For the past few years, a large part of the tech industry has run the same experiment: cut people, spend on AI instead, assume the tools absorb the difference. Two of the most documented cases — Meta and Klarna — show how that bet is actually playing out.
The layoffs:
CEO Mark Zuckerberg told staff the cuts were tied directly to rising AI infrastructure spending — compute and people are the company's two major cost centers, and more for one means less for the other.
The spending:
2026 capex guidance raised to $125–145 billion, up from $115–135 billion, and nearly double the $72.2 billion spent in all of 2025.
Q1 2026 AI infrastructure costs alone surged 35% year-over-year to $33.44 billion — driven by data centers, depreciation, and cloud spend.
The company was spending roughly $315–370 million per day on infrastructure.
The math that doesn't add up cleanly:
Wedbush estimated the 8,000-person layoff saves about $2.4 billion a year — a rounding error next to the AI spend.
Free cash flow was projected to fall from $43.6 billion (2025) to just $8.5 billion (2026) — an ~80% collapse.
Meta's own CFO, Susan Li, admitted: "We don't really know what the optimal size of the company will be in the future," and that Meta has "continued to underestimate our compute needs."
The takeaway: Thousands of jobs were cut for savings that were dwarfed by infrastructure spending the company itself says it can't fully predict.
The original bet (Feb 2024):
The reversal (about a year later):
CEO Sebastian Siemiatkowski admitted the aggressive AI-for-jobs swap had led to diminished service quality.
He acknowledged the company over-focused on cost-cutting at the expense of service, and that human interaction was still necessary for customer satisfaction.
Why it broke:
Customer satisfaction dropped specifically on complex, non-routine interactions.
Undoing the layoffs wasn't free — recruiting, onboarding, and training staff back up is expensive, and rarely modeled into the original AI business case.
The lesson that's now industry shorthand:
Humans are still needed for escalations, emotionally complex cases, and anything requiring judgment.
Hybrid models consistently beat full automation on both cost and satisfaction.
| Meta | Klarna | |
|---|---|---|
| Cut | ~8,000 jobs | ~700 agent roles |
| Bet | AI infra spend justifies smaller headcount | AI agent replaces customer support |
| Result so far | Layoffs done, spend still climbing, outcome uncertain | Reversed — rehiring humans after ~1 year |
| Company's own admission | "We don't know the optimal size" | "Overemphasis on cost-cutting hurt quality" |
Three things both cases have in common:
If you've been laid off, pushed into unsustainable hours because "AI should make everything faster," or asked to do more with less on the assumption a tool would absorb the gap — you're not an outlier. You're a data point in a pattern now documented at some of the largest companies in the world, run by people with far more information than any individual manager or founder had.
This isn't a rejection of AI industry-wide. It's a recalibration toward hybrid models — AI handles routine and volume, people handle judgment, escalation, and everything that was never really about typing speed to begin with. Companies that understood that early are ahead. The rest are learning it the expensive way — one earnings call, one rehire, one Klarna-shaped headline at a time.