Vercel presents Fluid, a unified compute layer for functions, sandboxes and builds

2026-09-01 · Publisher : Vercel · Domain : Cloud & DevOps

Vercel details Fluid, a compute layer that assembles machine, environment and storage on demand for each workload. Vendor figures: 15 million builds a day, 25 million sandboxes a week, 1 trillion requests a month.

### Background On 1 September 2026, **Vercel** presented **Fluid** as a unified compute layer that dynamically provisions machines tailored to each workload. It assembles hardware, environment and storage for three uses: functions, **sandboxes** (isolated environments, notably for AI agents) and builds. Vercel says the layer is already in production: "if you've shipped on Vercel, you've been on Fluid without knowing it". ### Key facts - Vendor figures: 15 million builds a day, 25 million sandboxes a week and 1 trillion requests a month. - **Hive**: hardware provisioning, with virtual machines started in milliseconds. - **Fluid images**: custom runtime environments, distributed through Vercel Container Registry. - **Vercel Drives**: portable storage independent of compute instances, in private beta. - This announcement extends Fluid compute, already offered for functions. ### Why it matters Vercel extends to its whole platform the model of compute that adapts to the workload, including for running code generated by AI agents, a fast-growing use. For teams, the question is the billing model of these heterogeneous resources and the degree of dependence on a platform that controls hardware, storage and registry. Usage figures are the vendor's. ### Takeaway After Fluid compute for functions, Vercel generalizes the approach to its whole platform.

Tags : Serverless, Edge, Agents, Cloud