Netlify Now Runs Long-Lived Node Servers, Except for Next.js Sites

Netlify put Netlify Server into early access on Nov. 10, letting teams run an Express, Hono, Fastify or Koa backend as a long-lived Node.js process with no adapter, though Next.js, Astro and Nuxt projects can’t use it yet (Netlify changelog, Nov. 10, 2026).

What Shipped

A project gets at most one server, defined in a single entry file at netlify/server/index.ts. Netlify’s docs show two styles: a default-exported handler that takes a Request and returns a Response, or a plain Express app that calls app.listen(process.env.PORT). Every deploy, including branch deploys and Deploy Previews, gets its own server version, and Netlify said it starts a new deploy’s server before the deploy goes live.

Unlike Netlify Functions, which handle one request per instance, a server handles many at once and keeps state in memory between them. That allows connection pools, in-process caches, WebSockets and Server-Sent Events. Edge functions, redirects, static files and functions still answer first; the server gets every request they don’t claim.

The docs list three instance sizes: small (1 GB, one shared vCPU, the default), medium (2 GB, one vCPU) and large (4 GB, two vCPUs). Free and Personal plans get small only. There are two regions, US East (Ohio) by default and Frankfurt. The server needs Node.js 24 or later, and a deploy fails on an older build version.

The Limits

During early access, a server runs at most 3 instances, or 1 on Free and Personal plans. Each instance stops after 5 minutes without requests, and there is no minimum instance count yet; Netlify said one is coming after early access. A small instance takes 32 requests at once, large 128; past that, requests get a 429.

A response must send its status and headers within 20 seconds, cold start included, or it fails with a 502. Netlify closes any outbound connection, database connections included, that sends no data for 60 seconds, so its docs tell teams to set a pool’s idle timeout under 60 seconds. Instances are also replaced every 8 hours, even when busy.

Netlify Server is free during early access and limited to credit-based plans and Enterprise plans with credits on. At general availability, Netlify said it will bill credits for reserved memory in GB-hours and active CPU in vCPU-hours, with rates announced beforehand.

Why Next.js Is Left Out

Netlify’s docs say full-stack frameworks, among them Next.js, Astro, Nuxt, SvelteKit, React Router and TanStack Start, still run their server code as functions through each framework’s adapter. Those adapters usually claim every non-static path, so a server added to the same project receives no requests at all. Netlify recommends the server only for projects that don’t use those frameworks.

The Take

This is Firerun’s read, not Netlify’s. Netlify is following Vercel toward always-on compute. Vercel opened its Daemons beta for long-running queue workers on Oct. 15 and made WebSockets in Functions generally available on Oct. 20, per its changelog. Netlify’s version needs one entry file and a stock Node server.

The framework gap decides who benefits. Teams with a standalone Express or Hono API behind a static frontend can move it over today and drop the function wrappers. Teams on Next.js or Astro get nothing until the adapters change.

Early access also means unknown pricing. Netlify pitches the server for evaluation and low-volume production use, so a production backend shouldn’t depend on it until the per-unit rates are public.

Key Takeaways

• Netlify Server entered early access Nov. 10, 2026, free until general availability.
• One entry file at netlify/server/index.ts runs Express, Hono, Fastify or Koa unchanged, with WebSockets and SSE.
• Limits: 3 instances (1 on Free and Personal), scale to zero after 5 idle minutes, Node.js 24 or later.
• Next.js, Astro, Nuxt and other full-stack framework projects can’t use it yet.