Apple’s artificial intelligence division continues to lose ground as another key researcher leaves for Meta. Bowen Zhang, a core member of Apple’s foundation models team, exited the company last Friday. He’s the fourth AI researcher to depart in just a month, all of whom are now heading to Meta’s newly formed superintelligence unit.
This exodus has left Apple’s AI models team in disarray. The Apple Foundation Models (AFM) group, responsible for developing the in-house technology powering Apple Intelligence, is now dealing with a mix of internal instability and external pressure. Engineers inside the unit describe growing uncertainty over the team’s future. With more employees actively interviewing elsewhere, Apple’s ability to retain top-tier AI talent is in serious doubt.
High-Profile Departures and Big Paychecks
Zhang’s move follows that of Ruoming Pang, the former leader of the AFM group, who was lured away by Meta with a compensation package reportedly worth over $200 million. Two other researchers, Tom Gunter and Mark Lee, also jumped ship to Meta. Another member, Floris Weers, recently left for a startup.
According to Bloomberg, Apple has responded to Meta’s aggressive recruiting by offering slight pay increases to AFM staffers, whether or not they’ve shown intent to leave. But insiders say these raises don’t match the kind of offers competitors are putting on the table. Apple and Meta both declined to comment on the departures.
Apple’s Shifting AI Strategy
The AFM group’s work underpins the company’s broader AI platform, but that foundation is starting to look unstable. Internal discussions now include potentially replacing in-house models with third-party systems, including those from OpenAI and Anthropic. According to sources familiar with the matter, this shift has rattled engineers who believe Apple is stepping away from its core strategy.
Apple has not made a final decision on whether future versions of Siri and other services will use external models. However, engineers have already begun prototyping versions of Siri that rely on ChatGPT and Claude, alongside parallel development using next-gen AFM models.
In another Bloomberg report, sources revealed that Apple’s internal policies may be holding back its AI performance. The company’s focus on privacy leads it to prioritize on-device processing rather than relying on cloud infrastructure. While this protects user data, it also limits computational power. Apple’s primary on-device model runs on just 3 billion parameters. Its internal cloud model reaches 150 billion. By comparison, rival systems hosted in the cloud can exceed a trillion parameters.
Internal Tensions and External Pressures
The AFM team now reports to Zhifeng Chen and is overseen by Daphne Luong, who leads AI research at Apple under senior vice president John Giannandrea. Executives have been trying to calm the internal unrest. They’ve told staff that in-house AI development is still a priority and tied to Apple’s larger strategy of building and owning key technologies, just as it has done with custom chips.
Still, morale appears shaken. The combination of unclear direction, aggressive poaching by competitors, and Apple’s own internal constraints has left the models team vulnerable. As Apple’s AI ambitions scale up, its ability to hold onto the talent building that future is in question.
