Artifact Registry pricing (GCP): storage + downloads + egress (practical estimate)

Reviewed by CloudCostKit Editorial Team. Last updated: 2026-01-27. Editorial policy and methodology.

Start with a calculator if you need a first-pass estimate, then use this guide to validate the assumptions and catch the billing traps.

Data Egress Cost Calculator API Response Size Transfer Calculator VPC Data Transfer Cost Calculator Cross-region Transfer Cost Calculator

Registries are "store bits, move bits". A reliable estimate separates three drivers: storage (GB-month), downloads (pull GB), and egress. Most surprises come from retention drift and peak pull storms (CI spikes, autoscaling, node churn).

0) Define your scope (what you are counting)

Artifact types: containers vs language packages vs build artifacts (sizes and churn differ).
Environments: prod/stage/dev repos (retention policies often drift by env).
Regions: multi-region clusters and cross-region runners change egress patterns.

1) Storage (GB-month)

Build storage from retention. Start with "how much do we keep" rather than "how much do we push".

Tool: Storage cost (GB-month).

Estimate average stored GB across the month, not just today's size.
For container images, layer sharing means storage is often less than "tags × size" (validate with actual usage).
Keep separate lines for base images vs app images (churn differs).

2) Downloads (pull volume)

Downloads scale with CI/CD and autoscaling. A good first model is: deploys/day × pulls per deploy × avg artifact size, plus a peak line for incident windows and node churn.

CI peaks: parallel builds and retries can create short pull storms.
Node churn: new nodes pull many images at once (worst during incidents).
Caching: node-local caching and image pre-pulling can reduce download volume dramatically.

3) Egress (split by destination)

Treat transfer as its own line item. Split by destination because pricing and billing boundaries differ.

Tools: Data egress cost, Cross-region transfer.

Same-region: often cheapest; keep clusters and registries co-located if possible.
Cross-region: multi-region clusters and remote runners can make this meaningful.
Internet: external users and third-party runners are the common egress driver.

Worked estimate template (copy/paste)

Stored GB-month = average stored GB across month (validate with registry usage)
Download GB/month = pulls/month × avg artifact size (baseline + peak)
Egress GB/month = subset of downloads billed as outbound (cross-region/internet)

Common pitfalls

Retention drift: old tags and layers kept forever.
Assuming tags × size equals storage (layer sharing changes the math).
Ignoring CI and scale-out peaks (pull storms are the peak scenario).
Cross-region pulls hidden inside "it’s internal" assumptions.
Not validating caching behavior (warm vs cold nodes).

How to validate

Validate actual stored usage and retention settings per repo.
Sample pull volume during peak build/deploy windows and scale-outs.
Validate which destinations are billed as outbound (same-region vs cross-region vs internet).

Sources

A driver-based Cloud SQL estimate: instance-hours (HA + replicas), storage GB-month, backups/retention, and data transfer. Includes a worked template, common pitfalls, and validation steps for peak sizing and growth.

Bigtable cost estimation: nodes, storage growth, and transfer (practical model)

A driver-based Bigtable estimate: provisioned capacity (node-hours), stored GB-month + growth, and network transfer. Includes validation steps for hotspots, compactions, and peak throughput that force over-provisioning.

Cloud Spanner cost estimation: capacity, storage, backups, and multi-region traffic

Estimate Spanner cost using measurable drivers: provisioned capacity (baseline + peak), stored GB-month (data + indexes), backups/retention, and multi-region/network patterns. Includes a worked template, common pitfalls, and validation steps.

GCP Cloud Storage Pricing & Cost Guide

Understand Cloud Storage cost drivers: storage class, operations, retrieval, and egress with estimation steps.

Google Kubernetes Engine (GKE) pricing: nodes, networking, storage, and observability

GKE cost is not just nodes: include node pools, autoscaling, requests/limits (bin packing), load balancing/egress, storage, and logs/metrics. Includes a worked estimate template, pitfalls, and validation steps to keep clusters right-sized.

Pub/Sub pricing: deliveries, retries, fan-out, and payload transfer (practical estimate)

A practical Pub/Sub estimate: publish volume, fan-out (subscriptions), delivery attempts (retries), retention/replay scenarios, and payload transfer. Includes a worked template, pitfalls, and validation steps.

Related calculators

Data Egress Cost Calculator

Estimate monthly egress spend from GB transferred and $/GB pricing.

API Response Size Transfer Calculator

Estimate monthly transfer from request volume and average response size.

VPC Data Transfer Cost Calculator

Estimate data transfer spend from GB/month and $/GB assumptions.

Cross-region Transfer Cost Calculator

Estimate monthly cross-region transfer cost from GB transferred and $/GB pricing.

FAQ

What usually drives Artifact Registry cost?

Storage (GB-month) is the baseline, but heavy CI/CD and cluster churn can make downloads and egress meaningful. Peak pull storms during deploys and scale-outs are the common surprise.

How do I estimate quickly?

Estimate stored GB-month from retention, then estimate download GB/month from pull volume and average artifact size. Split same-region vs cross-region/internet destinations for egress.

How do I validate?

Validate actual stored usage, validate retention policies, and sample download volume during peak build/deploy windows. Validate whether layer sharing reduces storage vs 'tags × size'.

What is the most common mistake?

Assuming storage equals tags × image size (layer sharing changes the math) and ignoring node churn and CI spikes that create short, expensive download peaks.

Last updated: 2026-01-27. Reviewed against CloudCostKit methodology and current provider documentation. See the Editorial Policy .