# Cloudflare Workers vs Vercel Functions in 2026

Last updated: 2026-11-10

Pick **Cloudflare Workers** when your code is light, should answer near every visitor and can live with 128 MB of memory per isolate. Pick **Vercel Functions** when you need full Node.js, Python or another runtime, several gigabytes of memory, or a Next.js app that leans on heavy Node.js dependencies. If both your users and your data are in Europe and you want a European company, **Eurobase** runs the Workers model on European infrastructure at a flat price.

## Workers and Vercel Functions at a glance

| Platform | What runs your code | Where it runs by default | Memory | Paid entry price |
|---|---|---|---|---|
| Cloudflare | V8 isolates | The data center closest to the request, in 348 cities | 128 MB per isolate | $5 a month per account, then by use |
| Vercel | Functions isolated in microVMs, with Fluid compute | Washington, D.C., iad1, until you change it | Up to 2 GB on Hobby, 4 GB on Pro | $20 a month on Pro, with $20 of usage credit |
| Eurobase | V8 isolates for framework apps, containers for servers | European infrastructure | 128 MiB of JavaScript heap | Flat €10 or €20 a month + VAT |

Prices and limits are from each provider's own pages, checked on 10 November 2026.

## What actually decides it

- **How much memory and CPU one request needs.** A Worker gets 128 MB and, on the paid plan, up to 5 minutes of CPU per request. A Vercel Function on Pro gets up to 4 GB and up to 800 seconds of run time, or 1,800 seconds in beta.
- **Which runtime your code expects.** Workers speak JavaScript, TypeScript, Python and Rust, plus WebAssembly, with Node.js APIs added through a compatibility layer. Vercel runs real Node.js, Bun, Python, Go, Ruby, Rust and containers.
- **How many calls a page makes to its database.** Code next to the visitor is fast only if it doesn't wait on a far-away database several times per page. Read [edge vs regional hosting](/i/edge-vs-regional-hosting) before you choose on latency alone.
- **What you pay for idle time.** Cloudflare charges for CPU time, not for waiting. Vercel bills active CPU only while code runs, but bills memory for as long as an instance has a request in flight.
- **Who the provider is.** Cloudflare, Inc. and Vercel Inc. are both US companies, and both sets of terms are governed by California law.

## 1. Cloudflare

Cloudflare creates each isolate inside an environment that's already up, instead of starting a virtual machine per function. Unless you configure placement, the Cloudflare location nearest the visitor handles the request. To sit next to your backend instead, turn on Smart Placement or give a hint that names a cloud region or a database host.

The paid plan includes 10 million requests and 30 million CPU milliseconds a month, then charges $0.30 per extra million requests and $0.02 per extra million CPU milliseconds. Requests to static assets are free, and there's no charge for egress. Time spent waiting on a fetch or a database doesn't count as CPU time, and an HTTP request has no wall-clock limit while the client stays connected. Cloudflare's own example puts a Worker serving 15 million requests at 7 ms of CPU each at $8 a month.

- **Best for:** APIs, middleware, auth checks and pages that render quickly for a worldwide audience, especially with D1, KV, R2, Durable Objects or queues.
- **Price:** a Free plan with 100,000 requests a day and 10 ms of CPU per request, then $5 a month per account plus usage.
- **What it lacks:** room for heavy work. Memory stops at 128 MB per isolate, a Worker bundle at 64 MiB, and startup must finish within 1 second.

## 2. Vercel

Vercel Functions turn your framework's server code into functions that run in one region. New projects run in Washington, D.C. until you change it. Hobby gets one region, Pro up to five, and Enterprise all of them. Its European regions are Dublin, Frankfurt, London, Paris and Stockholm.

Fluid compute, on by default for new projects since April 2025, lets one instance serve many requests at once. It also caches compiled bytecode on Node.js 20 and later and pre-warms production functions. You pay for three things: active CPU per hour, provisioned memory per GB-hour and invocations. In Frankfurt that's $0.184 an hour of active CPU and $0.0152 per GB-hour of memory. On Pro, invocations cost $0.60 per million. The Edge runtime still exists, but Vercel now recommends moving from edge to Node.js, and Next.js 16.3 drops `runtime = 'edge'`.

- **Best for:** Next.js and other full-stack apps that use Node.js libraries, large dependencies or long-running AI calls.
- **Price:** Hobby is free with 4 CPU-hours, 360 GB-hours and 1 million invocations a month, but only for personal, non-commercial use. Pro costs $20 a month with $20 of usage included, then usage on demand.
- **What it lacks:** code running near every visitor by default, since a function lives in the regions you pick. Functions are archived after 2 weeks without traffic in production, which adds at least a second to the next cold start.

## 3. Eurobase

Eurobase takes the same build a Worker uses, your Wrangler config plus Cloudflare adapters such as OpenNext for Next.js, and serves it from V8 isolates. Long-running servers go into containers, and Eurobase Functions cover small TypeScript handlers. The company is registered in Warsaw, Poland, and your apps run on European infrastructure run by Scaleway, a French provider.

Per request, an app gets 50 ms of CPU on the Free plan or 30 s on paid plans, and each instance has up to 128 MiB of JavaScript heap. On top of that, the workspace draws on one monthly CPU allowance: 50 minutes on Free, 500 on Starter and about 1,667 on Pro. There are no usage charges today. Work over a limit is refused, isolate apps and Functions stop until the monthly reset, and containers and static sites keep serving.

- **Best for:** Workers-style apps with a European audience whose data must stay in Europe, deployed by you or a coding agent through the MCP server and CLI.
- **Price:** a Free plan, then Starter at €10 or Pro at €20 a month + VAT for the whole workspace.
- **What it lacks:** KV, R2, Durable Objects and queues, a working Cache API, preview deployments and compute outside Europe. ISO 27001 and SOC 2 certification hasn't finished yet.

Moving over? There are guides to [migrate from Cloudflare Workers](/docs/guides/migrate-from-cloudflare-workers) and [migrate from Vercel](/docs/guides/migrate-from-vercel).

## How to choose

1. Measure one real request: its CPU time, its peak memory and how many database calls it makes.
2. If it fits in 128 MB and makes one database call or none, Cloudflare Workers will usually answer faster for a worldwide audience.
3. If it needs more memory, a native Node.js library or minutes of wall time, Vercel Functions fit better. Set the region next to your database.
4. If your users and database are in Europe and the contract asks for a European provider, compare both with [European Cloudflare Workers alternatives](/i/european-cloudflare-workers-alternatives) and [European Vercel alternatives](/i/european-vercel-alternatives).

## Frequently asked questions

### Is Cloudflare Workers cheaper than Vercel Functions?

For light, high-volume traffic it often is. Cloudflare's paid plan starts at $5 a month with 10 million requests included, and you never pay for time spent waiting on I/O. Vercel Pro starts at $20 a month and bills memory for the whole time an instance handles requests, so long AI or database calls can cost more there.

### Can I run Next.js on Cloudflare Workers?

Yes. Cloudflare now recommends vinext, which is in beta, and still documents the OpenNext adapter for existing apps. Node.js APIs come through Workers' compatibility layer, and the 64 MiB size and 128 MB memory limits apply, so test a large app before you move it.

### Do Vercel Functions run at the edge?

Not by default. They run in Washington, D.C. or the regions you choose, while static files come from Vercel's CDN. Vercel's Edge runtime runs closer to visitors, but Vercel now recommends Node.js instead.

### Which one has fewer cold starts?

Workers start an isolate rather than a machine, so their starts are short, though your own startup code still has to run within 1 second. Vercel cuts cold starts with bytecode caching and pre-warming on production deployments. See [serverless cold starts](/i/serverless-cold-starts) for the detail.

### Can I keep everything in Europe with either one?

You can pick European regions on Vercel and place a Worker near a European database, but both companies are based in the US. For a provider registered in the EU, look at [Eurobase vs Cloudflare](/compare/cloudflare) and [Eurobase vs Vercel](/compare/vercel).

## Sources

- [Cloudflare Workers pricing](https://developers.cloudflare.com/workers/platform/pricing/) and [Workers limits](https://developers.cloudflare.com/workers/platform/limits/)
- [How Workers works](https://developers.cloudflare.com/workers/reference/how-workers-works/), [Next.js on Workers](https://developers.cloudflare.com/workers/framework-guides/web-apps/nextjs/), [Workers placement](https://developers.cloudflare.com/workers/configuration/placement/), [Workers languages](https://developers.cloudflare.com/workers/languages/) and [Cloudflare network](https://www.cloudflare.com/network/)
- [Cloudflare terms](https://www.cloudflare.com/terms/)
- [Vercel Functions limits](https://vercel.com/docs/functions/limitations), [Vercel Functions pricing](https://vercel.com/docs/functions/usage-and-pricing) and [Vercel pricing](https://vercel.com/pricing)
- [Vercel runtimes](https://vercel.com/docs/functions/runtimes), [Fluid compute](https://vercel.com/docs/fluid-compute), [Edge runtime](https://vercel.com/docs/functions/runtimes/edge), [Function regions](https://vercel.com/docs/functions/configuring-functions/region) and [Vercel regions](https://vercel.com/docs/regions)
- [Vercel Hobby plan](https://vercel.com/docs/plans/hobby) and [Vercel terms](https://vercel.com/legal/terms)
- [Eurobase Workers docs](/docs/workers), [Eurobase plans](/docs/plans) and [Eurobase subprocessors](/legal/subprocessors)