What an MVP really costs in 2026 (with real numbers)

If you have looked up what an MVP costs, you will have read that AI divided the price by ten: €50,000-200,000 and half a year before, €5,000-20,000 and a month now. It is a comfortable claim and it travels well. The problem is that it does not appear in a single independent source — only in blogs written by providers selling exactly that.

The Spanish market data says something else. Hiberus Booster, publishing its ranges in April 2026, puts a basic MVP at €8,000-25,000 over 6-10 weeks; a medium-complexity one at €25,000-60,000 over 10-16 weeks; and a complex one at €60,000-120,000. They add a telling detail: 70% of the projects that reach them fall in the €25,000-60,000 band.

So where does the low figure come from? It is real, but it is a different product. A no-code prototype, or a single developer leaning on AI, can land between €0 and €8,000 in a few weeks. That exists and is sometimes exactly what you need. What it is not is a team-built MVP. Comparing the two is comparing apples to oranges.

The underlying premise — that AI has collapsed the cost — does not hold up either. Bain, in its 2025 report on generative AI in software development, measures productivity gains of 10-15% with assistants, and 25-30% only when the process is redesigned end to end. That is not a factor of ten.

There is an even more uncomfortable measurement. METR ran a randomised controlled trial with experienced developers working on repositories they already knew: they were 19% slower with AI. And they believed they had been 20% faster. Perceived speed and actual speed did not match.

The 2025 DORA report reaches a compatible conclusion: delivery throughput goes up and stability goes down. Their line sums it up well — AI does not fix a team, it amplifies what is already there. And in the Stack Overflow survey, 45% of developers say debugging AI-generated code takes them more time, not less.

The “55% faster” figure vendors quote comes from a lab experiment: writing an HTTP server in JavaScript, with around ninety participants. It measures a toy task, not the cost of a project.

None of this means AI is useless. It helps, just not where you are told. It helps with repetitive code, tests, migrations and exploring an API you do not know. It does not help with what actually eats the budget: deciding what to build, integrating with real systems, and sustaining the product once it is live.

So why can something ship in weeks? Because of scope, not tooling. A four-week MVP is an MVP with four weeks of scope. The honesty lies in saying what is in and what is out, not in promising that technology performs magic.

We work in short cycles with weekly deliveries. You see real progress every week and decide as you go, which means the first useful version arrives early and changes of direction stay cheap. That is not “the market now takes four weeks”: it is a way of working.

Before signing with anyone, ask four things: what is in and out of scope, what happens the day after launch, who maintains the product, and who owns the code and the infrastructure. The answers will tell you more about price than any published range.

The honest summary: the range has not collapsed, it has widened at the bottom. A new tier appeared that did not exist three years ago. But a product that has to hold real users still costs what it costs.