For users
Rent a GPU in seconds. Pay for the seconds you use.
Capacity from people with idle hardware, at a fraction of hyperscaler pricing. No quota request, no reservation, no minimum term.
Payments are not switched on yet, so nothing is charged today. Everything below describes how it works once they are.
Capacity
What is available right now
| Launch | |||||
|---|---|---|---|---|---|
| RTX 40908.9 | 24 GB | eu-nl | 0.94 | €0.61 | |
| L40S | 48 GB | us-east | 0.97 | €1.12 | |
| H1009.0 | 80 GB | eu-west | 0.99 | €2.89 |
sample. Sample figures, not live marketplace metrics. Nothing here is measured from real trading yet.Sample rows. 12,418 GPUs are listed across 40+ regions in the illustrative figures used on this site.
How it works
Four steps, and the meter runs for one
Pick from live capacity
Filter by card, VRAM, region and price. What you see is what providers have online right now, and a machine already rented is not offered to you.
Launch with your key
Paste an SSH public key and choose whether the container gets outbound internet. Kracht never handles a private key, and a launch that could not be reached is refused before you are charged.
Run whatever you want
Your own image, or the default that ships with OpenSSH and CUDA. SSH in, mount a volume, point a training script at it. It is a machine, not a sandboxed notebook.
Stop, and stop paying
Metering ends the second the container does. No minimum, no notice period, and nothing continues to bill after you press stop.
Isolation
What you are protected from
The card is yours alone
Kracht rents whole hosts, not slices. Nobody else's workload shares your GPU, your VRAM or your memory bus while you hold it, and the card is wiped and re-attested before the next renter.
The provider cannot read your work
Sessions run in containers the host user cannot enter, and a provider gets no shell into your session. What is on the disk while you run is yours.
Where we are honest about this
A determined host with physical access to their own machine is a stronger adversary than a container boundary. For work where that matters, read the isolation page before you decide, rather than taking a badge on a marketing page for it.
A host that dies mid-job
Billing stops at the last metered second, and you are not charged for the dead time. Reliability scoring pushes flaky machines down the match order so you meet them less often.
The full threat model, including what we do not defend against, is on the security page.
Billing
What you are actually charged
- Billed per second, from the moment the container starts to the moment it stops.
- The price is frozen when you launch. A provider re-pricing their listing cannot change what a running rental costs you.
- Credit is held when you launch and released when you stop, so a long run cannot quietly overdraw.
- Every charge is a line in a ledger you can read, tied to the rental it came from.
The other side
Have hardware instead?
The GPUs here belong to people who had them sitting idle. If that is you, the provider side pays 85% of every rented hour.
Two sides, one marketplace
Spin up the GPUs you need - or earn from the ones you already own
Rent on-demand compute by the second, or list idle hardware and let it pay for itself. Same marketplace, both directions.
Join 12,000+ builders / $5 free credit / no card required
Stay in the loop
Product updates and early access to new GPU classes. No spam.