Apple AI talent earthquake: Meta invests tens of millions of dollars in annual salary to poach core leaders

The battle for AI talent in Silicon Valley is intensifying, and the head of Apple's basic model team, Pang Ruoming, has been poached by Meta with a compensation package of over $200 million. The morale of the team has been dampened, and several core engineers have resigned along with him.

In July 2025, Apple suffered its most significant talent loss since advancing its self-developed artificial intelligence technology. According to multiple sources, Ruoming Pang, the head of Apple's Basic Model Team (AFM) and a distinguished engineer, has confirmed his resignation and will join Meta's newly formed Super Intelligence team. This computer PhD from Princeton University and a 15 year veteran of Google leads a team of about 100 people at Apple, responsible for developing core large-scale language models that support the Apple Intelligence system and AI functionality on iPhone and iPad devices. Meta has offered a top tier compensation package worth tens of millions of dollars annually to recruit this Chinese AI expert, with a total amount potentially exceeding 200 million dollars, showcasing the tech giant's aggressive strategy in the battle for top AI talent.

Pang Ruoming's resignation is not an isolated incident. His trusted assistant and senior big language modeling researcher Tom Gunter left Apple last month, and several team members have revealed to colleagues that they are considering switching to Meta or other competitors. Insiders have pointed out that Apple's management is evaluating the feasibility of using third-party models such as OpenAI and Anthropic to drive the next generation of Siri. This decision has led to significant strategic uncertainty for the base model team and severely weakened team morale. Pang Ruoming's departure is the most important personnel change since Apple launched its smart project, "said a source familiar with the situation." The team is currently in a state of turmoil, and the new management structure has added layers. Engineers need to report their work through multiple managers

Zuckerberg has recently continued to increase his AI talent strategy, successfully bringing in industry leaders such as Scale AI founder Alexandr Wang, renowned investor Daniel Gross, and former GitHub CEO Nat Friedman. This week, Meta also recruited researchers Yuanzhi Li from OpenAI and Anton Bakhtin, who previously participated in the development of Claude's model at Anthropic, with the aim of strengthening its technological layout in the field of general artificial intelligence.

Meta continues to attract top talent with a high salary strategy that exceeds the industry average, and is restructuring the talent competition landscape in the AI field. Zuckerberg regards AI as the top priority for the company's development and often personally contacts potential employees to offer highly attractive compensation plans.

Despite experiencing talent loss, Apple is still advancing two technological paths: testing external models to support Siri upgrades, and continuing to develop device side intelligent features based on self-developed models. In June of this year, Apple announced the opening of its core end-to-end basic model interface to developers, which is seen as a key measure to expand the iPhone and iPad ecosystem. These technologies support features such as email summaries, generative emojis, and priority notifications. At the same time, Apple quietly acquired two AI companies - TrueMeeting and WhyLabs - to enhance its technological reserves in the fields of Apple Vision Pro and Apple Intelligence.

TrueMeeting's technology can generate "super realistic digital twins" by scanning users' faces with an iPhone, which may be used to enhance Vision Pro's "Persona digital person" feature; WhyLabs focuses on monitoring the performance degradation of AI models to prevent security vulnerabilities or erroneous outputs in generative AI.

At the hardware level, Apple is developing high bandwidth memory (HBM) technology for the iPhone to be launched in 2027, which will be a major breakthrough in the field of mobile computing. The bandwidth of HBM can reach 512GB/s, which is 8 times that of traditional LPDDR5X memory, and can significantly improve the AI processing capability on the device side.

This technology is based on advanced 3D stacking packaging technology, which achieves high-speed interconnection by vertically stacking multiple layers of DRAM chips, meeting the massive data processing requirements for real-time inference of AI large models. This will support localized operation of large language models or visual AI models without relying on cloud computing. However, mobile HBM still faces challenges such as high costs (3-5 times that of traditional memory) and heat dissipation design, and the production schedule is uncertain.

Despite talent loss, Apple still maintains a certain level of strength in the field of AI basic research. In early July, Apple collaborated with the University of Hong Kong to release the DiffuDecoder model, which is the first AI programming assistant to use a diffusion model. It can generate code in parallel, breaking through the limitation of traditional models that can only generate linearly from left to right. The model achieved a 4.4% performance improvement in programming testing and faster code generation speed through the innovative "coupled GRPO" reinforcement learning algorithm. This study lays the foundation for the transformation of AI programming assistants from linear generation to intelligent parallel thinking. However, whether such research breakthroughs can be timely transformed into the core competitiveness of Apple products remains to be tested by the market. There is still a gap between technology research and development and product implementation.

After Pang Ruoming's resignation, the Apple Basic Model team will be taken over by Zhifeng Chen, and the team structure will shift from flat management to a multi-level reporting system. Chinese engineers such as Chong Wang, Zirui Wang, Chung Cheng Chiu, and Guoli Yin are expected to become the new managers. Apple will only launch Apple Intelligence in mid-2024, and its strategic advancement is already later than its competitors. The departure of the core leader casts a shadow over the strategic promotion of intelligent equipment, which may delay the process of independent technology iteration and force the company to rely more on external partners.

Industry analysts point out that the AI competition is not only a battle of models and computing power, but also a battle of talent. Meta's high salary strategy is changing the competitive landscape, and Apple needs to rethink its talent retention strategy, otherwise it may further widen the gap in the AI field. ”With Pang Ruoming's departure, the Apple Basic Model team will now be led by Zhifeng Chen. The team structure will also undergo changes: the original majority of engineers would report directly to Pang Ruoming, while the new organizational structure will include multiple managers reporting to Zhifeng Chen. Zuckerberg regards AI as the top priority for Meta's development and actively participates in the recruitment work of the AI department, often personally contacting potential employees. Although Apple attempts to compensate for the loss of talent through acquisitions and hardware innovation, the outcome of this talent battle will determine whether the future iPhone can be upgraded from a communication tool to a truly personal intelligent terminal.