What does it cost to build an MVP in 2026?

Every founder asks this question and almost every agency dodges it. The honest answer is that the range is enormous — and the reason it's enormous has very little to do with hourly rates, which is where most people start looking.

This is what we've seen across fifteen years of building first versions, and what changed in the last eighteen months.

The ranges, stated plainly

Broadly, MVPs fall into three bands. These are market ranges in 2026 for competent teams — not our prices, and not the cheapest quote you can find.

If a quote comes in dramatically under these, something is being left out — and it's usually the same things: testing, error handling, deployment, and whatever happens after launch. If it comes in dramatically over, you are probably being sold a scope you don't need yet.

What actually drives the number

Not the hourly rate. Not the country the team is in. Four things, in order of impact.

1. How many user types the product has

This is the single biggest multiplier and almost nobody prices it consciously. A product with one user type is one set of screens, one permission model, one onboarding flow. Add a second — an admin, an operator, a reviewer — and you have roughly doubled the surface area. Add a third and you have tripled it, plus the interactions between them.

Marketplaces are expensive for exactly this reason. Two-sided means two products that happen to share a database.

2. How much of it is genuinely novel

Authentication, payments, file upload, notifications, search, admin CRUD — these are solved problems. They should cost close to nothing now, and if you're being quoted meaningful money for them, that's worth a conversation.

The expensive part is whatever is actually specific to your idea: the matching logic, the pricing engine, the scheduling constraints, the thing you'd struggle to explain in one sentence. That's where the real hours go, and that's where you want your money going.

3. How settled the requirements are

Changing your mind in week one is free. Changing it in week eight, after the data model is built and three screens depend on it, is not. The cost of a decision rises roughly with the square of how late you make it.

This is why we spend the first two weeks of every Sprint on a spec with explicit acceptance criteria. Not because documentation is virtuous, but because it is the cheapest possible place to have the argument.

4. Whether anyone has said no to anything

The most expensive MVPs we've seen were not technically hard. They were scope that nobody was willing to cut. Every feature had a reason. Every reason was plausible. And the product shipped fourteen months late into a market that had moved.

What changed in the last eighteen months

The economics moved, and they moved unevenly. Roughly seventy percent of any build is predictable work — scaffolding, CRUD, schemas, migrations, API clients, integration glue, test suites. That portion is now close to free.

The remaining thirty percent — architecture, security, the judgment call on what to throw away — has not got cheaper at all. If anything it got more valuable, because AI will produce code that compiles, passes tests, and is quietly wrong, and someone has to be qualified to catch that.

Be sceptical of anyone claiming AI made them cheap while their team is unchanged and their timelines are unchanged. The saving is structural or it isn't real.

Three ways founders overspend before writing any code

1. 1Buying discovery. A four-week paid discovery phase that produces a slide deck is a way of billing for the conversation that should precede the engagement. Ask what artefact you get, and whether it's usable by another team if you walk away.
2. 2Paying for a design system before you have users. Custom illustration, a full component library, a brand refresh — all of it is real work, none of it tells you whether anyone wants the product. Ship with a clean, plain interface and spend that money after you have signal.
3. 3Building the roadmap instead of the MVP. If the plan includes things you'll need 'once we scale', you are paying today for a scale you don't have. Multi-tenancy, internationalisation, and role hierarchies are the usual suspects.

What to ask any agency you're considering

• What's the fixed price, and what specifically is excluded from it?
• Who owns the code and the IP, and when does it transfer?
• Whose cloud account and repository does it live in during the build?
• Who reviews generated code before it merges, and what is their experience?
• Is our code or data used to train any model?
• What happens in week ten if we want to take it in-house?

The answers to the last three tell you more than the price does. An agency that can't answer them crisply is one you'll have a difficult conversation with later.

The short version

Most MVPs should cost less in 2026 than they did in 2023, and should take less calendar time. If a quote reflects neither, ask what's changed in how the team works — because something should have.

And be more suspicious of a scope nobody has cut than of a number that looks high.