Mark Zuckerberg has spent the last three years telling investors, employees, and the world that Meta is no longer a social media company. It is an AI company. The metaverse was a distraction. The future is intelligence.
The company is spending $135 billion in 2026 to prove it.
That number is not a typo. One hundred and thirty-five billion dollars in capital expenditure in a single fiscal year. More than the entire GDP of Hungary. Nearly triple what Meta spent two years ago. Enough to buy Netflix twice with change left over.
And the flagship AI model that is supposed to justify all of that spending just failed its own internal tests.
What Happened With Avocado
Meta’s next-generation AI model, internally codenamed Avocado, was supposed to launch this month. It was supposed to be Meta’s first true “frontier-class” model, competitive with GPT-5.4 and Google’s Gemini 3.0 in every major capability category.
It is not launching this month.
The New York Times reported that internal evaluations showed Avocado underperforming competitors on reasoning, coding, and writing. Not marginally. Enough that Meta’s leaders have been internally debating whether to license Google’s Gemini temporarily to power Meta AI features while Avocado gets fixed. The launch has been pushed to at least May.
That is a significant moment. Meta has been positioning Avocado as the culmination of years of research and the reason a $135 billion capex budget was justified. The model was supposed to be the proof point. Instead it became the problem.
To understand the stakes, consider what Meta is spending this money on. Unlike Amazon, Microsoft, and Google, which operate cloud businesses that directly monetize the computing power they build, Meta has no cloud platform. It cannot rent out its data centers to enterprise customers. Every dollar it spends on AI infrastructure produces value only if Meta’s own AI products are good enough to drive advertising revenue and platform engagement.
If Avocado underperforms, the $135 billion does not earn its keep.
Fifteen Thousand Jobs Gone
The layoffs were announced alongside the capex guidance, which is not a coincidence.
Meta is cutting roughly 20% of its workforce across 2026, approximately 15,000 jobs. Zuckerberg’s internal framing is “Efficiency 2.0,” the second act of the cost discipline narrative that helped Meta’s stock rally sharply in 2023 after its disastrous metaverse spending period.
The logic is straightforward: free up payroll costs to fund AI infrastructure without alarming investors about overall expense growth. Pay fewer humans, build more compute.
But the execution is already generating legal liability. A lawsuit filed Tuesday in San Francisco County Superior Court alleges that Meta targeted older workers disproportionately in its layoffs. Nicolas Franchet, a 54-year-old former senior director who spent 13 years at the company before being let go last year, claims the data Meta provided to terminated employees shows workers 50 and older were 2.5 times as likely to be laid off as workers under 40.
Franchet had 16,353 unvested shares worth nearly $12 million on the day he was terminated. He is not alone in this claim. Age discrimination suits against Silicon Valley companies have been building for years. This one has specific internal company data attached to it, which makes it harder to dismiss than most.
The Metaverse Is Dead. Meta Just Won’t Quite Say It.
On March 18, Meta announced that Horizon Worlds, the virtual reality social platform that was supposed to be the centerpiece of its metaverse ambitions, would be removed from the Quest VR headset store at the end of this month and shut down entirely for VR on June 15.
It will survive only as a mobile app. A mobile app. The platform that was going to replace the internet as we know it is now a phone app.
The announcement came weeks after Meta cut more than 1,000 employees from Reality Labs, the metaverse division. Reality Labs has posted operating losses exceeding $70 billion since 2021. That is not a rounding error. That is seventy billion dollars lost chasing a vision that users declined to adopt.
Zuckerberg changed the company’s name from Facebook to Meta in 2021 specifically to signal the metaverse pivot. He called it “the next frontier.” The next frontier is now a mobile app with declining user numbers and a parent company that no longer wants to talk about it.
The pivot from metaverse to AI is not a course correction. It is a complete abandonment, dressed up as a natural evolution. Anyone who invested in Meta’s metaverse narrative paid for that $70 billion experiment.
The $27 Billion Infrastructure Deal That Raised Eyebrows
In the same week as the Avocado delay and the layoff announcements, Meta signed a $27 billion AI infrastructure partnership with Nebius Group, a cloud computing company that rebranded from its roots as Yandex’s international arm after Russia’s invasion of Ukraine.
Nebius offers a “GPU-as-a-Service” model, with high-density compute environments that general-purpose clouds from Amazon, Microsoft, and Google struggle to match for specific AI workloads. Meta chose it because it needed specialized infrastructure fast and the major cloud providers couldn’t build it quickly enough.
The deal makes sense on paper. The optics, given Nebius’s Russian origins, have made some observers uncomfortable even though the company has been explicitly distancing itself from its Yandex past. The DOJ and FTC are watching the AI infrastructure space closely for anti-competitive behavior, and a $27 billion deal between the world’s largest social network and a specialized GPU provider is exactly the kind of arrangement regulators are currently scrutinizing.
What the Stock Market Is Saying
META is trading around $613, down roughly 7% year to date as of this week.
The market is not panicking. But it is not cheering either. Investors cheered the Efficiency 2.0 layoff announcement when it first leaked. Then Avocado’s delay came out and took the edge off that reaction. The stock has been drifting lower since, caught between long-term optimism about Meta’s AI infrastructure build and short-term anxiety about whether the product that justifies the infrastructure is actually going to be ready.
Meta’s fiscal year 2025 results, released in January, were genuinely strong. Revenue of $200.97 billion, up 22% year-over-year. Net income of $60.46 billion. By most measures, the company’s core advertising business is extremely healthy.
But the 2026 capex guidance of $115 to $135 billion represents a near-tripling of spending from two years prior. That level of investment demands a proportionate return. Avocado falling short of competitors in its internal tests is not a reassuring signal that the return is on its way.
The Fundamental Problem With Meta’s AI Strategy
Every other major AI infrastructure spender has a direct monetization path for the compute it builds. Amazon rents it through AWS. Microsoft rents it through Azure. Google rents it through Google Cloud. Their AI spending is also their cloud business expansion. The revenue feedback loop is direct.
Meta’s compute generates value only if its AI products drive more engagement on Facebook, Instagram, and WhatsApp, which drives more advertising revenue. The chain from GPU procurement to ad dollar is longer, less direct, and more dependent on product quality than any of its major competitors’ models.
When Avocado works and outperforms, that chain functions. When Avocado delays and underperforms, every dollar of that $135 billion sits idle, waiting for a product that is not yet ready.
“Any further delays could sour investor sentiment, as the justification for the massive $135 billion capex depends entirely on Meta’s ability to stay at the forefront of generative AI performance,” one analyst wrote this week. He was being diplomatic. The reality is starker.
Zuckerberg bet the company on AI the same way he bet it on the metaverse. The metaverse cost $70 billion and produced a mobile app. AI is costing $135 billion in a single year. The model that was supposed to prove it was worth it just missed its deadline.
May is the new target. Wall Street is watching.
About the Author
