What is capex, and why AI needs so much of it
Capital expenditure is money spent on things that last. For Big Tech right now, that mostly means data centres full of AI chips.
Capex is short for capital expenditure: money a company spends on things it will use for years, such as buildings, machines and equipment. It is different from the day-to-day costs of running the business, like salaries or electricity.
Where to find it
Public companies in the US report capex every quarter in their filings with the SEC (the 10-Q each quarter and the 10-K once a year). It sits in the cash flow statement, usually as "purchases of property and equipment" or a similar line. Because the filings are public, anyone can check the number.
Why AI needs so much of it
Training and running large AI models takes data centres, and data centres are capex:
How to read a capex number
| Question | Why it matters |
|---|---|
| Is it rising faster than revenue? | The company is betting on future demand |
| What did management guide for next year? | Shows whether the build-out continues |
| How is it paid for? | Cash flow, or new debt and leases |
| Who receives the money? | Chip makers, builders and power companies |
Capex is not cost, yet
A server bought this year does not hit profit all at once. Its cost is spread over its useful life as depreciation. That is why a company can spend record amounts on data centres and still report strong profits now, with the bill arriving in later years. If AI revenue does not grow to match, that depreciation is where the pressure shows.
Sources
Read next
All explainersWhat is an AI agent?
A language model that can take steps on its own, check what happened and try again. Here is what that means and where it breaks.
Read · 5 min 02Chips · ExplainerWhy GPUs cost so much
The chips that train AI models are scarce, hard to make and wrapped in software nobody else has. Here is where the price comes from.
Read · 5 min