Hyperscalers: Capex vs. Headcount
This chart puts the two halves of the hyperscaler story on one denominator. It divides aggregate trailing-12-month capital expenditure by revenue, quarterly, and divides year-end headcount by fiscal-year revenue, annually, for Amazon, Microsoft, Alphabet and Meta, and rebases both to fiscal 2016. Capital per dollar of revenue climbs while people per dollar of revenue falls; the crossing is the split. It measures intensity, not returns on the spending or AI-only investment.
What does it show?
Since fiscal 2016, capex per dollar of revenue at Amazon, Microsoft, Alphabet and Meta has risen +216% (9 to 29 cents per dollar, trailing twelve months) while employees per dollar of revenue fell -18% by fiscal 2025. Revenue is now bought with machines rather than people.
Methodology
Two ratios for the same four companies — Amazon, Microsoft, Alphabet and Meta — each divided by revenue and rebased to fiscal 2016 = 100. Capex per dollar of revenue is aggregate trailing-12-month capital expenditure divided by aggregate trailing-12-month revenue, quarterly, from SEC XBRL facts: cash-flow capex is differenced from year-to-date values into fiscal quarters and revenue uses direct 10-Q quarters with fiscal Q4 derived from the 10-K. Employees per dollar of revenue is the sum of 10-K year-end headcounts divided by the sum of fiscal-year revenue, one point per fiscal year on the fiscal Q4 label, interpolated linearly across the quarters between and ending at the last year-end. Fiscal years differ (Microsoft ends in June), so the pooled quarter labels are fiscal, not calendar. Oracle is in the capex charts but not here, and Apple is in the headcount charts but not here: the split needs both series for the same company. Amazon's warehouse workforce dominates the pooled headcount, which is why the per-company figures are kept alongside. The chart measures intensity per dollar of revenue, not returns on capital, AI-only spending, or productivity.