Network Effect

Most things you buy are equally useful to you regardless of how many other people also own them. A bicycle doesn’t ride any better just because the neighbors have one too. With some products, it’s different: they become more valuable to each individual the more people use them. This is exactly what is called a network effect. A messenger app that only you have installed is completely useless. If all your friends are on it, switching to another app becomes almost impossible.

Why this creates market power

Network effects explain why, in many digital markets, one or two providers end up dominating in the end. The lead grows on its own. Whoever already has many users attracts new users more easily, because the product is worth more to them. Whoever has few users loses them to the bigger player. Economists speak of a self-reinforcing cycle.

For competitors, this is a tough obstacle. A new messenger app can be technically better, look nicer, and offer more privacy. Still, hardly anyone switches as long as their friends stay put. Experts call this blockage switching costs: switching doesn’t cost you money, but it costs you reachability. This is exactly why antitrust authorities are interested in network effects.

For investors, on the other hand, the effect is a reason to buy. A company with a strong network effect has a protective wall around its business. Analyst reports often use the English word “moat” for this. A high stock market valuation is frequently justified precisely on these grounds.

Direct and indirect networks

Two basic forms are distinguished. With a direct network effect, users benefit immediately from other users in the same group. Telephones, messenger apps, and social networks belong here: more participants mean more possible people to talk to.

With an indirect network effect, there are two different groups that attract each other. One example is an app store. Many users attract developers, and many apps in turn attract users. Delivery services work similarly, with restaurants and customers, as do game consoles, with players and game makers.

At the very beginning, there is therefore always a problem: no users, no value; no value, no users. This is called the chicken-and-egg problem. Companies often solve it with a lot of money, for instance through vouchers, free use, or paying the first providers. It’s also important to draw a distinction: larger companies often produce more cheaply, and that’s economies of scale. A network effect, by contrast, concerns the benefit to the customer, not the provider’s costs.

Network effects in AI products

In coverage of AI companies, this term comes up very often. The claim usually goes: whoever has many users collects a lot of data, uses it to improve their model, and thereby gains even more users. This is referred to as a data network effect. However, whether this effect is really strong in the case of language models is considered controversial.

A typical mistake is to assume that having many users alone constitutes a network effect. If a chatbot doesn’t get any better for you just because millions of others are also asking it questions, then there is no real network effect at play. It becomes clearly visible where additional offerings emerge: developers build extensions for the most widespread platform, and this makes it more attractive to new customers.

You encounter this principle constantly in everyday life. You stay on a group chat app because your class is there. You look for used items on the platform with the most listings. And you choose an operating system partly because the programs you want exist for it. In all these cases, it’s not just quality that decides, but also the number of other people involved.