Software headcount below its pre-Covid revenue scaling
Each of Microsoft, Alphabet, and Meta had a steady relationship between revenue and headcount before Covid. This chart fits that relationship on 2010-2019, holds it fixed, and asks how many people each company's later revenue would have predicted. All three sit below their own benchmark in 2025. The level depends on which pre-Covid window is used, so the method gives the range across every window fitted; the direction does not.
What does it show?
At its most recent annual filing each company employed fewer people than its own pre-Covid revenue-headcount relationship predicts: Microsoft 28% (2026), Alphabet 26% (2025), Meta 24% (2025).
Methodology
One log-linear model per company: log fiscal year-end headcount on log whole-company revenue, fitted by OLS on 2010–2019 for Microsoft and Alphabet and 2011–2019 for Meta. The fitted relationship is then held fixed and applied to later observed revenue; the chart plots reported headcount as a percentage above or below that benchmark, so zero means a company employs exactly what its own pre-Covid relationship predicts. Revenue and headcount for 2016 onward come from the SEC company-facts data behind the Big Tech scale chart; 2010–2015 is transcribed from each company's annual reports. The size of the gap depends on the baseline. Six windows were fitted and every one leaves every company below its benchmark, but the 2016-2019 window — four observations, the shortest publishable — moves Microsoft to -19.0% and Meta to -40.8%. Read the level as a range across the five publishable windows (a sixth, a three-observation 2020–2022 fit, is a stress test and is excluded here): Microsoft -28.1% to -19.0%, Alphabet -27.6% to -23.4%, Meta -40.8% to -24.5%. A companion chart, “How many workers would Big Tech's revenue predict?”, asks the same question of five companies on a 2016–2021 baseline and a pooled panel, and reports deeper gaps (Microsoft -30.7%, Alphabet -33.4%, Meta -37.2%, for 2025). It fits a single pooled revenue elasticity of 0.99; fitting each company separately over the longer window gives Microsoft 0.76, Alphabet 0.86, Meta 0.87, so the benchmark here grows more slowly with revenue and the shortfall is shallower. Same filings; the baseline window and the pooling are what separate them. This is a descriptive benchmark, not a break test: the fit is never re-estimated on the later years, so it cannot say whether the relationship changed or the companies moved along it. It identifies no cause — not productivity, AI, layoffs, contractors, outsourcing, or business mix — and is not an estimate of an optimal workforce. Acquisition years are not adjusted: fiscal-year revenue and year-end headcount can cover different ownership periods, which is why one sensitivity starts after the Nokia and Motorola transitions.