The platform sells classical compute through the same door as quantum compute: one wallet, one API, one evidence trail. You submit a container workload to POST /api/v1/platform/execution/jobs, the platform reserves the worst-case cost from your credits, provisions the machine on a major cloud, meters it per minute, and stops it when your runtime bound or reservation is exhausted. You never bring your own cloud account — capacity is platform-managed and cloud-neutral (provisioned across AWS, Azure and GCP).

​
Available targets

Every workspace’s default catalog activates these targets on first use: Rates are listed with the rest of the catalog in Backends & pricing: cloud list price plus the platform markup, converted to a per-minute rate.

​
Workload shape

A GPU/CPU workload is a container run. The fields the Execution Plane accepts: Example — one A100-80GB job with a hard 30-minute bound:

​
How billing protects you

  • Runtime bound is required. Per-minute targets refuse submission without max_runtime_seconds (422 runtime_bound_required).
  • Worst case is reserved up front. Rate × bounded minutes is held from your prepaid credits before anything is provisioned.
  • Metering is continuous. An enforcement sweep meters running GPU/CPU jobs every minute and stops and deprovisions the machine the moment the reservation is exhausted — a runaway job cannot outspend its bound.
  • Settlement is capped. The final charge never exceeds the reservation; unused reservation is released. Optional per-run guards (max_gpu_minutes, max_estimated_cost_usd, max_actual_charge_usd) and workspace daily/monthly budgets add further ceilings — the full list is in Run workloads.
  • Fail-closed availability. If GPU capacity is not wired in your environment, the submit is rejected with 503 gpu_provider_unavailable before any reservation — no money is ever held for a job that cannot run.

​
Who can run and who pays

Any workspace member can submit compute; buying credits and setting spend limits is restricted to owner/admin — see Roles & permissions. Every settled run produces a sealed Evidence Bundle with the usage and charge breakdown.

​
Pair it with quantum

The point of buying GPU/CPU here instead of a generic cloud: your classical stages share the wallet, budgets, idempotency and evidence trail of your quantum stages, and your containers can call the hosted QCOS runtime and QuantumLock APIs directly — see Run QCOS & QuantumLock with your compute.