Cloud-GPUs.com is a comparison site for cloud GPU instances. It provides access to a searchable database of over 5,000 GPU instance configurations across 30+ providers and 75+ GPU models. Prices are updated at least daily for most providers.
Just start typing to search across the entire database of thousands of instance configurations.
Narrow down the results list to only those you care about.
Select specific GPU models, individual cloud providers, total VRAM, hourly price, compliance certifications, and more.
For example: filter down to only instances with RTX PRO 6000 Blackwell GPUs.
To share or share a set of filtered results, bookmark the page's URL, or copy and share it.
All filters and search terms are stored in the link.
For example:
- Instances with RTX 4090 GPU(s)
- Instances with AMD GPUs
- Instances with 72GB+ total VRAM
- Spot instances with NVIDIA Blackwell GPUs
- Only ISO 27001-certified, EU-based providers
- Only GPUs compatible with CUDA 12.8 or 12.9
The GPU Price Analytics page provides a digestible summary of hourly pricing across all providers for the most popular GPUs pricing for each GPU model, and is updated daily.
This page is useful for getting an overview of the GPU rental market as a whole. In the screenshot below, the most expensive on-demand GPU is the NVIDIA GB300, with a median hourly price of $13.31.
The Cloud-GPUs.com database includes:
- Hyperscalers (AWS, Google Cloud Platform, Azure, Oracle)
- Other global cloud providers (Akamai, Vultr, DigitalOcean, ...)
- AI-focused Neoclouds (CoreWeave, Lambda Labs, Nebius, Together AI, ...)
- Regional cloud providers (Hetzner, Atlantic.net, Beyond.pl, ...)
- Compute marketplaces (Vast.ai, Runpod, ...)
Data is refreshed at least daily for most providers.
The results panel shows a summary of available instances, and clicking on the instance description will take you to the cloud provider's site.
This example shows two instance configurations available at the cloud provider Seeweb.it, with overall hourly prices per instance ranging from $1.23/h to $2.91/h. The first configuration has one NVIDIA RTX Pro 6000 Server GPUs attached, and the second has two. Both have pricing starting at $1.23/gpu/h.
The following info is also shown:
- This GPU uses the NVIDIA Blackwell architecture
- The amount of memory available on the GPU (VRAM) is 96GB for one GPU, or 192GB for both.
- The GPU's memory bandwidth is 1.597TB/s. This can be a useful performance proxy, especially for LLM inference.
- The interface on the GPU is PCIe
- The first instance has 16 CPU cores, 120GB RAM and 500GB disk space, and the second has twice as much of each.
- The servers are located in Italy ("IT")
- The $1.23/gpu/h price is for a 1-year reservation. 6- and 3-month reservations are slightly more expensive, and on-demand prices are $1.46/h.
Cloud-GPUs.com is one of the longest-running cloud GPU comparison portals. The original version, created by Harald Carlens, was a simple static website of listings driven by a .json file hosted in this repo.
In late 2025, Harald relaunched Cloud-GPUs.com with new interactive features and a daily data update process. These new features help users to search across thousands of GPU instances and dozens of providers to compare the latest prices and offerings.
Cloud-GPUs.com has been cited in The Verge, Cambridge University Press publications, DataCenter Knowledge, the Answer.ai blog, and Kerrisdale Capital's research notes.
Along with Cloud-GPUs.com, Harald also runs mlcontests.com (a directory for machine learning competitions and related research) and writes for joltml.com (an online magazine covering machine learning research and conferences).