Marginal Cost

Marginal cost answers a simple question: What does it cost to produce a single additional unit? A bakery that has baked 200 rolls needs a bit of flour, a bit of yeast, and a little electricity for the 201st roll. These additional costs are the marginal cost. The oven, the bakery itself, and the rent, on the other hand, don’t count, because they’ve already been paid for regardless. Economists therefore draw a strict distinction between costs that are incurred no matter what, and costs that arise anew with each additional unit. It is precisely this second kind that is meant.

Why software is so profitable

For a car, marginal costs are high. Every additional vehicle needs steel, tires, cables, and labor time. A car manufacturer therefore earns only a small share of the price for each car sold. For a computer program, it’s completely different. The first copy might cost millions in development work. The second copy costs almost nothing, because it merely needs to be downloaded.

This explains a large part of stock market history over the last thirty years. Companies with very low marginal costs can multiply their revenue without costs growing along with it. Beyond a certain point, almost every additional euro of revenue flows directly into profit. Investors are readily willing to pay high prices for such business models, because growth is barely held back.

A common mistake is to confuse low marginal costs with low total costs. Developing a drug often costs more than a billion euros. The pill itself then costs mere cents. Both are true at the same time: a huge upfront investment, and tiny marginal costs. It is precisely this combination that determines how a company sets its prices.

How to calculate it

You take the total cost at a certain production volume and the total cost for one unit more. The difference is the marginal cost. If producing 1,000 T-shirts costs 8,000 euros and producing 1,001 T-shirts costs 8,006 euros, the marginal cost is six euros. It’s important to look only at the change, and not to divide by the number of units.

Because that would actually be the average cost, and that’s something different. For 1,000 T-shirts, it comes to eight euros per unit here, because the factory hall and the machines are included in the calculation. Marginal cost and average cost rarely coincide. In practice, average cost falls with each additional unit as long as marginal cost stays below it.

Marginal cost also doesn’t stay constant. At first it often falls, because processes become more efficient and materials get cheaper when bought in larger quantities. At some point it rises again: overtime, rented extra machinery, more expensive suppliers. Economics textbooks depict this pattern as a U-shaped curve. Business economists use it to determine the most cost-effective production volume.

The special case of the AI data center

In news about tech companies, the term currently comes up particularly often, specifically in connection with artificial intelligence. A chatbot, after all, doesn’t behave like ordinary software. Every single response has to be computed anew, and this computation runs on expensive specialized chips in large data centers. This incurs electricity and hardware costs, every single time.

This means AI providers have real marginal costs per request — unlike, say, a social network that merely delivers already-finished text. With a subscription costing 20 euros a month, a very active user can therefore cost the provider money rather than bring in revenue. This is precisely why companies work so intensively on making their models compute more efficiently. If the cost per response falls, the profit margin improves immediately.

The term is also useful outside of technology. A streaming service, an additional passenger on a half-empty flight, one more concert ticket for a hall that isn’t sold out: in every case, it’s worth asking what that one extra unit really costs. Anyone who can answer that understands pricing decisions, discount promotions, and business models far better.