
Capex
Capex refers to a company's spending on long-lived assets such as buildings, machinery, or data centers. In the AI industry, capex has become the most important number, because building data centers now costs hundreds of billions of dollars.
Capex is short for the English term "Capital Expenditure," meaning investment spending. It refers to all the money a company spends on long-lived things. Examples include factory buildings, machinery, trucks, or computers. Such purchases remain worth something for many years after they are bought. This is distinguished from ongoing costs like salaries, electricity, or rent, which arise anew every month and create nothing lasting. These ongoing costs are called Opex in industry jargon, short for "Operational Expenditure."
Why investors stare at the investment figure
Capex shows what a company is betting on. Those who invest heavily are expecting growing demand in the coming years. Those who cut investment tend to expect stagnation. That's why the capex figure in quarterly reports is often more important than the current profit. It offers a glimpse into management's expectations for the future.
At the same time, high capex is a risk. The money is spent before it's clear whether the investment will pay off. A factory can't simply be returned if demand fails to materialize. This is referred to as tied-up capital: it's locked into concrete and technology and is no longer available flexibly. This is precisely why stock prices sometimes react negatively when a corporation announces higher investments than expected.
A common misconception is that capex is simply burning money. In accounting terms, it's the opposite. The money merely moves from one side of the balance sheet to the other, from cash into fixed assets. Only over the years is it depreciated as a loss in value, gradually weighing on profit.
From expenditure to depreciation
When a company buys something long-lived, it isn't allowed to deduct the costs immediately in full from profit. Instead, it spreads the amount over the estimated useful life. For a server, this is often five or six years. A purchase worth 600 million euros then weighs on profit by around 100 million per year. This spreading-out is called depreciation.
This leads to an effect that confuses many readers. A company can spend enormous sums and still report a solid profit. The bank balance drops immediately, but the reported profit declines only slowly. Anyone who wants to know how much money is actually flowing out therefore looks at free cash flow. Roughly speaking, this is the money from ongoing operations minus capital expenditures.
The assumed useful life also matters. If a corporation sets it longer, annual depreciation falls and profit looks better. With AI chips, this is contested. Do graphics cards in a data center really last six years, or are they technically obsolete after three? Billions of dollars in the figures hinge on this assumption.
Data centers as a multi-billion-dollar line item in the AI industry
In business news, the word capex now appears almost always in connection with AI. Microsoft, Amazon, Alphabet, and Meta are building data centers on a massive scale. Together, their annual investments now run into the hundreds of billions of dollars. The largest single item is specialized chips for training AI models. Added to that are buildings, cooling systems, and power connections.
An entire supply chain therefore depends on these corporations' capex planning. When investment rises, chip manufacturers, networking companies, and energy providers benefit. If a single major customer cuts its plans, several stock prices often fall on the same day. Some observers warn that a bubble is forming here, because money is being invested before it's clear how much can be earned with AI.
But you'll also encounter capex far beyond the world of technology. A railway ordering new trains, a city building a bridge, a baker buying a new oven: all of these are capital expenditures. Only the scale differs.