What is Capital expenditure?

Capital expenditure, usually shortened to capex, is money a company spends to buy or build things that last for years: buildings, land, servers and equipment. AI capex is the portion aimed at AI, mostly data centers, GPUs and other accelerators, networking gear, and the power and cooling to run them.

Capex versus operating costs

A company's ordinary costs, such as salaries and electricity, are expenses recorded in the period they occur. Capex is different. The cash goes out when the purchase is made, but the cost is spread over the asset's useful life through depreciation. This is why capex can shock investors: a company can spend tens of billions in a quarter while its reported profit moves much less.

A simple comparison: renting a kitchen is an operating cost, and building one is capex. The building bill arrives at once, and the benefit shows up over many years.

Why AI made the number so large

Training and serving large models requires enormous amounts of compute. Cloud providers must build capacity before customers use it, because data centers, chip orders and power connections take a long time to arrange. Large cloud companies report quarterly capex figures that have grown sharply, and they often give guidance for the year ahead. Because chips and memory are scarce, a large share of the spending flows to chip makers and to makers of high-bandwidth memory.

How to read the figures

Several measures are worth separating:

• Capex is the spending itself.
• Operating cash flow is the cash the business generates from its normal work.
• Free cash flow is operating cash flow minus capex. When capex exceeds operating cash flow, free cash flow turns negative and the company must draw on savings, borrow or raise money.
• Contracted backlog is revenue customers have committed to but the company has not yet recognized, which shows how much demand stands behind the spending.

Capex alone does not say whether the bet is good. The test is whether the new capacity fills with paying customers and earns back its cost.

Why it matters to you

Capex is the supply side of the AI market. It influences how much computing power exists, which affects what inference costs and how reliably you can get it. Providers that spend heavily may keep prices low to fill capacity, while a pullback could tighten supply or raise prices. If you depend on one provider, its capex and cash position are part of your risk picture.

Related ideas

Capex connects to neoclouds, which borrow heavily to build GPU capacity, and to the Tensor Processing Unit and other custom chips that large providers build to lower their own hardware costs.