Compute — The Three Runtimes

5. Compute — The Three Runtimes

Vercel Functions (the unified umbrella as of mid-2025) come in two runtimes. Understanding the difference is essential for every architecture conversation.

5.1 Edge Runtime

What it is: Lightweight execution environment built on V8 isolates — the same engine that runs Chrome. Not a full Node.js environment.

How it works: Code runs in V8 isolates at CDN Points of Presence (PoPs) — physically close to users. Cold starts are up to 9× faster than traditional serverless.

API surface: Web Standard APIs only (fetch, Request, Response, URL, crypto). No file system. No native modules. No arbitrary npm packages.

Execution limits:

• CPU time: 35ms (hard limit — measured in CPU, not wall clock)
• Memory: 128MB
• Response size: 4MB

Best for:

• Authentication checks (if (!token) redirect('/login'))
• A/B testing and feature flag routing
• Geolocation-based redirects (if (geo.country === 'DE') redirect('/de'))
• Request/response header manipulation
• Rate limiting (simple, IP-based)
• Lightweight personalisation at the network edge

Not for:

• Database queries (no persistent connections)
• Long-running tasks
• Node.js-specific libraries
• Anything with PHI/PII that must stay in a specific region

5.2 Serverless Functions (Node.js Runtime)

What it is: Full Node.js environment running in ephemeral containers (microVMs). Full npm package compatibility. Full file system access.

How it works: Each function invocation spins up a container, executes, and returns. With Fluid Compute (see Section 4), multiple invocations can share a single container instance.

Execution limits (plan-dependent):

• Hobby: 60 seconds max
• Pro: 15 seconds default, configurable up to 300 seconds (5 minutes) with Fluid Compute, up to 13 minutes for some configurations
• Enterprise: custom

Best for:

• Server-side rendering (SSR pages)
• API routes with database queries
• Authentication with JWT validation
• AI/LLM inference calls
• File processing, image generation
• Anything needing full Node.js APIs

Cold start reality: Traditional serverless suffers from cold starts (100–500ms). Fluid Compute mitigates this significantly via bytecode caching and instance reuse.

Runtime Comparison — Quick Reference