The Wire · Showcase
TURBOPACK SLASHES BUILD MEMORY OVERHEAD. NEXT.JS STREAMING GETS LEANER
By RepoJournal · Filed · About Vercel
Turbopack's serialization pipeline now drops data proactively during persistence, cutting memory consumption from 16.74 GB to 15.69 GB on canary builds.
The win comes from a fundamental rearchitecture of how Turbopack buffers data before writing to disk [1]. Because in-memory organization uses FxHash while disk output requires a different hash function and sorted order, the compiler was forced to hold everything in scratch buffers. Now it drops data as it flushes, treating persistence as a streaming operation instead of a batch job. Meanwhile, Next.js is simplifying its prerender pipeline [2] with a new chunk-callback pattern that eliminates redundant stream operations, making staged renders easier to implement without tee()ing streams twice. The App Router's route params also got smarter: IsRootLayout hints now generalize to IsRootLayoutOrAbove [3], letting the client correctly identify which params live above the topmost layout.tsx. On the Workflow side, hook conflict detection evolved from a boolean to a actionable handle [4] - getConflict() now returns the Run object that owns the token, not just a yes/no, so duplicate runs can inspect status, await results, or cancel competitors. Over in vercel/ai, the SDK removed deprecated experimental exports [5] as generateImage graduates from experimental to stable, and Fireworks now reports token usage in streaming responses [6]. Eve framework support shipped out of experimental [7], meaning teams can deploy Eve projects without the feature flag.
Action items
- → Test next build memory profiles against canary - persistence optimizations are live [ref:1] vercel/next.js [monitor]
- → Remove experimental_generateImage imports if you're on AI SDK 7 [ref:12] vercel/ai [plan]
- → Update workflow hooks to use getConflict() instead of hasConflict() [ref:6] vercel/workflow [plan]
- → Eve framework projects no longer need VERCEL_USE_EXPERIMENTAL_FRAMEWORKS flag [ref:17] vercel/vercel [monitor]
References
- [1] [turbopack] Drop data while serializing in `next build` (#94561) vercel/next.js
- [2] refactor: process streaming prerenders more efficiently ↗ vercel/next.js
- [3] Generalize IsRootLayout prefetch hint to IsRootLayoutOrAbove ↗ vercel/next.js
- [4] Replace hook.hasConflict with hook.getConflict() returning the conflicting Run (#2373) vercel/workflow
- [5] fix: Remove deprecated experimental generateImage exports (#16017) vercel/ai
- [6] @ai-sdk/[email protected] ↗ vercel/ai
- [7] [@vercel/frameworks] graduate Eve from experimental frameworks (#16612) vercel/vercel
FAQ
- What changed in Vercel on June 13, 2026?
- Turbopack's serialization pipeline now drops data proactively during persistence, cutting memory consumption from 16.74 GB to 15.69 GB on canary builds.
- What should Vercel teams do about it?
- Test next build memory profiles against canary - persistence optimizations are live [ref:1] • Remove experimental_generateImage imports if you're on AI SDK 7 [ref:12] • Update workflow hooks to use getConflict() instead of hasConflict() [ref:6]
- Which Vercel repositories shipped on June 13, 2026?
- vercel/next.js, vercel/workflow, vercel/ai, vercel/vercel