What it costs to keep an app online

About $57 for a product with sign-in, file uploads, a couple of reports and an office side, serving a business, not the public. Every figure below is a list price off the vendor's own page as of August 2026.

$25

Supabase Pro: database, sign-in and file storage

$26

Sentry Team, billed annually: knowing when it breaks

$5.70

Fly.io: one shared-CPU machine with 1 GB of memory

Push notifications are free on both platforms, at any volume: Apple and Google charge nothing to deliver them and there is no paid tier to graduate to. Nothing above is a starter tier you outgrow in a month either. The database plan carries 8 GB and 250 GB of traffic, and past that it bills by the gigabyte instead of cutting you off. The monitoring plan carries 50,000 errors a month, which is far more than a working product produces.

You can pay less. Both have free tiers, with a catch on each: Supabase takes no backups and pauses a project after a quiet week, and Sentry free is one seat and 5,000 errors a month, so exactly one person ever sees an alert. Compute is a few dollars a month whatever you do. Two more lines nobody puts in the budget: a domain is $12 to $15 a year, and a sender for password-reset email is free at this volume and about $20 a month past it.

Apple charges $99 a year and Google charges $25 once. That's on top of everything above, and only if you're shipping to the stores at all: the product priced on this page is a web application, which pays neither. The Apple one is a subscription in the strict sense: stop paying and your app comes out of the store.

Apple Developer Program

$99 a year, every year, and it's per organization, not per app. Lapsing it pulls every app you have off the App Store, which is a bad way to find out the card on file expired. What an iOS build costs covers the rest of Apple's side.

Google Play Console

$25 once, at registration, and never again. Google's ongoing requirement is engineering rather than money: apps have to target a reasonably recent Android version to stay updatable on the store, which the Android side goes into.

If you sell something used inside the app, both stores take a percentage of it, and that's a different order of number from anything on this page. What it costs to take payments has the rates and the cases where they don't apply.

Less than you'd think, and the money that growth does cost lands on the build instead. The same product described for forty customers and for forty thousand prices at $37,000 – $56,000 and $40,000 – $61,000, a difference of $4,000. The hosting bill over that range moves by a few tens of dollars a month.

$37,000 – $56,000

Built for forty customers

$40,000 – $61,000

The same product built for forty thousand

$4,000

What the difference costs, once, in engineering

Scale is a data problem before it's a server problem. The queries that stay quick at ten thousand records are a different piece of work from the ones that stay quick at ten million, and that work happens during the build whether or not the users ever arrive. Which is the argument for saying how many people will use it while somebody is still designing the thing.

Four things, and everything else is noise next to them. If your app has none of these, the bill above is roughly the bill forever.

• 01

  Video. Storing it, converting it into the formats phones want, and sending it out again. The traffic bill scales with how much people watch, which is exactly the number you were hoping would go up.

• 02

  Anything answered by an AI model. Charged per question asked, so a feature people like costs more than a feature they ignore.

• 03

  Files, when there are a lot of them and nobody ever deletes any. Storage is cheap and permanent, which is a slow way to get expensive.

• 04

  Sending email or text messages in volume. Priced per message, and a notification going to every customer is a different thing from one going to a staff member.

Two of those four also cost real money to build. Adding video to this product takes it from $37,000 – $56,000 to $41,000 – $62,000, and adding an AI feature takes it to $40,000 – $62,000. What AI actually costs in an app covers the running side of that one, which is the half that keeps going.

For any of the four, the thing worth asking for during the build is a spending cap and an alert. It costs almost nothing to put in at the start and it turns the worst case from an invoice you have to argue about into a feature that stops working for an afternoon.

Because the number in their head is maintenance, and the two get quoted as one line. Somebody says "running costs" and means the engineering to keep the app working, which runs to thousands a year. The servers underneath are the small part of that.

Hosting

$57 a month on this product. Infrastructure, monitoring and the store accounts. Predictable, itemizable, and flat until one of the four things above shows up.

Maintenance

Thousands a year, and it's engineering time. Annual OS releases, dependency upgrades, store policy changes and the defects that only appear at real volume. What app maintenance costs prices it properly.

Budget them as two lines, because they behave differently. One is a subscription that stays flat whatever you do. The other is a decision you get to make every year about how much the app changes, and in a quiet year you can decide it costs nothing.

Almost certainly not at this size. A server in your building costs less than $57 a month in electricity and more than that in the first evening somebody spends patching it.

The reasons to do it anyway are real and they aren't about price. Data that legally cannot leave a country or a building. A regulator who wants to know exactly where the disk is. An existing server room with somebody already paid to look after it. If one of those applies, say so early, because it changes how the thing gets built and not just where it ends up.

Everyone else is buying somebody else's operations team for the price of a phone contract. Backups that have been tested, patches that happen without you, and a status page to look at when something is wrong. That's what the line above is actually paying for.