Skip to content

SYSTEM Cited by 1 source

Vercel Fluid Compute

Fluid compute is the substrate Vercel Functions run on. Named architectural difference from per-request isolates and per-request containers: "Fluid compute handles multiple concurrent requests on the same instance."

Key property

One instance serves many concurrent requests. A classical FaaS substrate (AWS Lambda pre-2024, early Cloudflare Workers architecture) spawns one container / isolate per request; if a request waits on a database query for 500 ms, those 500 ms stall that entire execution unit from serving other work. Fluid compute allows the same instance to interleave multiple in-flight requests — while one is waiting on I/O, another can be executing on-CPU.

Why it shows up on this wiki

Enables Active CPU pricing structurally. If a single instance serves N concurrent requests, wall-clock-based billing (AWS Lambda GB-seconds) over-charges for I/O wait that parallelises across concurrency. Active CPU billing charges only for time spent executing code on-CPU — which aligns with the actual resource cost of concurrent-request serving.

Canonical disclosure (2026-04-21, Vercel):

The Bun runtime runs on Fluid compute, which handles multiple concurrent requests on the same instance. Active CPU pricing means you pay for time spent executing code, not wall-clock time waiting on responses. If your function is waiting for a database query or an API call, you're not being charged for that wait time.

Substrate properties (inferred + disclosed)

  • Multi-request per instance (disclosed).
  • Regional deployment (Vercel Functions target: iad1 / us-east-1 in the 2026-04-21 benchmark).
  • Runtime-agnostic: both Node.js and Bun run on it; the substrate does not emulate runtimes — each runs natively.
  • Vercel-managed observability + logging + monitoring integrates automatically.

Architectural peers

  • AWS Lambda (pre-2024 per-request model; 2024+ added per-container concurrency via runtime-managed concurrency limit) — now structurally closer but Vercel's billing inversion is distinct.
  • Cloudflare Workers — V8 isolates are multi-request-per- isolate by design; shares the concurrency property but not the OS-process substrate or the per-project runtime-choice axis.
  • Classical Kubernetes pods + a reverse-proxy — always-multi-request-per-pod; Fluid is the managed-serverless equivalent with per-invocation billing.

Seen in

Last updated · 476 distilled / 1,218 read