ARM runners: a low-effort cost lever in GitHub Actions

GitHub-hosted ARM Linux runners run on AWS Graviton hardware. AWS passes through a portion of the Graviton cost advantage: ARM instances are cheaper to operate per compute unit than x86 at cloud scale, and GitHub prices ARM below the x86 Linux rate ($0.005/min vs $0.006).

This is not a temporary promotion. The ARM discount reflects the real hardware economics of running Graviton vs x86 at scale. Graviton 3 (c7g, m7g) offers competitive performance to x86 on standard web workloads while consuming approximately 60% of the power. Note the gap narrowed after the January 2026 repricing, when GitHub cut x86 from $0.008 to $0.006 while ARM held near $0.005, so the saving is now roughly 17% rather than the steeper discount ARM launched with.

The ARM runner was generally available on GitHub-hosted runners for all orgs in 2025. ARM remains the cheaper of the two Linux options, just by a smaller margin than before.

The honest counterpoint: the ~17% per-minute saving evaporates if the workload runs slower on ARM. A job that takes 1.2x the wall-clock time on ARM would cost more than x86 despite the lower rate. Always benchmark before committing. In practice most standard web workloads perform within 5-10% of x86 on ARM64, so the ~17% rate advantage is largely realised, but it is a modest lever, not a halving.

jobs:
  build:
    # Before:
    # runs-on: ubuntu-latest

    # After: pick one of these
    runs-on: ubuntu-24.04-arm    # GitHub-hosted ARM
    # runs-on: depot-ubuntu-24.04-arm  # if using Depot
    # runs-on: blacksmith-2vcpu-ubuntu-2404-arm  # if using Blacksmith

    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: '22'
          cache: 'npm'
      - run: npm ci
      - run: npm test

For Docker builds, add multi-arch support with buildx:

- name: Set up Docker Buildx
  uses: docker/setup-buildx-action@v3

- name: Build and push
  uses: docker/build-push-action@v6
  with:
    platforms: linux/arm64,linux/amd64
    cache-from: type=gha
    cache-to: type=gha,mode=max
    push: true
    tags: ghcr.io/${{ github.repository }}:latest

One hour of YAML editing for a modest ongoing saving. The per-minute discount is smaller than it was when ARM launched, but it is still free money on a pure Linux workload. The only question is whether your workload has ARM compatibility issues.