Summary of Key Points
Apple's strategy in the AI era is not to "go all in" on developing its own large models. Instead, it focuses on controlling user interfaces, investing in hardware for real-time interactions on the user side, and competing for the invisible routing rights in the Agent era to maintain its core advantages. For instance, it purchased the Gemini model from Google as a "replaceable component" to avoid losing control of its platform, similar to what happened to IBM in the past. By appointing a hardware engineer as CEO, Apple strengthens its capabilities for real-time interactions (such as with smart glasses). It also attempts to seize the power to determine which models will execute user tasks in the Agent era through its System Orchestrator, but faces the challenge of organizational inertia that prevents it from taking risks and innovating.
1. Why Doesn't Apple Develop Its Own AI Models? – Fear of Repeating IBM's Mistakes
Why doesn't Apple create its own most advanced large models? It learned from IBM's lesson: In 1981, when IBM introduced the PC, it handed over the operating system to Microsoft, thinking it was just a "component." However, Microsoft turned DOS into a platform that became the standard for users. As a result, users asked whether a computer could run DOS software before considering whether it was from IBM, and IBM eventually lost its dominance in the PC market.
Now, Apple is smarter. It treats large models as "replaceable components." For example, the Foundation Models framework in iOS 27 allows users to switch between different models like Google Gemini and Claude. Industry trends, such as "routing," also support dynamic model selection (using a different model for simple tasks and more complex ones). Apple's strategy is to control the interfaces and scheduling rights, allowing model companies to compete with each other, thus always remaining the "decider" rather than being constrained by any one model company.
2. AI Needs to "Take Action," and Apple's Hardware Suddenly Becomes Valuable
AI used to be limited to "chatting," but now it needs to "act" – for example, to book a hotel directly without just providing suggestions. Real-time response is crucial; responses must occur within milliseconds, and the display must update instantly when the user turns their head. This requires "on-the-device" processing (done locally on the phone or glasses, not in the cloud).
On-the-device processing tests not just algorithms but also hardware capabilities such as batteries, cooling systems, sensors, and cameras – areas where John Ternus, Apple's hardware expert (who has led the development of the iPhone and Vision Pro), excels. Choosing him as CEO reflects Apple's belief that the more AI relies on real-time interactions, the more important hardware becomes. Smart glasses, for instance, are more critical than foldable screens because they enable AI to continuously monitor what the user is seeing, achieving true real-time perception.
3. In the Agent Era, Apps May No Longer Be Necessary? Apple Is Competing for the "Invisible Entrance"
Previously, users had to open apps to use services (e.g., Booking to book a hotel). In the Agent era, agents will complete tasks directly for users, and users may not even know which app is being used in the background. Apple's approach is to allow developers to expose app functions (such as checking hotels or making reservations) to the system, which then decides which app will be used based on the user's needs.
The ability to control this scheduling process has become a valuable asset. In the search era, Google spent $20 billion to acquire the default search engine in Safari to capture users' intent to search for something. In the Agent era, the power to decide which app (e.g., Booking or Ctrip) will be used to complete a task is even more valuable. While the App Store determines what users install once a year, the scheduler determines which app will handle each task every day.
4. Apple's Biggest Enemy Is Not OpenAI, but Its Own Tendency to Avoid Mistakes
Over the past 15 years, Tim Cook has turned Apple into a company that avoids mistakes, with strict processes and risk control, focusing on maintaining its existing user base (e.g., the iPhone ecosystem). However, the AI era requires taking risks. The Vision Pro was a major gamble that did not meet expectations.
John Ternus has been with Apple for 25 years and is skilled at optimizing existing hardware, but he may be hesitant to try unproven innovations. Apple's challenge is whether it can transition from a company that focuses on maintaining its status quo to one that is willing to innovate. If the failure of the Vision Pro makes the company even more risk-averse, it could miss out on new opportunities in the Agent era.
5. The Scarcest Resource in the AI Era Is the Power to Capture User Intent
In the past, Google's $20 billion acquisition of Safari was about gaining the right to be the first to respond to users' search queries. In the Agent era, the ability to directly assist users in tasks (e.g., booking a hotel) is even more valuable. Apple's goal is to retain this power: using hardware (phones/glasses) to keep users' data, senses, and identity, and using the scheduler to control which tasks are executed. Even if the models are developed by others, the user relationship remains with Apple. This is the essence of Apple's AI strategy – not to compete for models but to control the "entrance to users' intentions."
In Conclusion
Apple's approach to AI is not about creating the most intelligent models but about maintaining control over users' data, senses, and actions. By leveraging hardware and scheduling capabilities, it aims to be the "decider" in the AI era. Whether it will succeed depends on its ability to break free from its tendency to avoid mistakes.