Meta reported Q3 on October 21. It has 15% fewer employees than a year ago, yet it spent $34.91 billion on capex in a single quarter. If you run engineering, that's the planning signal. My read: budget is moving from payroll to compute, and the people who stay on the payroll are the ones who make that compute work.
The numbers
Per the earnings call transcript on Motley Fool, Meta ended the quarter with over 66,000 employees, down 12% from Q2 after the May 2026 headcount reduction. Meta's 8-K earnings release puts the exact count at 66,775, down 15% year over year.
Spending went the other way. Per the same release, capex including finance lease principal was $34.91 billion, against $19.4 billion a year earlier. Free cash flow was negative $2.12 billion. Revenue was $64.6 billion, up 26%. Operating income fell 37%, largely because of a $9.9 billion litigation charge.
CFO Susan Li narrowed 2026 capex guidance to $135 billion to $145 billion. On 2027, she said only that Meta is "still developing our budget and capacity plans for next year" and expects to "continue investing aggressively."
What management said about the constraint
Zuckerberg said compute would be "1 of the limiting factors over the next year," per the transcript. He said Meta expects to make its Muse models more than 10x more cost-efficient over the next year, and Li framed the same goal as compute cost efficiency.
Notice what that is. The stated limits are machines and the efficiency of running them.
What I take from it
First, the headcount cut and the capex raise are one decision, not two stories. A company can shrink headcount 15% and still badly need specific people. Those aren't contradictory facts. They're a reallocation.
Second, the roles that matter shift. Hitting a 10x cost-efficiency target means inference engineers, systems and kernel people, ML infrastructure, and the data center and power people. General headcount is what gets trimmed. Scarce infrastructure skill is what gets protected.
Third, be careful with the broader read. Meta is one company, with a one-time legal charge and a May layoff that explains much of the drop. Nothing in the call says every company should cut. Don't copy a hyperscaler's org chart because its stock moved.
What to do with it
If you're a CTO or VP of Engineering, three things.
- Know your compute-to-people ratio. If your AI spend is rising and your hiring plan hasn't changed, one of them is wrong.
- Decide which roles are infrastructure-critical and hire those first, before the big companies' efficiency programs start pulling the same people.
- Stop treating "we cut headcount" as proof of AI productivity. Meta's stated 10x target is about compute cost, not labor.
At VC5 we place engineers in the roles that sit behind these numbers.
Alphabet and Microsoft report October 28 and Amazon October 29, per the calendars I found. Watch whether their headcount lines move the way Meta's did.
The machines are getting the raise.
Sources: Meta Q3 2026 earnings release, 8-K Exhibit 99.1; Meta Q3 2026 earnings call transcript (Motley Fool).