Meta Spent $90 Billion Shutting Down the Metaverse, Spent $2 Billion Letting AI Move Into Your Computer

Author: Curry, Deep Tide TechFlow

On October 28, 2021, Zuckerberg stood next to a legless virtual avatar and announced that the company would rename itself from Facebook to Meta.

At that time, he said the metaverse would reach 1 billion people within ten years, support hundreds of billions of dollars in digital commerce, and provide job opportunities for millions of creators and developers.

That year, the metaverse was the hottest concept on Earth.

Microsoft announced plans for a metaverse version of Teams, Nvidia launched Omniverse, Nike opened a virtual store on Roblox… No one wanted to miss this ticket.

Meta didn’t just buy a ticket; it bought the whole ship.

Horizon Worlds, now seen as the core evidence of Meta’s rebranding story—you wear a headset, enter a virtual world, and explore, play, or hold meetings with others’ cartoon avatars.

When it launched at the end of 2021, it was Zuckerberg’s flagship product. But four and a half years later, fewer than 1 billion people have used it.

On March 17, Meta posted an announcement on its community forum: The VR version of Horizon Worlds will be shut down completely on June 15, with the app removed from Quest headsets and the virtual world no longer accessible. A mobile version remains operational.

It’s like a restaurant closing dine-in service and only offering takeout, but the restaurant was originally built for dine-in.

The department footing the bill is called Reality Labs. Over seven years, it has accumulated nearly $90 billion in operating losses. In the most recent quarter, it lost $6 billion, with revenue under $1 billion—less than one-sixth of its losses.

In January this year, the department laid off over 1,000 people, closed multiple VR content studios, and cut nearly all ongoing virtual world projects.

The ticket everyone feared missing in 2021 has now sunk, and the ticket is still clenched in their hands.

In mid-March, Reuters reported that Meta plans to cut about 20% of its staff, nearly 15,000 people. If implemented, this would be the largest layoff since 2022.

Meanwhile, Meta’s capital expenditure budget this year is between $115 billion and $135 billion, almost entirely invested in AI infrastructure.

Shutting down the virtual worlds, laying off 20% of staff, and redirecting the saved money and resources into AI.

On the day the news broke, Meta’s stock rose 3%. When Zuckerberg announced in 2021 that he would fully bet on the metaverse, the capital markets responded with applause.

Just a day before Horizon Worlds announced its shutdown, the answer was already on the table.

Virtual worlds close, personal computers take the stage

On March 16, Manus, acquired by Meta for $2 billion, launched a desktop version.

It features a “My Computer” function that allows AI to come down from the cloud and directly access your local computer: reading files, opening applications, running terminal commands.

This happened the day before Horizon Worlds announced its closure.

When Horizon Worlds launched, the experience was like this:

You spend two or three thousand dollars on a Quest headset, put it on, adjust the interpupillary distance, draw a safe boundary, and enter a cartoon-style virtual lobby. The avatars inside have no legs and float while walking. You can explore themed worlds, play mini-games, or chat with strangers’ virtual avatars.

After half an hour, the headset starts pressing against your face; after an hour, some people start feeling dizzy.

Meta spent four years and $900 million on this lobby. But it has never publicly disclosed active user numbers—not because of secrecy, but because it wouldn’t look good.

The Manus Desktop experience is like this:

You download an app, open it, and input a command. For example, “Organize the thousands of files in my Downloads folder by type.” It scans your hard drive, automatically creates subfolders, and sorts and archives files—all without you touching the keyboard.

In a demo, someone asked it to write a macOS app from scratch in a local development environment, which it did in 20 minutes. Remember, Manus was launched eight months ago, with over a million paying users and annual revenue exceeding $100 million.

When everyone said Meta’s acquisition of Manus was not worth it, compare it to the previously shut-down metaverse project Horizon Worlds.

One product, costing $900 million to bring you into a virtual world, nobody uses. The other, costing $200 million to bring into your real desktop, with real revenue and use cases—if it were you, which would you choose?

Same company, same week, shutting down the former and betting on the latter.

Previously, Meta built a world for you to come into; now, AI crosses the screen and comes to you.

But the right direction doesn’t mean the road is smooth. After turning around, Meta doesn’t seem to have become more confident.

Metaverse and AI might be the same kind of FOMO

If you only look at headlines, Meta now looks like a company making reckless moves.

The metaverse burned through $90 billion and shut down. The flagship AI model, Avocado, was scheduled for release in March but was delayed to May after internal testing showed its reasoning and coding capabilities lagged behind products from Google, OpenAI, and Anthropic.

The previous generation Llama 4, released last year, received a lukewarm response and didn’t stir much in the developer community. Reports suggest the company even discussed temporarily licensing Google’s Gemini to power its own products—after investing $135 billion in AI infrastructure, the company wants to use others’ models.

Chief AI scientist Yann LeCun has left to start his own venture; the new AI head, Alexandr Wang, recruited from Scale AI for $14.3 billion, has yet to deliver results…

A 20% layoff, shutting down the metaverse, missed model deadlines—all within a week—these headlines look like a company that doesn’t know what it’s doing.

But look beyond Meta, and you’ll see a pattern across the industry:

Everyone is doing the same thing—full throttle embracing AI.

In February, Block CEO Jack Dorsey announced layoffs of 4,000 employees, nearly half the company. The message was blunt: intelligent tools are changing how companies build and operate, enabling smaller teams to do more. The stock price jumped 25% that evening.

Shopify’s CEO issued a new rule: to request additional hires, you must first prove AI can’t do the job.

Amazon cut 16,000 jobs in January and also cut its robotics division in March. Atlassian laid off 1,600 people, saying it would focus all resources on AI enterprise software.

In the first 74 days of 2026, 166 tech companies laid off nearly 56,000 people.

Does this scene sound familiar?

It was the same in 2021. After Zuckerberg rebranded as Meta, Microsoft announced plans for a metaverse version of Teams, Nvidia launched Omniverse, Nike opened virtual stores on Roblox, Disney established a metaverse division, and Shanghai and Seoul released metaverse strategic plans…

Everyone was heading in the same direction, all afraid of missing out.

Five years later, the direction has changed, but the pursuit remains the same.

Last time, the consensus was “the metaverse is the next computing platform,” and Meta spent $900 billion proving that this was wrong. This time, the consensus is “AI can replace everything,” and all companies are laying off people, cutting budgets, and pouring the savings into AI.

The only difference is: last time, the consensus was proven false; this time, it hasn’t been yet.

But consensus is consensus. It’s characterized by everyone believing simultaneously, then everyone realizing it’s wrong at the same time. The time lag in between is the speed at which money burns.

Meta isn’t a dumber company than others. It just bets bigger each time, so when the consensus flips, it falls the hardest.

In 2021, the entire industry bet on the metaverse, and Meta changed its name. In 2026, the entire industry bets on AI, and Meta lays off a fifth of its staff.

Looking back five years later, did everyone bet correctly on AI?

No one knows. But we all know that in 2021, when asked that question, everyone’s answer was “Of course.”