What does it cost to run an AI-built SaaS app in production in 2026?

Most people budget for building the app and not for running it, which is understandable — the building part is what you were quoted for, and the running part arrives quietly, one twenty-dollar line at a time.

This is the whole list, with real 2026 prices, for a small SaaS that takes real money from real customers. Not an enterprise stack, not a minimum viable one: the actual set of things that are hard to argue against once you have paying users.

Prices below were checked against vendor pricing pages in August 2026. They change; the shape of the list does not.

The three buckets

There are three genuinely different kinds of cost here and mixing them up is where most budgets go wrong.

What you pay to build. Build-tool subscriptions and credits. This is the only bucket that is unbounded, because it is metered by how much you generate rather than by what you have.

What you pay to run. Database, hosting, email, monitoring, domain. Mostly fixed, mostly predictable, mostly non-optional once real money is involved.

What you pay to get paid. Card processing. Zero until you have revenue, then a percentage forever.

Bucket one: the build tools

In 2026 the entry paid tiers have converged almost suspiciously tightly. Lovable's Pro plan starts at $25 a month for 100 credits and scales upward on the same plan to $2,250 at 10,000 credits, with a Business tier from $50. Bolt's Pro is $25, with Teams at $30 per member. Replit's Core is $20 a month on annual billing — $25 if you pay monthly — and includes $25 of credits, with Pro at $95 annual. Cursor's Pro is $20, with Pro+ at $60, Ultra at $200 and team seats from $40.

The headline numbers are not the interesting part. The metering is.

Lovable's free tier gives five credits a day capped at thirty a month, and the documentation prices individual actions so you can see the shape: "Make the button gray" is 0.50 credits, "Remove the footer" 0.90, "Build me a landing page, use images" 2.00, "Add authentication with sign up and login" 1.20. That is generous for building and it is why the loop feels cheap — until you are in a loop that does not converge, where twenty attempts at an invisible problem cost the same as twenty real features. That mechanism is worth understanding before you budget for it, and it is set out in stuck at 80%.

There is also a line most people do not expect: deployed Lovable apps consume credits at runtime, not just during editing. So the app you built is not only a build cost, it is a running one, and the size of it depends on your traffic.

Bolt meters in tokens rather than credits — the free plan allows 300,000 a day and one million a month, and Pro starts at ten million a month with no daily cap. Replit and Cursor both bundle a credit allowance into the subscription and charge beyond it.

None of this is a criticism. It is just a different budgeting shape from a flat subscription, and it deserves to be modelled as variable rather than fixed.

Bucket two: what production actually needs

This is the part people underestimate, so I want to justify each line rather than just list it.

Database — Supabase Pro, from $25 a month. The free plan is genuinely good: 500 MB of database, 5 GB of egress, 50,000 monthly active users. It also has no backups at all, allows only two active projects, and pauses projects after a week of inactivity. Any one of those disqualifies it for something holding customer data; together they are decisive. Pro brings 8 GB of database, 250 GB of egress, 100 GB of storage, 100,000 monthly active users, seven-day backups and $10 of compute credit. Overages are metered — roughly $0.125 per GB of database, $0.09 per GB of egress, $0.0213 per GB of storage and $0.00325 per monthly active user — so growth costs money smoothly rather than in a cliff.

The backups line is the one to sit with. Free tier means that if something goes wrong with your data, there is no version of yesterday to go back to.

Hosting — Vercel Pro, $20 per user per month, including $20 of usage. The reason this is not optional is written on Vercel's own pricing page: the Hobby plan is for personal, non-commercial use. A product you charge for is commercial. A very large number of people running paid products on Hobby have simply never read that sentence, and it is a strange thing to be relying on.

Netlify is a reasonable alternative with a materially different failure mode: Free, Personal at $9, Pro at $20 flat with unlimited members, metered in credits — 300 a month on free, 1,000 on Personal, 3,000 and up on Pro. When you hit the ceiling Netlify pauses the site rather than billing you for the overage. Vercel bills, Netlify pauses. Neither is better in the abstract; they are different bad days, and you should know in advance which one you have signed up for.

Error tracking — Sentry Team, $26 a month. The free Developer plan allows 5,000 errors a month, counted per event and not per unique bug, which means one noisy loop firing on every page load can eat the month in an afternoon and then stop telling you anything. I would put this ahead of almost everything else on the list in value per dollar, because it is what converts every other problem on your app from anecdote into evidence.

Transactional email — free, or Resend Pro at $20. The free plan is 3,000 emails a month, which looks ample, and enforces a hard cap of 100 per day. Password resets, magic links, receipts and notifications share that budget. Cross it and the rest of the day's mail does not go out, with no error anywhere in your app. Pro removes the daily cap and lifts the monthly allowance to 50,000; Scale is $90; overage runs about $0.90 per thousand. If your volume genuinely fits in 100 a day, staying free is fine — just know exactly where the wall is.

Uptime monitoring — free is fine. This is the one place I will tell you the free tier is production-adequate rather than a compromise. UptimeRobot's free plan covers 50 monitors at five-minute intervals, which is more than a small SaaS needs. Paid options exist — UptimeRobot Solo at $12 a month on annual, BetterStack's Responder at $29 — and they buy you faster intervals and better alerting, not a fundamentally different service. Budget $0 here and spend it on backups instead.

Domain — about $1.50 a month. At renewal pricing rather than the first-year promotion: .com $18.48 a year, .net $18.58, .org $18.98, .dev $20.98, .app $22.98 — and .io at $75.98, which is the one that catches people who picked it because it looked like a startup. Cloudflare Registrar sells at cost, which is the cheapest legitimate way to hold a domain.

The total

Adding it up for a small SaaS taking real money: Supabase Pro $25, Vercel Pro $20, Sentry Team $26, Resend on the free tier, UptimeRobot free, and a .com at about $1.54 a month.

That is $72.54 a month.

Add Resend Pro when your email volume outgrows 100 a day and it is $92.54. Add one Cursor seat because somebody has to maintain it and it is $112.54.

So: roughly $70 to $115 a month in fixed subscriptions, and I would argue none of it is optional for an app holding customer data and taking payments. It is not a large number in absolute terms. It is a large number relative to the $0 most people are mentally budgeting, and it arrives at exactly the moment revenue is smallest.

Bucket three: getting paid, and the European correction

Stripe's standard American rate is 2.9% plus 30¢ per successful card charge, with roughly 1.5% added for international cards and about 1% for currency conversion. That number is quoted so universally that people outside the United States use it by default, and for European sellers it is meaningfully wrong.

Stripe's Austrian and wider EEA pricing is 1.5% + €0.25 for standard European Economic Area cards, 2.8% + €0.25 for premium European cards, 2.5% + €0.25 for UK cards, and 3.15% + €0.25 for other international cards.

The percentage on a standard European card is roughly half the American one. On $2,000 a month spread across 200 transactions, the American rate costs about $118 and the EEA standard rate about $80 — a difference of roughly $38 a month, or half your entire infrastructure bill. (Those two figures mix dollars and euros at rough parity, so treat them as a comparison of shape rather than an exact conversion.)

The general point survives the arithmetic: check your own region's published rate rather than the number everyone repeats, and if you sell across borders, know which of the four bands your typical customer falls into.

What is not on this list

A few things people expect to see and should not budget for yet.

A content delivery network, which is included in both Vercel and Netlify. A separate staging environment, which on both hosts is a preview deployment and costs nothing extra. Log aggregation, which your host already gives you enough of at this size. A status page, until you have customers who would look at one. And anything with the word "enterprise" in it.

And the largest line of all, which is not infrastructure: the person doing the work. I am deliberately not quoting a build price here, because the honest driver is how many kinds of user your app has rather than how many screens, and a number quoted before anyone has looked is a number that will be wrong. The only 2026 survey figure I am comfortable putting on the page is freelancermap's Freelancer Study 2026 — over 5,400 respondents, an average of €103 per hour — and that covers all freelance disciplines across the DACH region, so it is an order of magnitude rather than a rate card. Most published salary surveys, including Robert Half's, report permanent starting salaries rather than freelance rates, which is a different number for a different purpose.

The three that catch people

If you remember three things from this page, these are the ones that turn into an actual bad day.

Hosting on a plan that says non-commercial. It is free, it works, and it is not the plan for a product you charge for.

A database with no backups. Free-tier Supabase does not have them. That is fine for a prototype and unacceptable the day a customer's data is in it.

A hard daily email cap. Not a monthly one — a daily one. It stops your signup flow at user 101 and looks like nothing happened.

All three are cheap to fix and all three fail quietly, which is a combination worth being deliberate about. That is a large part of what launch and deployment work actually consists of: going through every environment-dependent value and every plan ceiling before real users find them for you.

If you want the numbers for your app specifically

The list above is the general shape. What it costs you depends on your data volume, your email volume and where your customers' cards are issued, and those are answerable in an hour with your actual dashboards open.

I do a Production-Ready Audit that includes a plan-and-ceiling review alongside the technical one: what you are currently on, what you will hit first, what it will cost when you do, and what is genuinely safe to leave on a free tier. From $499, back in five to seven days.

If you would rather just check one thing, send me the list of services your app depends on and I will tell you which of them has a ceiling you are close to. No charge — it is a short answer and finding out from a monitoring alert instead is a much worse way to learn it.

The wider picture of what tends to be missing at the same time is in the ten problems every AI-built app has in production, and if you are weighing this against starting over, that decision is worked through in rescue or rebuild.