CI/CD Cost Optimization: 12 Ways to Cut Your Build Bill in 2026

Cache node_modules, pip packages, Maven dependencies between runs. Most significant optimisation for dependency-heavy projects.

# GitHub Actions npm cache
- uses: actions/cache@v4
  with:
    path: ~/.npm
    key: ${{ runner.os }}-npm-${{ hashFiles('package-lock.json') }}

Only trigger CI on changes to relevant files. A docs-only commit should not run the full test suite.

# GitHub Actions path filter
on:
  push:
    paths:
      - 'src/**'
      - 'tests/**'
      - 'package.json'

Reuse unchanged Docker layers between builds. If only your app code changed, skip rebuilding base image and dependency layers.

# GitHub Actions BuildX cache
- uses: docker/build-push-action@v5
  with:
    cache-from: type=gha
    cache-to: type=gha,mode=max

Break-even for GitHub Actions self-hosted runners is ~16,000 minutes/month after the March 2026 platform fee. GitLab CE self-managed has no platform fee - break-even at ~5,000 minutes.

Lint and type-check (10-30 seconds) before integration tests (5-15 minutes). If lint fails, cancel before expensive steps run. At 15% fail rate and 8-minute tests: saves 1.2 minutes per failed run.

# Separate lint job runs first
jobs:
  lint:
    runs-on: ubuntu-latest
    steps: [lint-steps]
  test:
    needs: lint  # only runs if lint passes
    runs-on: ubuntu-latest

CircleCI: use Small (5 credits/min = $0.003/min) instead of Medium (10 credits/min = $0.006/min) for lightweight jobs like linting, static analysis, or unit tests on small codebases.

Switch non-time-sensitive pipelines (nightly integration tests, security scans, dependency audits) from every-commit to scheduled. A test suite running 20x/day could run once at night instead.

Only run deployment steps on specific branches. Only run expensive tests on PRs targeting main. Save staging deployment costs by not deploying feature branches.

# Only deploy on main
- name: Deploy
  if: github.ref == 'refs/heads/main'
  run: ./deploy.sh

GitHub Actions default retention: 90 days. For large build artifacts (Docker images, compiled binaries), reduce to 7-14 days. Storage costs $0.008/GB/day - a 10 GB artifact retained 90 days = $7.20 in storage.

AWS Graviton runners cost 20-30% less than x86 equivalent. GitHub now offers native Arm-hosted runners. Most Linux CI workloads (Node.js, Python, Go, Ruby) run unmodified on Arm. Java and some C++ projects may need Arm-specific builds.

Most engineers never check CI usage. Set spend alerts. Identify your top 5 most expensive workflows by minute consumption. In GitHub Actions: Settings > Billing > Usage. Often reveals a forgotten nightly job or misconfigured matrix build.

Parallel test jobs reduce wall-clock time but increase total minutes. A 30-minute serial test suite split into 3 x 10-minute parallel jobs = same cost but 3x faster feedback. Parallelisation is about feedback speed, not cost reduction. Only parallelise if developer wait time is impacting productivity.