How Cloudflare’s AI‑Rewritten Next.js Clone Achieves 4× Faster Builds and 57% Smaller Bundles

Dual Speed and Size Gains

Cloudflare measured build time and bundle size on a 33‑route App Router sample. Next.js 16.1.6 (Turbopack) took 7.38 seconds to build and produced a 168.9 KB gzip bundle. vinext built with Vite 7 + Rollup finished in 4.64 seconds (1.6× faster), while Vite 8 + Rolldown completed in 1.67 seconds (4.4× faster). The resulting gzip bundles were 74.0 KB (56% reduction) and 72.9 KB (57% reduction) respectively.

A Lightweight Alternative

Instead of wrapping Turbopack output, vinext re‑implements the full Next.js API on top of Vite’s plugin ecosystem, covering routing, server‑side rendering, React Server Components, Server Actions, caching, and middleware. Developers only need to replace the next command with vinext while keeping the existing project structure.

npm install vinext
# previously: next dev
vinext dev   # HMR‑enabled dev server
vinext build # production build
vinext deploy # one‑click deploy to Cloudflare Workers

How AI Achieved This

The rewrite was completed in less than a week. From the first line of code on February 13 to a working routing and streaming implementation, the AI wrote most of the code while a human engineer defined architecture and order of operations. The tool passed Next.js’s official test suite and reached 94% API coverage.

Four decisive factors made this possible:

Clear Next.js specifications: comprehensive documentation and abundant community discussion allowed the model to internalize the framework’s behavior.

Rich official test suite: the AI could validate its output against existing unit tests, ensuring correctness.

Robust Vite foundation: Vite’s fast HMR, module resolution, and bundling already provided a solid base.

Advanced large‑model capabilities: modern models can retain context for thousands of files, avoiding hallucinations that plagued earlier attempts.

During development, the engineer set the overall architecture and sequencing; the AI generated code and tests. Successful tests were merged, while failures were fed back to the model for self‑repair.

Key Feature: Traffic‑Aware Dynamic Pre‑Rendering (TPR)

Traditional Next.js pre‑renders every page, which can be prohibitive for sites with hundreds of thousands of routes. vinext introduces Traffic‑Aware Pre‑Rendering (TPR): Cloudflare analyzes the past 24‑hour traffic log, selects the most‑visited pages (e.g., the top 100), and pre‑renders them instantly. Less‑frequent pages are rendered on demand and then cached, dramatically cutting build time.

Conclusion: Rethinking Software Engineering Abstractions

Vinext is already in production on a U.S. government beta site (CIO.gov). Cloudflare argues that AI can now internalize an entire system’s architecture, eliminating the need for multiple layered frameworks as long as a solid runtime (Vite) and clear specifications (Next.js API) exist.

Try It Yourself

Vinext provides an automated migration tool (Agent Skill) compatible with Claude Code, Cursor, Codex, and other AI assistants. npx skills add cloudflare/vinext In a Next.js project, instruct the AI: migrate this project to vinext The AI will check compatibility, install dependencies, generate config files, and start the dev server, flagging any manual steps.

For a manual trial:

npx vinext init   # migrate existing Next.js project
npx vinext dev    # start dev server
npx vinext deploy # deploy to Cloudflare Workers

The source code is open‑sourced on GitHub for developers to explore.