Serious GPUs.
Sensible prices.
Rent a B200, RTX PRO 6000, A100 or RTX 4090 from $0.39/hr. Share a pooled card and pay up to 46% less for the same silicon. Running in minutes, no contract.
Billed hourly from your wallet · stop any time
What will you run?
We'll tell you what to rent.
The right GPU is the cheapest one your work fits on. Pick a workload to see what it needs and what it costs.
B200Dynamic
Memory figures are rules of thumb for weights at the stated precision. Context length, batch size and optimizer state add more.
Every GPU, one list.
No sales call needed.
Prices per GPU, with vCPU, RAM and local NVMe included. Dynamic shares a pooled card for less; Dedicated gives you the whole card.
| GPU | Memory | Dynamic | Dedicated | Monthly | |
|---|---|---|---|---|---|
B200New NVIDIA · Blackwell | 180 GB HBM3e | $4.25/hr−47% | $7.99/hr | $10,074/mo for 2× | Details |
| NVIDIA · Blackwell | 96 GB GDDR7 | $0.66/hr | — | $299/mo | Details |
| NVIDIA · Ampere | 80 GB HBM2e | $0.89/hr | — | — | Details |
| NVIDIA · Ada Lovelace | 24 GB GDDR6X | — | $0.39/hr | — | Details |
From sign-up to SSH
before your coffee cools.
No sales call and no quota request. Click through the console, or script the same launch from your terminal or CI.
With Google, GitHub or a work email.
Add credit by card. Hourly GPUs draw from it; budgets cap it.
Choose the card, mode and an app preset if you want one.
SSH, the web terminal or Jupyter. Terminate and the charges stop.
Everything around the GPU
comes with it.
No add-on tiers. Every account gets the tools a team needs to run GPUs without surprises.
Serve a Hub model on vLLM behind an OpenAI-compatible endpoint.
Jupyter with PyTorch, VS Code, vLLM, Text Generation WebUI, Langflow.
Volumes that outlive the instance. Keep datasets and checkpoints.
Root access from the browser or your own terminal.
Admins, members, finance managers and read-only seats.
Set a monthly limit. Spend stops there, not on your card statement.
Everything in the console is scriptable, from a laptop or CI.
Serverless LLM inference billed per token. Request access in the console.
packet.ai runs on hosted.ai's GPU cloud platform, the same software GPU providers use to run and share their own fleets. The scheduler that makes Dynamic possible is ours.
What is a Dynamic GPU?
Most jobs use a GPU in bursts: loading data, waiting for requests, saving checkpoints. On a Dedicated card you pay for every quiet moment. Dynamic shares a pooled card between isolated tenants, each with its own container, storage and enforced memory limit, so the idle time is filled and you pay up to 46% less.
Is a Dynamic GPU slower than a Dedicated one?
It's the same card and the same CUDA stack. When other tenants are busy at the same moment as you, your job can wait for its turn, so throughput varies more than on a Dedicated card. For steady benchmarks or long training runs, pick Dedicated.
How am I billed?
Hourly GPUs draw from a prepaid wallet you top up by card, and stop charging when you terminate the instance. Monthly plans are a flat monthly price. No contracts, no minimum term.
What happens to my data when I stop?
An instance's local disk goes with it. Put anything you want to keep on a persistent storage volume and attach it to your next instance.
Can my team share one account?
Yes. Invite people with roles, share one wallet, and set budgets so no one overspends.
Do you have H100, H200 or more than one GPU per machine?
H100, H200 and RTX 5090 are coming. B200 Dedicated comes with 1, 2, 3 or 8 GPUs per machine. For bare-metal servers and multi-node clusters, talk to our team.
Your next run
starts at $0.39.
Sign up, top up, launch. You'll be on a GPU in minutes.