Meta Poaches more Apple AI Talent for its Superintelligence Labs

Meta’s talent raid on Apple shows no sign of slowing. Bloomberg sources say Mark Zuckerberg has just lured senior machine‑learning engineers Mark Lee and Tom Gunter—both veterans of Apple’s foundation‑model group—into the company’s year‑old Superintelligence Labs unit on multi‑million‑dollar packages. Lee started at Meta last week; Gunter is due to follow within days. Their moves come barely eight weeks after their former boss, Ruoming Pang (Apple’s head of large language models), defected to Meta in a deal reportedly worth more than $200 million, and they cap a trio of high‑profile exits from Cupertino in one quarter.

Zuckerberg has personally championed the hiring spree, telling investors this month that Meta will spend “hundreds of billions of dollars” over the next decade on data‑centre capacity, advanced GPUs and specialist researchers to push beyond today’s large‑language‑model frontier. The company has already ordered around 350,000 Nvidia H100 accelerators and plans to scale to the equivalent of 600,000 H100s by year‑end, cementing its status as the world’s largest commercial AI cluster. Construction is also under way on Prometheus and Hyperion, two U.S. data‑centre campuses that together could draw five gigawatts of power.

Apple, meanwhile, is grappling with mixed reviews for its newly unveiled “Apple Intelligence” features, which lean on OpenAI’s ChatGPT to fill gaps in Cupertino’s in‑house model roster. Insiders say the strategic uncertainty—whether to buy, build or rent top‑tier generative AI—has bruised morale and left researchers receptive to Meta’s overtures. By dangling outsized equity grants and a shot at working on what Zuckerberg calls “post‑scale superintelligence,” Meta positions itself as the destination for engineers who want to train trillion‑parameter models without worrying about compute quotas.

The competitive stakes are enormous: whoever assembles the smartest, safest and cheapest next‑gen model will shape everything from voice interfaces and search to robotics and autonomous vehicles. Meta’s open‑source Llama family already powers thousands of start‑ups; a future “Llama 5” trained on its planned exascale cluster could challenge Google’s Gemini Ultra and OpenAI’s GPT‑5 for outright capability leadership. Recruiting the minds that once architected Apple’s private models is therefore both an offensive move—accelerating Meta’s roadmap—and a defensive one that slows a rival just as the generative‑AI platform wars enter their commercial phase.

Neither company commented on the defections, but one former Apple engineer put it bluntly on X – “Why fight for 40 GB of A100 time in Cupertino when Menlo Park hands you eight nodes of H100 on day one?” With cash‑rich tech giants offering eye‑watering pay and near‑unlimited silicon, the AI talent war is only heating up—and Meta’s latest victory suggests that, for now, money, compute and ambition are a hard combination to beat.