Most slow Next.js sites aren't slow because of the framework. They're slow because of a hundred small decisions — an unoptimized hero image, a client component that didn't need to be one, a third-party script loaded in the head. Here's the checklist we run on every project.
Start server-first
With the App Router, every component is a server component by default. Keep it that way. Add "use client" only to the leaves that need state, effects or browser APIs — a dropdown, a carousel, a form. Everything else ships as HTML with zero JavaScript.
- Push "use client" as deep into the tree as possible.
- Pass server-rendered content into client components as children.
- Never import a large library into a client component just for a helper function.
Images and fonts without layout shift
next/image and next/font eliminate most Cumulative Layout Shift for free — as long as you give images explicit dimensions and mark only the true LCP image as priority.
<Image
src="/hero.webp"
alt="Product dashboard"
width={1200}
height={800}
priority
sizes="(min-width: 1024px) 50vw, 100vw"
/>Stream the slow parts
Wrap slow data dependencies in Suspense boundaries so the page shell renders immediately. Users see content in milliseconds while the expensive query finishes in the background.
Enforce a performance budget
Performance regresses quietly. We run Lighthouse CI on every pull request with hard budgets for LCP, total blocking time and JavaScript size — a regression fails the build just like a broken test.
If performance isn't tested, it isn't a feature — it's a coincidence.
Key takeaways
- Server components by default, client components at the leaves.
- Explicit image dimensions and a single priority image.
- Suspense around slow data.
- Budgets enforced in CI.
- #Next.js
- #Performance
- #Core Web Vitals



