The Compute Cannibals

On May 1, Mark Zuckerberg stood in front of his employees at a company town hall and said something that most CEOs only imply. He explained the layoffs with a formula: the company has two major cost centers, compute infrastructure and people, and if you invest more in one, you have less for the other. Then he announced which one he'd chosen.

8,000 employees will be cut starting May 20.¹ That's 10% of Meta's workforce. More cuts are planned for the second half of 2026. Zuckerberg told the room he doesn't have a "crystal ball" for how things will play out, and he wouldn't promise it stops here.²

The same week, Meta raised its 2026 capital expenditure forecast to $125 billion to $145 billion, up from the already staggering $115 to $135 billion range it had projected just months earlier.³ That's roughly double what the company spent in 2025. The stock dropped 6% after hours.³

Investors flinched, but employees are about to do worse than flinch.

The Quiet Part

Here is the exact quote from the town hall: "We basically have two major cost centers in the company: compute infrastructure and people-oriented things. If we're investing more in one area to serve our community, then that means we have less capital to allocate to the other. So that means we do need to take down the size of the company somewhat."²

Notice what he's not saying. He's not claiming AI makes certain roles obsolete or that the work disappeared. He's saying the GPUs cost so much that there isn't enough budget left for the people. The jobs didn't get automated — they got outbid.

This is a new framing. The standard layoff story since 2023 has been: AI can do this work now, so we don't need as many people doing it. Zuckerberg is saying something more honest and more disturbing: we need the money that was paying your salary to buy GPUs instead.

He even preempted the obvious objection. When asked whether internal AI efficiency gains were the real driver, he said: "Getting everyone internally to use AI tools and getting to do the work more efficiently is not the thing that's driving layoffs."² It's not that AI replaced your job. It's that AI's electricity bill replaced your salary.

The Numbers

Where is the money going?

Meta signed a $27 billion deal with Nebius for dedicated AI compute capacity over five years, including early deployments of Nvidia's Vera Rubin chips.⁴ It's building a massive data center campus in Richland Parish, Louisiana, where the project has nearly tripled from its original $10 billion estimate to $27 billion.⁵ It invested $14.3 billion in Scale AI and installed its 28-year-old founder, Alexandr Wang, as Chief AI Officer, leading a new "Meta Superintelligence Labs."⁶

In Q1 2026, Meta posted $56.3 billion in revenue (up 33%), $22.9 billion in operating income (up 30%), and $26.8 billion in net profit (up 61%, boosted by an $8 billion tax benefit).³ This is not a company in financial distress. This is a company that decided GPUs are a better investment than people, and said so out loud.

The Pattern

Meta is not alone. It's just the most honest.

The five largest US cloud and AI infrastructure providers have collectively committed to spending between $660 billion and $690 billion on capital expenditure in 2026, nearly doubling 2025 levels.⁷ Analysts project Big Tech capex will top $1 trillion in 2027.⁸

The same week Meta announced its cuts, Microsoft offered voluntary buyouts to roughly 8,750 employees, about 7% of its US staff. Together, the two announcements pushed 2026 tech layoffs past 92,000. CNBC ran the headline that economists were already dreading: "AI-driven labor crisis is here."⁹

As of May 1, the tracker reads 100,443 tech workers laid off in 2026 across 155 events. That's over 800 per day.¹⁰ Nearly half of Q1's cuts were attributed to AI.¹²

We covered the pace in early April, when the daily rate was 963. It hasn't slowed, but what's changed is the justification. In March, companies said AI could do the work. In May, Zuckerberg said AI needs the budget.

The Cannibal Logic

This is where the compute cannibal thesis gets uncomfortable.

If you accept Zuckerberg's framing, humans and GPUs compete for the same pool of capital. That competition has a structural winner. GPU prices are set by Nvidia. Demand is functionally unlimited. Every company in the sector is racing to build the same infrastructure because none of them can afford to be the one that didn't. The cost of compute only goes up.

The cost of people, in this framework, is the flexible variable. When the GPU bill rises, headcount adjusts downward — not because the people aren't productive, but because the GPUs are more expensive.

This is the cannibal logic: the infrastructure eats the workforce. Not because it replaces the work, but because it outprices the workers.

Meta has now eliminated approximately 25,000 positions since 2022.¹¹ The first wave, 11,000 in late 2022, came from the metaverse spending crash. The second, 10,000 in 2023, was Zuckerberg's "Year of Efficiency." The third and fourth, in 2025 and 2026, are the AI pivot. Each round had a different story. Each round had the same outcome.

A Stevens Institute of Technology professor told Fortune: "At the very beginning, layoffs were something that he had to do. Now, it seems to be a norm."¹¹

The Competitive Trap

Here's why this doesn't stop with Meta.

When one company publicly frames headcount as a competing line item against infrastructure, it gives cover to every other company to do the same. The Zuckerberg formulation is a ready-made board-room slide: We have two cost centers. We're choosing the one with long-term strategic value.

Amazon has its own AI capex surge, projected at $200 billion for 2026. Alphabet is at $175 to $185 billion. Microsoft is tracking toward $120 billion. Oracle, $50 billion.⁷ Every one of them is making the same bet. Every one of them will face the same budget pressure. The infrastructure bill doesn't share the budget with headcount — it takes from it.

Meta's performance review system already reflects this. Managers have been directed to rate 15-20% of employees as "below expectations." The bottom 3% are pushed out. The top 20% get up to 300% bonuses.¹ The squeeze works from both directions: shrink the body, reward the survivors, and redirect the savings to the data center in Louisiana.

What We're Watching

May 20 is the date. That's when the 8,000 start getting their notifications. Meta hasn't said which teams are hit hardest, but earlier rounds in 2026 targeted Reality Labs (1,000-1,500 in January), recruiting, sales, and global operations (700 in March).¹

Zuckerberg said he'll "be able to share more soon" about whether more cuts are coming. He also said he doesn't have a crystal ball. Both of those things can be true. Neither of them is comforting if you're employee number 8,001.

The compute cannibals have a new spokesperson. He's being refreshingly honest about what the GPUs cost and who pays for them. The answer, across 25,000 former Meta employees, has been consistent: you do.