Simple rates. Exact accounting.
Every customer rate is calculated server-side from synchronized capacity, configured markup, and a historical price snapshot.
How customer pricing works
See your final rate
No surprise pass-through or browser-calculated charges.
Wallet-first deploys
Funds are reserved before a compute resource is requested.
Rates follow the resource you select
GPU rates vary by model, GPU count, region, topology, capacity class, and current availability. The GPU Marketplace is the source for currently available offerings and their displayed customer rates.
Compare offerings using the complete configuration rather than the GPU name alone. GPU count, memory, CPU, storage, region, environment compatibility, and capacity class can affect whether an offering fits a workload. The displayed Marketplace rate is the customer-facing basis to review before starting a deployment.
Before a deployment is submitted, the platform shows the applicable rate and reserves funds against the approved maximum. Billing begins only after the instance reaches its billable running state, and usage is recorded in proportional time windows. Terminating an instance stops future usage but does not reverse usage already completed.
Serverless requests use the same wallet-first principle with an estimated reservation followed by settlement against reported usage dimensions. Review Billing and Wallet for account behavior and the GPU Cloud guide for the compute lifecycle.