- Cloudflare says tsrs, a Rust port of TypeScript's compiler, is up to 8× faster and uses about 70% less memory than TypeScript 7.0.
- Cloudflare says AI models can now carry out large ports in days, which shifts the hard part to maintenance and matching the official compiler.
- Ask your team how much build capacity type checking uses, which baseline any benchmark cites, and who maintains a new tool.
When the checker becomes the bottleneck
A type checker reads code and flags mistakes before software ships. Owner.com, which sells software to restaurants, found that this routine step had grown into a real constraint.
Socket reports that Max Schwenk, who leads AI developer experience there, says the company's backend is about 37,000 TypeScript files that the checker treats as one unit. His figure for one pass on a single thread with TypeScript 7.0 is 80 seconds and 19 GiB of memory. Once linters, development servers and several coding agents ran together, even powerful laptops had no room left.
This shows a shift in where engineering cost sits. When AI agents write code, the checks around that code get multiplied too. The checker stops being a background utility. It becomes shared infrastructure that someone has to own.
What Cloudflare announced
Socket reports that on October 20, Cloudflare said it would fold tsrs, a Rust rewrite of the TypeScript compiler, into Oxc, its set of JavaScript tools. The VoidZero team, which Cloudflare acquired in June, will develop and maintain it.
Schwenk started the project. He used Claude coding agents to build the first Rust port, then tested and tuned it against Microsoft's Go compiler before handing it to Oxc.
The field is getting crowded. Microsoft's own native Go version, announced in March 2025, was billed as about ten times faster than the JavaScript original. Bun introduced bun check on October 10. Theo Browne's tsc-rs is being merged with tsrs.
Why agents strain a type checker
Socket's report explains the pressure point. When several agents work in separate worktrees (separate copies of the same project), each one runs its own type checking. That can use CPU and memory several times over.
Cutting that overhead makes room for more agents on the same machine. Jimmy Lai, according to Socket, is considering tsrs for the ChatGPT codebase for that reason.
Read the speed claims with care
Here are Cloudflare's headline numbers. With the same count of type-checking workers, tsrs runs up to 8× faster than TypeScript 7.0 and needs roughly 70% less memory. Out of the box, tsrs sizes itself to the host machine, spending more memory to gain speed.
Cloudflare says machine size and compiler settings can swing results widely. It says it set up like-for-like comparisons together with the TypeScript team, and it has published its benchmark tools and method.
The baseline matters. Cloudflare's announcement itself rates the TypeScript 7.1 beta as up to 2.5× faster than 7.0. The 8× number compares against 7.0, not against that beta. Bun's own tests, it says, show bun check running 3× to 6.4× faster than TypeScript 7.0.2.
TypeScript's Daniel Rosenwasser said on Hacker News that "a lot of the speed does not just come down to language choice." Some tsrs findings are already feeding into TypeScript 7.1.
The port was fast because the tests already existed
Socket says Schwenk began by keeping the Go version's algorithms and behavior, then used TypeScript's conformance tests to check the result. Cloudflare's view is that today's models can handle a large port with little human help, within days.
The lesson here is that AI did not invent a compiler. It translated one that came with a large test suite. A strong existing test suite is what made the translation checkable.
Why Rust, and where it stops
Cloudflare wants a Rust version partly for WebAssembly (Wasm), which lets code run in a browser or a Cloudflare Worker. Go can compile to Wasm, but it brings its garbage collector along. Cloudflare sees Rust as a better fit for small, constrained environments.
The goal is a full check loop close to the code. An agent could edit files and verify them inside a sandbox, with no round trip to a separate server.
The limits are real. The Wasm build runs on one thread. It lacks watch mode and editor language server (LSP) support. It has a 4 GiB memory ceiling per instance. The team is still working toward a stable release and still differs from the official compiler in places, including the order of error messages.
Who keeps it correct
Socket points to the harder problem. A port can be produced quickly, but someone must track every later TypeScript release. VoidZero's Michael Dong uses the word "slopforks" for hasty AI-made copies that lack that commitment.
Cloudflare says it will reuse a process built for Vinext. Agents review upstream changes, find compatibility gaps, run upstream tests and propose fixes. That is a stated plan, not yet a track record.
Questions to put to your team
Ask how much build and CI time and memory goes to type checking, and whether it grows with each agent you add.
When a vendor cites a speedup, ask which version and settings it was measured against.
Before adopting any compiler port, ask who maintains it, how it tracks upstream releases and what tests prove it matches the original. For tsrs, check whether any script depends on the order of error messages. Pilot it away from release-critical pipelines until a stable release exists.
Cloudflare says models can now produce ports like this within days. Trusting one takes a team that keeps showing up.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: Socket.
