Capital Expenditures

A company spends money in two very different ways. It pays salaries, electricity, and rent so that operations run today. And it buys things that will still be useful for many years to come: a factory building, a machine, a data center full of computers. The second kind is called capital expenditures. In English these are called Capital Expenditures, abbreviated Capex, and this abbreviation also appears in German business news. The difference is not merely linguistic: capital expenditures are treated differently in accounting than operating costs.

Why investors stare at the Capex figure

Capital expenditures reveal what a company believes in. Whoever pours billions into new facilities is betting on rising demand. Whoever cuts spending expects lean years ahead. That is why the Capex forecast in quarterly results is often more important than the profit figure itself. It regularly happens that a stock falls after good earnings numbers because the announced investments exceed expectations.

The second reason is risk. Capital expenditures tie up money for years. A machine cannot be returned the next day if the market collapses. A half-built data center is an expensive construction site, not an asset. Investors therefore always ask whether the investment pays off, that is, whether it will later generate more than it cost.

This question is especially open in the AI industry. The large technology corporations together invest triple-digit billions per year in computing capacity. Whether revenues from AI products will ever cover these sums is something nobody knows for certain today. This is exactly what is being debated on the stock market.

Depreciation: why the purchase doesn’t eat up profit immediately

Suppose a company buys computers for 500 million euros. The money flows out immediately. But it does not appear all at once in the profit statement. Instead, the amount is spread over the estimated useful life. Over five years, that is 100 million euros per year. This spreading out is called depreciation.

The idea behind this is simple. The computers work for five years and generate returns for five years. So they should also incur costs over five years. Otherwise the purchase year would look artificially catastrophic and the following years artificially brilliant.

This leads to a common misconception. A company can report high profits and still be short on cash, because the money is currently tied up in concrete and chips. Conversely, the estimated useful life can flatter the profit: whoever assumes that servers last six years instead of four depreciates less each year and reports more profit. Capital expenditures must be distinguished from operating expenses, in English Opex. These are the ongoing costs that count fully immediately.

From the chip factory to the cloud subscription

In the news, you usually encounter the term in connection with very large construction projects. A chipmaker announces a new factory for 20 billion dollars. A carmaker converts a plant to electric drive. A cloud provider announces that it is doubling its Capex next year. Power grids, railway lines, and fiber-optic cables are also capital expenditures, just made by states or network operators instead.

One detail is important for the AI industry. Whoever builds their own data centers has high capital expenditures. Whoever instead rents computing time from a cloud provider has operating costs. Many start-ups choose the second path because they lack the capital for the first. Large corporations build their own because it is cheaper in the long run and makes them independent.

This distinction also exists in private life. A purchased bicycle is an investment, a rental bike per ride is an operating expense. The bicycle costs a lot once and then almost nothing afterward. The rental bike costs little per ride, but again and again. Companies make the same decision, just with a lot more zeros.