Meta corta 8.000 pessoas para gastar US$ 135 bilhões em IA

On Wednesday, April 23, 2026, Meta’s Chief People Officer Janelle Gale sent a memo to staff. Roughly 8,000 jobs would be cut. Another 6,000 open roles would be closed. The departure date for the laid-off workers is May 20.

The memo’s framing did not say AI was replacing workers. It tied the cuts directly to the cost of what Meta is building. “We’re doing this as part of our continued effort to run the company more efficiently and to allow us to offset the other investments we’re making,” Gale wrote.

Those “other investments” have a number attached. Meta’s 2026 capital expenditure guidance is $115 to $135 billion. The 2025 actual number was about $72.2 billion. The 2024 actual number was $39.23 billion.

That is the trade. Meta is firing 8,000 people, closing another 6,000 open positions, and roughly doubling its capital budget for the second year in a row. Capex is not crowding out hiring as a side effect. The memo says it directly: the layoffs exist to offset the spending.

The Capex Curve

Meta’s capital expenditure trajectory does not look like a rounding adjustment. It looks like a regime change.

Year	Capex	YoY Change
2024 actual	$39.23 billion	n/a
2025 actual	~$72.2 billion	+84%
2026 guidance	$115–$135 billion	+59–87%

In two years, Meta has committed to spending roughly three times what it spent on capex in 2024. The 2024 number itself was already an all-time high. The 2026 number is roughly half the annual GDP of New Zealand, in capital spending, by one company.

It is worth pausing on what that money buys. Capex this large is not buying buildings or office furniture. It is buying GPUs (Graphics Processing Units, the parallel-compute chips that train large AI models), high-voltage substations, custom server racks, fiber, water rights, and land near nuclear plants. By the end of 2025, Meta said it would have more than 1.3 million GPUs operating. The 2026 guidance buys whatever the next training run requires.

The thing those capex numbers cannot buy is more time. Every dollar deferred is a competitive concession to Google, Microsoft, and Amazon, and the stretched depreciation schedules hyperscalers carry on their books quietly mask how fast Nvidia’s chip cadence obsoletes the silicon. Once Mark Zuckerberg committed to the AI fight, the spending cadence stopped being optional.

So the question becomes mechanical: how do you fund the incremental $43 to $63 billion of capex year over year? Meta is doing it three ways at once. It is issuing debt. It is closing 6,000 open roles. And it is removing 8,000 employees from the payroll while telling investors the result is “efficiency.”

What 16 Weeks Plus Two Buys

Meta is not being stingy with the exits. Affected U.S. employees receive 16 weeks of base pay, plus an additional two weeks for every year of service, plus 18 months of COBRA health-insurance premium coverage. International packages are described as similar.

For a senior engineer with five years at Meta, that is roughly six months of base salary plus a year and a half of paid health insurance. It is not a brutal severance. It is a managed one, designed to move quickly through HR review, generate minimal litigation, and clear desks before the next earnings call.

The total severance bill for 8,000 employees is meaningful but not enormous on Meta’s scale. At a rough loaded cost of $300,000 per person, eight thousand departures might cost the company on the order of $2.4 billion in one-time outlays. Set against $43 to $63 billion in incremental annual capex, the math does not close on payroll savings alone.

That is one reason the 14,000-position reduction matters more as a signal than as actual capex relief. Meta is already deep into AI debt financing. In October 2025, a Blue Owl Capital joint-venture vehicle named Beignet Investor LLC issued $27 billion in bonds, described in Fortune’s reporting as the largest private-credit transaction on record, to fund Meta’s “Hyperion” data center campus in Richland Parish, Louisiana, with Meta operating the facility as a long-term lessee while keeping most of the debt off its own balance sheet. Meta also issued roughly $30 billion in its own bonds the same month. At those scales, balance-sheet credibility, along with the credit ratings that ride on it, sits alongside cash flow as a binding constraint. Demonstrated headcount discipline is one of the cheapest signals a hyperscaler can send credit markets in that environment.

The Reality Labs Funeral

The April 23 announcement is not a one-off. It is the continuation of a strategic burial that started in mid-January.

On January 13, 2026, Meta’s Chief Technology Officer Andrew Bosworth posted internally that Reality Labs, the division built to deliver Mark Zuckerberg’s metaverse vision, would lose roughly 10% of its staff. The unit had about 15,000 people. More than 1,000 jobs were cut. Four internal VR game studios were closed.

The framing in January was a “pivot to AI wearables and phones.” The framing in April is a “push for efficiency.” Both rounds describe the same company decision: shrinking the headcount built up around the Metaverse pivot of 2021 while expanding the spend on AI infrastructure.

The historical numbers are stark. Reality Labs has accumulated roughly $74 billion in operating losses from 2020 through 2024 (about $6.6B in 2020, $10.2B in 2021, $13.7B in 2022, $16.1B in 2023, and $17.7B in 2024), with further multi-billion losses reported in 2025 quarters that push the cumulative figure toward $80 billion. Meta is reportedly cutting the Metaverse budget by roughly 30% to free up cash for AI. The April layoffs include further reductions across Reality Labs, Facebook social teams, recruiting, sales, and global operations.

The shape of this round echoes the 2022–2023 “Year of Efficiency,” which eliminated roughly 21,000 positions across two rounds (about 11,000 announced on November 9, 2022, and another 10,000 announced in March 2023). That earlier cycle ran alongside, and helped fund, the Metaverse pivot; the current cycle is running alongside the AI capex ramp. The pattern is consistent: each layoff cycle precedes or coincides with a major capital reallocation, and the previous strategic bet’s survivors are typically the ones cut. A similar dynamic is now visible at Oracle, which eliminated 30,000 jobs while accelerating its own AI buildout.

Capex Per Remaining Engineer

To see what is actually happening, divide the capex by the workforce that backs it.

Meta employed roughly 79,000 people at the start of 2026. After the May 20 cuts and the closed open roles, the company will run with closer to 71,000.

$\text{Capex per employee, 2024} = \frac{\$39.23\text{B}}{79{,}000} \approx \$497{,}000$

$\text{Capex per employee, 2025} = \frac{\$72.2\text{B}}{79{,}000} \approx \$914{,}000$

$\text{Capex per employee, 2026 (guided, mid-range)} = \frac{\$125\text{B}}{71{,}000} \approx \$1{,}760{,}000$

The capital intensity of each remaining Meta employee has roughly tripled in two years. Each remaining engineer, designer, salesperson, and lawyer is now responsible for backing roughly $1.76 million of physical infrastructure spend per year. The company is becoming far more capital-intensive than it is human-intensive.

That ratio used to be the signature of a steel mill or an oil refinery, not a social network. It is what AI as a business model looks like once the hyperscaler tier commits to building its own training stacks: software companies start to wear the financial profile of utilities.

The Steelman: AI Productivity Is Real

A credible counterargument deserves serious engagement. Parts of it are true.

AI tooling really has compressed the headcount needed for many software workflows. Coding assistants, agentic IDEs, and document-processing models have measurably accelerated specific tasks inside large engineering organizations. Bain & Company, in a February 2026 analysis, argued that some of the public software multiple compression reflects genuine concern that AI agents replicate “core functionality” of seat-based products. Meta’s leadership says, plausibly, that the company can run leaner now that fewer, more capable engineers can rely on internal AI agents to handle workflow that used to require a team.

The honest version of that case is that some of the 8,000 cuts are organizational redundancy that AI tooling has exposed: managers, recruiters, and middle layers that no longer need to exist when downstream productivity rises. That is consistent with the divisions reportedly affected, including recruiting, global operations, Reality Labs, and Facebook social teams.

The dishonest version of that case is the framing that this round is primarily a productivity-driven cut. It is not. It is paid for by the cost of building the infrastructure that produces the productivity. The story Meta is telling investors is that AI lets the company shrink. The story the capex curve tells is that AI demands the company spend until it bleeds. Both are true; only one is on the press release.

The Microsoft Mirror

The same week Meta cut 8,000 jobs, Microsoft announced something almost identical, dressed in different clothes.

On April 23, 2026, Microsoft confirmed it would offer voluntary buyouts to up to 7% of its U.S. employees. With Microsoft employing about 125,000 U.S. workers, that 7% threshold translates to roughly 8,750 eligible positions. The eligibility rule (workers at senior director level and below whose age plus years of service total 70 or more) concentrates the offer on long-tenured staff. CNBC reported the stated motivation as freeing up cash for the AI data center buildout.

Counted together with smaller rounds at Snap, Oracle, Disney, and Amazon, more than 96,000 tech jobs have been eliminated in the first four months of 2026. The pattern repeats: company by company, the same mechanism. Cut a percentage of the workforce, raise capex guidance, tell shareholders the result is “efficiency.” Buy GPUs.

The Year of Efficiency, Take Two

There is no mystery in what Meta is doing. There is only the question of whether the bet pays off.

The 2022–2023 cuts funded the Metaverse pivot, which by Meta’s own accounting cost about $70 billion and produced no consumer business of meaningful scale. The 2026 cuts fund the AI pivot: Llama models, in-house silicon, custom data centers, and the hyperscale energy footprint needed to run them.

Mark Zuckerberg’s history on capital allocation at this scale is mixed. He committed roughly $70 billion to a virtual world nobody wanted. Now he is committing $115 to $135 billion to AI infrastructure that, if it works, runs on a stack his rivals are also building. If it does not work, the 2027 layoffs will be larger than the 2026 ones, since the capex commitments have already been made.

This is the second-order problem with the “Year of Efficiency” framing. It has now happened twice in three years, each time in support of a different strategic bet. A company that periodically discards a tenth of its workforce to finance the founder’s next big idea is not running on efficiency. It is running on serial conviction. The cost is borne by the people who joined to build the last vision.

What Comes Next

Three things are likely in the next twelve months.

The pattern propagates. Microsoft is already mirroring it. Amazon, Google, and Apple have not announced cuts on this scale, but each is raising AI capex into 2026 against the same backdrop of credit-market scrutiny. If the bond market rewards Meta’s capex-funded discipline, hyperscaler CFOs now have a template to copy. More than 96,000 tech jobs eliminated through the first four months of 2026 are more likely a floor than a ceiling.

The labor market re-prices. Meta is expected to keep paying top AI engineers extreme compensation; that is the compute war’s premium tier. But the broader middle of the tech labor market is being squeezed, and CNBC has framed the combined Meta and Microsoft cuts as evidence of an “AI-driven labor crisis.” The reported divisions affected at Meta (recruiting, global operations, Reality Labs, sales, and Facebook social teams) are weighted toward middle-tier roles rather than AI research, and Microsoft’s eligibility cutoff at senior director and below points the same direction.

The infrastructure fight gets physical. The capex Meta is committing requires power, water, and land at a scale that local communities increasingly refuse to provide. The same week Meta announced the cuts, this site documented the drought-driven moratoriums that fourteen U.S. states have now passed against new data center construction. Capex on a balance sheet does not become a server rack until somebody approves a substation. The 2026 guidance number assumes that approval keeps coming. It may not.

Janelle Gale’s memo is notable for tying the layoffs and the capital plan together in the same sentence, rather than presenting them as separate decisions. The previous “Year of Efficiency” was framed as a return to discipline after the Metaverse hangover. This one is framed, more directly, as offsetting the AI infrastructure spend. The bond market funds the GPUs. The 8,000 employees pay with their jobs to make the discipline narrative credible.

The cooling towers are going up. The desks are coming down.