Column: AI crushes every moat except one — and Finland has it in abundance

On the stage of Y Combinator’s Startup School India, the founder of Indian payments giant Razorpay, Harshil Mathur, recently said something that would have sounded like a provocation just five years ago: AI is compressing every moat. It would be easy to wave away as another worn-out startup phrase, except that this man’s Razorpay processes more than 180 billion dollars in payments per year, was the first Indian company Y Combinator ever invested in (W15 batch), and has grown in just over ten years into the country’s largest payments platform. When that kind of founder says from the stage that all competitive moats are collapsing under AI, it pays to listen carefully.

But right after, it pays to ask the contrarian question: if that is true, where should we be looking? Which moat is the one AI does not break?

The answer is much better news for the Finnish founder than many of us dare to hope. But to see it, we first have to understand what Mathur is really saying.

▶

Source: Y Combinator — Harshil Mathur: AI Is Compressing Every Moat (Startup School India), 6 May 2026

What “compressing every moat” actually means

In the startup lexicon, a moat means the competitive edge that makes a company hard to copy — the water defense that keeps competitors at bay. Warren Buffett’s term has classically meant four things: brand, scale economies, switching costs and network effects.

Mathur’s thesis is that AI rapidly erodes precisely the version of the moat that the last 15 years have been built around in Silicon Valley. Better code can be built with generative AI faster and cheaper than ever — where 50 engineers were once needed, 5 will now do. A better interface is a week’s work, not a year’s. Better data is less useful, because a competitor can reach the same level with a few API calls. Faster brand-building is now possible with AI-produced content at scale that used to be the reserve of large players only.

This is a personal observation for the Razorpay founder. In ten years his company has survived many kinds of failure — the near-death moment described in the episode, where a bank pulled financing two weeks after Demo Day, is one of many. He built a payments giant in India inside a regulated industry at a time when the country’s UPI system was only emerging, and he succeeded before the country’s largest banks managed to respond. If anyone knows what it means to build a moat slowly, it is him.

And that is precisely why his message lands so heavily: even his own moats no longer hold if they rest only on technological supremacy.

But there is one moat AI does not break

This is where the column’s contrarian point begins to take shape. Mathur’s thesis is true — but only for technological moats. There is another category that AI cannot copy at any model size, and not in the years to come. And Finland has more of it, by nature, than most countries do.

This other category consists of three parts:

The first is personal trust, born of time. A ten-year relationship with a major customer is not knowledge available for purchase by an AI query. It is memory of who saved the last project at the last minute, who called when someone else would have just skipped the email, and who kept their word when they could have skipped that, too. A relationship like this is not synthesised — it is lived. And it is exactly what produces the early-morning call when a buyer is collecting bids from a new supplier, the call that says “do you want to take this”.

The second is deep local understanding. The twists of a Finnish electricity contract, the rhythm of pension-fund decision-making, the timetable of municipal procurement bureaucracy, the cycle of trade-union agreement renewals — this is knowledge a global AI model does not learn from training data. It is learned by living the field for years, and it is exactly what decides whether a foreign competitor can really enter Finland or stumbles at the first contract negotiation. The same logic holds in every country — but it is a structural advantage for the one who is already local.

**The third is a slowly built community.** Slush was not born overnight — building its atmosphere, ritual and meaning has taken close to two decades. Sixth Street’s NFL strategy rests on the same logic: the Patriots’ fan base was built generation by generation, not by campaign money. Wolt’s courier community, Supercell’s gaming culture, Reaktor’s expert network — all built with patient work that AI can assist but not replace. A community that feels part of something shared survives even a bad year. A community whose brand was synthesised in a month falls apart at the first tailwind.

Where the Finnish founder stumbles on this

It would be too easy to end the column here. But honesty demands also naming the pitfall. Many Finnish founders have in recent years tried to build an American moat, not their own natural one. That is: technological speed, viral growth and the “blitzscale” mindset — strategies the Silicon Valley playbook has taught for 15 years and that have been preached in Slush for a slightly shorter while.

This is understandable. When the only playbook on offer is American, it is natural to read it. But the outcome is that the Finnish growth company positions itself on exactly the field where Americans are stronger, and Mathur’s thesis adds that this whole field is going to level out further in the coming years. A losing strategy, multiplied by two.

The Finnish strength — Nordic trust, long-form professionalism, slow but deep customer relationships, knowledge of the local system — has lately been brushed aside as “old-fashioned”. It has not been good enough to talk about on Slush’s main stage, because it does not sound sexy enough. I have written earlier in Innohub about the structural value of trust as a competitive edge — and AI is now giving that thesis a hard confirmation that few expected.

What this means in practice

If you are a founder whose company is built on technological supremacy, it is time to take a look in the mirror. Is your moat really that your code is better, or is it knowing the customer in a way no one else knows them? If the answer is “both”, that is fine. If the answer is only the first, shift your weight — because that first is narrowing every month.

If you are a founder who builds slowly, locally and deeply, the lesson is different. Don’t try to hide a slow game behind speed. Name it honestly as a strength. The long game is the long-game player’s edge, and the AI era makes it more valuable, not less. For the Finnish founder this is a card unusually directly handed.

And if you are somewhere in between — like most of us — the most important question is not “how do we get better AI”, but “in which industry, with which customers and in which place does our ten-year relationship network give us an edge that AI has no business building”. Answer that question with care, and you have a strategy you can build for ten years on top of.

Closing

Mathur, of course, was not talking about Finland. He was speaking to an audience of Indian startup founders about how AI is changing the rules of building globally. But that is precisely why his message is more important to the Finnish founder than many local pep talks. When the founder of India’s largest fintech says that better technology is no longer enough, his words carry a weight that cannot be skipped.

And for us, the message is surprisingly energising: the world is moving in a direction in which Finland’s natural strengths are more wanted than they have been at any point this millennium. Nordic trust, local expertise, long relationships — all the things AI cannot synthesise.

It is left to us to use this edge boldly, instead of postponing it once again until “we have time to focus on it”. The time is now. It is not going to get better.