Skip to content
An honest comparison

See how we
actually compare.

Compare queues, pricing, access, and control against hyperscalers and GPU clouds.

€231
8×H100, 10 hours
€510
the same job, AWS
per-sec
billing
€0
egress

Side by side

Kracht vs. the field

The choices that actually move the bill, side by side. Switch to a single rival to score it round by round.

Feature
Kracht
Hyperscaler
Niche cloud
Per-second billing
Offered
Not offered
Offered
No egress fees
Offered
Not offered
Not offered
H100 €/hr on demand
€2.89
€5.99
€1.19
Instant provisioning
Offered
Offered
Not offered
Earn by listing hardware
Offered
Not offered
Not offered
Bring your own image
Offered
Offered
Offered
Global regions
6
30+
3-6

Competitor columns are live where a feed exists (Azure €5.99, Vast.ai €1.19, updated 2026-09-26), illustrative otherwise. Our €2.89 is the catalogue rate we bill, in EUR. Competitor prices converted at $1.1463 to €1, the ECB euro reference rate for 22 September 2026.

Compare any GPU

Every card, every provider, one grid

Pick a card, choose the providers, filter by region and sort by price: our rate, the big clouds, and the open market of independent hosts, all in one table. A blank is a provider that does not rent it.

37
GPUs priced
7
providers shown
25
market regions
Providers
6 of 150 cards
GPUVRAM
NVIDIA · GeForce RTX 40Ada Lovelace · 2022-24
RTX 4090Kracht24 GB€0.31not rentednot rentednot rentednot rentednot rented€0.18 (lowest in the row)
NVIDIA · A-seriesAmpere · 2020-21
A100Kracht40 / 80 GB€0.78€2.92€2.40€3.20€2.36€1.74€0.24 (lowest in the row)
A10Kracht24 GB€0.22€2.75€0.88not rentednot rented€1.13€0.21 (lowest in the row)
NVIDIA · L-seriesAda Lovelace · 2023
L40SKracht48 GB€0.64 (lowest in the row)not rentednot rentednot rented€1.96not rentednot rented
NVIDIA · H-seriesHopper · 2022-24
H100 (PCIe/SXM)Kracht80 GB€2.89 (lowest in the row)€5.99€6.00€9.65€5.37€3.48not rented
H200Kracht141 GB€3.74 (lowest in the row)not rentednot rentednot rented€5.50not rentednot rented
lowest in the row, where a row has two prices or more not rented by that providerprices per GPU-hour, EUR, open market updated 2026-09-26

Kracht is our per-second catalogue rate. Clouds are published list prices (Azure live); the open market is the cheapest live listing across independent hosts, peer hardware with no vendor SLA. USD converted at $1.1463 to €1, the ECB euro reference rate for 22 September 2026. Sources: Azure, AWS, Google Cloud, CoreWeave, Lambda, open market.

The whole stack

Everything a rental includes

One per-second rate, and this is what sits under it.

01

A whole GPU

H100 down to RTX 4090, the card is yours for the length of the rental.

02

An isolated session

Its own subnet, and the provider's own machine is blocked from reaching inside it.

03

Your own image

Bring any Docker image and SSH key. There is no runtime to lock you in.

04

Per-second metering

The meter starts when the box is running and stops the moment you do.

05

An append-only receipt

Every charge lands on one immutable ledger, traceable to the cent.

No spin

Where the others actually win

We would rather you pick the right tool than oversell ours. Here is where we honestly would not be our own first choice.

Hyperscalers / AWS, GCP, Azure

A managed everything-suite

If you want GPUs beside 200 first-party managed services, the big clouds win on breadth. We focus on compute, storage, and networking.

Dedicated single-tenant providers

A single contractual vendor

Some compliance regimes demand bespoke bare-metal contracts. A marketplace wins on price and flexibility, not owning every layer.

Niche regional clouds

An exotic region, today

We cover six regions, so a specific jurisdiction may well need a local specialist. We would rather say that clearly than imply a map we do not have.

Still think we're the right fit? Most teams do.

See the real numbers

Switching

Moving over takes one command

No re-architecting and no lock-in to unwind. Point your existing workload here and it runs the same way, for less. Three things do change, and they are below rather than in a support thread later.

  • Bring your existing Docker image; it runs unchanged
  • Same CUDA, drivers, and frameworks you already use
  • No egress fees to leave your current cloud
  • Anything inside your image works, because it is your image
kracht · deploy
# point your same image at Kracht
$ kracht launch --gpu 1 --image my-org/trainer:latest --command python train.py
Rented 8fa2c1d0 on h100-node-02 (eu-west) at 2.890000 EUR/hr
Billing has started. Stop it with: kracht stop 8fa2c1d0
# same job, €2.89/hr instead of €6.09/hr
one GPU, there8 lines
aws ec2 run-instances \  --instance-type p5.48xlarge \  --image-id <ami> \  --key-name <key-pair> \  --subnet-id <subnet> \  --security-group-ids <sg> \  --block-device-mappings <json> \  --count 1
one GPU, here1 line
kracht launch --gpu 1 --region eu-west
01

Two keys, not one

SSH gets you in; a WireGuard key is what publishes the route at all. Providers sit behind NAT, so a rental missing it bills without ever being reachable.

02

Checkpointing is yours

Nothing snapshots your progress for you. On a long run, write checkpoints to your own storage on a schedule you choose, because a host that stops takes its scratch with it.

03

Billing is prepaid, in euros

You top up a balance and it meters down per second, rather than receiving an invoice at month end. There is no committed spend to negotiate and no term to unwind.

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.

No card required

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