Skip to content
FinOps workspace
COST LEDGER / DECISION GUIDE

Artifact Registry cost optimization

Updated: 18 November 2026Request a registry cost review

Connect repository costs to the artifacts your teams need to build, deploy and recover applications. Compare retention and location choices before changing production repositories.

Server cabinets in the mint and dark green Cost Ledger visual style
Retain the artifacts required to operate and recover each application.

Map the repository bill to the application

Start with the repository boundary rather than a deletion target. Artifact Registry stores software packages and container images and integrates with build and deployment workflows. Its repositories can supply workloads such as GKE and Cloud Run. Ask application owners which artifacts remain necessary for deployment, investigation and rollback before proposing retention changes.

The official pricing page separates storage, data transfer and applicable vulnerability scanning. Standard and remote repositories store artifacts; virtual repositories instead incur usage through their upstream repositories. The storage free allowance applies across projects attached to the billing account. Confirm the currency, current rates and agreement before building an estimate.

Review lineEvidence to collectDecision to make
Retained artifactsRepository inventory, storage trend and ownershipWhich versions have a continuing business purpose?
DownloadsRepository and consumer locations, billed transferWhich routes deserve a location comparison?
Security featuresEnabled scanning and corresponding billed usageWhich cost belongs to the approved security scope?
Operating workPolicy reviews, exceptions and engineering effortWho will maintain the controls after handover?

Use the GCP cost review to reconcile billing-account scope and project ownership. For container applications, retain the separate GKE workload cost analysis or Cloud Run billing review. Count a registry charge once when assembling the application total.

Test cleanup policies against release requirements

Google's cleanup-policy documentation describes delete rules, keep rules and dry-run testing. These policies apply to standard repositories. A keep rule overrides a matching delete rule; a keep rule alone does not remove anything. Tagged artifacts cannot be deleted when repository tags are immutable. For Docker images, cleanup also preserves images referenced by a parent manifest; after the parent is deleted, related images are eligible for a subsequent cleanup run. Review the applicable repository configuration before expecting a rule to remove a version.

  1. List repositories, consuming applications and accountable owners.
  2. Agree retention objectives and artifacts that must remain available.
  3. Draft delete and keep rules for a bounded repository scope.
  4. Run a dry run and compare candidate removals with release requirements.
  5. Approve the policy, verify required deployments and record the resulting storage trend.

For dry-run evidence, Google requires the data-write type of Data Access audit logging for Artifact Registry. Cleanup runs as a periodic background process. Treat the candidate results and subsequent billed storage as separate observations. Do not report a forecast reduction as realized savings merely because a policy was saved.

Request an exception register that identifies the artifact, owner, reason and review trigger. A supported release may need longer retention than a disposable development build. Have your team demonstrate that it can explain one retained artifact and one removal candidate from the supplied evidence. Use the realized savings guide to distinguish policy intent from the financial result.

Compare locations using actual download routes

Artifact Registry transfer pricing depends on the source repository and destination. For destinations within Google Cloud, Google's price sheet lists transfers within the same location as free. It also lists free region-to-multi-region and multi-region-to-region routes on the same continent. Traffic to non-Google clients, such as a local server, follows the applicable outbound Internet pricing; geographic proximity alone does not make it free. Review actual deployment and build consumers before assuming that all downloads stay within one region.

The repository-location guidance recommends balancing latency, availability and bandwidth costs. Region and multi-region are different location choices. For GKE Image streaming, the documentation specifies compatible repository and node locations. Preserve application and organization requirements when comparing a different repository arrangement.

  • Identify where builds, deployments and external downloads originate.
  • Keep production and development consumption distinguishable.
  • Compare any additional copies and operating effort with the transfer difference.
  • Have application owners accept the deployment and recovery behavior.

Compare the whole delivery path before choosing a cheaper-looking route. Prepare a diagram of repositories, consumers and billed transfer lines. Record which evidence supports each line and which traffic remains unallocated. Use the cloud data movement review for a wider network boundary and keep repository retention decisions distinct from runtime capacity changes.

Compare review proposals and acceptance evidence

A useful proposal names the repository scope, read access required, customer effort and production changes included. Ask candidates to demonstrate relevant GCP billing and release-engineering experience with an authorized sample finding. Compare the method used to reconcile an invoice line to a repository action rather than a headline savings claim.

DeliverableWhat to requestWho should accept it
BaselineInventory, billing reconciliation and unexplained amountsFinance and platform owners
Retention proposalDraft policies, dry-run evidence and exceptionsRelease and recovery owners
Location optionsConsumer map and cost comparison with assumptionsApplication and architecture owners
HandoverEditable policy record, queries and review instructionsThe team operating the repositories

Agree the evidence needed to close the engagement. Separate diagnostic fees, implementation work and recurring support in the quote. State who authorizes deletion and how unresolved dependencies are escalated. Compare candidates with the FinOps proposal checklist; include internal effort in the cloud ownership cost comparison.

Get my FinOps quote