Apple has lost one of its senior artificial intelligence figures in robotics, with Jian Zhang leaving to join Meta’s Robotics Studio, according to Bloomberg. The move underscores an intensifying talent war in embodied AI—where software meets hardware—at precisely the moment Apple is trying to redefine its strategy around generative models and intelligent assistants.
Zhang’s work at Apple centered on automation technologies and applying machine learning to real-world tasks, the kind of research that tends to blend perception, motion planning, and on-device inference. His team’s remit was distinct from a separate group exploring a robot “virtual companion,” but both efforts pointed to a larger ambition: making Apple’s devices more capable of understanding and acting within physical environments.
Meta steps up its embodied AI push
Meta has been recruiting aggressively across AI, pairing open-source model momentum with expanded robotics work. The company’s research units have invested in embodied AI tooling such as Habitat, a simulation platform for training agents to navigate and manipulate in realistic 3D spaces, and in egocentric perception research via datasets like Ego4D. Bringing in a seasoned robotics AI lead from Apple gives Meta more depth in turning these research assets into product-ready systems.
Compensation has been a decisive lever. Bloomberg previously reported Meta offered Apple’s Foundational Models leader Ruoming Pang a package valued around $200 million. Recruiters say top-of-market AI offers now hinge on outsized equity grants and long-vesting stock refreshers, a structure that’s difficult for rivals to counter when their strategy or product timelines appear uncertain.
This is not an isolated departure. Bloomberg and other outlets have tracked at least a dozen AI experts moving from Apple to Meta, OpenAI, and others since the start of the year. In parallel, three researchers from Apple’s foundation models group are said to be heading to OpenAI and Anthropic, amplifying concerns about continuity inside Apple’s AI ranks.
Apple’s AI direction faces scrutiny
A report from the Financial Times described the recent exits as a “crisis of confidence” in Apple’s AI trajectory. Internally, Apple has been working on a more capable, LLM-driven version of Siri, while its highly publicized Apple Intelligence features for the assistant slipped from an earlier software release. Senior software leadership has told employees that a second-generation Siri architecture is in the pipeline with major improvements planned further out.
Apple has also explored using third-party models to power elements of Siri and system intelligence, holding discussions with OpenAI, Anthropic, and Google. That consideration is pragmatic given the cost and pace of frontier model training—industry analysts estimate cutting-edge runs now reach into the hundreds of millions of dollars in compute—but it has also reportedly created friction within Apple’s large language model group as engineers weigh platform clarity against dependence on outside stacks.
The immediate implication of Zhang’s departure is sharper for robotics than for chatbots. Embodied AI demands a tight loop between model design, sensor fusion, simulation, and real-world evaluation, plus ruthless optimization for on-device silicon. Apple’s strengths—custom chips, power-efficient inference, and privacy-centric design—are well suited here. Losing leadership in precisely this nexus risks slowing a field where iteration speed is a competitive moat.
What this means for Meta and Apple
For Meta, the hire bolsters a strategy that links foundational models, AR/VR platforms, and embodied agents. The company’s long-term bet is that assistants won’t just answer queries; they’ll perceive surroundings, manipulate objects, and support real-world tasks through glasses, cameras, and eventually home or workplace robots. Senior talent with consumer device experience can help translate research into reliable, shipping features.
For Apple, the message must be coherence and cadence. Clear decisions about in-house versus partner models, visible momentum in Siri’s next generation, and a focused robotics roadmap could steady recruiting and retention. There’s precedent: when Apple drew a line around custom silicon, it unlocked a decade of performance gains. A similarly resolute stance on where Apple will lead in AI—on-device agents, multimodal perception, or household robotics—would help counter Meta’s pitch of outsized upside and open research.
What to watch next
Signals to monitor include additional leadership moves between Apple and frontier AI labs; whether Apple formalizes a third-party model partnership for Siri; progress in Apple’s on-device multimodal capabilities; and Meta’s pace integrating robotics research into consumer-facing assistants. In an era when AI agents are expected to see, plan, and act, the winner won’t just ship the biggest model—it will ship the smartest stack that works in the messy, physical world.
