Zuckerberg, Unfiltered: How Llama’s Chaos Triggered Meta’s AI War
Meta’s Llama didn’t just stumble—it awakened a new version of Mark Zuckerberg. One that’s ready to fight.
On June 12, 2025, Meta shook the AI world by acquiring a 49% stake in Scale AI for $14.3 billion and bringing on its CEO, Alexandr Wang, to lead a new “superintelligence” initiative. Wang was given full authority to steer Meta’s AI transformation strategy.
But this wasn’t a standalone play—it was the first volley in Zuckerberg’s all-out AI war.
Just eight days later, CNBC reported that Meta had poached Daniel Gross, co-founder and CEO of Safe Superintelligence (SSI), one of the most talked-about new AI labs. Earlier this year, Meta had tried to acquire SSI outright for $8 billion. When the deal was rejected, Zuckerberg pivoted to a more direct strategy: take the people.
And he didn’t stop there. Meta has also attempted to recruit former GitHub CEO Nat Friedman, and is reportedly looking to invest in NFDG, the venture firm Friedman co-founded with Gross. Behind the scenes, Meta has been courting other heavyweights as well—like OpenAI’s Noam Brown and DeepMind’s Koray Kavukcuoglu—though no confirmations have emerged yet.
These aggressive moves all point to a singular goal: to build a world-class, execution-driven team for Meta’s newly established Superintelligence Lab, one capable of changing the rules of the AI race—not through research publications, but by reshaping the industry itself.
From Llama’s Collapse to a Strategic Reset
Meta’s flagship open-source model family, Llama, has underperformed dramatically. Once hyped as a serious contender, Llama 4 dropped to 32nd place on LM Arena and fell behind competitors by as much as 50% in benchmark capability. Worse, Meta was accused of using an unreleased “optimized” version to inflate its scores—damaging its credibility. Internally, the Llama team imploded: 11 of the 14 original core members left, with five joining French rival Mistral.
Zuckerberg saw the writing on the wall: Meta’s internal research-first approach wasn’t delivering. He decided to pivot, fast and hard, away from idealistic R&D and toward ruthless industrial execution.
And that meant moving on from Yann LeCun—Meta’s long-time Chief AI Scientist and public AI visionary.
While LeCun wasn’t directly responsible for Llama’s engineering, his philosophy shaped Meta’s AI priorities. LeCun championed foundational science over short-term wins, arguing that human-like intelligence would emerge only through deep, long-term research. That slow-burn mindset clashed with Llama’s product-driven ambitions—and ultimately, the pressure to commercialize proved overwhelming.
Now, Zuckerberg is backing a completely different archetype.
Goodbye Researchers, Hello Operators
Alexandr Wang and Daniel Gross aren’t academic theorists—they’re builders.
Wang didn’t rise by inventing cutting-edge algorithms. He built the world’s largest data labeling empire and a world-class RLHF (Reinforcement Learning from Human Feedback) stack, positioning Scale AI as a critical supplier in the generative AI boom.
Gross, on the other hand, excels at ecosystem engineering. He helped launch YC’s AI track, invested early in GitHub, Airtable, and Character.AI, and now aims to build tightly aligned, vertically integrated AI companies. He understands how to wire up talent, compute, and capital into self-reinforcing networks.
In other words, Zuckerberg isn’t just collecting smart people—he’s assembling a wartime cabinet.
And the other AI giants are already feeling the heat.
Meta’s Stake in Scale Disrupts the Data Supply Chain
Meta’s partial acquisition of Scale has already caused tremors in the industry. Google—previously Scale’s biggest client—was slated to spend nearly $200 million on labeled data in 2025. But following Meta’s announcement, it abruptly cut ties. OpenAI followed suit, shifting to more neutral vendors. Even Elon Musk’s xAI reportedly froze its talks with Scale.
Labelbox, Snorkel AI, and other “neutral” players are now seeing a surge in demand. In a matter of days, the data supply chain for AI training has begun to fracture.
Just like his acquisitions of Instagram and WhatsApp, Zuckerberg is once again reshaping the infrastructure. This time, the battleground is compute, talent, and data. If Meta can’t lead in model quality, it can still choke its competitors’ pipelines—and change the tempo of the war.
A New AI Playbook, Led by Zuckerberg Himself
What’s most telling isn’t just the hires or the money. It’s how directly Zuckerberg is now involved.
Unlike Meta’s old FAIR lab—built around academic freedom and long-term inquiry—Superintelligence Lab reports straight to him. No layers, no committees. Just a founder-CEO pulling the strings.
This lab isn’t about publishing papers. It’s about shipping product: AI assistants, smart glasses, generative content tools embedded across Meta’s ecosystem. The strategy isn’t to beat OpenAI at abstract reasoning—it’s to out-deploy them.
And this new team is built for exactly that: an industrial pipeline of talent, data, models, and deployment.
Zuckerberg isn’t looking for a LeCun 2.0. He’s throwing out the old “scientist-led” model altogether.
The Real Zuckerberg Has Entered the Game
Ironically, Llama’s failure may be what finally gave Meta a shot at winning.
By imploding, it gave Zuckerberg the cover he needed to strip out the old AI leadership and rebuild from scratch. He no longer has to pretend that research alone will win the AI war. Now, he’s playing a game he knows better than anyone: resource orchestration, platform control, PR warfare, and relentless product execution.
This isn’t a lab-driven AI company anymore.
It’s a Zuckerberg-led war machine.
And for the rest of Silicon Valley, that changes everything.