Hyperscaler capex, depreciation, and AI monetisation

What this actually is

Five companies — the largest cloud and search businesses — are spending hundreds of billions of dollars a year building data centres for artificial intelligence. That spending is revenue for chipmakers, power companies and builders, and it is depreciation on the spenders' own income statements for years afterwards.

Why it matters to a person

This is the largest single driver of index earnings growth since 2023, and the largest single concentration risk. If the revenue from AI arrives more slowly than the depreciation from building it, the most valuable companies in the market take an earnings hit at the same time, and so does everything they buy from.

For scale

Combined capex went from roughly $150 billion in 2022 to several times that by 2026. The closest historical analogue is telecom fibre in 1998–2001: the demand forecast was directionally right and the timing was wrong by a decade, which bankrupted the builders while the users prospered.

How to read a release

The mechanism

Five companies spend more than the rest of the index combined; if revenue lags the depreciation, the largest earnings engine in the market stalls.

What to watch: Capex guidance; depreciation schedules; AI revenue disclosures; free cash flow. Arrives quarterly; the effect typically lasts years. Impact tier 1 — Regime-setting: Moves everything and sets the direction for months or years. These are the discount rate, the level of earnings, the availability of credit, and the shocks large enough to stop the economy. If one of these turns, nothing else in the list matters much.

Primary source — sec.gov ↗