A Vercel bill that jumps from $20 to $180 does not mean the platform suddenly became bad. It means one resource changed scale, almost always data transfer, image transformations or function CPU. The first move is not to migrate; it is to open the Usage page and find the line that doubled. If the spike came from a traffic burst, a bot or images served at full resolution, a cache fix can bring the bill back to normal within days.

If the bill stays high after those fixes, the right alternative depends on your workload. Cloudflare Workers starts at $5 per month, does not charge for egress and handles a mostly static site well, provided you accept that Next.js runs through vinext, still in beta, or through OpenNext. Netlify supports Next.js with zero configuration through its OpenNext adapter, on a $20 Pro plan with 3,000 credits included. Render and Railway bill containers by resource, with less operational work than a server but not always a lower price. Fly.io fits containerized apps served in several regions. A $20 VPS running Coolify becomes the better calculation once you host several services and accept running the machine.

Here are the prices, limits and traps we checked in November 2026, with a comparison table and a pick for each profile.

Why the Vercel bill climbs#

The Pro plan costs $20 per month and includes one deployment seat plus $20 of monthly credit to spend across billable resources. Once that credit is gone, Vercel switches to on-demand billing with no cap by default. That transition is what surprises people: the advertised price stays $20, but the real invoice follows traffic, functions and storage. The Pro plan documentation says the credit expires at the end of the month and that alerts fire at 75% of the credit.

The heaviest line is usually Fast Data Transfer, the bandwidth between Vercel's CDN and the visitor's browser. Pro includes 1 TB per month, then charges $0.15 per GB (pricing page). A site that serves 3 TB of video, uncompressed images or PDF files therefore pays for 2 TB extra, about $300 before any function cost. Extra CDN requests start at $2 per million, and image transformations cost $0.05 per 1,000 beyond the included allowance.

Function billing changed with Fluid compute. The pricing page lists $0.128 per hour of active CPU, $0.0106 per GB-hour of provisioned memory and $0.60 per million invocations beyond the included quotas. A function that stays alive waiting on a slow database consumes provisioned memory even when its CPU is low. An API route that queries a remote database without caching can therefore cost more than a heavily visited static page. Builds cost $0.007 per minute on basic machines, and retained deployments occupy storage billed per GB-month.

The Usage page in the dashboard is the starting point. It breaks consumption down by project and region. On the command line, Vercel's documentation gives:

bash
vercel billing usage --since 30d --group-by project
vercel billing usage --since 30d --group-by region

Look at the per-project split first: a forgotten project, a branch preview or an old client site can account for half the bill. Then look at the region: transfer billed in a region far from your visitors costs more. Finally, compare cached and uncached requests. A low cache ratio on static assets usually means a bad Cache-Control header or a URL that changes on every visit.

Before changing the architecture, set a budget. Vercel applies a default $200 on-demand budget for new teams, with notifications at 50%, 75% and 100%. You can lower it and ask Vercel to pause production deployments automatically:

bash
vercel billing budgets set --amount 100 --pause --webhook https://example.com/alert

The Spend Management documentation adds two caveats. Checks run every few minutes, so pausing is not instant. And pausing cuts production traffic with a 503 DEPLOYMENT_PAUSED error, but does not delete stored resources: deployment storage, Blob storage and container registry charges keep running. A budget is a brake, not a cure.

Five levers before you switch hosts#

1. Check cache headers. Next.js assets include a hash in their filename and can be cached for a long time. Dynamic pages send private, no-cache, no-store as soon as they read a cookie or a header. If your homepage is dynamic for no good reason, it invokes a function on every visit. The Next.js self-hosting guide explains these headers. Test with:

bash
curl -sI https://your-domain.com/ | grep -i cache-control

2. Serve images at the right size. A 4,000 px photo displayed in a 400 px thumbnail costs ten times more transfer than needed. Use next/image, but check the requested sizes and the cache TTL. On Vercel, transformations are billed beyond the included allowance. A separate image CDN or images generated at build time can remove that line entirely.

3. Reduce active CPU in functions. Move work to build time when possible, cache responses that do not depend on the user, and avoid sequential network calls. In Next.js, unstable_cache, revalidate and cache directives avoid recomputing the same page. A function that initializes a heavy client on every invocation also costs more: keep initialization off the critical path or reuse the client.

4. Clean up builds and deployments. Every branch preview and retained deployment occupies storage. A short retention policy for previews and deletion of old deployments reduce the "Deployment Storage" line. Builds themselves are billed per minute on Pro: do not rebuild the site for a typo in a Markdown file if you can trigger a targeted build.

5. Keep media away from compute. Video, audio and archives do not belong in the application bundle. Put them in object storage or a CDN with cheaper egress, then serve them from a stable URL with a long cache. This cuts both Vercel transfer and function memory.

If those five actions do not bring the bill down, the problem is no longer a setting: it is the billing model. Move to the comparison.

The comparison at a glance#

CriterionVercelCloudflare WorkersNetlifyRenderRailwayFly.io Notre choixVPS + Coolify
ModelManaged PaaS built around Next.jsGlobal serverless platformManaged PaaS with an OpenNext adapterContainer PaaSContainer PaaSContainers and virtual machinesRented server plus self-hosted PaaS
Entry priceHobby $0 (non-commercial); Pro $20/month with $20 credit$5/month minimum per accountFree 300 credits; Personal $9/month; Pro $20/monthHobby $0 + compute; Pro $25/month + computeHobby $5/month with $5 usage; Pro $20 with $20No platform fee; 256 MB machine at $2.19/monthServer from €6.49/month; Coolify free
Billing unitTransfer, requests, CPU, memory, buildsRequests, CPU, storageCredits: deploys, compute, bandwidth, requestsCPU, RAM, bandwidth, disksCPU, RAM, volume, egressCPU, RAM, volumes, egressFixed server price
Next.jsNative verified adaptervinext (beta) or OpenNext; partial image optimizationOpenNext, App Router, SSR, ISR and Server ActionsNode service with next startDocker or NixpacksDockernext start or Docker: every feature
Included traffic1 TB on Pro, then $0.15/GBNo egress fees; static assets free20 credits per GB, about $0.13/GB at the Pro rate25 GB on Pro, then $0.15/GB$0.05/GB egress$0.02/GB in North America and Europe20 TB included on Hetzner in the EU
LimitsFunctions 300 s on Hobby, 300 s default on Pro, up to 800 s128 MB memory, 64 MiB Worker, 5 min CPU maximumFunction limits depend on plan and adapter512 MB on the free plan; paid tiers from 512 MBService size depends on planMachine tiers from 256 MB to several GBServer RAM, CPU and disk
OperationsNoneLowLowLowLow to mediumMedium: Docker, volumes, backupsHigh: OS, Docker, backups, monitoring
Best forA complete Next.js app with no opsA static site or high-traffic APIA complete Next.js app with minimal migrationA container or a small serviceOn-demand containerized servicesAn app close to users, multi-regionSeveral services and data control
Public prices checked in November 2026, before tax. Vercel, Cloudflare, Netlify, Render, Railway and Fly.io bill in US dollars; Hetzner bills in euros. Render and Railway prices are added to actual resource consumption.