Zero Marginal Cost

When a bakery bakes one more bread roll, that costs flour, electricity, and labor time. Economists call this amount for a single additional unit the marginal cost. For digital goods, this amount shrinks to nearly zero: copying a music file one more time costs the provider a fraction of a cent. That is exactly what the English term Zero Marginal Cost means, literally “marginal cost of zero”. An important distinction: developing the product may have cost millions. Only every further copy is practically free.

Why digital products can grow so fast

A bakery that wants to sell ten times as much needs more ovens, more flour, and more staff. Its costs grow at roughly the same pace as revenue. A software company, by contrast, can grow from a thousand to a million users without costs growing along with it. The profit per additional customer then comes close to a hundred percent of the price.

This explains a large part of the stock market valuations of tech companies. Investors pay high prices for firms whose revenue can rise sharply without expenses following suit. For software, the so-called gross margin—that is, the share of revenue remaining after direct production costs—is often 70 to 90 percent. For a car manufacturer, it tends to be more like 15 to 25 percent.

The flip side is a high upfront effort. Whoever develops an operating system or a game invests for years first, without any revenue. If the product finds no users, that money is lost. Zero Marginal Cost thus rewards winners extremely strongly and punishes losers just as harshly.

What makes copying so cheap

The reason lies in the nature of information. A digital product is a sequence of zeros and ones. Duplicating this sequence just means writing to storage and sending data through a cable. Both have become dramatically cheaper over the decades, while raw materials and labor time have remained expensive.

Still, “zero” is never meant literally. Every download consumes electricity, network capacity, and a piece of data center. Experts therefore speak more precisely of “near zero”. Only once these residual costs are so small that a company can ignore them when setting the price does the economic logic of the term kick in.

For artificial intelligence, this logic only applies to a limited extent, and that is a common misconception. Every chatbot answer has to be freshly computed. This computational work is called inference and runs on expensive graphics chips. A single answer costs the provider anywhere from a fraction of a cent to several cents, depending on the model. That is little, but clearly more than zero, and with billions of requests it adds up.

From free apps to AI subscription models

In everyday life, you encounter this principle everywhere things are offered for free. Map services, search engines, and messengers don’t charge money because one additional user barely burdens the provider. Instead, money is made through advertising or paid add-on features. This model is called freemium: basic version free, extras cost money.

Streaming follows the same logic. Netflix pays once for a series and can then deliver it as often as it likes. The ten-millionth stream costs almost nothing. That’s why flat rates are possible here, whereas a gym would run into real space problems with too many members.

In business news, marginal cost comes up almost every time analysts talk about scaling. For AI providers, the term is being scrutinized critically: some companies sell subscriptions at fixed prices even though heavy users generate high computing costs. This is exactly where it becomes clear that digital does not automatically mean free.