud | 80 | 1.39 | https://www.runpod.io/pricing | | RunPod | A100 SXM | A100 80GB | Secure Cloud | 80 | 1.59 | https://www.runpod.io/pricing | | RunPod | L40S | L40S | Community Cloud | 48 | 0.79 | https://www.runpod.io/pricing | | RunPod | L40S | L40S | Secure Cloud | 48 | 1.09 | https://www.runpod.io/pricing | | RunPod | RTX 6000 Ada | RTX 6000 Ada | Community Cloud | 48 | 0.74 | https://www.runpod.io/pricing | | RunPod | RTX 6000 Ada | RTX 6000 Ada | Secure Cloud | 48 | 0.84 | https://www.runpod.io/pricing | | RunPod | A40 | A40 | Community Cloud | 48 | 0.35 | https://www.runpod.io/pricing | | RunPod | A40 | A40 | Secure Cloud | 48 | 0.49 | https://www.runpod.io/pricing | | RunPod | L40 | L40 | Community Cloud | 48 | 0.69 | https://www.runpod.io/pricing | | RunPod | L40 | L40 | Secure Cloud | 48 | 0.82 | https://www.runpod.io/pricing | | RunPod | RTX A6000 | RTX A6000 | Community Cloud | 48 | 0.33 | https://www.runpod.io/pricing | | RunPod | RTX A6000 | RTX A6000 | Secure Cloud | 48 | 0.53 | https://www.runpod.io/pricing | | RunPod | RTX 5090 | RTX 5090 | Community Cloud | 32 | 0.69 | https://www.runpod.io/pricing | | RunPod | RTX 5090 | RTX 5090 | Secure Cloud | 32 | 0.99 | https://www.runpod.io/pricing | | RunPod | L4 | L4 | Community Cloud | 24 | 0.44 | https://www.runpod.io/pricing | | RunPod | L4 | L4 | Secure Cloud | 24 | 0.49 | https://www.runpod.io/pricing | | RunPod | RTX 3090 | RTX 3090 | Community Cloud | 24 | 0.22 | https://www.runpod.io/pricing | | RunPod | RTX 3090 | RTX 3090 | Secure Cloud | 24 | 0.5 | https://www.runpod.io/pricing | | RunPod | RTX 4090 | RTX 4090 | Community Cloud | 24 | 0.34 | https://www.runpod.io/pricing | | RunPod | RTX 4090 | RTX 4090 | Secure Cloud | 24 | 0.74 | https://www.runpod.io/pricing | | RunPod | RTX A5000 | RTX A5000 | Community Cloud | 24 | 0.16 | https://www.runpod.io/pricing | | RunPod | RTX A5000 | RTX A5000 | Secure Cloud | 24 | 0.27 | https://www.runpod.io/pricing | | Lambda | NVIDIA B200 SXM6 | B200 | 8-GPU instance | 180 | 6.69 | https://lambda.ai/pricing | | Lambda | NVIDIA H100 SXM | H100 | 8-GPU instance | 80 | 3.99 | https://lambda.ai/pricing | | Lambda | NVIDIA A100 SXM | A100 80GB | 8-GPU instance | 80 | 2.79 | https://lambda.ai/pricing | | Lambda | NVIDIA A100 SXM | A100 40GB | 8-GPU instance | 40 | 1.99 | https://lambda.ai/pricing | | Lambda | NVIDIA Tesla V100 | V100 | 8-GPU instance | 16 | 0.79 | https://lambda.ai/pricing | | Lambda | NVIDIA B200 SXM6 | B200 | 4-GPU instance | 180 | 6.79 | https://lambda.ai/pricing | | Lambda | NVIDIA H100 SXM | H100 | 4-GPU instance | 80 | 4.09 | https://lambda.ai/pricing | | Lambda | NVIDIA A100 PCIe | A100 40GB | 4-GPU instance | 40 | 1.99 | https://lambda.ai/pricing | | Lambda | NVIDIA A6000 | RTX A6000 | 4-GPU instance | 48 | 1.09 | https://lambda.ai/pricing | | Lambda | NVIDIA B200 SXM6 | B200 | 2-GPU instance | 180 | 6.89 | https://lambda.ai/pricing | | Lambda | NVIDIA H100 SXM | H100 | 2-GPU instance | 80 | 4.19 | https://lambda.ai/pricing | | Lambda | NVIDIA A100 PCIe | A100 40GB | 2-GPU instance | 40 | 1.99 | https://lambda.ai/pricing | | Lambda | NVIDIA A6000 | RTX A6000 | 2-GPU instance | 48 | 1.09 | https://lambda.ai/pricing | | Lambda | NVIDIA B200 SXM6 | B200 | | 180 | 6.99 | https://lambda.ai/pricing | | Lambda | NVIDIA GH200 | GH200 | | 96 | 2.29 | https://lambda.ai/pricing | | Lambda | NVIDIA H100 SXM | H100 | | 80 | 4.29 | https://lambda.ai/pricing | | Lambda | NVIDIA H100 PCIe | H100 | | 80 | 3.29 | https://lambda.ai/pricing | | Lambda | NVIDIA A100 SXM | A100 40GB | | 40 | 1.99 | https://lambda.ai/pricing | | Lambda | NVIDIA A100 PCIe | A100 40GB | | 40 | 1.99 | https://lambda.ai/pricing | | Lambda | NVIDIA A10 | A10 | | 24 | 1.29 | https://lambda.ai/pricing | | Lambda | NVIDIA A6000 | RTX A6000 | | 48 | 1.09 | https://lambda.ai/pricing | | Lambda | NVIDIA Quadro RTX 6000 | Quadro RTX 6000 | | 24 | 0.69 | https://lambda.ai/pricing | | DigitalOcean GPU Droplets | AMD Instinct™ MI325X | MI325X | | 256 | 3.8 | https://www.digitalocean.com/pricing/gpu-droplets | | DigitalOcean GPU Droplets | AMD Instinct™ MI300X | MI300X | | 192 | 2.59 | https://www.digitalocean.com/pricing/gpu-droplets | | DigitalOcean GPU Droplets | NVIDIA HGX H200 | H200 | | 141 | 4.47 | https://www.digitalocean.com/pricing/gpu-droplets | | DigitalOcean GPU Droplets | NVIDIA HGX H100 | H100 | | 80 | 4.41 | https://www.digitalocean.com/pricing/gpu-droplets | | DigitalOcean GPU Droplets | NVIDIA RTX 4000 Ada Generation | RTX 4000 Ada | | 20 | 0.76 | https://www.digitalocean.com/pricing/gpu-droplets | | DigitalOcean GPU Droplets | NVIDIA RTX 6000 Ada Generation | RTX 6000 Ada | | 48 | 1.57 | https://www.digitalocean.com/pricing/gpu-droplets | | DigitalOcean GPU Droplets | NVIDIA L40S | L40S | | 48 | 1.57 | https://www.digitalocean.com/pricing/gpu-droplets | | Paperspace (DigitalOcean) | H100 | H100 | | | Not stated | https://www.paperspace.com/pricing | | Paperspace (DigitalOcean) | A100-80G | A100 80GB | | 80 | Not stated | https://www.paperspace.com/pricing | | Paperspace (DigitalOcean) | A4000 | RTX A4000 | | | 0.76 | https://www.paperspace.com/pricing | | Paperspace (DigitalOcean) | A6000 | RTX A6000 | | | 1.89 | https://www.paperspace.com/pricing | | Paperspace (DigitalOcean) | V100 | V100 | | 16 | 2.3 | https://www.paperspace.com/pricing | | Paperspace (DigitalOcean) | A5000 | RTX A5000 | | | 1.38 | https://www.paperspace.com/pricing | | CoreWeave | NVIDIA GB300 NVL72 | GB300 | 4-GPU instance | 279 | Not stated | https://www.coreweave.com/pricing | | CoreWeave | NVIDIA GB200 NVL72 | GB200 | 4-GPU instance | 186 | 10.5 | https://www.coreweave.com/pricing | | CoreWeave | NVIDIA HGX B300 | B300 | 8-GPU instance | 270 | Not stated | https://www.coreweave.com/pricing | | CoreWeave | NVIDIA HGX B200 | B200 | 8-GPU instance | 180 | 8.6 | https://www.coreweave.com/pricing | | CoreWeave | NVIDIA RTX PRO 6000 Blackwell Server Edition (High Memory) | RTX PRO 6000 Blackwell | 8-GPU instance | 96 | 2.5 | https://www.coreweave.com/pricing | | CoreWeave | NVIDIA RTX PRO 6000 Blackwell Server Edition (Standard Memory) | RTX PRO 6000 Blackwell | 8-GPU instance | 96 | Not stated | https://www.coreweave.com/pricing | | CoreWeave | NVIDIA GH200 | GH200 | | 96 | 6.5 | https://www.coreweave.com/pricing | | CoreWeave | NVIDIA HGX H100 | H100 | 8-GPU instance | 80 | 6.155 | https://www.coreweave.com/pricing | | CoreWeave | NVIDIA L40 | L40 | 8-GPU instance | 48 | 1.25 | https://www.coreweave.com/pricing | | CoreWeave | NVIDIA HGX H200 | H200 | 8-GPU instance | 141 | 6.305 | https://www.coreweave.com/pricing | | CoreWeave | NVIDIA L40S | L40S | 8-GPU instance | 48 | 2.25 | https://www.coreweave.com/pricing | | CoreWeave | NVIDIA A100 | A100 80GB | 8-GPU instance | 80 | 2.7 | https://www.coreweave.com/pricing | | Hyperstack | NVIDIA H200 SXM | H200 | | 141 | 3.99 | https://www.hyperstack.cloud/gpu-pricing | | Hyperstack | NVIDIA H100 SXM | H100 | | 80 | 3.2 | https://www.hyperstack.cloud/gpu-pricing | | Hyperstack | NVIDIA H100 NVLink | H100 | | 80 | 2.6 | https://www.hyperstack.cloud/gpu-pricing | | Hyperstack | NVIDIA H100 | H100 | | 80 | 2.5 | https://www.hyperstack.cloud/gpu-pricing | | Hyperstack | NVIDIA RTX Pro 6000 SE | RTX PRO 6000 Blackwell | | 96 | 1.85 | https://www.hyperstack.cloud/gpu-pricing | | Hyperstack | NVIDIA A100 SXM | A100 80GB | | 80 | 1.6 | https://www.hyperstack.cloud/gpu-pricing | | Hyperstack | NVIDIA A100 NVLink | A100 80GB | | 80 | 1.4 | https://www.hyperstack.cloud/gpu-pricing | | Hyperstack | NVIDIA A100 | A100 80GB | | 80 | 1.35 | https://www.hyperstack.cloud/gpu-pricing | | Hyperstack | NVIDIA L40 | L40 | | 48 | 1.0 | https://www.hyperstack.cloud/gpu-pricing | | Hyperstack | NVIDIA A6000 | RTX A6000 | | 48 | 0.5 | https://www.hyperstack.cloud/gpu-pricing | | Hyperstack | NVIDIA A4000 | RTX A4000 | | 16 | 0.15 | https://www.hyperstack.cloud/gpu-pricing | | Hyperstack | NVIDIA B200 | B200 | | 192 | 6.0 | https://www.hyperstack.cloud/gpu-pricing | | Hyperstack | NVIDIA B300 | B300 | | 288 | 7.4 | https://www.hyperstack.cloud/gpu-pricing | | Nebius | NVIDIA GB300 NVL72 | GB300 | | | Not stated | https://nebius.com/prices | | Nebius | NVIDIA HGX B300 | B300 | | | 9.5 | https://nebius.com/prices | | Nebius | NVIDIA GB200 NVL72 | GB200 | | | Not stated | https://nebius.com/prices | | Nebius | NVIDIA HGX B200 | B200 | | | 8.5 | https://nebius.com/prices | | Nebius | NVIDIA HGX H200 | H200 | | | 5.4 | https://nebius.com/prices | | Nebius | NVIDIA HGX H100 | H100 | | | 4.5 | https://nebius.com/prices | | Nebius | NVIDIA RTX PRO 6000 | RTX PRO 6000 Blackwell | | | 1.8 | https://nebius.com/prices | | Nebius | NVIDIA L40S with Intel CPU | L40S | | | from 1.55 | https://nebius.com/prices | | Nebius | NVIDIA L40S with AMD CPU | L40S | | | from 1.82 | https://nebius.com/prices | | Modal | NVIDIA B300 | B300 | | | 7.0992 | https://modal.com/pricing | | Modal | NVIDIA B200 | B200 | | | 6.2496 | https://modal.com/pricing | | Modal | NVIDIA H200 SXM | H200 | | | 4.5396 | https://modal.com/pricing | | Modal | NVIDIA H100 SXM5 | H100 | | | 3.9492 | https://modal.com/pricing | | Modal | NVIDIA RTX PRO 6000 | RTX PRO 6000 Blackwell | | | 3.0312 | https://modal.com/pricing | | Modal | NVIDIA A100, 80 GB | A100 80GB | | 80 | 2.4984 | https://modal.com/pricing | | Modal | NVIDIA A100, 40 GB | A100 40GB | | 40 | 2.0988 | https://modal.com/pricing | | Modal | NVIDIA L40S | L40S | | | 1.9512 | https://modal.com/pricing | | Modal | NVIDIA A10 | A10 | | | 1.1016 | https://modal.com/pricing | | Modal | NVIDIA L4 | L4 | | | 0.7992 | https://modal.com/pricing | | Modal | NVIDIA T4 | T4 | | | 0.5904 | https://modal.com/pricing | | Crusoe | NVIDIA GB200 186GB NVL72 | GB200 | | 186 | Not stated | https://www.crusoe.ai/cloud/pricing | | Crusoe | NVIDIA B200 180GB HGX | B200 | | 180 | Not stated | https://www.crusoe.ai/cloud/pricing | | Crusoe | NVIDIA H200 141GB HGX | H200 | | 141 | 4.29 | https://www.crusoe.ai/cloud/pricing | | Crusoe | NVIDIA H100 80GB HGX | H100 | | 80 | 3.9 | https://www.crusoe.ai/cloud/pricing | | Crusoe | NVIDIA A100 80GB SXM | A100 80GB | | 80 | 2.3 | https://www.crusoe.ai/cloud/pricing | | Crusoe | NVIDIA A100 80GB PCIe | A100 80GB | | 80 | 2.0 | https://www.crusoe.ai/cloud/pricing | | Crusoe | NVIDIA L40S 48GB | L40S | | 48 | 1.5 | https://www.crusoe.ai/cloud/pricing | | Crusoe | AMD MI355X 288GB | MI355X | | 288 | Not stated | https://www.crusoe.ai/cloud/pricing | | Crusoe | AMD MI300X 192GB | MI300X | | 192 | 3.45 | https://www.crusoe.ai/cloud/pricing | | JarvisLabs | NVIDIA H200 SXM | H200 | | 141 | 4.59 | https://jarvislabs.ai/pricing | | JarvisLabs | NVIDIA H100 SXM | H100 | | 80 | 3.49 | https://jarvislabs.ai/pricing | | JarvisLabs | NVIDIA RTX Pro 6000 Blackwell | RTX PRO 6000 Blackwell | | 96 | 2.19 | https://jarvislabs.ai/pricing | | JarvisLabs | NVIDIA A100 80GB | A100 80GB | | 80 | 1.49 | https://jarvislabs.ai/pricing | | JarvisLabs | NVIDIA A100 40GB | A100 40GB | | 40 | 0.89 | https://jarvislabs.ai/pricing | | JarvisLabs | NVIDIA A30 | A30 | | 24 | 0.41 | https://jarvislabs.ai/pricing | | JarvisLabs | NVIDIA L4 | L4 | | 24 | 0.44 | https://jarvislabs.ai/pricing | | Voltage Park | NVIDIA HGX H100 | H100 | | | starting at 1.99 | https://www.voltagepark.com/pricing | | AWS EC2 | NVIDIA B300 | B300 | p6-b300.48xlarge, us-east-1, 8-GPU instance | 268 | 17.802 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA B200 | B200 | p6-b200.48xlarge, us-east-1, 8-GPU instance | 180 | 14.2416 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA H200 | H200 | p5en.48xlarge, us-east-1, 8-GPU instance | 141 | 7.912 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA H200 | H200 | p5e.48xlarge, us-east-1, 8-GPU instance | 141 | Not stated | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA H100 | H100 | p5.48xlarge, us-east-1, 8-GPU instance | 80 | 6.88 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA H100 | H100 | p5.4xlarge, us-east-1 | 80 | 6.88 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA A100 | A100 80GB | p4de.24xlarge, us-east-1, 8-GPU instance | 80 | 3.4309 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA A100 | A100 40GB | p4d.24xlarge, us-east-1, 8-GPU instance | 40 | 2.7447 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA L40S | L40S | g6e.xlarge, us-east-1 | 48 | 1.861 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA L40S | L40S | g6e.12xlarge, us-east-1, 4-GPU instance | 48 | 2.6232 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA L40S | L40S | g6e.48xlarge, us-east-1, 8-GPU instance | 48 | 3.7664 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA L4 | L4 | g6.xlarge, us-east-1 | 24 | 0.8048 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA L4 | L4 | g6.48xlarge, us-east-1, 8-GPU instance | 24 | 1.6688 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA A10G | A10G | g5.xlarge, us-east-1 | 24 | 1.006 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | AWS EC2 | NVIDIA A10G | A10G | g5.48xlarge, us-east-1, 8-GPU instance | 24 | 2.036 | https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/20261006172454/us-east-1/index.csv | | Google Cloud | NVIDIA B200 | B200 | a4-highgpu-8g, us-central1, 8-GPU instance | | Not stated | https://cloud.google.com/products/compute/pricing/accelerator-optimized | | Google Cloud | NVIDIA H200 SXM | H200 | a3-ultragpu-8g, us-central1, 8-GPU instance | | 10.6009 | https://cloud.google.com/products/compute/pricing/accelerator-optimized | | Google Cloud | NVIDIA H100 SXM | H100 | a3-megagpu-8g, us-central1, 8-GPU instance | | 11.6751 | https://cloud.google.com/products/compute/pricing/accelerator-optimized | | Google Cloud | NVIDIA H100 SXM | H100 | a3-highgpu-8g, us-central1, 8-GPU instance | | 11.0613 | https://cloud.google.com/products/compute/pricing/accelerator-optimized | | Google Cloud | NVIDIA A100 80GB | A100 80GB | a2-ultragpu-1g, us-central1 | 80 | 5.0688 | https://cloud.google.com/products/compute/pricing/accelerator-optimized | | Google Cloud | NVIDIA A100 80GB | A100 80GB | a2-ultragpu-8g, us-central1, 8-GPU instance | 80 | 5.0688 | https://cloud.google.com/products/compute/pricing/accelerator-optimized | | Google Cloud | NVIDIA A100 40GB | A100 40GB | a2-highgpu-1g, us-central1 | 40 | 3.6734 | https://cloud.google.com/products/compute/pricing/accelerator-optimized | | Google Cloud | NVIDIA A100 40GB | A100 40GB | a2-highgpu-8g, us-central1, 8-GPU instance | 40 | 3.6734 | https://cloud.google.com/products/compute/pricing/accelerator-optimized | | Google Cloud | NVIDIA RTX PRO 6000 Blackwell Server Edition | RTX PRO 6000 Blackwell | g4-standard-48, us-central1 | | 4.4999 | https://cloud.google.com/products/compute/pricing/accelerator-optimized | | Google Cloud | NVIDIA RTX PRO 6000 Blackwell Server Edition | RTX PRO 6000 Blackwell | g4-standard-384, us-central1, 8-GPU instance | | 4.4999 | https://cloud.google.com/products/compute/pricing/accelerator-optimized | | Google Cloud | NVIDIA L4 | L4 | g2-standard-4, us-central1 | | 0.7068 | https://cloud.google.com/products/compute/pricing/accelerator-optimized | | Google Cloud | NVIDIA L4 | L4 | g2-standard-96, us-central1, 8-GPU instance | | 1.0004 | https://cloud.google.com/products/compute/pricing/accelerator-optimized | | Microsoft Azure | NVIDIA H200 | H200 | Standard_ND96isr_H200_v5, eastus2, 8-GPU instance | 141 | 10.6 | https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20%27Virtual%20Machines%27%20and%20priceType%20eq%20%27Consumption%27%20and%20armRegionName%20eq%20%27eastus2%27%20and%20armSkuName%20eq%20%27Standard_ND96isr_H200_v5%27 | | Microsoft Azure | NVIDIA H100 | H100 | Standard_ND96isr_H100_v5, eastus, 8-GPU instance | 80 | 12.29 | https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20%27Virtual%20Machines%27%20and%20priceType%20eq%20%27Consumption%27%20and%20armRegionName%20eq%20%27eastus%27%20and%20armSkuName%20eq%20%27Standard_ND96isr_H100_v5%27 | | Microsoft Azure | NVIDIA H100 NVL | H100 | Standard_NC40ads_H100_v5, eastus | 94 | 6.98 | https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20%27Virtual%20Machines%27%20and%20priceType%20eq%20%27Consumption%27%20and%20armRegionName%20eq%20%27eastus%27%20and%20armSkuName%20eq%20%27Standard_NC40ads_H100_v5%27 | | Microsoft Azure | NVIDIA H100 NVL | H100 | Standard_NC80adis_H100_v5, eastus, 2-GPU instance | 94 | 6.98 | https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20%27Virtual%20Machines%27%20and%20priceType%20eq%20%27Consumption%27%20and%20armRegionName%20eq%20%27eastus%27%20and%20armSkuName%20eq%20%27Standard_NC80adis_H100_v5%27 | | Microsoft Azure | NVIDIA A100 80GB | A100 80GB | Standard_ND96amsr_A100_v4, eastus, 8-GPU instance | 80 | 4.0963 | https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20%27Virtual%20Machines%27%20and%20priceType%20eq%20%27Consumption%27%20and%20armRegionName%20eq%20%27eastus%27%20and%20armSkuName%20eq%20%27Standard_ND96amsr_A100_v4%27 | | Microsoft Azure | NVIDIA A100 40GB | A100 40GB | Standard_ND96asr_v4, eastus, 8-GPU instance | 40 | 3.3996 | https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20%27Virtual%20Machines%27%20and%20priceType%20eq%20%27Consumption%27%20and%20armRegionName%20eq%20%27eastus%27%20and%20armSkuName%20eq%20%27Standard_ND96asr_v4%27 | | Microsoft Azure | NVIDIA A100 80GB PCIe | A100 80GB | Standard_NC24ads_A100_v4, eastus | 80 | 3.673 | https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20%27Virtual%20Machines%27%20and%20priceType%20eq%20%27Consumption%27%20and%20armRegionName%20eq%20%27eastus%27%20and%20armSkuName%20eq%20%27Standard_NC24ads_A100_v4%27 | | Microsoft Azure | NVIDIA A100 80GB PCIe | A100 80GB | Standard_NC96ads_A100_v4, eastus, 4-GPU instance | 80 | 3.673 | https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20%27Virtual%20Machines%27%20and%20priceType%20eq%20%27Consumption%27%20and%20armRegionName%20eq%20%27eastus%27%20and%20armSkuName%20eq%20%27Standard_NC96ads_A100_v4%27 | ### Hyperscalers (AWS, Google Cloud, Azure) Whole-instance on-demand prices in one region each (AWS EC2 us-east-1, Google Cloud us-central1, Microsoft Azure eastus), divided by the GPU count. A hyperscaler GPU instance includes vCPUs, system memory, local storage and fast networking, and most hyperscaler GPU capacity is bought with reservations, savings plans or committed-use discounts, which are not recorded. Compare them with GPU-cloud prices with that in mind. ### Providers with no recorded prices - Vast.ai (https://vast.ai/pricing): A marketplace: each host sets its own price, so there is no single list price. We record a figure only when Vast.ai's public pages state one in their HTML. Live marketplace prices load in the browser; the page HTML contained no prices, so none are recorded. Raw data: https://gpucostlab.com/tools/gpu-price-table/prices.json ## What the prices are - **On-demand list prices**, the rate anyone can pay by the hour without a contract, before tax and credits. Reserved, committed, spot and promotional prices on the same pages are left out, and the provider notes say where they exist. - **Per GPU-hour.** When a provider lists a whole instance (CoreWeave lists 8-GPU machines, and so do the hyperscalers for their largest GPUs), the price is divided by the number of GPUs, and the note shows the instance price. Lambda's 1, 2, 4 and 8-GPU instances are listed separately because the per-GPU price differs by instance size. - **GPU only, where the provider bills that way.** Most instance prices include CPU, RAM and local disk. Modal bills CPU and memory separately, and its listed GPU prices are base prices: its plan table puts non-preemptible execution at 3x. Storage and egress are extra almost everywhere. - **Read from the provider's own page**, once per update, with the date on every row. If a page does not state a price, the row says "Not stated" and the note says why: "Contact sales", a 3-year-commitment price shown where an hourly one would be, or a page that contradicts itself. ## Same name, different GPU An "H100" is not one product, and the differences matter for inference speed: | Variant | Memory | Bandwidth | GPU-to-GPU | |---|---|---|---| | NVIDIA H100 SXM | 80 GB HBM3 | 3.35 TB/s | NVLink 900 GB/s; PCIe Gen5 128 GB/s | | NVIDIA H100 PCIe | 80 GB HBM2e | 2.00 TB/s | NVLink bridge (2 GPUs) 600 GB/s; PCIe Gen5 x16 | | NVIDIA H100 NVL | 94 GB HBM3 | 3.90 TB/s | NVLink bridge 600 GB/s; PCIe Gen5 128 GB/s | | NVIDIA H200 SXM | 141 GB HBM3e | 4.80 TB/s | NVLink 900 GB/s; PCIe Gen5 128 GB/s | | NVIDIA B200 (HGX B200) | 180 GB HBM3E | 8.00 TB/s | NVLink 5 1.8 TB/s GPU-to-GPU (NVLink Switch) | Decode speed is mostly limited by memory bandwidth, so an H100 PCIe has about 60% of an H100 SXM's single-stream ceiling for the same model (see the [VRAM calculator](/llm-vram-calculator/)). PCIe cards link at most through NVLink bridges in pairs or small groups, so tensor parallelism across more of them crosses PCIe, which SXM and HGX baseboards avoid. When a provider names the form factor (SXM, HGX, PCIe, NVL), the table records it. When it does not, the row keeps the provider's label and makes no guess. ## What is missing, and why - **Vast.ai.** It is a marketplace: thousands of hosts set their own prices, and its pricing pages load the live rates in your browser. The pages' HTML contained no prices when we read them, so nothing is recorded here rather than a number from memory. Check [vast.ai/pricing](https://vast.ai/pricing) for the live rates. - **Providers whose public pages state no on-demand price**, such as Voltage Park's on-demand table, which says "Contact for pricing". Its FAQ gives a "starting at" price for one H100, and that is what the table shows, labeled as such. ## AWS, Google Cloud and Azure The hyperscalers sell GPUs as whole virtual machines, priced per instance, per region and per purchase model. The table records their **on-demand Linux price in one US region each** (AWS us-east-1, Google Cloud us-central1, Azure eastus) for the instance types people rent GPUs in: P6, P5, P4 and G6e, G6 and G5 on AWS; A4, A3, A2, G4 and G2 on Google Cloud; ND and NC sizes on Azure. Each row names the instance and the region, shows the whole-instance price in the note, and links both to the pricing page and to the machine-readable file or API response it was read from. Read those rows with three things in mind: - **The instance is more than the GPUs.** An 8-GPU p5.48xlarge comes with 192 vCPUs, 2 TiB of memory, eight 3.84 TB NVMe drives and 3,200 Gbps of EFA networking between machines, and Google's a3-highgpu-8g with 208 vCPUs and about 1.9 TB of memory. A GPU cloud's price per GPU may or may not include as much. - **Few people pay on-demand.** Most hyperscaler GPU capacity is bought with reservations, savings plans, committed-use discounts or short-term capacity blocks, which can cost far less than the list price. Those prices are not in the table. - **Some sizes come only as 8-GPU machines.** The per-GPU figure assumes you use all eight. In the October 2026 read, the cheapest hyperscaler on-demand price per GPU-hour was 1.6 times (L4) to 3.5 times (H100) the cheapest GPU-cloud listing for the same GPU. The summary under the table shows the current figures side by side. Where a hyperscaler states a form factor (Google Cloud says SXM for its H100 and H200 machines; Azure says H100 NVL and A100 PCIe for its NC sizes), the row records it; AWS does not name it (its P5 page lists 900 GB/s NVSwitch between the GPUs), so its H100 and H200 rows make no guess. One AWS size (p5e.48xlarge) has no on-demand price in the us-east-1 price list, Google Cloud lists no on-demand price for its B200 machine (a4-highgpu-8g), and Azure's ND H200 v5 has no eastus price, so its row uses eastus2 and says so. ## How the table is kept current A script ([`update_prices.py`](/methodology/#gpu-prices)) re-reads each provider's public pricing page once a week, checks robots.txt first, parses the on-demand prices and prints a diff of every price that changed, appeared or disappeared. For AWS it streams the public EC2 Price List file for us-east-1; for Azure it queries the public Retail Prices API; for Google Cloud it reads the accelerator-optimized pricing page. None of them needs an account. If a page fails to load or no longer parses, the previous prices stay, with their original dates, and the failure is reported for a person to look at. Nothing is ever filled in by guessing. ## Cheapest per hour is not cheapest per token A slower GPU at a lower hourly price can cost more per million tokens. Cost per token is the hourly price divided by the throughput you actually get, and throughput depends on the model, precision, batch size and memory bandwidth. Take two or three candidates from this table to the [break-even calculator](/self-host-vs-api-cost/) with throughput you measured on each. For training, the [fine-tuning cost estimator](/llm-fine-tuning-cost/) turns the same prices into the cost of a LoRA, QLoRA or full fine-tuning run. ## Renting The prices in this table come from each provider's own page whether or not they pay this site anything. [RunPod](https://www.runpod.io/), [Vast.ai](https://vast.ai/), [Lambda](https://lambda.ai/) and [DigitalOcean](https://www.digitalocean.com/) run referral or affiliate programs. The others in the table do not pay us. The order is always by price or by whichever column you sort on, never by commission.