The idle-GPU problem
Compute is scarce for builders while useful hardware sits idle elsewhere, so we started building the market we wanted to rent from.
Two people in Europe building the compute market we wanted to rent from: idle hardware earning its keep, renters paying by the second, and both sides reading the same ledger.
Two sides, one market
Rent by the second. Earn from idle hardware.
The working half, today. Per-second, in EUR.
Explore computeTwo more surfaces on top, still early.
What we are building
Teams rent by the second; providers list GPUs they own. We handle the middle: matching, isolation, metering, payout.
Our story
Compute is scarce for builders while useful hardware sits idle elsewhere, so we started building the market we wanted to rent from.
The first rental ran end to end on real hardware: launched, metered per second, and settled at the 85/15 split. Then the connect hop, reading files out of the running container.
The first SSH login into a rental, one card scoped to it and the host's own sshd blocked from inside. Snapshot and restore proven the same day.
Each rental on its own session network, single-tenant, its container destroyed at teardown. The provider bootstrap runs unattended from one command.
No real money has moved yet: every balance is credit granted by hand. The ledger arithmetic is proven; the rail is next.
The flywheel
Owners turn dormant hardware into income.
More supply on the market pushes the going rate down.
Cheaper compute pulls in more builders and bigger runs.
Higher utilisation makes listing even more worth it.
Compute makes GPU capacity available. Studio is the planned workflow surface above it. Autopilot is the planned approval layer that reads what the work costs.
Rent GPUs by the second from idle hardware, or earn from the cards you own. Priced at match, settled in euro.
Draw the workflow and attach hardware to each step, then read a finished run back step by step.
See what the work costs and where it's leaking. Simulates a change against what you already paid, applies nothing until you approve.
What we stand for
The parts that matter are written into the schema, where changing our minds later is a migration and a broken test, not a quiet edit.
How the rest is enforcedProviders keep 85% of everything their hardware earns. Not a launch offer, and not a number that quietly moves once there is enough hardware here to move it.
How it holds
The rate you agreed is the rate you pay, for as long as the machine runs. A provider putting their prices up cannot reach a job that is already going.
How it holds
There is no balance field for anyone to edit. Ask why a number is what it is and the answer is the list of things that made it, not somebody's recollection.
How it holds
Per-second billing multiplies a rounding error by every tick, so there is no rounding to multiply. One currency, six decimal places, and the two halves of every split are required to add back to the whole.
How it holds
By the numbers
No dashboard of vanity metrics. Every figure here traces to the code.
people
the whole company
to the provider
of what a card makes
products
on one market
decimals
the euro ledger is exact to
Where we actually are
Everything above this is what we are trying to build; this is what is actually running today, and what is not.
Running today
Not yet
The team
Kracht is two people. That is the whole company.
Edition 001
Co-founderBuy side
The side you buy from
Owns the product and everything a customer sees: the marketplace, the pricing, and the claims.
Edition 002
Co-founderRun side
The side that runs it
Owns the machinery: placement, the provider agent, and the settlement that moves the money.
Rent on-demand compute by the second, or list idle hardware and let it pay for itself. Same marketplace, both directions.
No card required