Why Vercel Does Not Replace a VPS: Five Technical Limits of Serverless

Vercel can deploy your app, but it cannot replace a VPS by architecture. Here are five concrete technical reasons, and they apply to the entire serverless category, not just Vercel.

What it is

Vercel is a serverless platform, not a traditional server. It handles each incoming request as a new invocation in a managed sandbox, scales automatically, and tears down the runtime when the request completes. A VPS, by contrast, gives you an always-on process with full OS-level control. These are two different architectural models serving two different needs.

The distinction matters because many people assume that since Vercel can deploy a Next.js app, it can host any backend. It cannot, at least not without friction. The same limits apply to Netlify Functions, Cloudflare Workers, AWS Lambda, and Google Cloud Functions, because they share the same serverless model.

Why it matters

• Maximum function duration is capped: 60 seconds on Hobby, 300 seconds on Pro, 900 seconds on Enterprise. Background workers, queue consumers, and media pipelines get cut mid-run.
• Each invocation runs in a fresh sandbox, so in-memory state, cache, and connection pools are wiped between calls.
• The local filesystem is ephemeral. You can write to /tmp during an invocation, but it disappears on the next call.
• WebSocket connections are supported but bound by the function duration limit, so realtime connections close periodically and reconnect to a different instance.
• There is no root access. You cannot install system packages, run custom daemons, or set up crontab inside the function runtime.

How it works

On a VPS, a process stays alive, memory persists, the disk is durable, and you control the OS. On Vercel, a request triggers an invocation, the sandbox runs, state is held only for that invocation, and everything is cleaned up when it ends. The next request may land on a completely new instance with no memory of the last one.

This is not a quality problem. It is the nature of the serverless model. The platform is optimized for request-driven, stateless, web-native workloads. For those workloads, it is excellent. For long-running processes, persistent state, or custom system software, it is the wrong tool.

Caveats

Vercel is strong where it is designed to be: frontend, SSR, BFF, webhooks, and thin APIs. It has improved over time with Fluid compute, cron jobs, and WebSocket support. But WebSocket connections still close at the duration limit, and durable state still belongs in an external database, Redis, or queue. The practical answer is usually a hybrid architecture: Vercel for the web layer, Cloudflare Workers for edge, AWS Lambda for heavy compute, and a VPS for workers, daemons, and media pipelines.

Who it's for