Summary of Key Points
This was Tim Cook's last WWDC as Apple's CEO, and the focus was not on new hardware but on Apple's "make-up" in the field of AI: a complete overhaul of Siri. Apple decided to abandon its own efforts to develop cutting-edge AI models and instead collaborate with Google to use the customized Gemini model, which has 1.2 trillion parameters and costs approximately $1 billion per year. However, Apple will continue to maintain its own infrastructure for computing power. Additionally, the most advanced AI features require 12GB of memory, which means older models will not be able to utilize these capabilities, and they are not available to users in some regions for now. This event served as both a farewell for Cook and a pragmatic compromise by Apple in the wave of generative AI—acknowledging that it cannot create the most advanced models, but betting that its infrastructure (chips, power, data centers) will be its long-term advantage.
Breakdown and Interpretation
1. Cook's Farewell: Emotionality Behind a Urgent Need for AI Improvement
The opening of WWDC was filled with sentiment as Cook used a video featuring a celebrity guest to say goodbye to his iconic "Good morning" welcome speech, signaling that this would be his last time hosting the event as CEO. Beyond the emotion, Apple's anxiety about its AI capabilities was evident. Over the past year, there have been multiple delays in the updates for Apple Intelligence and Siri, and internally, it has even been acknowledged as an "AI strategy crisis." With Cook stepping down, WWDC had to present a significant AI achievement to address external doubts. After all, during his tenure, Apple's stock price rose by 2000%, yet it has often been viewed as lagging behind in the AI sector. The overhaul of Siri represents Apple's attempt to catch up more quickly.
2. The "Rebirth" of Siri: From a "Voice Remote" to a "System-Level AI Assistant"
The previous version of Siri was more like a "voice remote"—limited to simple commands such as setting alarms or checking the weather. The new Siri, however, has been completely reimagined:
- Underlying Reconfiguration: It no longer relies on an old architecture but uses a new model developed in collaboration with Google, enabling it to understand context (for example, suggesting tasks while chatting or recognizing locations when viewing Instagram posts).
- System-Level Integration: It has been integrated into apps like passwords, emails, and calendars (for instance, automatically updating weak passwords or identifying contacts/locations in calendar events).
- Improved Interaction: There is a dedicated app that allows users to review past conversations, and the interface has evolved from colorful borders to a dark theme with Dynamic Islands.
In short, the new Siri can now understand users in a way similar to ChatGPT, rather than simply executing commands mechanically.
3. Apple's Pragmatic Compromise: Why Rent Google's Model Instead of Developing One Own?
Apple is known for developing everything from chips to software in-house, but this time it chose to rent Google's Gemini model. The reason is straightforward: creating cutting-edge AI models is extremely costly.
- Cost Considerations: OpenAI has an operating margin of -122% (losing $1.22 for every dollar earned), and Anthropic costs $1.25 billion per month in computing power. Training a new model can cost hundreds of millions or even billions of dollars. Renting the model from Google is much more economical for Apple.
- Long-Term Investment: Apple understands that models will eventually become commodities that everyone can use, but the infrastructure (chips, data centers, power) that supports them is what provides a competitive advantage. Therefore, Apple retains its own private cloud servers and only rents the model weights, keeping the computing power under its control.
This is not a sign of defeat; it's about using the least amount of money to address its AI weaknesses while maintaining its long-term strengths.
4. Hardware Requirements: Older Devices May Not Be Compatible
The most advanced AI features of the new Siri require 12GB of memory, which means:
- The standard iPhone 17 (with 8GB of memory) will not be able to use these functions; only the iPhone Air, 17 Pro, and 17 Max are compatible.
- iPads need an M4 chip with 12GB of memory, and Macs require an M3 chip with 12GB of memory.
Apple's approach is clear: the more powerful the AI features, the higher the hardware requirements. This is both a technical limitation (large models need more memory) and a business strategy to encourage users to upgrade to more expensive devices, thereby boosting hardware sales. In the future, when users update their systems, they will not only wonder if an upgrade is possible but also whether they can fully utilize all the AI features.
5. The Limits of Cooperation: How Does Apple Maintain Privacy and Control?
Although Apple is collaborating with Google, it has not handed over its core assets:
- It does not use Google's client-side code or infrastructure.
- The model is customized for Apple Intelligence and is not directly tied to Google's Gemini app.
- Privacy is maintained as conversation history is synchronized privately through iCloud, without passing data through Google's servers.
- For complex tasks, Apple uses its own AFM Cloud Pro model in combination with Google and NVIDIA GPUs, all within its private cloud.
In other words, Apple is renting Google's "brain" but keeping its own "body" (computing power, data, privacy).
Final Conclusion
This WWDC marked a pragmatic shift for Apple in the AI era: acknowledging that it cannot develop the most advanced models on its own, but using collaboration to address its shortcomings. By setting hardware requirements and strategically deploying its infrastructure, Apple aims to turn AI into a tool to drive sales and strengthen its competitive position. For users, if they want to experience the best AI capabilities, they will either need to upgrade to a newer device or wait for further updates. Regardless, Apple's AI era has truly begun.