Vercel alternatives for Next.js without surprise bills

Vercel Pro now offers a flat rate for traffic: $20 a month covers 50 TB and 10 million requests, so bandwidth rarely blows up a Vercel bill. Function CPU and memory, image transforms, cache writes, builds and seats still run on a meter. The Vercel alternatives without surprise bills are the ones where you pay for a server.

That's us, Hetzner and DigitalOcean Droplets, and on ours traffic isn't metered at all. Leaving Vercel to save on bandwidth is now often a mistake: Render, Netlify, AWS Amplify and Railway all charge more for 5 TB than Vercel's flat rate does.

Vercel pricing in 2026: what's still metered

Vercel Pro is $20 a month. You get the same amount back as usage credit, plus one deploying seat, and every extra paid seat is another $20 a month.

The Flat Rate CDN is a setting in Billing: you turn it on and pick a tier. Pro includes the lowest tier free: 1 TB of data transfer and 1 million CDN requests a month. The $20 tier raises that to 50 TB and 10 million requests. The $100 and $300 tiers keep the same 50 TB and only raise the request count. A one-day traffic spike doesn't change the bill. The catch is the fair-use list: bulk file distribution, download services and sites where media is most of the bandwidth aren't eligible, and Vercel says it may move those projects to its Flex network, require another plan or limit service. If your traffic is mostly video or downloads, it's still a reason to leave.

Everything else is billed by use. In Washington, D.C., on-demand rates are:

• Active CPU, $0.128 an hour, counted only while your code runs
• Provisioned memory, $0.0106 per GB-hour, counted while a function handles requests, waiting on I/O included
• Function invocations, $0.60 per million
• Image transforms from $0.05 per 1,000, ISR writes from $4 per million, and builds on Standard machines $0.014 a minute

Here's where the meter crosses a fixed server. Our 2 vCPU, 4 GB server is $43.13 a month. On Vercel, $43.13 buys about 337 hours of Active CPU, roughly 11 a day, with memory and invocations billed on top. The server runs two vCPUs for all 720 hours of the month, busy or idle. Once your server-rendered pages and API routes keep a CPU busy more than 11 hours a day, the meter is the expensive way to buy that time.

The meter also counts traffic you didn't want. One user on Vercel's community forum reported "a spike of over $1,900" in function charges in one month from an AI scraper crawling a staging deployment, at 923% of a $200 budget.

That $200 is the default spend management budget for new teams, and it only alerts you. To stop usage, you switch on Pause Production Deployments. Vercel checks spend every few minutes, so the pause can land minutes after you cross the line. Paused sites return a 503, and each project has to be resumed by hand.

Best Vercel alternatives for Next.js hosting: what each one meters

Prices are list prices for US regions on October 20, 2026. The traffic column is what 5 TB (5,000 GB) going out in a month adds on top of the plan. On a phone, swipe the table sideways to see every column.

Where a platform sells servers, the row uses its dedicated-CPU size closest to 2 vCPU and 4 GB; Render's page doesn't say whether its CPUs are shared. Netlify is priced at its $10 add-on rate. "Not stated" means we didn't find it on the vendor's pricing page.

American Cloud

Pick us if you want the bill set the day you choose the server, and a person to talk to when something breaks. Our median support reply time is under 2 minutes.

You run Next.js as a normal Node server on a virtual machine you pay for by the month, and outbound traffic isn't metered on servers, Kubernetes, load balancers or object storage. The same account has managed PostgreSQL, MySQL and Redis, S3-compatible object storage at $0.02 per GB a month for uploads and a shared cache, and managed Kubernetes for when one server isn't enough. If you want git-push deploys, add Coolify. We run three US regions: us-west-0, us-west-1 and us-central-0.

The Migrate from Vercel recipe has your AI assistant read the repo, map each Vercel feature, price the server, deploy it and switch DNS while Vercel keeps serving.

Hetzner Cloud pricing in the US

The CCX13 in Ashburn or Hillsboro is $50.99 a month for 2 dedicated vCPU, 8 GB and 80 GB, not counting an IPv4 address. US servers include 1 TB a month where the same server in Europe includes 20 TB, and each extra TB is $1.20. Hetzner's traffic alerts at 75% and 100% don't stop the traffic.

DigitalOcean Droplets and App Platform pricing

Two products with different meters. A CPU-Optimized Droplet with 2 vCPU and 4 GiB is $42 a month with 4,000 GiB of transfer pooled across your team, then $0.01 per GiB. App Platform builds and runs your container for you: the dedicated 2 vCPU, 4 GiB size is $78 a month with 500 GiB of transfer, then $0.02 per GiB. Managed PostgreSQL starts at $15.15 a month. DigitalOcean's spend alerts email you when spending crosses a budget you set.

Cloudflare Workers with OpenNext

Cloudflare runs your app as a Worker, not on a server. The Paid plan is $5 a month minimum, and Cloudflare charges nothing for data transfer. You pay for requests past 10 million ($0.30 per million) and CPU time past 30 million milliseconds ($0.02 per million ms). Next.js runs there through the OpenNext adapter, which supports Next.js 16 and 15, but not Node middleware yet, and a Worker is capped at 10 MiB compressed on the Paid plan. Postgres lives elsewhere: Workers reach it through Hyperdrive.

Fly.io

Fly.io bills machines and traffic separately. A performance-2x machine with 4 GB in Ashburn is $66 a month, and outbound traffic in North America is $0.02 per GB with no free allowance stated. New organizations get no free tier. Managed Postgres starts at $38 a month.

Railway pricing and usage limits

Railway Pro is a $20 monthly minimum that all counts toward usage, and everything is billed by the second: about $20 per vCPU-month, $10 per GB of memory a month, and $0.05 per GB of traffic. Railway gives you two controls: an email alert, or a hard limit that takes all your workloads offline when you hit it.

Render pricing: bandwidth and compute

Render's Pro workspace is $25 a month plus compute, and a 2 CPU, 4 GB web service is $85 a month. Traffic is the line to watch: Pro includes 25 GB, then it's $0.15 per GB. Render Postgres with 1 GB of RAM is $19 a month.

Netlify

The closest to Vercel's workflow. Pro is $20 a month with unlimited members and 3,000 credits. Everything draws on the same credits: a GB of traffic is 20, a GB-hour of compute is 10, and a production deploy is 15. Extra credits are $10 for 1,500, about $0.13 per GB of traffic. When one project goes past its limit, every project on the account pauses until the next cycle, unless you've turned on auto recharge.

AWS Amplify

Amplify Hosting has no plan fee; you pay $0.15 per GB served after the first 15 GB, $0.30 per million server-rendered requests plus their duration, and $0.01 per build minute. The database is a separate AWS service with its own bill.

Coolify pricing: self-hosted or Coolify Cloud

Coolify is the open-source answer to "I want Vercel's workflow on my own server." Self-hosted it's free, with every feature. Coolify Cloud runs the control panel for you at $5 a month for two servers and $3 for each one after that, and you still bring the servers. Either way you get git-push deploys, preview deployments for pull requests, HTTPS, databases and one-click services. Dokploy does a similar job.

Coolify runs on a standard Linux server. On ours, the one-click marketplace image installs it when the VM is created, and the Coolify recipe has your AI assistant size the server and set it up. Then every app and teammate you add lands on the same server bill.

Self-hosting Next.js with Docker: what changes past one server

Next.js doesn't need Vercel. The deployment docs say a Node.js server "supports all Next.js features," and so does Docker. On one server it's simple: next start with a persistent disk handles the ISR and data cache on its own, and next/image optimizes images with no setup, though on glibc-based Linux it may need tuning to keep memory down.

The self-hosting guide adds one step for any production setup and three once you run several instances:

• Put a reverse proxy such as nginx in front of the server, and turn off its buffering so streaming works.
• Several instances each keep their own cache, so configure a shared cacheHandler backed by Redis or S3.
• Give every instance the same NEXT_SERVER_ACTIONS_ENCRYPTION_KEY, or Server Actions fail across instances.
• Set a deploymentId so a rolling deploy doesn't mix old and new builds.

What you give up is Vercel doing this for you: previews, the build pipeline and the proxy become yours. Coolify covers the previews and builds. Our Next.js deploy recipe covers the server, nginx and HTTPS.

To get your own number, run your stack through our pricing calculator.

Frequently asked questions