What a GPU-hour costs

2,500 prices · 16 providers

Providers: clouds, neoclouds and marketplaces

ProviderH100 on-demand
AWS

Amazon's cloud, the largest hyperscaler: GPUs in EC2 P and G instances, priced per region.

$6.88
Microsoft Azure

Microsoft's cloud: GPUs in ND and NC virtual machines, with spot and 1- and 3-year reservations.

$6.98
Google Cloud

Google's cloud: NVIDIA GPUs in A2, A3, A4 and G2 machines, and Google's own TPUs.

$11.06
Oracle Cloud

Oracle's cloud: one list price per GPU-hour, the same in every region.

$10.00
CoreWeave

A neocloud built for AI, renting NVIDIA GPUs in multi-GPU servers by the hour.

$6.16
Lambda

A neocloud for AI training, with on-demand GPU instances and 1-Click Clusters.

$3.79
Nebius

A neocloud from Amsterdam, with on-demand and preemptible NVIDIA GPUs.

$4.50
Crusoe

A neocloud that runs its data centers on stranded and clean energy.

$3.90
Together AI

An AI cloud for inference and training that also rents GPU clusters.

$3.99
Runpod

A GPU cloud with a Secure Cloud and a Community Cloud of third-party hosts.

$1.99
DigitalOcean

A developer cloud that rents GPU Droplets by the hour.

$4.41
Akamai (Linode)

Akamai's cloud, formerly Linode, with GPU instances for inference and video.

–
Vultr

A global cloud with on-demand NVIDIA and AMD GPUs.

$2.99
Scaleway

A French cloud priced in euros, converted here at the ECB rate.

$3.21
Verda (DataCrunch)

A Finnish GPU cloud, formerly DataCrunch, whose prices move with demand.

$4.94
Vast.ai

A marketplace where hosts set their own GPU prices; the lowest listings are shown.

$2.60