On May 20, 2026, roughly 8,000 Meta employees receive their layoff notifications. That same week, CFO Susan Li announces that the 2026 AI capex is climbing to a range of 145 billion. The two numbers are not independent. They are connected by a phrase from the April 30 town hall: Mark Zuckerberg speaks of "two big cost centers" competing for the same capital. This article breaks down that equation, its winners, its losers, and the counter-narrative that few outlets picked up.
Susan Li's equation: 14,000 jobs vs. $145 billion
During the April 29, 2026 earnings call, Susan Li did not dance around the issue. The 2026 capex is raised by roughly $10B versus prior guidance, driven by a mix of higher component costs and datacenter needs for future capabilities. In the same sequence, she announced a headcount reduction plan "in May."
“We plan to reduce the size of our employee base in May. A leaner operating model will allow us to move more quickly while also helping to offset the substantial investments we are making.
”
The phrase "to offset the substantial investments we are making" is not a rhetorical flourish. It's an explicit budget transfer. At the April 30 town hall covered by Bloomberg, Zuckerberg put it even more bluntly: "two big cost centers, compute infrastructure and humans. If we invest more in one, we have less capital for the other."
The numbers land against a certain narrative. Meta reported a record quarter: 26.77B in net profit, 3.56 billion daily active users. It is precisely in this record-setting context that 10% of the workforce is going. The logic isn't survival, it's arbitrage.
Markets did not applaud the record. Meta stock fell 6% in after-hours following the earnings call, signaling structural concern about the ROI of capex that doubles in a year. Asked about expected returns, Zuckerberg replied: "That's a very technical question."
The Wang bet: the end of LeCun's monopoly
To understand why Meta accepts this budget transfer, you need to go back to June 2025. Meta acquires 49% of Scale AI for $14.3B and recruits its founder Alexandr Wang, 28, as the group's first Chief AI Officer. Wang leads a new entity called Meta Superintelligence Labs.
This hire triggered an internal earthquake. In November 2025, Yann LeCun, longtime Chief AI Scientist and Turing Award recipient, left Meta after being required to report to Wang, who is 41 years his junior. LeCun went on to found AMI Labs with $1.03B in funding.
- June 2025Scale AI acquisition
Meta invests $14.3B for 49% of Scale AI and recruits Alexandr Wang.
- Nov. 2025Yann LeCun departs
The deep learning pioneer leaves Meta after 12 years, refuses to report to Wang. Launches AMI Labs.
- April 2026Muse Spark
First model from Meta Superintelligence Labs, presented as a step toward "personal superintelligence."
- Apr 22, 2026MCI revelation
Reuters exposes the Model Capability Initiative, a program recording employees' keystrokes and screenshots.
- Apr 23, 2026Janelle Gale memo
Bloomberg publishes the internal memo announcing 8,000 layoffs effective May 20.
- May 20, 2026Effective date
The 8,000 affected employees receive their notifications.
Add to this the Llama 4 benchmark manipulation scandal in early 2026. The legacy AI team lost credibility, accelerating consolidation around Wang. The internal structure became dual: Superintelligence Labs (Wang) on the frontier side, Applied AI Engineering (Maher Saba) on existing products. Several analysts see it as internal hedging against the risk of Labs failure.
When a company pays $14.3 billion to hire a 28-year-old as division head, that's not a recruitment. That's a declaration of intent about who will decide the next decade.
Why four hyperscalers are burning $700 billion at the same time
Meta's move can't be explained in isolation. The four American hyperscalers, together, plan $650-700B in capital expenditures in 2026. That's the largest concentrated infrastructure investment cycle in modern tech history.
About 75% of that capex is dedicated to AI, roughly $450B in AI-specific spending. Meta's capex-to-revenue ratio at 54% is the highest of the four. Concretely: more than half of what Meta's advertising business takes in goes back out to datacenters, Nvidia/AMD/Broadcom GPUs, and network infrastructure.
CNBC frames the ratio differently: Meta's AI capex represents four to five times its total annual payroll, estimated at around $27B. Humans still cost money, but they're no longer the decisive variable in the model.
The counter-narrative: Gartner, Acemoglu, and the missing ROI
This is where the official story starts to crack. In May 2026, Gartner publishes a report covering 350 global executives piloting or deploying AI agents.
“Workforce reductions may create budget room, but they do not create return.
”
The detail matters. About 80% of companies in the sample reduced headcount, but ROI rates don't correlate with reduction rates. In other words: those who cut more don't get more return. Gartner positions 2028 or 2029 as the horizon at which autonomous AI would become a net job creator, provided new forms of work emerge in time.
On May 11, 2026, MIT Technology Review publishes an interview with Daron Acemoglu, 2024 Nobel Prize in Economics. His reading is even more cautious.
“There's a huge amount of uncertainty about AI's actual economic trajectory, despite confident rhetoric from industry leaders.
”
Acemoglu estimates that AI will deliver only a modest productivity boost to the US economy, and that adoption will be measured in decades, not years. Capability gaps, hallucinations, and the difficulty of organizational integration are slowing the real pace. This voice stands out precisely because it doesn't come from an AI skeptic, but from a Nobel laureate who has spent his career modeling technological transitions.
If Gartner and Acemoglu are right on timing, then the trade-offs Meta, Microsoft, Amazon, and Alphabet are making in 2026 rest on a bet: that ROI will arrive before markets lose patience. Meta's -6% post-earnings suggests that patience is already thin.
The MCI program: monitoring workers to replace them
The story takes a more uncomfortable turn on April 22, 2026, the day before the Bloomberg memo. Reuters reveals the existence of the MCI program, for Model Capability Initiative. Meta records keystrokes and screenshots of its own employees to train the AI agents that could, eventually, automate their work.
This sequence fueled the April 30 town hall, where several employees challenged the logic of a record quarter ending with 10% of the workforce cut. Zuckerberg acknowledged having no three-year plan: "I don't have a crystal ball." That phrase, unremarkable on its surface, leaves a lot of uncertainty for the 70,000 remaining employees.
For the 8,000 departing, the calendar is known: notification on May 20, California WARN Act requiring 60 days of paid notice for California-based employees, final payment around August 14, 2026. H-1B visa holders have 60 days to find a new sponsor or face forced departure. This tension between US immigration policy and AI-driven layoffs has yet to find a political answer.
Going further
The Q1 2026 earnings call on April 29 is the most direct source for hearing Susan Li explicitly connect layoffs to capex. The first 15 minutes contain the core sequence of numbers and strategy.
Sources used for this analysis:
The question that remains
If you're an engineer, ops, or product person at a tech company, the Meta signal isn't "Meta is laying people off." The signal is "even a record quarter no longer protects against a capital arbitrage decision." For teams managing their trajectory in 2026, two questions have become priority: where does my work create irreducible value, and how do I make that value legible to finance.
Discuss your stack and AI trajectory with Blokby