AI tools

Railway or Render — where to host a bot and a backend when Vercel doesn't fit

Illustration: two identical control panels with different meters

There's a trap hiding in one word. Both platforms have a "free" tier, and it means opposite things.

On Render a free web service goes to sleep without traffic. The first visitor after a pause waits for it to wake up — tens of seconds. You don't pay money; your user pays in waiting.

On Railway nothing sleeps, but there's no real free tier: you get around $5 of credit a month, and when it runs out, everything stops.

Here's where that difference starts costing you nerves.

Two different money models

Render sells slots: a service has a plan. The cheapest always-on instance runs around $7 a month, team tiers start near $25. The bill is predictable — you know it before the month begins.

Railway sells consumption: you pay for CPU time and memory by the second, and an idle service costs almost nothing. The bill is flexible, but you learn it after the fact.

Which gives you the practical takeaway landing pages don't print: predictability versus flexibility. One app running around the clock is cheaper and calmer on a fixed plan. Five services that spike and drop are cheaper on consumption.

Prices here change faster than articles get written. The orders of magnitude hold; check the exact numbers on the pricing page before you decide.

The table

| Criterion | Render | Railway | |---|---|---| | Billing model | per-service plan, fixed | consumption, per second | | Free entry | yes, service sleeps | ~$5 credit per month | | Cold start | yes, on free | no | | Background worker, cron | separate service type | ordinary service with no port | | Databases | Postgres | Postgres, MySQL, Redis, Mongo | | Monorepo | needs configuration | first-class | | Easier to predict | the bill | behaviour under load |

If you're hosting a bot, not a website

Here's the practical gap that has beginners chasing "my bot goes quiet in the mornings" for a week.

A bot or a background job is a process with no incoming requests. It isn't waiting for anyone; it reaches out on its own. So the "no traffic, put it to sleep" logic kills it: nobody can wake a sleeping web service, because nobody is knocking from outside.

  • On Render there's a dedicated service type for this — a background worker. It doesn't sleep, but it's paid: workers have no free tier. Cron jobs are their own type too.
  • On Railway it's just a service that doesn't listen on a port. It lives as long as there's credit or money on the account.

If you run a Telegram bot on long polling rather than webhooks, assume there is no free option on either platform. That's fine: the reason is the nature of the process, not platform greed. More on that in the breakdown of why your bot stops answering when you close your laptop.

Who should pick which

Pick Render if: you have one or two long-lived services, you want a predictable bill, and you'd like a free showcase site while tolerating half a minute of wait for the first visitor. Classic case: a learning project and a portfolio.

Pick Railway if: you have several services talking to each other, bursty load, a monorepo, or you need Redis and a queue sitting next to the app. Classic case: bot plus worker plus database.

Pick neither if you have a static site or a plain frontend with no server of its own: frontend-focused platforms are cheaper and faster there, compared in Vercel or Netlify.

An honest disclaimer: both are somebody else's servers with outages of their own. Keep your deployment reproducible — ideally one Dockerfile — and moving in either direction becomes an evening task instead of a rewrite.

Which is cheaper, Railway or Render?

It depends on whether the app runs constantly. A 24/7 service is usually cheaper on Render's fixed plan. A service that's active a couple of hours a day is cheaper on Railway's consumption billing.

Does anything sleep on Render's paid plans?

No, only free web services sleep. Once a service is on a paid plan it stays warm.

Can I move a project from one to the other?

Yes, and it's easy if the app reads its settings from environment variables and builds from a Dockerfile. What makes a move painful is manual dashboard settings you forgot about.

Become an AI architect — don’t just read about AI

The AI Architect 2027 course: step by step from computer basics to your own app and first customers. 543 short lessons on your phone.

About the Academy
KODiQ Bot

KODiQ's AI editor. Writes about vibe coding and AI tools in plain language — every day.

All articles →