The math is stark: 8,000 people out, $145 billion in.

That’s the deal Mark Zuckerberg laid out at a company town hall last week. No euphemisms about “restructuring for the future.” No corporate doublespeak about “aligning resources.” Just a blunt admission: compute infrastructure and people are Meta’s two major cost centers. With AI hardware costs exploding, something had to give.

That something is 10% of Meta’s workforce, starting May 20th. And Zuckerberg wouldn’t rule out more cuts later this year.

The Numbers That Made Wall Street Flinch

Meta raised its full-year 2026 capital expenditure guidance to between $125 billion and $145 billion — up from an already staggering $115–$135 billion range. The midpoint of roughly $135 billion is nearly double what Meta spent on capex in all of 2025 ($72.2 billion) and more than it spent in 2024 and 2025 combined.

To put that in terms humans can process: Meta’s AI budget could buy Ford, General Motors, and Chrysler. Combined. With change to spare.

Investors weren’t thrilled. Meta stock dropped more than 6% in after-hours trading following Q1 earnings — even as the company reported phenomenal results. $56.3 billion in revenue (up 33% year-over-year). $26.8 billion in net income. Q1 capital expenditure alone hit $19.84 billion.

Meta is printing money and burning it at nearly the same rate, all in pursuit of AI dominance.

A $135 Billion Bet on Vibes

What makes this different from the usual corporate layoff announcement is Zuckerberg’s unusual directness. When an analyst asked what signposts he’s watching to ensure healthy returns on this massive investment, his response was… not reassuring.

“That’s a very technical question,” he said, before pivoting to talk about building “leading models and leading products” without offering specific metrics or timelines. He admitted Meta doesn’t have “a very precise plan for exactly how each product is going to scale month over month.”

CFO Susan Li went further, telling investors she couldn’t predict the company’s “optimal long-term workforce size” given how quickly AI capabilities are evolving. The chief financial officer of one of the world’s most valuable companies is saying she genuinely doesn’t know how many humans the company will need going forward.

Read that again. Let it sink in.

Humans for Hardware: The Great AI Trade

Meta isn’t alone in this calculation, but it’s the most transparent about it. Across the broader tech industry, more than 95,000 workers have lost their jobs in 2026 so far, with nearly half of Q1’s 80,000+ layoffs officially linked to AI or automation.

Here’s the nuance that matters: these layoffs aren’t primarily about AI replacing workers. They’re about AI infrastructure costing so much that companies are cutting headcount to fund the buildout. There’s an important distinction.

Nvidia’s VP of applied deep learning, Bryan Catanzaro, said earlier this week that compute already costs more than the employees on his team. A 2024 MIT study found AI automation was economically viable in only 23% of vision-related roles.

So we’re in a bizarre transitional moment: companies are firing humans not because robots can do their jobs, but because the robots are too expensive to run alongside the humans. The savings from layoffs flow directly into data center budgets, GPU purchases, and power infrastructure.

What $135 Billion Actually Buys

Meta is building a multi-vendor chip strategy that reads like a silicon arms race:

  • Nvidia’s latest systems remain the backbone, but Meta is diversifying away from Nvidia dependency
  • AMD chips in “significant” quantities as a complement
  • Custom MTIA silicon developed with Broadcom — more than one gigawatt of proprietary chips, with four new chip generations planned over the next two years
  • Broadcom CEO Hock Tan is even stepping down from Meta’s board to take a direct advisory role on the custom silicon roadmap

The shift toward custom chips is strategic. By co-designing its own accelerators, Meta aims to optimize performance and total cost of ownership for its specific workloads — ranking algorithms, content recommendation, and increasingly, generative AI for products like Meta AI and the Llama model family.

Zuckerberg also pointed to “memory pricing” as a cost driver. In the AI chip world, high-bandwidth memory (HBM) has become the new gold — expensive, scarce, and essential for running large language models efficiently.

The Broader Arms Race

Meta isn’t spending in isolation. Alphabet and Amazon both reported earnings the same week, pouring comparable sums into AI infrastructure. The difference? Their stocks went up because they showed AI-driven revenue growth in cloud services.

Meta doesn’t have a cloud business to monetize its AI investments through. Alphabet can point to Google Cloud AI revenue. Amazon has AWS. Meta has… better ad targeting and a chatbot.

The total across Big Tech likely exceeds half a trillion dollars in 2026 AI infrastructure spending. That’s not a typo.

Why This Matters Beyond Tech

For workers everywhere: Meta just established a precedent. The most profitable company in social media history is explicitly trading human jobs for AI infrastructure. Every CEO in every industry is watching. The “AI augments workers” narrative just got a lot harder to sell.

For AI development: This spending is what drives rapid improvement in AI capabilities. Every dollar poured into compute means more powerful models and faster training runs. The pain is concentrated now; the benefits will be distributed later. Cold comfort if you’re one of the 8,000.

For startups: When the giants are spending $135 billion on infrastructure, competing on raw compute becomes impossible. Meta’s open-source Llama strategy gets more interesting here — building infrastructure smaller companies can’t afford, then giving away the models to create an ecosystem.

The Uncomfortable Question

What if it doesn’t work?

Meta is making the largest capital allocation bet in corporate history based on the assumption that AI capabilities will continue improving and eventually generate returns that justify the spending. But Zuckerberg himself admits he doesn’t have precise plans. The ROI question gets a hand-wave.

If scaling laws hit diminishing returns, if regulation constrains deployment, or if consumer appetite for AI features plateaus — Meta will have spent the GDP of a small country on infrastructure it can’t fully utilize, while having fired the humans who used to drive its business.

Then again, if it does work — if AI agents become the primary way people interact with digital services — then $135 billion might look like a bargain in hindsight. And the companies that didn’t make the bet will be the ones writing layoff memos.

That’s the trillion-dollar question of 2026: is this the biggest infrastructure investment in history, or the biggest misallocation of capital since the dot-com bubble?

We probably won’t know for another two to three years. But 8,000 people are paying the price for the bet right now.