Meta's superintelligence lab loses key recruits to OpenAI

Two months ago, Mark Zuckerberg went on a jaw-dropping poaching spree. Meta shelled out $14.3 billion for a 49% stake in Scale AI, brought in Alexandr Wang as chief AI officer, and handed out athlete-level pay packages reportedly reaching up to $300 million over four years to poach researchers from OpenAI.

Now, the golden handcuffs are already slipping off.

The revolving door at TBD Lab

WIRED reported that at least three researchers have resigned from the newly formed Meta Superintelligence Labs (MSL). Notably, Avi Verma and Ethan Knight both left after stints of less than a month to return to OpenAI. A third researcher, Rishabh Agarwal, announced on X that he was departing Meta after 7.5 years across Google Brain, DeepMind, and Meta to pursue a different kind of risk.

Meta is also losing veteran leadership. Chaya Nayak, who led generative AI product management at Meta for nearly a decade, is heading to OpenAI as well.

However, Meta pushed back on the narrative. Meta spokesperson Dave Arnold told The Verge that while MSL is a massive umbrella org of thousands, the elite frontier unit dubbed the TBD Lab has only seen one actual departure: Knight. Arnold clarified that Verma and Agarwal never formally started roles within TBD Lab.

Semantics aside, the talent churn inside Meta's premier AI division is impossible to hide.

Big checks, bigger growing pains

Zuck's aggressive recruiting initially spooked OpenAI. Sam Altman even penned a staff memo calling Meta's poaching tactics distasteful, quipping that he had lost count of how many OpenAI staff Meta tried to recruit as chief scientist.

Meta even secured former OpenAI researcher Shengjia Zhao as MSL chief scientist, but only after Zhao reportedly signed paperwork to return to OpenAI before Meta managed to lock him down.

Yet money cannot solve culture shock. Insiders told The Verge that top-tier AI talent often prioritizes alignment and research freedom over pure cash. Meanwhile, Meta has been wrestling with bureaucracy, shuffling its AI workforce into four teams covering research, product, and infrastructure, while shelving its underwhelming Behemoth model.

Meta also issued an internal memo temporarily freezing hiring across MSL to plan its 2026 budget. While Alexandr Wang insists investment is only accelerating—pointing to a fresh technical tie-up with Midjourney—the frenzied talent blitz has clearly paused.

Why it matters

We think this early exodus is a reality check for Big Tech: you cannot simply buy your way to superintelligence with nine-figure checks.

For top researchers, compute density and compensation matter, but so do organizational stability and research culture. When scientists jump ship back to OpenAI within weeks, it suggests Meta's corporate structure is struggling to offer the agile environment frontier research requires.

Zuck's strategy looks increasingly reactive. Buying nearly half of Scale AI, rebranding divisions overnight, and offering wild equity grants made great headlines, but frequent reorganizations and abandoned models show how hard it is to manufacture research breakthroughs through sheer capital.

OpenAI, on the other hand, just proved its gravitational pull. Retaining talent against blank checks shows that mission, momentum, and researcher autonomy remain moats that billions in capital cannot instantly replicate.