Moat

When a company makes a lot of money, other firms try to do the same. Usually this drives prices down and shrinks profits. Yet some companies remain profitable for decades because they possess something imitators cannot easily copy. In the world of finance, this protection is called a moat, from the German Burggraben. The image comes from the castle: the water-filled trench keeps attackers away, even if they come in large numbers. The term was popularized by investor Warren Buffett and is now standard in business journalism.

Why investors ask about the moat, not the product

Anyone who buys a stock is betting on profits five or ten years from now. A good product today says little about that. What matters is whether competitors can replicate the product next year. Without a moat, a market quickly settles at prices barely above cost. This happens, for instance, with airlines or memory chips, where profits swing wildly.

For AI companies, the question is especially pressing. Developing a large language model costs hundreds of millions of dollars. Such sums can only be recouped if protection exists afterward. Otherwise a cheaper provider with similar quality will scoop up the customers. In 2023, a leaked Google memo became famous, titled: “We Have No Moat” — referring to the pressure from freely available models.

That is why the term underlies many stock price movements. When a company unveils a new model that rivals an expensive competing product, the competitor’s share price often falls. The model itself is not the news — the signal that the moat is narrower than assumed is.

What a moat actually consists of

The most common type is network effects. A service becomes more valuable the more people use it. A social network without friends is useless, so hardly anyone switches to a newcomer alone. A second type is switching costs: if a company has built its entire accounting on one piece of software, switching is expensive and risky. A third type is economies of scale, meaning lower cost per unit through sheer volume.

Then there are patents, licenses, and brands. A patented drug cannot be copied by anyone for twenty years. A strong brand allows higher prices for a technically comparable product. Exclusive data can also protect a company, for example when only one provider has access to certain measurements.

A common mistake is to mistake a technical lead for a moat. A lead is just a lead, not a moat. It holds only as long as no one catches up, and in AI development that often happens within months. Size alone is not enough either: Nokia was the world’s largest handset maker and still lost the market within a few years.

The term in quarterly reports and AI headlines

The word comes up regularly in analyst calls after quarterly earnings. Executives are asked how they intend to defend their moat. Rating agencies like Morningstar even assign formal classifications ranging from “no moat” to “wide moat.” The term is also everyday language in market commentary and investor letters.

In the AI sector, it currently comes up in three places. For chipmaker Nvidia, the moat is considered to be less the hardware than the CUDA software, on which almost all research projects are built. For cloud providers, it is the switching costs of large corporate customers. For chatbot providers, it remains unclear whether habit and brand name are enough, since switching costs the user only a click.

As a reader, a simple test question is worthwhile. What exactly would a well-funded competitor need to do to replace this offering within two years? If the answer is “a lot of money and time,” a moat is plausible. If the answer is “train a better model,” then probably not.