Alexander Wang revealed that, before its frontier models were validated, Zuckerberg forcibly approved a $100–200 billion bet on AI.

Article author and source: Wall Street Journal

At the table of tech giants, no bet has ever been as wild as this one.

Recently, Alexander Wang, Meta’s current head of AI, revealed in an interview for the first time that Meta is making a bold gamble.

The stakes for this bold gamble range from $100 billion to $200 billion.

This massive investment was made by Zuckerberg himself, despite the fact that Meta’s cutting-edge models had not been verified at all!

If the new model fails, Meta will face a severe financial crisis.

The trillion-dollar Meta empire is now strapped to an AI train racing full speed toward a cliff. Success will crown it with a throne; failure will plunge it into an abyss of irrecoverable ruin.

The AI competition has really escalated to this level?

Hawaii's "Terrifying Team-Building": CFO Issues Desperate Warning as Zuckerberg Bets $200 Billion!

What shocking secrets lie behind this trillion-dollar life-or-death game?

Recently, Stripe co-founder John Collison invited Zuckerberg and Alexandr Wang into an Irish-style pub for his podcast, “Cheeky Pint.” Over two hours of drinks, several revealing insights came to light.

Let’s go back to the end of last year.

Back then, Hawaii was sunny. Meta’s core executive team gathered there for their annual executive retreat.

On the surface, it was a relaxed gathering over cocktails. But throughout the event, tensions simmered beneath the surface—so much so that you could say people were breaking out in cold sweats.

Alexandr Wang, recently poached by Meta in a massive hiring move, experienced an unforgettable moment here.

During a team-building break, Meta's CFO Susan Li quietly pulled me into a secluded corner.

Alexander Wang revealed: In private, Susan Li repeatedly issued him a stern warning in an extremely serious, almost desperate tone: “Listen, if this cutting-edge AI model development fails, the financial consequences will be extremely, extremely severe!”

The CFO’s panic is not without reason, as she has just learned of CEO Zuckerberg’s reckless decision.

Around the same time, Xiao Zha called Alexandr Wang over. At that point, Muse Wonder and the entire cutting-edge model architecture were still in an extremely early stage, with even the most fundamental technical validation yet to be completed.

But Zuckerberg can't wait! The AI race is heating up, and the rapid advances by OpenAI and Google are putting immense pressure on him.

Zuckerberg looked at Wang and personally asked for his final "ace in the hole": "Are you sure this will work?"

Faced with his boss’s questioning, Wang gave a positive response.

Upon hearing this, Zuckerberg didn’t hesitate and immediately issued a check for $100 billion to $200 billion in infrastructure investment!

This money will become GPUs, data centers, cooling towers, and astonishing amounts of electricity capable of powering several cities.

Alexander Wang's revelation has fully exposed Meta's all-or-nothing gamble.

This is not rational planning at all—it shows that Meta has bet the entire company’s future on an unproven AI project. It’s a make-or-break gamble.

If you mess up, the company faces immediate disaster; if you win the bet, the stock price can take off.

From laughingstock to a breakthrough in "monster-level" large models

Bearing the CEO’s “hundred-billion-dollar bluff,” Alexandr Wang and his team embarked on months of grueling development under immense pressure.

What did this money actually achieve?

A year ago, when Meta announced it was spending a huge sum to acquire Wang’s team, everyone online mocked Zuckerberg as a "sucker."

However, Muse Wonder's benchmark scores are truly impressive.

In comprehensive intelligence, the full version of Wonder Full achieved an impressive score of 56, surpassing Fable 5.1 and ranking just below Opus 5.5.

Wonder demonstrates a dominant performance in long-term coding capability, with a DeepSWE pass rate of approximately 79%.

From open-source Llama, Meta directly enters the closed-source top-tier Agent arena, officially declaring its dominance in cutting-edge long-term software engineering!

As a netizen brilliantly commented: “Good heavens, you really need to study Zuck’s current strategy—right before what could be the biggest victory of his life, he’s casually grilling! Just a year and a half ago, Meta was the biggest joke; now they’ve got a monster model that can beat Fable and Astra!”

However, immediately after its release, Muse Wonder faced a fierce backlash.

Many people have said that its actual performance is terrible—one word to describe it is “benchmaxxed.”

In the lava lamp test, there were no lamps, no glass, no lava—not even a red cylinder.

This result is truly astonishing!

In response, Alexandr Wang personally stepped in to confront it head-on.

He struck back mercilessly on X:

When developing Muse Wonder, our goal was to build a model that is useful for real-world, real humans performing real tasks. We focus on things that may not seem as "glamorous" on Twitter—such as avoiding flashy 3D game demos.

This is a polite admission of being outclassed!

Big V Token Gremlin said: "Muse was terrible when writing Three.js! Even creating a simple Minecraft clone was disastrous. Really disappointing."

Sharper criticism came from the pricing strategy.

Developers have pointed out that the currently available "low-health" version of Muse Wonder, which has a benchmark score of only 52, is priced identically to Opus 5.5 with a score of 58 ($4 per million tokens for input, $20 per million tokens for output).

The true, fully realized Wonder Full is available only to subscribers.

The real game-changer: Active Memory and Active Focus

On the Cheeky Pint podcast, Zuckerberg and Wang thoroughly break down the real reasons why Muse Wonder is called a "monster"—Active Memory and Active Focus.

As some developers have said: “What truly matters isn’t isolated benchmark scores—it’s how terrifyingly powerful its Active Memory becomes when the model actively engages with your emails, calendar, and personal documents. That’s where cutting-edge models truly diverge.”

Active Memory: It possesses super memory across time. Meta officially states: “We’ve invested tremendous effort to ensure the model remembers what matters to you—days, weeks, or even months later.” It is no longer a chatbot with a “goldfish memory,” but a digital twin that deeply remembers all your preferences and habits.

Active Focus: Previous AIs would give up when encountering errors. But Muse Wonder has been trained on a vast amount of "real-world failure scenarios." When it gets stuck during a task, it proactively seeks alternative solutions; before performing critical actions, it pauses to confirm with you.

During the demo, Muse Wonder continuously scours the web for rental listings, filters them according to personalized criteria, and even contacts landlords autonomously!

This is a true "super personal assistant" that works for you 24/7.

Zuckerberg’s contrarian view: AI won’t be a winner-takes-all market!

Given the massive investment, what is Meta really aiming for?

In the podcast, Zuckerberg stated his opinion: "AI will absolutely not be a winner-takes-all scenario."

He doesn’t believe the world will be monopolized by a single “god model” like OpenAI. Instead, he is convinced that the future will belong to countless individual superintelligences.

This also explains why Meta established the "Superintelligence Lab" and why they proposed the visions of "Agent Commerce" and "Beyond Smartphones" in hardware.

Imagine a future where you’re wearing Meta’s smart glasses, and inside lives Muse Wonder with Active Memory.

Your personal agent will communicate with the merchant's agent, negotiate prices, and purchase items.

If truly realized, Meta would completely reshape the global digital business model. This is the future that $200 billion seeks to buy!

In Silicon Valley today, you either bet hundreds of billions on a future, or sit on hundreds of billions in cash and wait to die.

Zuckerberg chose the former. He bet Meta on the table, loaded the gun, and pulled the trigger with Muse Wonder.

At this point, the bullet has hit the bullseye.

Meta did not head toward financial disaster; instead, it successfully entered the top tier of frontier models, bringing personal superintelligence to the fingertips of billions of users.

As for how well Muse Wonder works, let’s wait until the model is released.