01
GPU rental
02
Image and video
03
Voice and 3D
04
Agent access
Live GPU market
38 NVIDIA GPU types
From $0.13 an hour · billed per second
Running hot. Paid by the second.
Molten rents whole NVIDIA GPUs and ready AI models on Robinhood Chain. Top up in USDG from your wallet, pay for the seconds you run, and stop whenever you want.
Built on
01
Docs
Molten turns the GPUs everyone is fighting over into something you rent from a wallet. Pick an H100 or a B200, connect over SSH, and pay only for the seconds it runs. Or skip the machine and pay per image, clip and voice. Your agent can do all of it on its own.
Billed by the second
SSH and Jupyter
Paid in USDG
Built for agents
02
Products
Four ways to put a GPU to work
One balance in USDG pays for all four. No subscription, no seats, no minimum term.
01
GPU Rental
A whole NVIDIA GPU with SSH and Jupyter, from an RTX 4090 to a B300. Start it, use it, stop it: you pay for the seconds in between.
02
Ready Models
Images, video, voice and 3D from models that are already running. Send a prompt, get the file, and pay for that one output.
03
Agent Access
An API key with its own spend cap. Your agent rents the GPU it needs, runs its job and shuts the machine down by itself.
04
One Balance
Top up once in USDG on Robinhood Chain. Every rental and every output comes out of the same balance, to the cent.
03
Billing
Billed by the second, from the moment SSH is ready to the moment you press Stop.
NVIDIA GPU types to rent, from an RTX A4000 to a B300
Billing step on every rental, never rounded up to the hour
Of VRAM on a single B300, the biggest GPU we rent
04
Film
Start to finish
From wallet to GPU, start to finish
Connect a wallet, top up in USDG, rent an H100 and SSH in. No sales call, no waitlist, no contract.
05
GPUs
Every GPU at its rate per hour
Rates follow the live GPU market and are charged by the second. A GPU that never starts costs nothing.
38 GPU types on the market right now
06
How it works
From wallet to GPU in four steps
No account, no email, no sales call. The whole path takes about two minutes, most of it the GPU booting.
07
Principles
How we run your machines
You pay for running seconds
The boot, a failed start and every second after you press Stop are free. Only the time the GPU works for you is billed.
You hold the switch
One button stops any GPU at once, and a spend cap stops your agent before it goes further than you want.
Your wallet is your account
No email, no password, no card. Sign in with a signature and pay from the wallet you already use.
Nothing stays behind
Each machine is wiped when it ends. Copy off what you need first: we keep no disk, no prompt and no file.
08
Stack
Runs the tools you already use
PyTorch, CUDA and Jupyter come installed, models come from Hugging Face, and payments settle on Robinhood Chain.
Framework
Framework
Framework
GPU
Models
Workspace
Workspace
System
System
Chain
Wallet
Shell
09
Examples
What a few hours buy
Of balance is all a GPU needs to start
Fine-tune an 8B model on your own data overnight, then shut the machine off before breakfast. Six hours on one H100.
H100 SXM
About 6 hours
Meridian Health
Run a ComfyUI workflow and render a few hundred images for a product launch, for less than a coffee.
RTX 4090
About 2 hours
Carrow Logistics
Serve a 70B model on a single B200 for a weekend demo, then hand the card back on Monday morning.
B200
48 hours
Veldt Insurance
10
Guides
Guides
Short, practical guides to renting GPUs, running models and letting agents pay their own way.
11
FAQ
Common questions
Everything about renting a GPU on Molten. Still stuck? Ask us on X, a human answers.
What is Molten?
A GPU cloud on Robinhood Chain. You rent whole NVIDIA GPUs by the second, or pay per output for images, video and voice, all from one USDG balance tied to your wallet.
How do I pay?
Send USDG on Robinhood Chain from your wallet to top up your balance. Rentals and outputs come out of it, to the cent. The transfer itself pays its gas in ETH.
What comes with a GPU?
The whole GPU, a disk for your files, SSH with your keys and a Jupyter server, with PyTorch and CUDA installed. You are root on the machine.
What if I forget to stop it?
A GPU stops on its own when your balance runs out, and an agent key stops at its cap. When a machine ends its disk is wiped, so copy your files off first.
Can my agent use Molten?
Yes. Create an API key with a spend cap and hand it to your agent. It can list GPUs, rent one, run its job, stop it and make outputs, all inside that cap.
12
X
Ready to run something hot?
Connect a wallet, top up a few USDG and have a GPU answering on SSH in about two minutes.
Molten on X
New GPUs, new models and price drops, as they land.