Updated 30 April 2026|Vendor-neutral|Last verified April 2026

Monorepo vs Polyrepo:
A Decision Guide for 2026

Not another glossary. Real benchmark data, honest tradeoffs, and an interactive decision tool that maps your team profile to a specific recommendation.

Quick Decision MatrixFull decision tree
If you have...Pick
< 10 engineers, 1-3 servicesPolyrepo
10-50 engineers, 5-20 services, JS/TS stackMonorepo + Turborepo or Nx
50-500 engineers, polyglot, frequent cross-cutsMonorepo + Nx (JS) or Bazel (polyglot)
500+ engineers, multi-language, separate product domainsHybrid: monorepo per domain

3-Question Decision Tool

1 / 3

How many engineers on your team?

GoogleMonorepo

2B lines of code

Piper VCS + Bazel. ~40k commits/day. source

MetaMonorepo

Unified iOS/Android/web/server

Mercurial fork (Sapling). Custom VFS. source

SpotifyPolyrepo

Large distributed eng org

Services isolated by team boundary. Internal tooling. source

AirbnbMigrated to mono

JS monorepo migration

Mid-2010s consolidation of JS codebase. source

320
teams studied
Faros AI benchmark — faros.ai
19h
median PR cycle (mono)
vs 2h in polyrepo
18%
CI build time reduction
post-migration avg. — developers.dev
~450
JVM services merged
Cash App / Block — engineering.block.xyz

The Honest Take

Monorepo wins for cross-cutting changes and unified tooling. Polyrepo wins for isolation and independent deployment cadence. The decision is dominated by team size and language mix more than codebase size. A 10-engineer JS/TS team consolidating into a Turborepo monorepo is a two-day project. A 50-engineer polyglot team moving to Bazel is a 6-month platform project.

The worst outcome is migrating without a clear pain trigger. Most "we should consolidate" conversations are about code sharing, which is solvable in polyrepo with a private npm registry. The real monorepo triggers are: frequent atomic cross-service changes, unified dependency management breaking down, or CI costs scaling with repo count.

Real Benchmark

Faros AI studied 320 engineering teams: median PR cycle time 19 hours (monorepo) vs 2 hours (polyrepo). Caveat: monorepo PRs are larger. The delta signals that PR discipline matters more than repo layout. Source: Faros AI

If CI cost is driving this conversation: CI/CD cost calculator | CI/CD cost guide

Definitions (the short version)

Monorepo

One repository containing multiple projects, packages, or services. Cross-package changes are a single commit. Shared tooling, one CI config.

Polyrepo (Multirepo)

Each project lives in its own repository. Independent release cycles. Code sharing requires published packages. Also called multirepo.

Hybrid

Monorepo per bounded domain, isolated across domain boundaries. Common at 500+ engineers. One monorepo per product area.

Tool Comparison Matrix (2026)

Full comparison with deep dives
ToolBest forLanguagesCachingDistributionStatus 2026
NxJS/TS platform teams 20+JS/TS + partial polyglotLocal + Nx CloudYes (Nx Agents)Active
TurborepoJS/TS, fast setup, < 50 pkgsJS/TS onlyLocal + Vercel RemoteNoActive (Vercel)
BazelPolyglot, correctness-criticalAll majorLocal + remote (BuildBuddy)Yes (BuildFarm)Bazel 7+ stable
RushMicrosoft-stack enterpriseJS/TS + .NETLocal + cloudYesActively maintained
PantsPython-heavy polyglotPython, Go, JavaLocal + remoteYesPants 2.x active
LernaLegacy publishing onlyJS/TSVia Nx (post-2022)Via Nxv9 Sept 2025: maintenance

Lerna v9 (September 2025) dropped bootstrap/add/link. Actively maintained by Nrwl but use case is narrower than v3-v5. Is Lerna still maintained?

Who Uses What at Scale

All case studies

Google

Monorepo

2 billion LOC, 40k commits/day

Piper VCS + Bazel. 25k+ engineers. The case most teams cite but cannot replicate.

Potvin & Levenberg, CACM 2016 →

Meta (Facebook)

Monorepo

Unified Android/iOS/web/server

Mercurial fork (Sapling). Custom VFS for partial-clone at scale.

Meta Engineering Blog →

Cash App / Block

Monorepo

~450 JVM services consolidated

Multi-quarter migration. ~2 years to feel the full win. Bazel-based.

Block Engineering Blog →

Spotify

Polyrepo

Large distributed org

Primarily polyrepo. Services isolated by team boundary. Custom internal tooling.

Spotify Engineering →

Frequently Asked Questions

What is the difference between monorepo and polyrepo?+
A monorepo stores all projects in one repository. A polyrepo (multirepo) gives each project its own repo. The storage layout choice cascades into CI pipeline design, code sharing strategy, release coordination, and access control. Neither is inherently superior; the correct choice depends on team size, language mix, and deployment cadence.
Is monorepo faster than polyrepo?+
Build time can be faster with incremental/affected builds and caching (Nx affected, Turbo --filter). However, PR cycle time data from Faros AI (320 teams) shows median PR cycle time of 19 hours in monorepos vs 2 hours in polyrepos. Monorepos have faster builds when caching works; PR review cycles are slower because PRs are typically larger.
What companies use monorepos?+
Google (2 billion lines, Piper VCS, Bazel), Meta (Mercurial-based Sapling), Twitter/X (Pants build tool origin), Microsoft (Rush Stack per product), Uber (Go backend), Airbnb (JS migration), and Cash App/Block (~450 JVM services consolidated). Each company built or adopted custom tooling to handle their scale.
Should startups use a monorepo?+
For 1-10 engineers: usually no. The tooling overhead of a monorepo exceeds the benefit at small scale. For 5-15 engineers with a JS/TS-only stack and fewer than 5 deployable services, Turborepo is a low-friction choice. Avoid Bazel or complex Nx setups at startup scale.
Is Lerna still maintained in 2026?+
Yes. Lerna has been maintained by Nrwl (the Nx company) since May 2022. Lerna v9 was released in September 2025 and dropped the bootstrap, add, and link commands. Its use case is narrower than v3-v5: primarily package version management and changelog automation. Most new projects should start with Nx or Turborepo.

Where to Go From Here

Decision Tree

Walk 5-6 questions to your recommendation

When Polyrepo Wins

The counter-narrative most sites skip

By Team Size

Startup / mid-size / enterprise

Pros and Cons

Honest table with citations

Tool Comparison

Nx, Turborepo, Bazel, Rush, Pants, Lerna

Code Samples

Same task, side-by-side in mono vs poly

Migration Cost

Real numbers from real teams

Case Studies

Google, Meta, Cash App, Proton, Airbnb

Cross-portfolio: If team size is your main input, see engineering hiring cost for the hiring side of this decision. For platform team context, see platform engineering cost.