Introducing Fluid compute
By Steven Van ·
Fluid compute bills active CPU rather than idle wait time, and Vercel says it cut costs by up to 85% for early adopters of its Functions.
Vercel Functions can now run on Fluid compute, which runs many requests on a single function instance and bills only the CPU a function actually uses, not the time it spends waiting. Vercel describes it as a model between servers and serverless, and says it has helped thousands of early adopters cut compute costs by up to 85%.
Under traditional serverless, a function waiting on an API call or database response still occupies its instance. With Fluid, that waiting function can handle other requests. Vercel says this removes charges for I/O wait and network latency. A customer quoted on the page reports cutting costs by over 50% with in-function concurrency and no code changes.
- In-function concurrency: multiple invocations share one function instance.
- Cold-start reduction: functions are pre-warmed and use bytecode caching.
- Streaming: data is sent to users as it becomes available, aimed at AI, media and real-time apps.
- Cross-region failover: traffic is rerouted to backup regions during outages.
- Dynamic scaling: concurrency and resources adjust to real-time demand.
- Post-response tasks: functions can keep running after a response is sent, for work like logging or database updates.
Vercel says Fluid scales functions before instances, so it can handle tens of thousands of concurrent invocations. It also lists built-in recursion protection, which stops infinite loops before they run up usage. The company says the whole setup needs zero configuration and no maintenance overhead.
