0
Challenge · 24Rendering & Caching

Partial Prerendering (PPR)

Renderingused in ~72% of apps

For a decade you had to pick one label per page: static (fast, cached, but can't be personalized) or dynamic (personalized, but rendered from scratch on every request). PPR deletes that choice. One page can ship a prerendered static shell in the first byte AND stream personalized, per-request holes into it. Interviewers use this to separate people who memorized 'SSG vs SSR' from people who understand the Next.js 16 rendering model.

Core idea. PPR is not a switch you flip — it's a model. A single route is rendered as a static shell (everything that doesn't read request data, prerendered at build and served instantly) plus dynamic holes (anything that reads cookies / headers / searchParams, rendered per request). You mark each hole with `<Suspense>`. The deadly trap: reading `searchParams` (or `cookies()`) in the page body pulls request data into the shell, so there is no static shell left — the WHOLE page becomes dynamic. The fix is to push that read down into a `<Suspense>` child.
Playground, this isn't a screenshot
Where do the Suspense boundaries go?
<Suspense fallback={<Spinner/>}><PanelA /></Suspense>
<Suspense fallback={<Spinner/>}><PanelB /></Suspense>
<Header />static — never suspends
Suspense
<PanelA />fast data
loaded
Suspense
<PanelB />slow data
loaded

One boundary per panel = independent reveals. Each panel shows its own spinner and pops in the moment it is ready — the slow panel no longer holds the fast one hostage. The header sits outside both boundaries, so it never disappears.

↑ play a little first, then dig in