The AI talent war raging in Silicon Valley has reached unprecedented intensity. Meta is making headlines by aggressively poaching Apple's core AI talent, dealing serious blows to Apple's AI strategy. In this post, we'll dive deep into the current state of the AI talent war, the crisis Apple faces, and what the future might hold.
When One Genius Is Worth More Than Ten Thousand
When Ilya Sutskever, OpenAI's co-founder, left the company in 2024 to establish his startup "Safe Superintelligence (SSI)," it was valued at $32 billion (approximately $43 trillion Korean won) right after launch. That's comparable to Hyundai Motor Company's market cap.
This phenomenon perfectly illustrates the nature of the AI era. Tasks that once required massive workforces are now determined by a handful of brilliant developers. As AI advances, the value of average developers relatively declines, while a small number of innovators with groundbreaking ideas and exceptional technical skills create enormous value.
Apple's AI Team Exodus
Apple's crisis isn't about declining iPhone sales. In fact, the iPhone topped global smartphone sales in Q1 2024. The real problem is the mass exodus of key talent from Apple's AI division.
The most shocking event was Romain Pang's departure to Meta. This Princeton-educated AI genius led Apple's AI team and was the mastermind behind Siri, the heart of Apple Intelligence. As the head of the Foundation team comprising over 100 developers, his departure was described by international media as "symbolizing Apple's AI brain drain."
Meta's offer to Romain Pang was a staggering $2 billion (approximately 2.7 trillion Korean won). This wasn't just a simple job change—it was a strategic attack designed to shake a competitor's core project from its roots. Beyond Romain Pang, Apple's AI Senior Vice President John Giannandrea was removed from his position as head of Siri development in March, and deputy team leader Tom Gruber also left the company.
Zuckerberg's Revenge and Meta's All-Out War
Mark Zuckerberg's aggressive approach stems from the disappointing performance of Meta's Llama model released in April 2024. After harsh media criticism, Zuckerberg was furious and declared he was returning to "founder mode." He disbanded existing teams, demoted leaders, and implemented massive organizational restructuring.
Meta employees are currently working weekends and pulling all-nighters daily. Zuckerberg's goal isn't simply to win the AI competition—it's to end the competition entirely. He's personally emailing hundreds of AI researchers, reviewing their papers, and inviting them to his home to make recruitment offers.
The New Nature of AI-Era Warfare
"Genesis: Artificial Intelligence, Hope, and the Human Spirit," co-authored by former Google CEO Eric Schmidt, former U.S. Secretary of State Henry Kissinger, and former Microsoft CTO Craig Mundie, presents a fascinating perspective. They warn that nations possessing superintelligence that substantially exceeds human cognitive abilities across all domains could weaponize existing viruses to devastating levels.
Like Stuxnet, the malware that destroyed Iran's uranium centrifuges in 2010, they predict the ability to completely destroy competitors' scientific facilities without leaving a trace. The authors categorically state: "In the AI market, second place is meaningless."
Apple's Dilemma and Limitations
Apple's biggest problem is that AI talent no longer chooses Apple. Brilliant developers want to learn from people even more brilliant than themselves. Who would want to join Apple after its key talent has left?
Some suggest "Apple has money, so they can just acquire startups," but modern warfare isn't determined by resources alone. Victory in the AI war comes from buying genius minds with money. No matter how much money Apple has, bringing back departed talent won't be easy.
Tim Cook doesn't personally approach AI researchers like Zuckerberg does. Past iPhone success doesn't guarantee future success.
The Promise and Limitations of Small Models
Apple still has opportunities. Instead of building massive AI models, they could combine smaller models and apply the meticulous attention to detail Apple excels at to maintain competitiveness.
Research using blending techniques has shown that combining small models with around 6 billion parameters can outperform GPT models with 175 billion parameters. User satisfaction was about 50% higher, with rejection rates about 2.5% lower.
However, this approach has limitations. In actual commercialization, variables like data bias can worsen hallucination problems, and API response speed and costs may be disadvantageous compared to single models. Looking at Korean AI company Wrtn's case, using multiple mixed models faces real profitability challenges due to API costs.
Google: Another Formidable Competitor
Even if Apple beats Meta, Google awaits. Google's Gemini 2.0 shows superior performance compared to most major LLMs. The authoritative IEEE in electronics has assessed Google as ahead of both Meta and OpenAI.
Recent reports suggest OpenAI has stagnated while trying to balance technological innovation with corporate profits. Strategies focusing on one goal versus trying to achieve multiple objectives simultaneously yield different results—just like how SK Hynix's focus on HBM beat Samsung Electronics, which tried to do everything.
Declining Morale and Leadership Crisis Within Apple
Most critically, even Apple's executives admit their AI technology is inferior to competitors. Apple Senior Vice President John Giannandrea and Apple Vision Pro development leader Mike Rockwell have both acknowledged Apple AI's inferiority to competitors.
If leaders don't believe in their own weapons, what soldier would risk their life fighting? In June 2024, Apple's Silicon Machine Learning Framework team even threatened mass resignation, shaking the company.
Meta's Continued Offensive
After failing to acquire Sutskever's SSI, Meta recruited co-founder Daniel Gross. They've committed to investing $14.3 billion in data labeling company Scale AI and recruited Silicon Valley notable Alessandro Wang.
Zuckerberg positions himself not as head of a media company, but as someone striving to be first to create AGI that surpasses human intelligence. Meanwhile, Apple has been forced to consider outsourcing Siri, actively reviewing contracts with OpenAI and Anthropic.
Apple's Structural Limitations
Apple's success formula relied on vertical integration—controlling everything from hardware to software, app distribution, and cloud services. However, creating the best AI models requires massive cloud computing and user data, which directly conflicts with Apple's privacy-first policies and on-device AI processing approach.
While iCloud exists, data storage clouds and AI computation clouds are different concepts. Unable to create proper AI due to their principles, Apple began considering integrating OpenAI or Anthropic's AI models into Siri.
Yann LeCun's World Models and Next-Generation AI
More seriously, Meta has Yann LeCun, one of the fathers of deep learning. He criticizes LLMs as not the path to AGI, arguing that true AGI comes from "world models" that learn about the world through visual data rather than text.
World models don't simply memorize data, learn patterns, and predict results. Instead, they model how the world works to predict future situations and make plans. They learn by asking "What happens if I press this button?" and develop accordingly.
This means Meta is preparing for the post-LLM war. The LLM era will end within 3-5 years, and the next phase will see governments and big tech transition to world model-based AI.
Existing LLMs are like people who've read many books—knowledgeable but lacking real experience and imagination. World models are like people with rich imagination who excel at mental simulation. The latter can respond appropriately in complex situations with numerous variables.
Concerns About Korea's AI Strategy
Korea is just now starting to build LLMs, but by the time we produce globally competitive LLMs, big tech companies will have already evolved to world models, likely leaving us still behind.
This is due to differences in technological development speed and investment scale. Global big tech companies are already investing in next-generation technology while we're still trying to catch up with current-generation technology.
Apple's Remaining Options
What paths remain for Apple? Based on "Genesis," Apple's chances of claiming the AI throne seem virtually nonexistent.
Apple's best option is to take AI engines created by OpenAI, Anthropic, or Meta and package them with their superior UI/UX. This mirrors Apple's past success strategy—rather than developing revolutionary technology directly, making existing technology more user-friendly and beautiful was Apple's strength.
However, this alone won't maintain the overwhelming advantage Apple once enjoyed. In the AI era, technology itself is the core of differentiation.
Conclusion: The Rise and Fall of Civilizations and the New Order
Henry Kissinger's words from his 1999 book "Diplomacy" aptly describe Apple's current situation: "Every civilization, however magnificent, is ultimately temporary."
Apple, once the world's greatest, faces crisis before the new paradigm of AI. Meta's aggressive talent acquisition and next-generation technology development, Google's technical superiority, and Apple's internal structural limitations are combining to cast dark clouds over Apple's future.
Of course, the battle isn't completely decided yet. Technological development is unpredictable, and Apple still possesses powerful brand strength, ecosystem, and massive capital. However, if current trends continue, Apple will likely cede AI-era leadership to other companies.
The AI era represents not just technological change but a complete restructuring of industrial frameworks. Past success formulas no longer work in this emerging new order. Which companies will survive and thrive in this wave of change will be determined within the next few years..