Economies of Scale

Anyone who manufactures something has two types of costs. Some are incurred once, no matter how much is produced: a factory building, a machine, the development of a product. Others are incurred per unit, such as materials and electricity. If you produce more, the one-time costs are spread across more units. This makes each unit cheaper. This exact relationship is called economies of scale.

An example: A bakery has an oven costing 20,000 euros. If it bakes 10,000 loaves of bread, each loaf carries 2 euros of oven costs. If it bakes 100,000 loaves, it’s only 20 cents. The bread is the same, the oven too. Only the math looks different.

Why large providers are hard to catch up with

Economies of scale explain why a few giants dominate the market in some industries. Whoever already sells a lot produces more cheaply. Whoever produces more cheaply can offer lower prices. Lower prices bring in even more customers. So the lead keeps growing on its own.

For new providers, this is a real obstacle. They would have to start out very large from the beginning in order to keep up. This costs a lot of money before the first euro of revenue arrives. Experts therefore speak of a market entry barrier, meaning a hurdle to entering a market.

An important distinction: economies of scale are not the same as network effects. With network effects, a product becomes better for users the more people use it — like a messenger app that all your friends are already on. Economies of scale only affect the provider’s costs, not the benefit to customers. In practice, both often occur together.

Where the savings really come from

The most important mechanism is the spreading of fixed costs. Software companies show this in the extreme. Developing a program costs millions. Delivering one additional copy costs almost nothing. So the ten-millionth user causes hardly any additional cost, yet pays the full price.

There are further effects as well. Large buyers get volume discounts from suppliers. Large operations can divide labor more thoroughly, so that each employee gains routine in one task. And special machines only pay off above a certain volume, because otherwise they run too rarely.

A common misconception is that growth always makes things cheaper. This only holds up to a certain size. After that, it often tips over: coordination becomes cumbersome, decisions take longer, transport routes become longer. Then the cost per unit rises again. This is called negative economies of scale.

Economies of scale in the AI industry

In the news about artificial intelligence, economies of scale come up almost constantly, often without the term being mentioned. Training a large language model, meaning having it learn from enormous amounts of text, costs hundreds of millions of dollars. This sum is incurred once. After that, the model can answer billions of queries. The more users there are, the smaller the training cost share per answer.

The effect also applies to data centers. Whoever buys tens of thousands of specialized chips at once gets better prices than a small start-up. Whoever builds their own server halls pays less per unit of computing power than someone who rents capacity by the hour. That’s why corporations like Microsoft, Google, and Amazon are investing billions in their own infrastructure.

But you also encounter the term outside of technology. Discount retailers push down prices through sheer purchasing volume. Car manufacturers use the same chassis platform for multiple models. And when analysts write that a company will “become profitable as volume increases,” they mean exactly that: the fixed costs are already in place, now only the volume needs to follow.