Vercel vs Cloudflare: Where to Host Your SaaS in 2026
Vercel vs Cloudflare for a SaaS: Node functions and Next.js on Vercel, or Workers, R2, and TanStack Start on Cloudflare. How to pick in 2026.

Vercel vs Cloudflare is the hosting decision for a Next.js or TanStack Start SaaS: where the server code, the files, and the preview deploys should live. Vercel runs Fluid compute, full Node.js, in a region you choose. Cloudflare Workers runs V8 isolates on its own network, with D1, R2, KV, and Durable Objects on the same account, and no Workers egress fee.
Prices below were read on 16 November 2026 from vercel.com/pricing and the Workers pricing page. Limits are from Vercel Functions limits and Workers limits. Where two Vercel pages disagree, both figures are in the pricing section.
Short answer
Choose Vercel when the app is Next.js, you need full Node.js and more than 128 MB of memory, and a small team can live with a per-seat Pro plan. Choose Cloudflare when the work is mostly I/O, you want R2 without an egress line, and the server fits in a 128 MB isolate (TanStack Start, or Next.js through an adapter). Paid entry is Vercel Pro at $20/month or Workers Paid at $5/month per account. Those are plan fees, not a forecast of the bill.
| Choose Vercel if | Choose Cloudflare if |
|---|---|
| Next.js is the app and you want its native deploy path | Requests spend their time on fetch, the database, and files |
| You need full Node.js, a bundle over 64 MiB, or more than 128 MB of memory | Download egress would dominate the bill (R2 lists internet egress as free) |
| One or two people deploy, and Postgres can stay an external database | You are starting on TanStack Start, Astro, or React Router, or you want one account fee instead of a seat per deployer |
Quotable definition: Vercel hosts the app on its delivery network and runs server code as Fluid compute functions (full Node.js, one region by default). Cloudflare Workers runs that code in V8 isolates across its network and bills CPU time, not wall-clock duration.
Vercel vs Cloudflare at a glance
| Vercel | Cloudflare Workers | |
|---|---|---|
| Runtime | Fluid compute, full Node.js, region iad1 by default. Pro and Enterprise can set more than one region | V8 isolates on Cloudflare's global network (how Workers works) |
| Next.js | Native | vinext (beta, Cloudflare's recommended path) or the OpenNext adapter |
| Other frameworks | A Node server can run here. Next.js is the path Vercel documents most deeply | TanStack Start, Astro, React Router, and others via automatic configuration |
| Memory | Default 2 GB / 1 vCPU. Max 2 GB on Hobby, 4 GB / 2 vCPU on Pro | 128 MB per isolate, on Free and Paid |
| Time limit | Default 300s. Max 300s Hobby, 800s Pro. An 1800s extended maximum is in beta | No HTTP wall-clock cap while the client stays connected. CPU time: 10 ms Free, 30s default on Paid, 5 min max on Paid |
| Bundle | 250 MB uncompressed. Large functions (beta) go to 5 GB | 64 MiB uncompressed |
| Data | Blob, Global Config, Marketplace integrations | D1, R2, KV, Durable Objects, Queues, Workflows, Hyperdrive. Workers AI is a binding in the Next.js guide |
| Paid entry | Pro, $20/month | Workers Paid, $5/month per account |
| Bandwidth | Fast Data Transfer: 100 GB Hobby, 1 TB Pro, then starting at $0.15/GB | No Workers egress charge. R2 egress is free. Static asset requests are free and unlimited |
| Seats | Extra deploying seats are $20/month. Viewer seats on Pro are free | The Workers fee is per account, separate from Cloudflare zone plans |
| Previews | Unlimited deployments on Hobby and Pro | Workers Builds: preview command defaults to npx wrangler preview and does not promote to production |
| Hard to leave | Blob, Global Config, image optimization, the Vercel cache | D1, Durable Objects, KV, and cloudflare:workers bindings |
A third path is running Next.js yourself. The Docker self-host guide is that option. The rest of this post stays on the managed choice.
Runtime: Fluid compute vs Workers isolates
Vercel Functions and Fluid compute
The numbers are in the table above, from function limits. What the table does not say: those size limits are enforced by AWS, and Vercel's compliance docs say the CDN and deployment platform primarily use AWS, across 19 regions. Active CPU excludes I/O waits. Provisioned memory includes them, for the whole time the instance is up, at 2 GB unless you change it. Bodies are capped at 4.5 MB, except streams. A Postgres driver or a native addon fits. You still pay for that 2 GB while the handler waits on the database.
Workers: isolates, CPU time, 128 MB
How Workers works uses V8 isolates on Cloudflare's network. An isolate starts inside an existing process and can be evicted, so global mutable state is a bad session store. Limits (13 November 2026) match the glance table, plus a subrequest cap of 50 on Free and 10,000 on Paid.
CPU time ignores fetch(), KV, and database waits. Cloudflare says most Workers use about 2.2 ms, and that auth, SSR, or a large payload typically uses 10 to 20 ms. That is their description, not a number we measured, and it sits on top of the Free plan's 10 ms cap. waitUntil() lasts 30 seconds after the response. Crons, queues, and Durable Object alarms cap wall time at 15 minutes.
Node here is a compatibility layer. From compatibility date 2026-08-04, nodejs_compat is on by default (docs). Many built-ins are real. Polyfills throw when called. A native addon, node:child_process, or a huge heap does not move unchanged. Crossing 128 MB returns error 1102: stream the body, and keep large objects in R2, KV, or D1.
Framework support, including Next.js on Cloudflare
Next.js on Vercel
Next.js on Vercel is the native case: git deploy, server functions, image optimization, and ISR are platform features, each with a meter on the pricing page. Our Vercel deployment docs are that path for the Next.js kit.
Next.js on Cloudflare: vinext and OpenNext
Cloudflare's Next.js guide names vinext as the default way to run Next.js on Workers. It is a Vite plugin that keeps app/, pages/, next.config.js, and public/, and the page says it is in beta. Run npx vinext check before a production app. The guide marks App Router, Pages Router, server components, server actions, streaming SSR, static generation, ISR, and middleware as supported. Image optimization is partial. Bindings come from cloudflare:workers. The same page still documents the OpenNext adapter for apps that cannot move yet, and a static export on Pages when you specifically want Pages.
Automatic configuration (updated 5 November 2026) splits the tools:
| Tool | Next.js 16 and later | Next.js 15 and earlier |
|---|---|---|
wrangler setup or wrangler deploy | OpenNext | OpenNext |
cf init | vinext | OpenNext |
"Cloudflare recommends vinext" and "Wrangler still wires OpenNext" are both current. To use vinext with Wrangler, that page says to run vinext init yourself.
TurboStarter's Next.js kit follows OpenNext today. The Cloudflare docs and the Workers deploy guide use OpenNext, Hyperdrive, and an R2 bucket for the incremental cache. We have not moved that kit to vinext.
TanStack Start, Astro, and React Router
TanStack Start is a first-party Workers guide: the Cloudflare Vite plugin, main set to @tanstack/react-start/server-entry, and wrangler deploy. The Rsbuild variant is not supported. The same automatic-configuration page lists Astro (@astrojs/cloudflare) and React Router (Cloudflare Vite plugin). TanStack Start vs Next.js is the framework tradeoff under our two kits. TanStack Start does not need a Next adapter. Next.js on Workers does.
Vercel vs Cloudflare pricing for a SaaS
No invented monthly total below. Dynamic pages, database wait, deployer count, and where files live all move the bill. The list prices show which line.
What you pay to walk in the door
| Vercel Hobby | Vercel Pro | Workers Free | Workers Paid | |
|---|---|---|---|---|
| Plan fee | $0 | $20/month, with 1 deploying seat and $20 of usage credit (Pro plan) | $0 | $5/month per account |
| Extra people | 1 seat | $20/month per extra Owner or Member. Viewers are free | No Workers seat line | No Workers seat line |
| Requests | 1M function invocations | Starting at $0.60 per 1M, against the credit first | 100,000/day | 10M included, then $0.30 per million |
| Compute | 4 Active CPU hours, 360 GB-hours provisioned memory | Active CPU from $0.128/hour. Provisioned memory from $0.0106 per GB-hour | 10 ms CPU. No duration charge | 30M CPU ms included, then $0.02 per million CPU ms. No duration charge |
| Egress | 100 GB Fast Data Transfer | 1 TB, then from $0.15/GB on the pricing page | No Workers egress charge | No Workers egress charge |
A Vercel team of four deployers is an $80 seat line before a request is served. Workers Paid stays $5 for the account. That Workers fee is separate from any Cloudflare zone plan (Free, Pro, Business, Enterprise).
Vercel's two CDN numbers
On 16 November 2026 the pricing page lists Pro CDN requests as 10 million included, then from $2 per million, and Fast Data Transfer as 1 TB, then from $0.15/GB. The Pro plan docs describe the included Flat Rate CDN tier as 1 million requests and 1 TB. Both agree on 1 TB of transfer. They disagree on included requests. Check the tier on your team before you model a bill.
Pro rates are "starting at" because the region changes the price. The default region is iad1.
What moves as traffic grows
On Vercel the moving lines are Active CPU (code only, not the Postgres wait), provisioned memory for the whole instance lifetime at 2 GB unless you change it, invocations at $0.60 per million after the credit, and Fast Data Transfer after the included terabyte. Fast Origin Transfer, between the delivery network and your functions, starts at $0.06/GB on Pro (10 GB on Hobby). Seats ignore traffic: the second deployer costs as much as the platform fee.
On Cloudflare the Workers bill moves on requests and CPU milliseconds. Duration is not a meter, and static asset requests are free, so a marketing site in front of a small API does not pay for files that never enter the Worker. Cloudflare's own example: 15 million requests at 7 ms of CPU is $5 + $1.50 of requests + $1.50 of CPU = $8. That is their example, not a model of your app. The Free plan stops at 100,000 requests a day (error 1027) or 10 ms of CPU (error 1102). A spending limit can block new invocations account-wide. The $5 fee itself sits outside that limit.
Files are the sharp split. R2 standard storage is $0.015 per GB-month after 10 GB-month free, with free internet egress. Class A operations are $4.50 per million and Class B are $0.36 per million after 1 million Class A and 10 million Class B free. Vercel Blob on Pro is $0.023 per GB, $0.40 per million simple operations, and $5 per million advanced operations, with transfer included in Flat Rate CDN. Large downloads favor R2. HTML-heavy apps feel CPU, memory, and seats first.
For a worked Vercel-plus-Supabase bill, use Supabase and Vercel pricing. This post does not repeat those totals.
If vercel alternative means the seat line or the bandwidth line, Workers is the usual candidate. If it means "I need a real Node process and a large heap," a container host is the closer fit. Workers does not become that by spending more.
Data and platform primitives
On Vercel
On the pricing page, Blob is object storage (Hobby: 1 GB, 10,000 simple operations, 2,000 advanced operations, 10 GB transfer. Pro prices are above). Global Config is the config store (Hobby: 100,000 reads and 100 writes. Pro: from $3 per million reads and $1 per 100 writes). Marketplace integrations are how other vendors attach. This post does not inventory that catalog. Workflows include 50,000 events on Hobby and cost $20 per million on Pro. Postgres is yours to bring, usually as DATABASE_URL. Nothing on that page is a SQL database.
On Cloudflare
From the Workers pricing page:
- D1 is SQLite. Paid includes 25 billion rows read, 50 million rows written, and 5 GB stored, then $0.001 per million rows read, $1 per million rows written, and $0.75 per GB-month. No egress charge. Limits cap one database at 10 GB on Paid and 500 MB on Free, and the 10 GB cap cannot be raised. Each database is single-threaded. See the D1 and Drizzle guide.
- R2 has an S3 API and free internet egress. KV on Paid includes 10 million reads (then $0.50 per million) and 1 million writes (then $5 per million).
- Hyperdrive pools connections to an outside database: 100,000 queries a day on Free, unlimited on Paid. Queues bill about three operations per message. Paid includes 1 million, then $0.40 per million.
- Durable Objects have their own request and duration prices, billed as 128 MB per object. Workflows share Workers request and CPU rates, plus storage and steps. Cloudflare says it will not bill step and storage before the date in the Workflows changelog. Workers AI is a binding in the Next.js guide. No inference price here.
Vercel expects the database, and often the files, to be another vendor. Cloudflare sells SQLite, objects, KV, queues, and stateful objects on the same account, plus Hyperdrive when you still want Postgres.
Cloudflare Pages vs Workers
Start new projects on Workers. The Pages overview says Workers covers most Pages use cases, has the broader feature set, and is the primary applications platform. Static files ship as Workers static assets, and those requests are free and unlimited. Pages Functions are already billed as Workers.
The migration guide adds Durable Objects, Cron Triggers, and a wider observability surface. Routing that Pages inferred (an SPA versus 404.html) becomes explicit config, and each Worker version has its own assets, so a rollout can briefly mix versions. Keep Pages for a project you have not moved. A new SaaS should not start there.
Developer experience
Git and previews. Vercel includes automatic CI/CD, environment variables, build logs, and unlimited deployments on Hobby and Pro. A custom preview suffix is $100/month, so the base URL is vercel.app. Workers Builds connects Artifacts, GitHub, or GitLab. A push runs an optional build, then npx wrangler deploy by default (configuration). The dashboard name must match name in Wrangler. Previews use npx wrangler preview and do not promote. A repo with no Wrangler file gets a pull request plus a preview.
Logs and local dev. Workers Logs keep 200,000 events a day for 3 days on Free, and 20 million a month (then $0.60 per million) for 7 days on Paid. On 1 December 2026 they move to Cloudflare Observability pricing. Vercel includes build logs. A Vercel function-log price is not on the pricing page, so it is not quoted here. Locally, Workers use wrangler dev (wrangler dev --remote caps a zone at 50 routes). Vercel uses its CLI against the git project. A laptop Next.js dev server will not catch a missing Node API or a 128 MB blow-up. Test in the runtime you will ship.
Lock-in, and moving later
From Vercel to Cloudflare
A Next.js app on ordinary Node, with Postgres behind a connection string, redeploys: DNS, env vars, Hyperdrive in front of the database. The Workers deploy guide is that move for our kit, with R2 for the OpenNext cache. You replace Blob and Global Config, and you re-test image optimization and ISR. A 2 GB, 250 MB function does not fit a 128 MB isolate and a 64 MiB script. Run wrangler dev, and npx vinext check if you take Cloudflare's recommended path. Our OpenNext guide also says to remove export const runtime = "edge".
From Cloudflare to Vercel
If the only Cloudflare-specific piece is the deploy, change the command and attach Postgres. D1 stays behind: SQLite, 10 GB paid cap, and Vercel will not host it. Durable Objects, KV, and cloudflare:workers bindings become env vars and HTTP clients. R2 can stay via the S3 API, but bytes a Vercel function sends to users count as Fast Data Transfer. External Postgres and S3-compatible files keep both doors open. D1 and bindings mean the app lives on Workers.
How this maps to TurboStarter
We ship both shapes.
Core Kit is the Next.js monorepo (web, plus Expo and a browser extension). It deploys to Vercel from the Vercel docs and to Workers from the Cloudflare docs, using OpenNext, Hyperdrive, and R2. Auth and billing sit on your Postgres, so the host is a deploy target. The Docker guide is the exit that keeps Node.
Edge Kit is the kit built for Cloudflare: TanStack Start, D1, R2, Queues, one Worker. The introduction lists what is in the box. Use it when the product is web-only and the data fits D1, or Hyperdrive to Postgres later. Skip it if you need Core's mobile app and extension.
Choosing a starter, rather than a host, starts at the kit comparison. The Vercel-plus-Postgres bill is the Supabase and Vercel breakdown.
Core Kit is the Next.js kit when the same app should deploy to Vercel, Cloudflare, or a container.
Questions people ask before they pick
The entry fee is lower: Workers Paid is $5 per account, Vercel Pro is $20 with one deploying seat and $20 of credit, and each extra Vercel deployer is $20. Workers and R2 list no egress charge. Vercel Pro includes 1 TB of Fast Data Transfer, then starts at $0.15/GB. Which bill is lower still depends on CPU, static vs dynamic traffic, 2 GB instance time, and seats. See Vercel pricing and Workers pricing.
Yes. Cloudflare recommends vinext and marks it beta: run npx vinext check first. wrangler deploy still configures OpenNext, which is the path in our Cloudflare deploy guide.
AWS. Security docs say the CDN and deployment platform primarily use Amazon Web Services, across 19 regions. The 20 October 2025 incident write-up says compute, CDN, and firewall run on those regions, and that the outage started in AWS us-east-1. The trust center also names Azure and Google Cloud, including Azure Cosmos DB for vercel.com. Those pages do not describe Cloudflare as the network under Vercel.
Workers, for a new project. The Pages docs say so. Static files are Workers static assets, and those requests are free. Pages Functions are billed as Workers. Keep Pages only for a project you have not migrated.
Vercel, when you want Next.js on full Node.js, more than 128 MB, and the short preview path. Cloudflare, when you want isolates, R2 without egress fees, D1 or Durable Objects, and a per-account price. TanStack Start is a first-class Workers target. Next.js on Workers goes through vinext (beta) or OpenNext.
Yes, if the app is Next.js and Postgres can sit behind Hyperdrive. Re-test 128 MB, the 64 MiB script, and Node compatibility, and replace Blob, Global Config, and Vercel image and cache behavior. Our deploy guide is that move. Leaving is harder once data is in D1 or coordination is a Durable Object.
Cloudflare's own guide runs it with the Vite plugin and Wrangler, with D1 and R2 as bindings. That is what Edge Kit ships. Use it for a web-only product. Use Next.js when you also need the Node-shaped kit. See TanStack Start vs Next.js.
Vercel vs Cloudflare, the decision
Vercel vs Cloudflare is the runtime you will own for the next six months. Pick Vercel when Next.js needs full Node and a $20 seat per deployer is acceptable. Pick Cloudflare when the request mostly waits on I/O, 128 MB is enough, and you want storage plus D1 or Hyperdrive on one account with no Workers egress line. Pick a container when you need Node and neither meter. If it is still open, ship the Next.js app on Vercel first. The Workers move is cheaper once you know the memory and the Node dependencies.
Turborepo - the ultimate guide to modern JavaScript monorepo management
Master Turborepo for monorepo management with setup tips, Turborepo vs Nx guidance, caching strategies, and best practices for faster JavaScript builds.
What Is an AI Wrapper? How to Build One That Isn't Just a Wrapper (2026)
AI wrapper meaning, examples, and defensibility in plain language, plus a production architecture with Vercel AI SDK, auth, credits, and one backend for web, mobile, and extension.



