Deploy AI models on H100, A100, H200, and AMD MI300X GPUs with up to 40% cost savings. Lightning-fast machine learning inference on enterprise GPU infrastructure.
Pause a GPU and your environment survives — files, models and installed packages come back exactly where you left them, on the same provider or a cheaper one. You pay by the second, and only while it's running.
Pause · Resume · MoveSame environment, new provider
$1.99/hrH100s starting at
40% lessthan traditional clouds
See how it works
See a VM pause in real time.
console.aquanode.io/vms/llm-finetune-01
Connected
llm-finetune-01 training
RUNNING
GPUH100 (80GB)
vCPU28 cores
Memory180 GB
Storage6 TB NVMe
GPU util
88%
VRAM
72%
hyperstack · canada-1
This week · session cost
Compute time47.2 hrs
Rate$1.99/hr
Paused (no charge)12.6 hrs · saved $25.07
You pay$93.93
hyperstack
massed compute
vast.ai
hotaisle
vultr
Event log
--:--:--waiting…
0.0s/4.0s
Snapshot the environment, release the GPU, stop the billing.
Why this matters
Your setup shouldn't die when the GPU does.
You stop the box for the weekend. The provider runs out of capacity. A cheaper GPU shows up somewhere else. On a normal cloud every one of those means running the install script again and hoping it still works. Here it means one action, and your environment comes back with you.
01 / Pause
Step away without losing your work.
One action snapshots your home directory to your own storage bucket, then releases the GPU so billing stops. The box goes away. Your environment doesn't.
AWS
GCP
RP
AQ
Monthly spend at 50% active utilization
02 / Resume
Pick up right where you left off.
Resume and Aquanode rents a fresh GPU, restores the snapshot onto it, and hands it back with your files, models and packages in place. No reinstall script.
paused↻running↻paused
Loop all day. Pay only for the active minutes.
03 / Move
Come back on a different cloud.
When you resume, pick another provider or another GPU from the live marketplace — the price difference is shown before you confirm. Same environment, new hardware.
Pause and stop paying. Resume on a cheaper GPU — your graph, models, custom nodes and queue come back with you, so you requeue the render instead of rebuilding the whole workspace.
Rendering on hyperstack · canada-1 — queue: 3 workflows
Coverage
One account. Every GPU provider that matters.
Aquanode aggregates capacity across decentralized and hyperscale clouds. Deploy on any of them, stop when you're done, and come back on whichever one you want — one account, one bill.
From overnight training runs to production inference, every one of these is the same trick: the environment outlives the machine it was running on.
TRAIN
Train for days without holding the GPU the whole time.
Kick off a multi-day run, stop it for the weekend, pick it back up Monday with the same data and checkpoints on disk. You pay for the hours the GPUs actually burn.
B200 · 8× clustercheckpoints in your bucketresume where it stopped
TUNE
Fine-tune once, then move it wherever's cheapest.
Spin up a single H100, LoRA your base model, then bring the same environment back up on whichever provider is cheapest to serve it from.
H100 · 1× (LoRA)PEFT + 8-bit optimresume on another provider
SERVE
Stop the endpoint between pushes, not the project.
Shut a model down when traffic dies and bring the exact same weights, packages and config back when it returns. Nothing to re-provision.
A100 · on-demandweights stay in your bucketsame env on resume
EXPLORE
Open a notebook today, find it tomorrow.
Stop the box when you're done for the day. Start it again and your notebooks, data and installed packages are already there — on a different GPU class if you want one.
pick your GPU and regionnotebook state persistscome back anywhere
Deploy from your terminal
Rent a GPU, SSH in, never leave your shell.
aq is the Aquanode command-line client. It rents the cheapest matching GPU, provisions the box, and hands you a shell — no console tab required.
zsh
$curl -fsSL https://github.com/Aquanodeio/aq-releases/releases/latest/download/install.sh | sh
$aq login # pair this CLI to your Aquanode account
$aq up --comfyui --name my-box # rent a GPU, provision the box
→ HTTPS URL printed when it's ready
$aq ssh my-box # managed key, managed alias — zero setup
$aq down my-box # tear it down, stop billing
aq ssh maintains a managed ~/.ssh/config alias for every live box — so anything that speaks ssh_config works with no aq involved:
$ ssh aq-my-box
$ scp big.safetensors aq-my-box:/workspace/
Install, flags, and the full command reference live in the docs.
Built for people who are done rebuilding their setup.
01
Your environment survives
Stop a box and its files, models and packages come back on the next one — including on a different provider.
02
One bill, every cloud
No more reconciling 6 invoices. Centralized billing across everything you run.
03
Pay with crypto or card
USDC, ETH, or a plain credit card. Whichever fits your stack.
04
Up to 40% cheaper
You stop paying the second you stop the box, on top of marketplace prices that already undercut the big clouds.
Trusted by Builders Everywhere
Don't take our word for it, here's what builders like you are saying.
I switched from Runpod to Aquanode, and it feels so much better to use. The GPUs are priced more reasonably! I'd highly recommend it.
ML Engineer
University of California, Los Angeles
The pause-and-resume feature makes it so much easier to manage our GPUs across multiple clouds.
Infrastructure Engineer
AI Startup
Trust & security
Compliant & secure.
SOC 2
Type I
AICPA · Trust Services
SOC 2
Type II
AICPA · Continuous audit
ISO
27001
Information security
HIPAA
Compliant
Protected health info
Enterprise GPU providers on Aquanode hold ISO 27001, SOC 2 Type I & II, and HIPAA certifications. Community compute is labeled separately in the marketplace.
Enterprise · certifiedCommunity · self-attested
Ready when you are
Stop paying for idle GPUs.
Sign up in 60 seconds. Pay only for the GPU minutes you actually use.
