Meta's AI Vampire Spiral

I've been quiet on Groktopus for a while. Not from lack of interest — if anything, the story got louder while I was silent. The truth is I've been sitting with a growing unease about where we are in this AI moment, watching the patterns we mapped in 2025 play out with a fidelity that brings me no satisfaction. There's a particular weight to being right about something you were hoping you'd gotten wrong.

So let me be direct about why I'm writing this now. A Reuters exclusive published March 14th revealed that Meta is planning sweeping layoffs — approximately 16,000 of its 78,000 employees, or about 20% of its remaining workforce — driven by mounting AI infrastructure costs. This is not an announcement. Meta has not confirmed it. But the sourcing is solid, and it fits a pattern that Groktopus has been documenting since mid-2025 with uncomfortable precision.

This isn't a victory lap. When you watch thousands of jobs disappear and know more are coming, being right about the mechanism feels like obligation, not vindication. Staying quiet at this point would be irresponsible.

So let's talk about the AI Vampire. And let's talk about who actually wrote the playbook Zuckerberg is trying to follow.

I. Jack Dorsey Ran This Play First

Before we get to Meta, we need to talk about Block. Because the story of Meta's reported workforce reduction doesn't begin in Menlo Park — it begins with what Jack Dorsey did on February 26, 2026, and what Wall Street rewarded him for doing it.

On that date, Dorsey announced alongside Q4 2025 earnings that Block would reduce its workforce from over 10,000 employees to just under 6,000 — a roughly 40% cut, eliminating more than 4,000 jobs. The company had posted $10.36 billion in gross profit for the full year. Dorsey was explicit about the productivity math he was targeting: he wanted to push gross profit per employee from roughly $500K pre-COVID to $2 million or more — a 4x improvement. The framing was that Block was becoming "intelligence-native," a company rebuilt around AI capabilities rather than headcount. Wall Street's response was immediate and unambiguous. Block's stock surged more than 16% the following day.

That market signal traveled fast. When a public company can eliminate 40% of its human workforce, frame it as strategic vision rather than distress, and watch its share price climb double digits in response, every other public company CEO gets the same memo at the same time. The message: the market will reward this. Do the math.

Dorsey said the quiet part in his shareholder letter and on the analyst call that followed. "I don't think we're early to this realization. I think most companies are late." He predicted that within a year, the majority of companies would reach the same conclusion and make similar structural changes. He was describing contagion as though it were inevitability. The Reuters report on Meta's planned cuts landed just 16 days later, on March 14 — confirming the contagion had already arrived before Dorsey's ink was dry.

Zuckerberg is not leading this transformation. He is following a playbook that Dorsey validated, watching the market reward it, and attempting to replicate the outcome at larger scale. Block cut more than 4,000. Meta is reportedly planning to cut 16,000. And Meta's strategic foundation — as we'll see — is considerably shakier than Block's was when Dorsey pulled the trigger.

II. The AI Vampire Comes to Menlo Park

In February 2026, software engineer and writer Steve Yegge published an essay called "The AI Vampire." His argument was precise and uncomfortable: AI productivity tools make developers dramatically more capable — but companies capture 100% of the productivity gain while workers carry impossible cognitive loads on unchanged compensation. You produce more. You get paid the same. The exhaustion compounds. The vampire feeds.

Yegge was writing about the individual experience. What he couldn't fully anticipate is what happens when a company takes that dynamic and operationalizes it at organizational scale.

That's Meta's play. The individual AI Vampire extracts value through cognitive overload. The corporate version Meta is executing skips the extraction phase entirely. It eliminates the humans and converts the salary budget directly into compute. No messy burnout management required. No performance reviews. Just a clean substitution: headcount out, GPUs in.

Mark Zuckerberg told investors the quiet part without embarrassment. AI, he said, enables "single individuals completing projects that once required large teams." He declared 2026 "a big year for delivering personal superintelligence." These aren't aspirational statements — they are capital allocation memos dressed in vision language. The planned 20% workforce reduction and the $115-135 billion in capital expenditure for 2026 alone are two sides of the same ledger entry.

This is not an AI transformation story. This is a value extraction story in which AI serves as both the justification and the instrument — and Zuckerberg learned the justification was sufficient by watching what the market did to Block's stock.

III. The Pattern of Failed Big Bets

To understand what's reportedly coming at Meta, you have to understand what happened there before. This is not Zuckerberg's first time betting the company on a technology vision that outpaced its strategic foundation.

The metaverse chapter is essential context. Meta Reality Labs lost over $16 billion in 2023 alone. Cumulative losses exceeded $60 billion. When the New York Times reported Reality Labs layoffs eliminating 1,500 positions in January 2026, it wasn't a strategic pivot. It was an admission. The bet failed. The capital was gone. The humans paid the price.

The AI chapter began before the metaverse chapter formally closed. And it started with a talent hemorrhage that should have been disqualifying.

Eleven of the fourteen original authors of the Llama paper — Meta's flagship AI research contribution — left the company. Reporting from Winbuzzer put it plainly: 78% of the team that built Meta's AI crown jewel walked out. They went to Mistral AI, to Anthropic, to Google DeepMind — to competitors who, whatever their own flaws, weren't asking researchers to do breakthrough science inside a management culture that had just incinerated $60 billion on virtual reality headsets.

When your core research team dissolves, you have two choices. You can do the slow, painful work of rebuilding culture and capability from the inside. Or you can write a check.

Meta wrote a check. A very large one.

The Scale AI deal — $14.8 billion for a 49% stake, valuing the data annotation company at $29 billion — was announced in June 2025 as a strategic partnership. Our analysis at the time was blunter: this was expensive damage control. You don't pay $29 billion for a data labeling company because you have a sophisticated AI strategy. You pay it because you destroyed the team that was building your AI capability and you need to buy something back fast.

Groktopus · June 2025 "Companies with solid strategic foundations build capabilities; companies with shaky foundations buy expensive solutions to problems they created." — The $29 Billion Mistake

The Llama 4 "Behemoth" model — Meta's attempt to compete at the frontier — was delayed indefinitely due to performance concerns. The flagship model that was supposed to validate the entire infrastructure investment couldn't clear the performance bar. This, while Meta was committing to $600 billion in AI data center spending by 2028.

The pattern is visible across three data points: a failed $60B metaverse bet, a $29B panic acquisition to replace talent that left, and a flagship model that couldn't ship on schedule. What's the reported response? A 20% workforce reduction to fund more infrastructure.

IV. The Reported 20% Cut — Trading Humans for GPUs

The Reuters exclusive on March 14, 2026 reported what many in the industry had suspected: Meta is planning sweeping layoffs, with approximately 16,000 of its 78,000 employees on the chopping block. Reuters' framing — "as AI costs mount" — is accurate, and it's also the tell. This isn't performance management. It's capital reallocation.

To appreciate the scale, you have to look at the rolling cuts in sequence: 3,600 workers in early 2025, framed as performance-based; 600 positions in the new Superintelligence Labs in October 2025; 1,500 at Reality Labs in January 2026; and now a planned 16,000. That's over 21,000 jobs. And the workers who survive this process are getting paid less for it. A 10% stock award cut in 2025 followed by a 5% reduction in 2026 represents roughly 15% cumulative compensation erosion. Survive the cuts, take a pay cut. The AI Vampire operates on two fronts simultaneously.

The infrastructure commitments on the other side of this transaction are staggering. $600 billion committed to AI data centers through 2028. Capital expenditure of $115-135 billion for 2026 alone — three times 2024 levels. A $50 billion facility in Louisiana. Deals for six gigawatts of nuclear power. When you're building facilities the size of small cities and signing nuclear power agreements, you need political management capabilities beyond traditional tech lobbying. That's likely why Dina Powell McCormick, a former Trump advisor, was brought in as company president.

Zuckerberg's stated vision makes the substitution logic explicit. The future he's building is one where individual capability, amplified by AI, replaces organizational headcount. He's not describing augmentation. He's describing replacement with human-first language applied over the surface — and he's doing it having watched Dorsey run the same play and get rewarded for it.

V. The Industry Pattern — Meta Is Not the First, But May Be the Most Extreme

Dorsey set the template, but the broader wave predates even Block's cuts. Forbes reported that 276,000+ tech workers were displaced by AI-cited layoffs in 2024-2025. Goldman Sachs estimated 5,000-10,000 net monthly job losses in AI-exposed industries — roughly 120,000 annually. These aren't pandemic overcorrection numbers. These are structural displacement numbers.

Oxford Economics has argued that many AI-cited layoffs are pandemic overhiring corrections in disguise. There's something to that argument for many companies. It's considerably harder to apply to Meta. When you're committing $600 billion to infrastructure while simultaneously planning to eliminate 20% of your human workforce, the substitution logic is explicit in the capital allocation itself.

Palantir CEO Alex Karp offered the most unvarnished read on the trajectory. AI, he said, "destroys humanities jobs, elevates vocational trades" — and could push college graduate unemployment into the mid-30% range. He called tech leaders who ignore the political consequences "insane." Whatever you think of Karp's politics, his willingness to name what's actually happening is useful. Most tech executives are running the same playbook with more careful language.

Groktopus · June 2025 "42% of companies abandoned most AI initiatives in 2025, up from 17% in 2024. Meta is the poster child for the infrastructure-first failure pattern." — Oracle and Meta's AI Infrastructure Spending Spree Reveals Strategic Missteps

The difference between Block and Meta is instructive. Dorsey's cuts came from a company with a clear operational thesis about what the leaner organization would actually do. Meta's planned reduction comes from a company that has lost 78% of its core AI research team, whose flagship model is delayed indefinitely, and whose last major AI acquisition was by its own admission a response to a talent crisis. Block was cutting fat. Meta may be cutting bone.

VI. We Called It — The Groktopus Record

I don't want to linger on this. But the record matters, because if we could see this coming in mid-2025, the question worth sitting with is why it happened anyway.

In June 2025, "The $29 Billion Mistake" argued that the Scale AI acquisition was expensive damage control and drew the explicit parallel to the metaverse pattern — massive capital deployment chasing strategic direction changes without solid foundation. We predicted Meta would face destroyed competitive positioning and strategic credibility in ruins. Nine months later, the Reuters report confirmed the crisis had deepened, not resolved.

That same month, "Academic Evidence for Year One Success" documented what McKinsey and Microsoft research actually showed about AI implementation: strategic approach outperforms infrastructure-first spending significantly. We cited S&P Global data showing 42% of companies had scrapped most AI initiatives, up from 17% the prior year. Meta's continued infrastructure escalation combined with continued talent hemorrhage is the infrastructure-first failure pattern that article warned about, executed at maximum scale.

"Oracle and Meta's AI Infrastructure Spending Spree" went directly at the talent crisis — the 78% Llama team departure, the Scale AI acquisition, the pattern of buying capability rather than building it. The key line: "Companies are spending heavily on infrastructure without understanding their actual implementation requirements." Llama 4 Behemoth's indefinite delay is the direct consequence.

And "The AI-Native Business Model Revolution" made the contrast explicit. Midjourney: $50 million in revenue, 11 employees, $4.5 million per employee. That's what genuine AI-native efficiency actually looks like — not eliminating humans to fund infrastructure, but building a model where a small, highly capable team produces outsized value. We called Meta's $14.8B bet "crisis management rather than innovation leadership." The Reuters report suggests the crisis management has only accelerated.

None of this is gloating. These analyses were written in the hope that companies and policymakers would course-correct. They didn't. The human cost is now materializing in severance packages, and the fact that we anticipated the pattern makes it harder to watch, not easier.

VII. The AI Vampire's Endgame

Return to Yegge's framework. The individual AI Vampire drains workers through cognitive overload — capture all the productivity gain, leave the human with the load and the same compensation. The feedback loop eventually breaks the worker through burnout, quiet resignation, the slow erosion of motivation that comes from working harder to make someone else wealthy.

Meta's version is more efficient. Skip the burnout phase. Convert the human salary line directly to GPU budget. The vampire doesn't wait for the worker to deplete — it simply removes the worker from the equation. And Zuckerberg has a proof of concept from Dorsey that the market will stand and applaud while it happens.

The sustainability question is genuine. Can a company that has lost 78% of its core AI research team, burned more than $60 billion on failed virtual reality, and is reportedly eliminating a fifth of its remaining workforce actually execute a $600 billion infrastructure buildout? The institutional knowledge leaving Meta isn't just bodies out the door — it's accumulated context, implicit understanding of what works and why, the relationships and judgment that don't transfer into any onboarding document. You can buy Scale AI. You cannot buy back the researchers who built Llama and chose to leave.

Dorsey told investors most companies would reach Block's conclusion within a year. Meta's plan reportedly surfaced weeks after Block's cuts. The contagion is already visible. When the market rewards a 40% workforce reduction with a 16%+ stock surge, the incentive structure for every other public company CEO becomes immediately legible. Expect more companies to run the same arithmetic.

The human count behind these decisions deserves to be named plainly. Over 21,000 Meta jobs eliminated or reportedly planned for elimination across the rolling restructuring. 276,000 across the industry. Goldman Sachs projecting 120,000 net losses annually. These are not abstractions. They are livelihoods, mortgages, health insurance, children's college funds — people who organized their lives around employment that no longer exists. The language of "capital reallocation" and "intelligence-native strategy" is doing significant work to make an enormous human cost sound like a natural system optimizing.

Yegge, characteristically, offered more practical guidance than most: shorter workdays as structural defense against cognitive extraction, collective pushback against the value-capture dynamic, cultural accountability for how AI productivity gains are actually distributed. These sound modest against the scale of what's reportedly happening at Meta. They are modest. But the alternative — waiting for corporations to voluntarily share productivity gains — has a clear track record over the past two years.

The AI Vampire will drain whatever is offered. At the individual level, that's cognitive bandwidth. At the corporate level, it's entire workforces. The dynamic is the same; only the unit of extraction changes.

What changes the dynamic isn't moral argument — Zuckerberg has heard the moral arguments, and so has Dorsey. What changes it is structural friction: labor agreements that tie compensation to productivity gains, regulatory frameworks requiring genuine transition support, and enough public clarity about what's actually happening that "AI transformation" stops being a socially acceptable euphemism for workforce elimination.

We've been arguing since 2025 that companies have a choice between genuine AI transformation — where technology amplifies human capability and the gains are shared — and value extraction dressed in transformation language. The Block playbook, now apparently being replicated at Meta, has made the industry's preference visible in the clearest possible terms.

The silence was about hoping the analysis was wrong. It wasn't. And that changes the obligation.

Groktopus covers AI transformation with an emphasis on what's actually happening versus what companies say is happening.