Meta has been one of the most aggressive players in the global artificial intelligence race. Earlier this year, the company launched a bold new division called Meta Superintelligence Labs (MSL), with a mission to push beyond large language models and pursue what CEO Mark Zuckerberg described as “personal superintelligence.”
The announcement generated huge buzz. Meta attracted top AI researchers with eye-watering pay packages—some reportedly worth hundreds of millions of dollars. The division was split into four units covering advanced research, product applications, infrastructure, and oversight of models like LLaMA. For a moment, it looked like Meta might challenge OpenAI, Anthropic, and Google DeepMind head-on.
But only months after its formation, MSL is showing signs of turbulence. Several high-profile researchers have already left. Avi Verma and Ethan Knight, both respected scientists, returned to OpenAI. Rishabh Agarwal, another senior figure, resigned citing a desire to take different risks. More recently, Chaya Nayak, who directed generative AI product management at Meta for nearly a decade, also exited—moving to none other than OpenAI.
Why Talent Is Walking Away
The departures highlight that in AI, money alone doesn’t secure loyalty. Reports suggest that many long-time Meta staffers feel overshadowed by the attention and resources flowing into the superintelligence unit. While newcomers were rewarded with extraordinary salaries and influence, existing AI teams working on generative products felt sidelined. That cultural imbalance has sparked frustration and, in some cases, open resentment.
Another factor is organizational churn. Meta has repeatedly restructured its AI operations, recently splitting them into four separate groups and even shelving a large model project code-named Behemoth after disappointing performance. For researchers who thrive on stability and clear direction, such constant change can be discouraging.
(You can read AI & Big Data Expo Europe 2025 Set to Take Place in Amsterdam This October)
What It Means for Meta—and the AI Race
For Meta, the timing is sensitive. Competitors are scaling fast:
- OpenAI continues to dominate the public imagination with ChatGPT and enterprise adoption.
- Anthropic is attracting major corporate partners.
- Google DeepMind is quietly advancing multimodal models with deep infrastructure backing.
Meta still has world-class resources, talent, and infrastructure. LLaMA remains one of the most widely adopted open-source large language models, giving Meta strong influence in the developer community. Yet the early instability at its superintelligence division suggests that building cutting-edge AI is not just about GPUs and cash—it’s about trust, focus, and mission alignment.
The Bigger Picture
The exits from Meta’s new lab underscore a simple truth: the global AI race is as much about people as it is about technology. Talented researchers want to feel that their work matters, that their teams are stable, and that they are part of a mission that aligns with their values. Without that, even billion-dollar offers can’t stop defections.
Meta’s bold experiment with superintelligence isn’t over. But its early struggles show how fragile even the most ambitious AI ventures can be if culture, clarity, and collaboration aren’t carefully managed.
In short: Meta tried to buy speed by throwing money at AI talent. What it learned is that innovation runs on people, not just paychecks.
You can checkout AI in Focus: Why August 2025 Was a Turning Point, for daily tec, and AI update follow on Instagram.
