Is Bun Switching Its Core from Zig to Rust? What It Means for Node.js

On May 5, a tweet by @LukeParkerDev featuring a GitHub commit titled phase-a: draft batch big-1 (12 files) ignited speculation that Bun might abandon Zig for Rust. The commit, authored by Bun founder Jarred Sumner, added nearly 28,000 lines of code without deletions.

Founder’s Clarification: An Experiment, Not a Decision

Jarred posted on Hacker News, stating the reaction was exaggerated. He emphasized that the code does not compile yet and that the branch may be discarded entirely; the goal is a side‑by‑side comparison of a runnable Rust version against the existing Zig implementation to let data drive the decision.

"This is my own branch. The whole post’s reaction is overblown. 302 comments discuss code that doesn’t even run. We have not committed to a rewrite. These codes may all be thrown away."

Why Rust?

Jarred explained that Zig lacks destructors, constructors, and a borrow checker, whereas Rust provides these features, catching memory‑safety issues at compile time instead of after users file GitHub issues. He denied any link to Zig’s AI‑code‑contribution policy, citing purely technical reasons.

28,000 Lines Generated by an Coding Agent

The massive code addition was not handwritten. Bun’s docs/PORTING.md guides an AI coding agent to translate .zig files to .rs in two phases:

Phase A : For each .zig file, generate a draft .rs file that mirrors the logic; compilation is not required.

Phase B : Incrementally make each crate compile.

Thus, the agent mechanically translated the 28k lines, producing a non‑runnable draft that will be refined step by step.

Node.js Community Reacts

Following Bun’s news, Node.js core contributor Matteo Collina asked whether Node.js should attempt a similar Rust rewrite, noting the massive exposure the question received.

"Should @nodejs attempt a rewrite to Rust like @bunjavascript is doing by burning unlimited tokens?"

Matteo highlighted three points:

The performance bottleneck in Node.js lies in glue C++ code on the stream path, not Zig.

He has a Tokio‑based HTTP prototype that runs about twice as fast as current Node.js, close to Bun’s benchmarks, but it sacrifices backward compatibility, making a full rewrite infeasible.

Node.js is already adopting Rust for parts of its codebase, though a complete rewrite is unlikely.

These observations suggest a trend: Zig’s niche is pressured as AI‑assisted development favors languages like Rust that support destructors, borrow checking, and have strong tooling.

Implications

The experiment shows that AI‑assisted large‑scale language migration is now a practical engineering approach, even if the resulting code still requires human review. While Bun’s outcome remains uncertain, the community sees Rust emerging as the preferred choice for low‑level runtime rewrites, and Node.js is moving toward incremental Rust integration.