Skip to main content
resource · gcp

Vertex AI Custom Job

schedulable
no
category
ai-ml-services

Does ZopNight manage Vertex AI Custom Job?

Vertex AI custom jobs bill for the machines, GPUs, and TPUs they hold for the duration of the run, and accelerator choice can move the cost by orders of magnitude. ZopNight lists them from the live aiplatform API across 32 Vertex regions, because Cloud Asset Inventory reports transient jobs with stale state.

Rules that fire on Vertex AI Custom Job

no live rules

No active rule family targets Vertex AI Custom Job today. Rules that used to are retired, and retired rules publish no pages and fire no findings. Scheduling and permissions coverage are unaffected.

Browse every live recommendation for this platform →

A custom job runs arbitrary training code on Vertex AI-managed compute, including GPUs and TPUs, billed for the duration of the run. GPU-heavy custom jobs are among the most expensive short-lived resources on GCP.

Run-duration billing on managed training compute

The meter on a custom job is the machine specification it requests (vCPUs, memory, and any accelerators) multiplied by how long the job holds them. Nothing persists after the job finishes, so there is no standing cost; the risk concentrates in what the job asks for and how long it runs. Accelerator selection moves the price by orders of magnitude, which is why two jobs running identical code can differ wildly in what they cost.

Live-API discovery for custom jobs

Cloud Asset Inventory reports transient Vertex job types with stale state: a finished job can sit PENDING in the index long after completion. ZopNight therefore lists custom jobs from the live aiplatform REST API across 32 Vertex regions, mapping each through the same enrichment path as asset-inventory resources so machine specs, duration, and display names come out correct. Accelerator spend folds into ML cost attribution. Custom jobs are not schedulable because they end on their own. The platform’s role here is attribution and anomaly visibility rather than stop and start.

Custom-job cost traps

The expensive failures: a job that hangs and holds its GPUs until someone notices; retry loops resubmitting a failing job against the same oversized machine; and accelerator settings habitually copied from a previous experiment that genuinely needed far more power. Duration times machine spec is the entire bill, and both factors deserve review before submission.

Console location for custom jobs

Google Cloud console → Vertex AI → Training → Custom jobs shows each job per region with state and duration. Long-running entries making no progress are the first thing worth checking.

See it fire on your bill.

Connect an account read-only. The first findings land in minutes.

417 rule families across 353 resource types on 22 platforms. Every threshold, metric, and IAM action is documented on these pages before you grant anything.

417 rule families documented
353 resource types covered
read-only default access level
Multi-cloud automation· Production-ready in 30 min· SOC 2 · ISO 27001· 20–60% off the bill, first month· 4 platforms · 1 console· Multi-cloud automation· Production-ready in 30 min· SOC 2 · ISO 27001· 20–60% off the bill, first month· 4 platforms · 1 console·