Turbopack in Next.js 16.3 adds development memory eviction, persistent file cache for builds, experimental Rust React Compiler, and `import.meta.glob` support.

The Next.js Blog provides insights, updates, and tutorials on Next.js, a popular React framework for building server-side rendered (SSR) and static websites. Developers can learn about Next.js features, routing mechanisms, and data fetching strategies, as well as explore real-world examples and use cases of Next.js applications.

Next.js

Next.js 16.3 brings significant Turbopack performance improvements. Key highlights include up to 90% reduction in dev server memory usage via in-memory cache eviction backed by filesystem persistence, persistent disk cache for `next build` (yielding up to 5.5× faster cached builds), and an experimental Rust-native React Compiler integration offering 20–50% compilation speedups over the Babel version. The release also adds Vite-compatible `import.meta.glob` API support with file-watcher integration, over 15% faster HMR cold start on complex apps, smaller runtime bundles by shipping only needed features, and per-package PostCSS configuration support for monorepos. Several compatibility fixes round out the release.

Turbopack: What's New in Next.js 16.3

The upgrades nowadays is either rust-based something or ai enhancement

The wildest part of Next.js 16.3 isn’t just the 90% RAM drop—it’s how they did it. Moving active cache data out of memory and dumping it onto the filesystem (disk) without destroying performance is a massive engineering win. Our laptops are finally safe. 💻🔥

It seems Next.js set the scene on fire in version 16.3🔥👌

Love seeing the focus on developer experience. Faster HMR and cached builds save way more time than people realize.

Worth reading past the headline number: the 90% dev-memory drop isn’t pure optimization, it’s eviction — Turbopack now drops cached results out of RAM and leans on the filesystem cache from 16.1, so the win depends on that cache staying enabled (both are default-on in 16.3, with a turbopackMemoryEviction escape hatch). The build-side cache has a similar catch for CI: copying .next between runs works, but on ephemeral runners the cache restore/save transfer competes with the compile time it saves, and these caches grow with the route graph. Measure the whole pipeline, not just the compile step — a 2.3x compile speedup can wash out if the cache upload costs more than the compile it avoided.