Vercel introduces Fluid compute, a serverless model built on "mini-servers"

2025-02-04 · Publisher : Vercel · Domain : Cloud & DevOps

Vercel launches Fluid compute: instead of one invocation per instance, "mini-servers" handle several requests in parallel. Vercel claims up to 85% compute savings for early adopters.

### Background On 4 February 2025, **Vercel** introduced **Fluid compute**, which it describes as an evolution of classic serverless. Instead of starting one function per invocation, Vercel uses "high-performance mini-servers" that combine serverless efficiency with server-like flexibility. The feature is available to all users, with no migration or code change; it is enabled in project settings. ### Key facts - **In-function concurrency**: a "many requests per instance" model that reuses existing resources before creating new ones and handles tens of thousands of concurrent invocations. - **Bytecode caching**: function code is precompiled to reduce cold starts. - **waitUntil**: logging, analytics or database updates can continue after the response is sent to the client. - Vercel announces compute costs reduced "by up to 85%" for early adopters (vendor figure). ### Why it matters Serverless bills each invocation and often waits for the response of an API or an AI model: waiting time costs money. Fluid compute targets this waste, particularly for AI workloads with long waits. For a services firm, it is a criterion for comparing serverless pricing models, to be checked on one's own workloads rather than on the announced figure. ### Takeaway Fluid compute brings serverless closer to the server model to fix its waiting costs.

Tags : Serverless, FinOps, Edge, Cloud