Image Credit: Leonardo AI
News Summary
- 25,000 jobs eliminated at Meta since 2022, across four rounds of cuts
- 700 employees were let go across five divisions in March 2026 alone
- Reality Labs posted a $6.02 billion operating loss in 2025 on $955 million in revenue
- Meta plans to spend between $115 billion and $135 billion on AI in 2026
- AI engineers at Meta now earn up to $240,000 per year
- Mark Zuckerberg paid $14.3 billion to bring a single AI strategist from Scale AI onto his team
Twenty-five thousand jobs. Gone. Most people missed it because it happened slowly, one quiet announcement at a time.
Something seismic is happening inside Meta. This is not a routine cost-cutting exercise. This is a company actively dismantling what it spent five years and tens of billions of dollars building, and replacing it with something entirely different. The metaverse is fading. Artificial intelligence is winning. And thousands of real people are losing their jobs in that transition.
In this Article
The Full Story: What Actually Happened
On March 25, 2026, Meta began notifying hundreds of employees across five divisions that their roles no longer existed. The affected teams spanned Reality Labs, Facebook core social, global operations, recruiting, and sales. A Meta spokesperson confirmed the cuts with a carefully worded statement noting that teams across Meta regularly restructure to ensure they are in the best position to achieve their goals. That is polished language for a decision affecting thousands of real people and their families.
This March round was not the beginning. It was the most recent chapter in a four-year pattern of workforce reduction that most coverage has failed to connect as a single, coherent strategy.
The Layoff Timeline That Nobody Assembled
Late 2022 brought 11,000 job cuts, the largest single layoff event in Meta's history at that time. The year 2023 added another 10,000 positions eliminated under what Mark Zuckerberg publicly labelled the Year of Efficiency. Early 2025 saw a further 3,600 workers let go. January 2026 saw roughly 1,000 Reality Labs cuts alongside the permanent closures of three VR game studios: Armature Studio, Twisted Pixel, and Sanzaru Games. March 2026 added 700 more. The cumulative total reaches approximately 25,000 positions removed from one of the world's largest technology companies in under four years.
At every step, the stated rationale was the same: the company needed to move faster on artificial intelligence. What made each round feel isolated was the careful separation of announcements, different divisions affected each time, and different spokespeople delivering the news. Assembled in sequence, however, the pattern reveals a deliberate and long-running strategic reallocation, not a crisis response.
Reality Labs: The Billion-Dollar Wound
Reality Labs reported a $6.02 billion operating loss in 2025 on revenue of just $955 million. That figure is not a temporary dip or a short-term investment trough. The division had been posting losses of comparable scale for several consecutive years. Meta's total cumulative losses from Reality Labs since 2019 exceed $50 billion, a figure confirmed in the company's quarterly filings with the United States Securities and Exchange Commission. Leadership kept funding it because Zuckerberg genuinely believed the metaverse represented the next evolution of the internet. The consumer adoption data consistently disagreed.
Horizon Worlds, Meta's flagship virtual world and the centrepiece of its metaverse ambition, never attracted more than a few hundred thousand monthly active users. Roblox, a platform competing for broadly the same concept of a social virtual environment, reported 150 million daily active users over the same period. That comparison is not incidental. It is the clearest single illustration of why the metaverse failed as a consumer proposition. This pattern of entire product categories being displaced by AI-adjacent shifts is something we tracked in our report on how artificial intelligence erased millions of jobs while simultaneously creating new ones.
Why This Matters: The Bigger Picture
One number sits at the centre of everything Meta is doing right now. That number is $115 billion to $135 billion. It is the capital expenditure the company has forecasted for 2026, representing almost double what it spent in 2025, with virtually all of it directed at AI chips, data centres, and large language model research.
To place that figure in context, Meta posted $59.89 billion in revenue in Q4 2025, a 24 percent increase year over year, with net income reaching $22.8 billion. The company is financially strong. Yet its leadership chose to redirect the bulk of future investment into infrastructure for a technology whose full commercial return remains unproven at this scale. That is either an act of exceptional strategic confidence or a response to competitive pressure so intense that standing still is simply not possible.
The $14.3 Billion Hire That Said Everything
In June 2025, Zuckerberg agreed to pay $14.3 billion to bring Alexandr Wang, the founder of Scale AI, onto his team to lead Meta's artificial intelligence strategy. This was not a conventional executive hire. It was structurally closer to acquiring a company and retaining only its founder. Wang now directs AI strategy across the entire Meta portfolio, a role that did not exist at that seniority level two years earlier.
In October of the same year, Vishal Shah, who had spent four years leading the metaverse initiative, was reassigned to Vice President of AI Products. The leadership structure itself is inverted. The individuals who designed and executed the old vision are now working to build the new one. For a sense of how similar multi-billion-dollar AI commitments are playing out across the broader industry, our coverage of Microsoft's $10 billion Japan AI deal and its quietly significant data conditions offers a directly relevant parallel.
What the Independent Research Actually Shows
Research published by McKinsey and Company confirms that AI adoption is accelerating across every major industry sector globally. The Stanford AI Index, an independent annual report tracking AI development, has documented how the computational cost of training frontier AI models has grown by orders of magnitude over five years. That trajectory directly explains why companies at Meta's scale must spend at this level simply to remain in the competitive conversation.
A practical illustration of the cost involved: training a single advanced AI model can require thousands of specialised graphics processing units running without interruption for several weeks. Each GPU costs tens of thousands of dollars to procure. The electricity required to run a large-scale AI training operation can reach millions of dollars per month. A report published by the International Energy Agency in 2025 projected that data centres supporting AI workloads could account for more than 4 percent of global electricity consumption by 2026, up from under 2 percent in 2022. These costs do not decrease as models grow more capable. They compound. This is precisely why workforce restructuring and infrastructure investment are proceeding simultaneously across the industry, and why the consequences for individual workers are not peripheral to this story but central to it. The deeper implications for careers and personal financial planning are explored in our reporting on why technology executives simultaneously fear AGI and continue building toward it.
The Angle Nobody Covered: One Stat That Changes Everything
Here is the finding that most coverage of Meta layoffs failed to include. According to research published by Gartner, only 1 percent of technology layoffs in the first half of 2025 actually resulted from artificial intelligence increasing employee productivity in a measurable and documented way. Read that figure carefully.
The remaining 99 percent of job cuts were made in anticipation of AI gains that have not yet fully materialised. Companies, including Meta, are reducing their headcount based on projected future efficiency rather than demonstrated present performance. Kaelyn Lowmaster, a director in Gartner's HR practice, stated that in some cases organisations will need to rehire for the precise roles they have recently eliminated.
This is the story beneath the headline. Meta is not cutting jobs because artificial intelligence has already replaced those employees. Meta is cutting jobs to fund the construction of the AI that it hopes will eventually justify those cuts. The workers being let go are, in an economic sense, subsidising the technology that is intended to eventually replace them. That tension is not a peripheral observation. It is the defining ethical question sitting at the centre of how the technology industry is conducting itself in 2026.
The Metaverse Did Not Fail Because of AI. It Failed Because of People.
Consumer behaviour sealed the metaverse's fate before the AI wave arrived. Meta spent years and tens of billions of dollars attempting to persuade ordinary people to put on a VR headset and attend meetings, socialise, or work inside a virtual environment. The adoption figures never materialised at a scale that justified the investment. The experience also generated a practical credibility problem: early versions of Meta's avatars drew mockery rather than enthusiasm, and internal demos that leaked to the press reinforced the impression that the product was not ready for mainstream audiences.
The contrast with AI adoption is stark and instructive. A smartphone user in 2024 could ask a conversational AI model a question, generate an image from a text prompt, or summarise a lengthy document in seconds, without purchasing any new hardware, downloading a specialised application, or changing a single element of their daily routine. The barrier to adoption was effectively zero. The metaverse required people to change their entire computing behaviour, their physical posture, their social environment, and their expectation of what productivity felt like. AI asked nothing of the user. That asymmetry determined the outcome more decisively than any product design decision either side made. The real-world displacement of conventional software tools by AI is documented in our piece on how Claude AI is replacing traditional software tools in 2026.
The One Reality Labs Bet That Actually Worked
Not every product inside Reality Labs is in retreat. The Ray-Ban Meta smart glasses represent the single hardware bet from the division that gained genuine, measurable consumer traction. Luxottica's chief financial officer confirmed in late 2025 that the company expects to reach 10 million unit production capacity ahead of the originally stated 2026 schedule. For a consumer hardware product in a category that most analysts considered marginal, that figure carries real significance.
The reason these glasses succeeded where the Quest headsets consistently struggled is directly relevant to Meta's broader strategic lesson. Smart glasses ask nothing different of the user. They look and function like ordinary eyewear in everyday social settings. Future iterations will integrate AI voice assistants, live language translation, and real-time visual recognition, all delivered through a form factor that requires no behavioural adaptation on the part of the person wearing them. This is Meta's viable hardware future: not an immersive world you must choose to enter, but an intelligence layer quietly and unobtrusively attached to the world you already occupy. What that ongoing connectivity means for household AI spending is something we explored in our analysis of what a monthly AI bill might realistically look like for ordinary consumers.
The Road Ahead: Where Meta Goes From Here
Meta's restructuring is not a conclusion. It is the preparation for the next phase of the company's development, and that phase has several distinct directions already observable from current data and verified company disclosures.
AI Features Deployed Across Three Billion Users
Meta's most significant near-term advantage in the AI race has nothing to do with model benchmarks or research publications. It is a distribution. Facebook, Instagram, and WhatsApp collectively serve over three billion active users. No dedicated AI company currently possesses a comparable deployment surface. More intelligent content recommendations, AI-assisted messaging, generative tools for content creators, and personalised discovery features are all in active development and phased rollout across these platforms.
Zuckerberg stated in a January 2026 public post that AI would have a significant impact on the business that year, observing that projects previously requiring large teams could now be accomplished by a single talented person. Whether that observation eventually translates into further workforce reduction or simply a changed hiring model is one of the most consequential open questions in the industry. The data privacy dimension of deploying AI across billions of user interactions also warrants serious attention, as explored in our coverage of Google's web and app activity settings and what they reveal about the AI data trade-off facing consumers.
The Talent Market Has Been Entirely Repriced
Meta is actively recruiting AI engineers at annual salaries reaching $240,000. Researchers from competing firms have received sign-on bonuses reported to exceed $100 million in individual cases, a figure that captures precisely how acutely the industry feels the shortage of genuinely capable AI researchers. This is not hyperbole. It reflects verified reports from technology media covering specific compensation packages offered to researchers moving from OpenAI, Google DeepMind, and academic institutions to Meta's AI division.
Roles in machine learning engineering, large language model optimisation, AI safety research, and data infrastructure are experiencing the strongest demand in Meta's current hiring activity. By contrast, positions in product design, general marketing, and broad management functions that do not require demonstrated AI competency face a contracting internal market. The Organisation for Economic Co-operation and Development has projected that artificial intelligence will transform more professions than it eliminates outright over the coming decade, but the transformation period itself carries genuine hardship for those caught in the transition without the resources to adapt quickly. The gap between promise and measurable delivery in transformative technology is a recurring theme examined in our reporting on quantum computing hype versus measurable reality in 2026.
The LLaMA Open-Source Strategy and Why It Differs
Meta's approach to large language model development represents one of the most strategically distinct decisions the company has made in this period. The LLaMA model family was released as open-source software, allowing researchers, independent developers, and startups worldwide to build directly on Meta's underlying architecture at no cost. That decision was not a gesture of generosity. It was a calculated move to build ecosystem breadth that Meta does not need to fund internally.
By early 2026, thousands of independent projects had been built using LLaMA as their foundation. Academic papers had advanced the model's capabilities in directions Meta's own research teams had not prioritised. Startup products built on LLaMA had generated press coverage and user attention that indirectly reinforced the credibility of Meta's AI work. This outcome stands in deliberate contrast to the closed-model strategies pursued by OpenAI, which restricts access to GPT-4 and GPT-4o behind paid APIs, and Anthropic, which similarly gates access to its Claude model family. Meta is wagering that ecosystem scale will ultimately prove more durable than model exclusivity in determining long-term competitive positioning. The commercial dynamics of that bet, and how investors are pricing the different strategic approaches across the AI industry, are examined in depth in our reporting on Anthropic's IPO plans and what they reveal about the commercial AI economy in 2026.
Key Facts in One Place
- 25,000 total jobs eliminated at Meta since 2022 across four rounds of restructuring
- 700 employees were removed across five divisions in March 2026 alone
- Reality Labs posted a $6.02 billion operating loss in 2025 on $955 million in revenue
- Meta's cumulative Reality Labs losses since 2019 exceed $50 billion, per SEC filings
- Capital expenditure of $115 billion to $135 billion forecasted for 2026, nearly double that of 2025
- Ray-Ban Meta smart glasses are approaching 10 million unit production capacity ahead of schedule
- AI engineers at Meta earn up to $240,000 annually, with select research hires receiving bonuses above $100 million
- Only 1 percent of technology layoffs in the first half of 2025 resulted from verified AI productivity gains, per Gartner
- Horizon Worlds peaked at hundreds of thousands of monthly users versus Roblox's 150 million daily active users
- Meta platforms collectively serve over 3 billion users, providing AI features with unmatched deployment scale
The human cost behind these numbers is not abstract. Engineers who spent years developing virtual environments and immersive hardware are now searching for new positions in a market that has rapidly re-priced their skills. The generational and societal dimensions of that disruption, particularly as it affects younger workers entering the workforce, are examined in our piece on why governments are moving to restrict social media for young users and what alternatives are genuinely available.
The Question Silicon Valley Refuses to Answer
Meta is not unique in this story. It is simply the most visible and largest-scale example of a pattern repeating itself across every major technology company operating today. Budgets are concentrating on artificial intelligence. Workforces are contracting to fund that concentration. The stated justification, that AI will eventually deliver efficiency, new employment categories, and broader economic value, remains a projection rather than a demonstrated and verified outcome at this scale of investment.
The Gartner data makes this deeply uncomfortable. If only 1 percent of current technology layoffs reflect AI-driven productivity gains that have actually been realised, then the remaining 99 percent represent a social cost being extracted in advance of a technological benefit that has not yet arrived. Organisations are essentially borrowing against a future that is still being constructed, and the people losing their jobs are the collateral against that borrowing.
Whether the AI infrastructure Meta is spending $135 billion to build in 2026 ultimately delivers for its users, its shareholders, and the workers it intends to hire once the platform is operational remains a genuinely open question. The honest position, as of April 2026, is that the outcome is not known. What is known is that the bet is enormous, the commitment appears irreversible, and the real-world consequences for millions of employees across the technology sector will be substantial regardless of how the technology eventually performs. For a broader perspective on where AI regulation, public investment, and national strategy currently sit across different economies, our reporting on Anthropic's IPO and the commercialisation of AI safety research provides essential context for understanding the full competitive landscape.
Is Meta making the most consequential strategic bet in its history, or funding an efficiency revolution on the backs of the very people who built it? That question deserves a direct and honest answer, and right now, the technology industry does not have one to give.