Prices as of early 2026, and they will move
Treat every number here as an order of magnitude, not a quote. What will not change is the shape: the tools are cheap, the hosting is cheap, and the thing that surprises people is the bill your own app generates once strangers use it.
Building
The coding tool. Entry subscriptions sit around 20 US dollars a month — Cursor, Claude, Copilot, Replit are all near this. That is roughly ₹1,700, ₦30,000, R$110, or 380 pesos. Heavy agent use runs 100 to 200 dollars a month on higher tiers.
If you are paying per use rather than by subscription, watch the shape of it. An agent working through a large codebase reads a great deal of text, and a focused hour can cost several dollars. Long autonomous runs are where it adds up, not chat.
The free tiers are real. You can learn everything in this course on them. They are limited by volume, and you will notice.
Running
Hosting. Free until you have meaningful traffic. Vercel, Netlify, Cloudflare and Render all mean it. Paid plans start around 5 to 25 dollars a month.
Database. Free tier from Neon or Supabase covers a small app comfortably. Paid starts near 19 to 25 dollars a month.
Domain. 10 to 15 dollars a year. About ₹1,000, or ₦18,000, or 250 pesos. Buy it, and set auto-renew, because a lapsed domain that someone else buys is a genuinely bad day.
Email. Sending signup and reset emails: free up to a few thousand a month with Resend or Postmark.
Phone verification. SMS is the expensive one. Roughly 0.5 to 8 US cents per message depending on country, and India and Nigeria are not the cheap end. A thousand signups can cost more than everything above combined. This is why so many apps use email codes instead.
A typical small live app: zero to 30 dollars a month, plus the domain. That is the honest floor.
The one that surprises people
If your app calls an AI model — a chat feature, a summariser, anything — you now have a bill that grows with usage. Not with revenue. With usage.
Work an example. A chat feature where an average exchange costs about half a US cent. Two hundred conversations a day is roughly 30 dollars a month. Fine. Then someone posts your link somewhere busy, you get five thousand conversations in a day, and that day costs 25 dollars on its own. Still survivable.
Now the actual danger. One person writes a script that calls your endpoint in a loop, because it is public and there is nothing stopping it. Overnight, unattended, that is hundreds or thousands of dollars, and the money is genuinely spent.
Two limits, and you need both
A hard spending cap at the provider. Not a budget alert. An alert emails you after the money has gone; a cap refuses the request. Set it on day one, before the feature is live, at a number you could actually afford to lose.
A per-user limit in your own code. Something like: this person gets 20 AI requests a day, counted in your database, checked before you call the provider. Ten lines of code. It is the difference between a bad night and a bad month.
AI tools will not add either of these unless you ask. "Add an AI chat feature" produces a working chat feature with no ceiling, because a ceiling was not in the request.
Free tiers end quietly
Most free tiers do not stop when you cross the line. They start charging, or they suspend the service, and which one it is depends on whether a card is on file. Read that specific detail for every service you sign up to. It takes two minutes and it is the difference between a dead app and an unexpected invoice.
Also: set a calendar reminder for the day each trial converts. Trials that quietly become subscriptions are a large part of what people mean when they say building an app was expensive.
The real cost
It is not the money. Twenty to fifty dollars a month runs a real product. The real cost is the afternoon spent chasing a bug that a save point would have made irrelevant, and the week spent on a feature nobody asked for. Money is the cheap input here.
Before you move on