Cost and billing
Serverless compute
What is serverless compute?
Serverless compute means capacity is acquired when work exists and released when it does not, with no cluster to provision or maintain. You are billed for the time your work actually occupies hardware rather than for reserved capacity.
For training this is a natural fit: a run is a bounded job with a definite end, and holding a GPU between runs is pure waste.
For serving it is a choice with consequences. Releasing capacity when idle saves money and introduces cold starts, so the right setting depends on whether a human is waiting for the response.
Sourcing capacity across multiple providers rather than one makes availability less correlated with any single vendor’s inventory, which matters most exactly when demand is high everywhere.
Related terms
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.