8× NVIDIA H100
Ready$15.20/hr
llama-finetune•
640GB VRAM960GB RAM96 vCPU🇺🇸
Compute infrastructure, designed for researchers.
Rent compute, scale training runs and run hundreds of experiments in parallel with Cloudy.
cloudy.so/compute
/My Team$2,847.20
N
psst... the demo above is interactive ↑
“Managing GPUs and memory for my experiments has always been a hassle. With cloudy, I finally have a much simpler and cost-effective solution.”
Shyam Gopal
Researcher at Genmo.ai
“Managing disk state during training is a frustrating experience. Cloudy's storage abstractions makes this much less painful.”
Rick Lamers
AI Researcher at Groq
“Cloudy made doing experiments so much smoother and way more flexible. Can't wait to see where this project goes!”
Neel Kant
Machine Learning at Prometheus
“Cloudy makes it so much easier to rent GPUs, always available, storage is synced across regions, feels seamless.”
Sahil Chaudhary
Founder of Glaive.ai
Platform Features
Everything you need to train at scale
From compute to storage, we provide the infrastructure you need to run training experiments efficiently.
1. Rent GPU Compute
Access H100s, H200s, and A100s from 13+ providers.
| GPU | VRAM | $/hr | |
|---|---|---|---|
| 1× H100 | 80GB | $1.90 | |
| 2× H100 | 160GB | $3.80 | |
| 4× H100 | 320GB | $7.60 | |
| 8× H100 | 640GB | $15.20 |
2. Persistent Storage
Browse and navigate your volumes.
llama-weights
training-data
2.4 TBcheckpoints
890 GBlogs
12 GBoutputs
340 GB3. Volume Forking
Clone terabytes of storage, and iterate on your experiments.
llama-data
Source→
4. Public Volumes
Share storage volumes with the community.
openllama-weights4.2 TB
cloudy.so/v/openllama-weights
Anyone can view & fork
FAQs
We support H100s and H200s and we never run out of them. The GPUs are sourced from 13 different cloud providers. We unify the experience with simple Docker images and storage volumes, so you can continue exactly where you left off, no matter which provider you're on.
Storage is priced at $0.00004/GB/hr. You can store up to 1PB in a single volume. But, you only pay for the storage you use.
Our storage volumes are heavily cached and designed for storing large amounts of data, especially large files like models and datasets. They're accessible from any region, ensuring your data is always where you need it.
Yes! You can make any volume public for read-only access. Anyone with the link can browse the files and fork the volume to their own account. Private volumes are only accessible to your team members.
Forking creates an instant, independent copy of a volume. It runs in the background and doesn't affect the original. This is useful for creating snapshots, experimenting with data, or sharing a starting point with others.
Yes! Our built-in file browser lets you view, download, and explore your volume contents directly from the dashboard—no need to mount to an instance. You can preview text files, images, and navigate directories.
Yes! You can invite team members to collaborate. Instances and volumes are shared across your team, making it easy to work together on the same projects and infrastructure.
We're here to help! Email us at support@cloudy.so and we'll get back to you as soon as possible.
Ready to get started?
Rent compute and get started with Cloudy
Go to Dashboard