From June to July 2025, the tech industry witnessed a seismic AI talent war. Social media giant Meta lured away four core AI experts from Apple, including the head of the foundational model team, Pang Ruoming, with sky-high salaries. This move not only triggered internal upheaval in Apple’s AI team but also highlighted the stark differences in AI strategy between the two tech giants. The event has far-reaching implications on global AI talent distribution and technological development.
Talent Earthquake: Apple’s Core Team Defects to Meta
From June to July 2025, a dramatic AI talent war unfolded in Silicon Valley. Meta successfully poached Pang Ruoming, head of Apple’s foundational model team, along with three other core team members, offering them a compensation package worth over $200 million over four years (including a $100 million signing bonus). This sum far surpasses the $74.6 million annual salary of Apple’s CEO, Tim Cook. Pang, a seasoned scientist with a PhD from Shanghai Jiaotong University, previously led Google’s speech recognition framework development before joining Apple in 2021. At Apple, he assembled a 100-person AFM team, responsible for key iPhone features such as smart summaries and emoji generation. He also led the development of the 2025 WWDC AI model interface. His departure triggered a chain reaction: Tom Gunter, head of large model inference optimization; Mark Lee, a multimodal interaction expert; and Zhang Bowen, a real-time environmental sensing researcher, followed suit and joined Meta. This mass departure left Apple’s AFM team in a technical vacuum.
The deeper cause of this “quake” lies in Meta’s strategic precision. Through its Super Intelligence Lab (MSL), Meta employed a “high salary + Chinese talent network” strategy, effectively building a talent siphon. 70% of MSL’s core members are of Chinese descent, many from top universities like Tsinghua, Peking University, and the University of Science and Technology of China (USTC). For example, Pang’s partner at Meta, Yu Jiahui (a graduate of USTC’s elite class), was brought in through a network of alumni, forming a “familiar faces bring familiar faces” talent chain. This network effect not only reduces communication costs but also continuously attracts potential talent through academic collaborations. In contrast, Apple engineers are paid 30% less on average than their counterparts at Meta, and the company’s lengthy decision-making processes and reluctance to approve open-source plans have demotivated its team. Pang, during his exit interview, bluntly stated: “We were dancing with shackles on, while Meta offered a stage for exploring AGI.”
Meta’s poaching strategy has a notable leverage effect. In addition to Apple, Meta also lured eight experts from OpenAI, including Huiwen Chang, co-creator of GPT-4o’s image generation system, and Shengjia Zhao, head of the synthetic data team, further intensifying the talent imbalance in the industry. This “salary ceiling” strategy not only directly removes key figures but also destabilizes the talent base of competitors by causing industry-wide disruption. Some engineers from Apple’s AFM team have secretly begun seeking opportunities elsewhere, while Meta’s $100 million signing bonus initiative is reshaping the career value assessment system for AI talent.
Strategic Contest: The Clash of Edge Constraints and Cloud Acceleration
At its core, this talent battle represents the fundamental differences in AI strategic approaches between Apple and Meta. Apple adheres to a “device-first” principle, requiring all AI models to be compatible with devices like iPhones, which limits the scale of model parameters (with the largest model only having 30 billion parameters). This results in task accuracy being over 20% lower than cloud-based models. Pang’s team repeatedly proposed relaxing these limitations, even outlining a roadmap for an open-source, lightweight model called “Ajax LLM,” but it was outright rejected by Apple’s Senior VP of Software Engineering, Eddy Cue, due to concerns over user privacy. This “performance-for-size” strategy confines Apple’s AI features to simple device-side interactions, while Meta, with its Llama 4 model (400 billion parameters), has already achieved multimodal interactions with performance close to GPT-4.5. Coupled with AI glasses and other hardware, Meta is reshaping the social entry points of the future.
The talent drain directly forced Apple’s technology roadmap into a passive state. Siri 2.0, which was initially planned for release in 2025, was delayed to 2026 due to the team’s departure, with internal discussions even considering replacing the self-developed model with Anthropic’s Claude, potentially causing Apple to lose its technological independence. Worse still, in order to fit its models to the A-series chips in iPhones, Apple’s models need to be quantized and pruned down to a tenth of their original size. This results in more than a 20% decrease in accuracy for complex tasks like multi-turn conversations and long-text generation. If external researchers develop more efficient compression methods, Apple’s edge in device-side AI could be completely undermined. In contrast, Meta’s “cloud + hardware” dual-track strategy has created a synergistic effect. The integration of Llama 4’s multimodal abilities with AI glasses’ real-time environmental awareness places Meta at least 18 months ahead of Apple in social AI interactions.
This strategic difference also highlights contrasting organizational cultures. Apple engineers describe their work environment as “like conducting R&D in a closed castle,” while Meta’s MSL team is referred to as “the AGI testbed.” Members of Meta’s team have access to enormous computational resources and can quickly verify technical breakthroughs through open-source collaboration. For example, Meta’s latest Llama 4 model has achieved multimodal interactions, whereas Apple, due to its talent drain, has been forced to slow down technical iteration, keeping its AI features limited to simple Q&A. Ironically, Apple’s “hybrid AI” project (an edge-cloud collaborative architecture), which had been secretly developed, is progressing slowly due to talent loss, while Meta has already applied similar technologies in its products.
Industry Disruption: The Butterfly Effect of the AI Talent War
This talent war reflects a profound shift in the global AI talent landscape. Top researchers are no longer confined to traditional tech giants but are increasingly choosing sides based on high salaries, clear goals, and technical freedom. Within Meta’s MSL team, 70% of the core members have Chinese backgrounds, with many coming from top universities like Tsinghua, Peking University, and USTC. This talent network is reshaping the global AI technology map. For example, Pang’s partner Yu Jiahui at Meta was introduced through alumni networks, forming a “familiar faces bringing familiar faces” chain of talent. Meanwhile, the trend of AI talent “decentralization” is intensifying—top researchers are shifting their career choices from “seeking stability” to “seeking breakthroughs,” and Meta’s “high salary + Chinese talent network + clear goals” strategy is becoming a model for other companies to emulate.
Apple is now facing a triple crisis: the technical gap may make it difficult to regain its self-developed capabilities within the next 2–3 years; brand trust has declined due to the delay of Siri 2.0; and investor confidence has been dampened by a 1.5% drop in its stock price. The more severe concern is that if Apple ultimately adopts external models, the uniqueness of its device ecosystem will be compromised. Industry observers note that abandoning edge models may cause Apple to lose its core competitive advantage in data privacy. In contrast, while Meta has quickly built a technological advantage through poaching, its aggressive strategy carries risks: the cultural clash between the high-salary team and the original FAIR institute could lead to internal friction, and the massive investments need to yield AGI breakthroughs in the short term, or Meta will face shareholder pressure.
This talent war also signifies a shift in the battle for AI dominance, entering a new stage: the “talent pricing power” phase. Meta’s $200 million contract for Pang Ruoming not only shattered industry salary records but also reshaped the rules of the talent market with its “salary ceiling” strategy. OpenAI has already been forced to adjust its compensation system to cope with poaching, and companies like Google and Amazon are beginning to adopt Meta’s “high salary + Chinese talent network” strategy. Meanwhile, Chinese scientists are playing a key role in this shift. In Meta’s MSL team, 70% of the core members have Chinese backgrounds, and many come from prestigious Chinese universities, which is reshaping the global AI technology landscape.
Looking ahead, if Apple does not clarify its AI roadmap (device-first or cloud collaboration) by the end of 2025, the advantages it accumulated during the mobile internet era may gradually erode. Meta’s “high salary + Chinese talent network + clear goals” strategy is becoming a model for other companies, signaling that future AI talent wars will become even more intense. This smoke-free war is not only about the technological dominance of two companies but marks the global AI competition entering a new phase of “talent pricing power” battles.
