03 · AI
The money at the top of the chain.
AI infrastructure is bought before it is built. This page shows the capital spending of the companies doing the buying, and the revenue of the equipment makers a slice of it eventually reaches, both taken from the same filings. It draws the two series and stops there.
01
Both ends of the chain
Quarterly, reconstructed from SEC company facts by differencing the year-to-date figures a 10-Q actually prints.
02
Who is spending
Capital expenditure per filer. Total property, plant and equipment, because no filer breaks out data centres separately.
Latest quarter
USD billion
03
What this does and does not show
- Capital expenditure is not AI spending. It is every building, server, vehicle and fit-out a company capitalised that quarter. No filer publishes an AI line. This is the ceiling, not the measurement, and treating it as the measurement is the most common mistake made with these numbers.
- No relationship is asserted between the two series. They are drawn on separate axes, in separate charts, deliberately. Any lag or ratio you read into them is yours, not this page's.
- The equipment side covers SEC filers only. Tokyo Electron, Screen, Advantest and Kokusai report to their home regulators and are absent, so that total is partial and understates the industry.
- The quarters are reconstructed, not taken as printed. Capital expenditure is a cash-flow line that most filers report year-to-date in a 10-Q, so each quarter here is the difference between consecutive year-to-date figures from the same reporting period. Nothing is interpolated: a quarter that cannot be derived from what the filer published is absent.
- Each total covers a fixed panel of filers. Adding up whoever happens to have filed makes the newest quarter look like a collapse in spending. The note under each chart names the filers in that total, says how many quarters were dropped, and names anyone left out.
- Nothing here is a forecast, and no company is ranked on anything but the arithmetic on the axis.