Skip to content
Use cases
Computeruns today
ComputeStartups

Ship a product before you can afford hardware

Rent the card for the hours the model runs, and keep the money that would have bought it. Nothing to depreciate and nothing to explain to a board.

A GPU is a capital decision. A product that has not launched yet cannot justify one, and does not have to.

Why it is hard

  • The purchase comes before the revenue

    Buying a card means committing money to a guess about how much inference you will need, made before anybody has used the thing.

  • Guessing wrong costs either way

    Buy too small and the model does not fit. Buy too big and most of the card is idle while the runway shortens.

  • It is money you cannot get back

    Hardware is the one line item that does not scale down when a month goes badly.

What answers it

  • An expense instead of a purchase

    Per-second metering means the cost follows the usage. A quiet week is a small bill rather than the same bill.

  • Change the card, not the plan

    Seven models to move between. Finding out that a smaller card is enough costs an hour, and acting on it costs nothing.

  • A ceiling you set yourself

    A spend limit refuses new rentals past a number you choose, while work already running finishes rather than being destroyed at a threshold.

How billing works

usage_eventsbilled per second, EURlaunchmeter starts with the containerheartbeat10s at the price frozen at matchstopmeter stops, nothing after
heartbeatsent twiceusage_eventsledger_entriescharge · provider_share 85%account_balancesa view: sums, never a columnone fresh row, summedthe retry: unchangedunique(instance_id, seq)

per second, to the microeuro · no minimum · no egress fee · idle costs nothing

Metering starts when the container does and stops when it stops. The rate is frozen the moment you match, so a provider re-pricing cannot change a running rental, and an aborted run costs what it used.

Start on a 4090, move up when something makes you.