CI/CD artifact storage cost in 2026: per-GB pricing and retention math

Storage is the line item teams ignore until the bill arrives. CI artifacts (build outputs, test reports, log bundles, coverage data) accumulate quietly. The default retention is generous, the default upload is "everything", and the per-GB price is small enough to feel rounding-error. At 200 builds/day across 30 services for two years, that rounding error becomes a four-figure monthly line. This page walks through what each vendor charges, what default retention does to your bill, and the two levers that consistently cut artifact storage spend 70-90%.

Per-vendor artifact storage rates (2026)

Worked example: a 30-service backend platform

A platform team running 30 microservices, each with a CI pipeline that uploads roughly 80 MB of artifacts per build (test reports, coverage, the built binary, log bundles). Average 15 builds/service/day across 20 working days = 9,000 builds/month per service x 30 services... no, scale that down: 15 builds/service/day x 20 days x 30 services = 9,000 builds/month total. Storage created per month: 9,000 x 80 MB = 720 GB.

With the GitHub Actions default 90-day retention, peak storage is 720 GB x 3 months = 2,160 GB stored at any given time once steady-state is reached. Less the 1 GB Team-tier inclusion: 2,159 GB billable at $0.25/GB-month = $540/month. The same workload on S3 Standard would be $50/month. Same workload with retention dropped to 14 days: peak storage 720 GB x (14/30) = 336 GB stored, billed at $0.25 x 335 = $84/month on GitHub Actions, or $8/month on S3.

Two changes (cut retention from 90 days to 14 days, push to S3 instead of vendor-hosted) take the bill from $540/month to $8/month, a 67x reduction. Most teams in the 25-100 dev range have a meaningful artifact storage line and have not made either change.

Retention policy: the simplest lever

Retention can be set globally on most platforms but the per-workflow override is more powerful. GitHub Actions lets you set retention-days per artifact upload step. GitLab supports artifacts.expire_in per job. CircleCI does not expose retention as a per-job setting (it is plan-wide), which is one of the reasons larger teams drift toward GitHub Actions or self-hosted GitLab.

The triage to use: classify artifacts into "deployable" (release binaries, signed bundles, container images you might roll back to), "diagnostic" (test reports, coverage, log bundles you keep for trend analysis), and "intermediate" (build caches, object files, anything you never look at after the next push). Set retention to 365 days for deployable, 30-60 days for diagnostic, 1-7 days for intermediate. Most teams have far more intermediate than they realise.

When to push to S3 instead

The vendor-hosted artifact convenience comes with an 11x markup on per-GB-month versus S3. The break-even where rolling your own becomes worth the integration effort is roughly $50/month in vendor artifact spend, which corresponds to about 200 GB of stored artifacts at the GitHub Actions rate. Above that threshold, an S3 bucket with a lifecycle policy (delete after N days, transition to Intelligent-Tiering after 30) plus a small upload step in CI is straightforward to maintain.

The integration cost is real: you need an OIDC trust between your CI and AWS, an IAM policy for the bucket, a CI step that uploads with the right key naming convention, and a cleanup script if your lifecycle rules are not enough. Most teams take 1-2 days to set up and then forget about it. The S3 lifecycle policy documentation is on the AWS docs page.

The bandwidth trap

Storage is the headline. Bandwidth is the gotcha. GitHub Actions includes a transfer allowance and then charges $0.50/GB out beyond it. S3 charges $0.09/GB out to the public internet. If you fan out artifact downloads to many concurrent jobs (a common pattern for matrix testing where each job needs the build output), the egress can outpace storage. Cache locally in the runner where possible, and consider using a CloudFront or vendor-CDN distribution if your CI pulls the same large artifacts repeatedly.

Cross-region downloads add a separate AWS egress charge ($0.02/GB inter-region transfer plus the destination egress). Pin your CI artifacts to one region and pull from runners in the same region wherever the workflow allows. We cover the egress dimension in detail on the cross-region egress cost page.

Container registry storage: a sibling line

If your CI builds container images, those images live in a container registry (GHCR, GitLab Container Registry, ECR, Docker Hub). The pricing math is similar: per-GB-month, with vendor markup over raw S3. ECR Standard is $0.10/GB-month. GitHub Container Registry is $0.25/GB-month for private images (free for public). Docker Hub paid plans bundle storage with seat counts.

The cost driver is layer churn. Container images pile up image manifests as each build produces a new tag. A retention policy on the registry that keeps only the most recent N tags per service plus all tags matching v* for releases is often sufficient. ECR has lifecycle policies built in. GHCR exposes retention through GitHub Packages settings. Either way, the discipline is the same as artifact storage: classify, retain, prune.

Frequently Asked Questions