How does Turborepo's caching work and what makes it fast?

Turborepo speeds up monorepo builds through content-addressed caching and parallel execution.

Caching mechanism:

1. Before running a task, Turbo computes a hash from: source files in the package, env variables listed in turbo.json, the task's dependencies' output hashes.
2. If a cache entry exists for that hash, Turbo replays the cached output (logs + output files) instead of running the task.
3. Cache hits are instantaneous — CI re-runs of unchanged packages take milliseconds.

// turbo.json
{
  "pipeline": {
    "build": {
      "dependsOn": ["^build"],  // build deps first
      "outputs": ["dist/**"],
      "env": ["NODE_ENV"]       // include in hash
    },
    "test": {
      "dependsOn": ["build"],
      "outputs": []
    }
  }
}

Remote cache — Turborepo can push/pull cache to Vercel Remote Cache or a self-hosted HTTP cache. First CI run populates it; all subsequent machines (and developer laptops) benefit.