Big Tech Headcount vs Capex: Q3 Filings Show Capex Up 243%, Staff Up 4.9%

In the quarter to 30 September 2026, Alphabet, Meta, Microsoft and Amazon spent $201.9 billion on property and equipment. Two years earlier the same quarter cost them $58.9 billion. Over those eight quarters their combined big tech headcount rose 4.9%. Put the two lines side by side and capex per employee more than tripled, from about $29,000 a quarter to about $95,000.

The short answer. Across Alphabet, Meta, Microsoft and Amazon, Q3 2026 capex was $201.9 billion against 2.13 million employees, about $94,700 per head for the quarter. Meta led at $508,000 per employee, Alphabet $251,000, Microsoft $177,000 and Amazon $47,000. Headcount moved between minus 7.8% and plus 13% in two years.

The big tech headcount and capex table, from the Q3 filings

This table is the core of the post. Every number in it was taken from an earnings release or annual report filed with the SEC, and every derived figure is simple division you can repeat. The window is eight quarters: the quarter ended 30 September 2024 to the quarter ended 30 September 2026, with Q3 2025 as the midpoint.

Capex here means cash spent on property and equipment in that one quarter. It is the most consistent line across all four cash flow statements. Headcount is the company's own reported employee count at the end of the quarter, with one exception covered below.

Capex basis: Alphabet and Meta purchases of property and equipment; Amazon gross purchases of property and equipment, before proceeds and incentives; Microsoft additions to property and equipment for its fiscal Q1 2025 and Q1 2027. All exclude finance leases. Microsoft headcount is its 30 June figure from the 10-K, the nearest disclosure to each September quarter. Capex per employee is that quarter's capex divided by period-end headcount, rounded.

Read the change column against the capex columns. Capex roughly tripled or quadrupled at every company. Headcount moved by single digits at three of the four and fell at two of them.

The midpoint tells you when the gap opened. In Q3 2025, combined capex was $97.3 billion. So the four companies added $38.4 billion of quarterly spend in the first year and $104.7 billion in the second. Most of the acceleration is recent, and most of it happened while Meta and Microsoft were shrinking. The cash consequences of that acceleration are tracked separately in the quarterly ratio of hyperscaler capex to operating cash flow.

Capex per employee, company by company

Capex per employee is a crude ratio, and I think it is a useful one for exactly that reason. It strips out scale. It asks one question: for each person on the payroll, how much physical infrastructure is the company buying this quarter?

Meta: the steepest swap of the four

Meta reported 66,775 employees at 30 September 2026, down 15% year on year. A year earlier it had 78,450, and two years earlier 72,404. So the company grew 8.4% and then cut 14.9%.

Capex moved the other way throughout. Purchases of property and equipment were $8.3 billion in Q3 2024, $18.8 billion in Q3 2025 and $33.9 billion in Q3 2026. Meta's own capex measure adds finance lease principal and comes to $34.91 billion for the latest quarter, guided to $135 billion to $145 billion for the full year.

The result is $508,000 of capex per employee in a single quarter. Free cash flow was negative $2.12 billion in the quarter by Meta's own reconciliation. Of the four, Meta is the cleanest case of a company that has fewer people and far more machines than it did two years ago.

Alphabet: hiring and building at the same time

Alphabet breaks the simple story. It had 204,757 employees at quarter end, up from 190,167 a year earlier and 181,269 two years earlier. That is 23,488 more people, the largest absolute addition outside Amazon.

Its capex rose faster still: $13.1 billion in Q3 2024, $24.0 billion in Q3 2025, $51.4 billion in Q3 2026. Operating cash flow was $42.3 billion, so free cash flow came in at negative $9.1 billion, by Alphabet's own reconciliation.

Capex per employee went from $72,000 to $251,000. Alphabet is not swapping people for data centres. It is buying both, and the data centres are growing about 30 times faster in percentage terms.

Microsoft: falling headcount and an exit offer

Microsoft's 10-K for the year to June 2026 states that it employed approximately 223,000 people on a full-time basis. That compares with about 228,000 in both 2024 and 2025. Product research and development stood at 77,000 of the total.

In its fiscal Q1 2027, the quarter to 30 September 2026, additions to property and equipment were $39.4 billion against $19.4 billion a year earlier and $14.9 billion two years earlier. CFO Amy Hood said on the call that total capex including finance leases was $45.2 billion, and that roughly three quarters was for short-lived assets, primarily CPUs and GPUs.

Hood also said the next quarter's operating expense guidance includes an estimate for the recently announced employee voluntary exit offer. Press reports put the offer at about 60,000 eligible employees. Microsoft has not said how many will accept.

Amazon: the only big employer still adding

Amazon's employee table shows 1,638,000 full-time and part-time employees at 30 September 2026, up 4% year on year and 5.6% on two years earlier. Much of the workforce sits in fulfilment and logistics, which is a different labour model from the other three.

Gross purchases of property and equipment were $77.3 billion in the quarter, against $22.6 billion in Q3 2024. Amazon still expects 2026 cash capex of about $220 billion. Trailing twelve month free cash flow was negative $42.4 billion.

Amazon's capex per employee is the lowest of the four at $47,200, simply because the denominator is so large. Its growth rate in the ratio, about 3.2 times, is in line with the others.

Eight quarters of big tech headcount: who added and who cut

Strip the capex out and look at the people line on its own. The headline figure of plus 4.9% hides a split. Amazon added 87,000 people. The other three, combined, went from 481,673 to 494,532, an increase of 12,859 or 2.7%.

Inside that 2.7%, Alphabet added 23,488 while Meta lost 5,629 and Microsoft about 5,000. Two of the four companies that spend most on AI infrastructure employ fewer people today than they did in 2024.

The path matters too. Meta and Alphabet both grew between Q3 2024 and Q3 2025. Meta then reversed hard. Amazon's count stayed nearly flat from Q3 2025 through Q1 2026, at 1,578,000, 1,576,000 and 1,575,000, then rose by 63,000 across the next two quarters.

The wider question of whether AI is behind job cuts in the economic data gets a fuller treatment in the analysis of what payroll data shows about AI job losses. Four companies are not the labour market, and nothing in this post should be read as a claim about it.

Has big tech capex replaced hiring? What the filings support

The tempting headline is that big tech has swapped engineers for GPUs. The ratio supports a narrower claim. Spending on physical assets rose about 50 times faster than combined headcount, measured in percentage terms. That is a change in the mix of what these companies buy, and it is large.

It is not proof of substitution. Substitution would mean the money that went to data centres would otherwise have gone to salaries, or that machines are doing work people used to do. Neither appears in a cash flow statement.

The evidence that points towards a swap

Two of the four companies cut staff while their capex rose. Microsoft tied its voluntary exit offer to the same quarter in which it reported record infrastructure spend. Meta lowered expected Reality Labs losses for next year to around $15 billion, describing the move as reallocating investment toward AI. In both cases the company itself connects the people decision to the budget decision, at least loosely.

A second signal is the cash position. Alphabet and Meta posted negative quarterly free cash flow, and Amazon's trailing twelve months were negative too. When the build consumes all operating cash, every other cost line, including headcount, is competing for what is left.

The evidence that points against it

Alphabet added 13% to its workforce while quadrupling quarterly capex. If capex simply displaced hiring, the company with the second-highest capex growth should not be the fastest hirer in the group.

Scale argues against it too. Meta's entire headcount reduction of 11,675 people over the latest year, priced at even a generous loaded cost, is a small fraction of the $15.1 billion increase in one quarter's capex. The headcount cuts are not what funds the build. The operating cash flow and, increasingly, the debt markets are. My read is that the cuts are a margin defence, not a financing source.

How the capex per employee numbers were built

Every ratio in this post comes from two printed figures and one division. The method is laid out here so you can check it or extend it to the next quarter.

Three choices drive the result. First, cash purchases rather than total capex including leases, because it is the only line all four print the same way. Including finance leases would raise Meta and Microsoft and would not change the ranking. Second, period-end headcount rather than an average, because only Amazon publishes enough points to average. Third, one quarter of capex rather than a trailing year, because the point of the comparison is how fast the ratio is moving now.

Annualise the latest quarter and the numbers become harder to read past. Meta's Q3 run rate is roughly $2.0 million of capex per employee per year, and Alphabet's about $1.0 million. Treat those as illustrations of the current pace, not as forecasts, because quarterly capex is lumpy.

Where this comparison is weakest

A ratio this simple invites over-reading. Here is where it breaks.

Amazon's workforce is a different kind of headcount

Amazon's 1.64 million employees are mostly in fulfilment, delivery and stores, not engineering. Dividing AWS data centre spend by warehouse staff mixes two businesses. A cleaner comparison would use AWS headcount alone, and Amazon does not disclose it. The combined figures are therefore dominated by one company whose labour model has little to do with the AI build.

Microsoft counts its people once a year

Microsoft publishes headcount only in its annual 10-K. Its Q3 figures here use the June number, three months stale, and they cannot reflect the exit offer announced in October 2026. If a meaningful share of the 60,000 eligible employees accept, the next annual figure will show a much steeper fall than the 2.2% here.

Capex and payroll are not the same kind of money

Capex is spent once and depreciated over years. Salaries recur every month. Comparing a quarter of capex with a headcount count compares a stock decision with a flow of people, and it says nothing about returns. Some of the 2026 rise is also price rather than volume, since memory costs pushed several guides higher. Where the depreciation lands is covered in the analysis of server useful life and hyperscaler profit.

None of these breaks the core finding. Capex per employee tripled or more at every company, whichever definition you choose. What they do break is any claim that the ratio measures automation, productivity, or job replacement. It measures where the cash went.

What big tech capex per employee means for your hiring plan

You are not running a hyperscaler, so the useful question is what this shift does to the companies you buy from and compete with for talent.

First, the largest buyers of engineering talent are not expanding their total demand for it. Three of the four big employers in this table ended the period with flat or falling headcount outside Alphabet. That changes the bargaining position on senior hires, and it is part of the context for the argument that AI layoffs are partly an overhiring correction.

Second, your vendors' cost base is shifting from people to depreciation. A software supplier whose platform runs on this infrastructure will price to recover capital, not salaries. Expect usage meters and credit packs rather than seat discounts, a pattern traced in the comparison of revenue per employee at AI-native companies.

Third, resist the copycat move. A founder who reads this table and freezes hiring to fund GPUs is copying a strategy built on hundreds of billions of operating cash flow. I would not. The lesson I take is narrower: measure what each new hire and each compute commitment produced, on the same sheet, and fund whichever wins.

Frequently asked questions

How to use these numbers this week

Put two lines from your own business on one page: last quarter's spend on people and last quarter's spend on compute and software. Then write down what each produced. If you can quantify the output of one and not the other, that is your next measurement project.

When the next round of filings lands in late January, rerun the table above. It takes four documents and eight divisions. Microsoft's buyout uptake and Meta's 2027 budget are the two inputs most likely to move the ratio again.

Related reading