Skip to content
New here? Rent a GPU by the second
Kracht
Sign inSign up
Pick the workload, not the machine

GPU solutions forevery workload.

Build, train, and deploy across compute, storage, networking, and serving without the lock-in.

Start free
Talk to sales
Train
whole hosts
Fine-tune
from a checkpoint
Serve
near your users
Earn
85% of the hour

The problem

GPU access is broken. We fixed the market.

Compute is the scarcest resource in AI, yet most of the world's GPUs sit idle. Kracht connects both sides and sets a fair price in real time.

FIG 01 routing

jobplacementa100catalogue7900 xtxh100

job · model + region

The request: a GPU model and a region, both catalogue codes, never a machine. Where it lands is placement's decision.

The old way

Renting GPUs today

  • H100s sold out, with waitlists before you can train
  • Hyperscaler rates above the silicon's worth, with long lock-ins
  • Egress fees and opaque minimums you only find on the invoice
  • Quota requests and sales calls just to get started
  • No idea which step of a pipeline is holding the expensive card
  • You find the waste on the invoice, a month after it happened
  • Your own GPUs sit idle, depreciating, earning nothing
With Kracht

A two-sided compute market

  • Listed GPUs you can launch on now, running in under 20 secondsclaim: asserted by us, with nothing behind it yet
  • Market-set prices, 53% to 68% under list for the exact same card
  • Per-second billing, no egress, no minimums, no lock-in
  • Sign up and kracht launch, with no tickets or gatekeeping
  • Each workflow step holds its own card, and an idle step holds none
  • The metering behind your invoice, read back as where it leaked
  • List unused GPUs and turn idle hardware into income

One platform, three layers

Rent the metal. Build on the rest.

Compute is live today. Studio and Autopilot stack on top of it, same account, same meter. Click a layer to open it.

Compute

Live
Compute

Rent GPUs by the second from idle hardware, or earn from the cards you own. Priced at match, settled in euro.

Explore Compute
StudioPlanned
AutopilotPlanned

The comparison

The honest math on a rental.

The same card, priced by the market instead of a margin. The 15% we keep is the same number on every page and in the ledger.

the old way

Renting elsewhere

recommended

Kracht

The same H100

list price, plus margin
€2.89 / hr · 53% lower

Billing

billed by the hour, rounded up
billed by the second

Egress and minimums

egress fees, opaque minimums
no egress, no minimums

Time to start

quota request, sales call
self-serve, under 20sclaim: asserted by us, with nothing behind it yet

Your runtime

locked to their image
bring your own image and key

Where the money goes

margin you never see
85% to the provider, 15% to us

illustrative comparison · H100 rate from the one price list

one rental, settledinst-91bc
metered per second€8.94
provider payout€7.60
Kracht take€1.34
85% provider15% kracht
15%is all we keep. Exact-integer, so not a cent leaks in the rounding.
See how billing works
append-only ledgeridempotent
+usageseq 6042+€0.000803
+usageseq 6043+€0.000803
+takeplatform+€0.000120
+payoutacct-77a1+€0.000683
[€] a 41-minute job, per second€1.97
[↑] the same job, billed hourly€2.89
// X-AXIS: JOB LENGTH// Y-AXIS: EUR · 1 ■ = €0.48
  • A 7m job on an H100: €0.34 per second, €2.89 billed hourly.
  • A 20m job on an H100: €0.96 per second, €2.89 billed hourly.
  • A 41m job on an H100: €1.97 per second, €2.89 billed hourly.
  • A 52m job on an H100: €2.50 per second, €2.89 billed hourly.
  • A 1h10 job on an H100: €3.37 per second, €5.78 billed hourly.
  • A 1h34 job on an H100: €4.53 per second, €5.78 billed hourly.
  • A 2h04 job on an H100: €5.97 per second, €8.67 billed hourly.
  • A 2h30 job on an H100: €7.22 per second, €8.67 billed hourly.
  • A 2h52 job on an H100: €8.28 per second, €8.67 billed hourly.
per secondwhat hourly rounding adds

illustrative · list rate × job length, from the one price list · 1 square is 10 minutes of the card

live meter
€0.0000
billed this run
the market, metered live
illustrative mesh · move your cursor
€0.000803
rate · per second, H100
per second
billing granularity
€0
cost while idle
microeuro
settled precision

Use cases

Built for the way your team works.

One market, four ways in. Pick the seat that is yours to see what Kracht does for it.

Researchers

Experiments that end when you do.

Per-second capacity with transparent per-team billing, so you never pay for an idle card.

per sec
metered, not rounded
For researchers

Startups

Skip procurement, keep the runway.

No quota forms, no sales calls, no lock-in, on GPUs 53% to 68% under a hyperscaler's list price.

under 20s
signup to running
claim: asserted by us, with nothing behind it yetFor startups

Teams

Endpoints that scale to zero.

Traffic-following endpoints that cost nothing while idle and stop the meter when demand does.

€0
while idle
For teams

Providers

Idle GPUs become income.

List spare hardware and earn 85% of what it makes, paid per second onto an append-only ledger.

85%
provider share
For providers

One CLI, every surface

One command. Every surface.

Describe the job once and launch it from the CLI, the API, the dashboard or your pipeline. Same meter, same ledger, wherever it starts.

launch.sh
$ kracht launch --gpu 1 --region eu-central \
    --image pytorch/pytorch \
    --command python train.py
CLIone command
$ kracht launch --gpu 1 --ssh
Waiting for it to finish provisioning...
root@3f9a1c2b7d4e:~#
APIrest · sdk
POST /v1/me/instances
{ gpu_count: 1 }
200 · billed / sec
DASHBOARDpoint · click
pick a card →
pick an image →
Launch · billed per second
CIheadless
- run: kracht launch --gpu 1
    --command ./train.sh
✓ ends when it exits

Illustrative snippets. Every command is one the kracht CLI accepts.

kracht launchzsh
$ kracht launch --gpu 1 --region eu-central --max-price 3.00
Rented 8fa2c1d0 on h100-node-02 (eu-central) at 2.890000 EUR/hr
Billing has started. Stop it with:  kracht stop 8fa2c1d0
kracht launch [flags]
Rent a machine
--gpu is how many cards, not which. The scheduler picks the machine, and the price it prints is frozen onto the rental at match.
--gpu N  GPUs required (default 1)
--image REF  container image
--region CODE  restrict to a region
--max-price EUR  refuse anything dearer, per hour
--ssh  open a shell once it is ready
US Eastonlineus-east · capacity pooledUS Westonlineus-west · capacity pooledEU Westonlineeu-west · capacity pooledEU Centralonlineeu-central · capacity pooledNetherlandsonlineeu-nl · capacity pooledAsia Pacific Southonlineap-south · capacity pooled

Global by default

One network, six regions

See how placement works
6
regions
85%
provider share
per sec
metering
€0
to start

Trust and security

Security your team can sign off on.

Every rental carries the same guarantees, so legal, finance and security can evaluate Kracht without a call.

PROVIDER HOSTtheir machine, their networkDATA ISOLATEDBLOCKEDyour rentalSESSION VERIFIEDhost sshdother cardshub

the fence · isolated by default

The rental has its own session subnet. The host's sshd has a route toward it, drawn here as the grey wire, and that route ends at the fence.

Visit the trust center
EU data residency
Encrypted in transit + rest
GDPR-aligned handling
Session isolation
Bring-your-own SSH key
Append-only audit

Frequently asked questions

No. You rent a whole host, launch it with one command, and tear it down when the job is done. There is no cluster to babysit and nothing to provision ahead of time.

Two sides, one marketplace

Spin up a GPU, or earn from your own

Rent on-demand compute by the second, or list idle hardware and let it pay for itself. Same marketplace, both directions.

Start renting
List your hardware

No card required

85%
revenue to providers
Per-second
billing, live
EUR
metered to the microeuro