CITIC Securities: Apple releases Apple Intelligence with impressive cross-app information integration capabilities

CITIC Securities released a research report stating that based on the demonstration of Apple's (AAPL.US) AI-related features at WWDC, Apple's AI landing process on the terminal side focuses on information integration and access across apps, making the positioning of system-level personal assistants clearer and successfully widening the gap with existing Android AI phones. Looking ahead, if a manufacturer can achieve the ultimate in AI phone form, the bank believes it may be Apple, which has integrated advantages in chips, models, terminals, and operating systems.

CITIC Securities believes that AI will be an important direction for Apple's future product innovation, and WWDC has officially sounded the horn for Apple's AI landing. With WWDC as the boundary, the curtain is rising on Apple's great year of innovation, firmly bullish on Apple's software and hardware innovation cycle.

CITIC Securities' main points are as follows:

What is Apple Intelligence?

At the 2024 Apple Worldwide Developers Conference (WWDC 2024), Apple introduced its core principles for landing generative AI, which are powerful enough, intuitive and easy to use, deeply integrated, personalized, and privacy-protected. Apple launched the personalized intelligent system Apple Intelligence, which can understand and generate language and images, simplifying the interaction process across apps. According to Apple's official website, the model architecture of Apple Intelligence consists of three main parts: Apple's large-scale model on the terminal side (with about 3 billion parameters, capable of generating 30 tokens per second on the iPhone 15 Pro, with an initial response delay of about 0.6 milliseconds) + Apple's large-scale model on the cloud side (private cloud compute, based on Apple's official evaluation results, the processing capability of Apple's cloud-side model is comparable to GPT-3.5) + Chat GPT (called after Apple's self-developed model), depending on the terminal model when the computing power is sufficient, and using private cloud computing or Chat GPT for more complex scenarios. In terms of specific cooperation with OpenAI, users can ask questions related to text, documents, photos, PDFs, etc., and Siri will determine whether the query request is processed by ChatGPT. In this process, Siri will use GPT-4o for free, users do not need to create an account, and the usage process will not be recorded.

With Apple Intelligence, what are the new features of Siri?

1) Core Highlight: Cross-App Information Integration: During use, by creating semantic indexes for content such as photos, calendar schedules, and files, combined with incoming emails and messages, Siri can discover and understand information across apps (we believe that early on, native apps will be the main focus, with some third parties also supporting calls). It has cross-platform information processing capabilities: for example, when receiving an email notification of a meeting delay, you can directly ask Siri if you can still make it to the movie you previously arranged with a friend. In this process, Siri knows who the user's friend is, what time the movie starts, where the movie theater is, the time needed to reach the destination, and integrates tools such as messages, emails, and maps We believe that the new Siri's ability to integrate information across apps benefits from Apple's integrated advantages in system + chip + model + terminal. Due to the difficulty for Android phone manufacturers to integrate system (Android), chip (Qualcomm, MediaTek), model (currently, except for Samsung's adoption of Gemini, generally using self-developed or open-source large models), and terminal (various brands), existing AI applications on Android phones are limited to specific apps (such as for recording, calling, photo editing), while Apple can target specific scenarios. We believe that the core highlight of the new Siri is the cross-app information integration, allowing Apple to successfully widen the gap with existing Android AI phones.

2) Support for continuous conversation and context understanding: Supports more natural semantic understanding, can connect context, can analyze incoherent language or corrected language (this means Siri can have continuous conversations with users, providing a more natural experience);

3) Ability to understand screen content: For example, you can tell Siri to add address information sent by a friend to the friend's address. On the iPhone, you can also set the way to interact with Siri (voice or text), and freely switch between text and voice depending on the situation.

From product innovation to demand implementation, what is the logic of transmission?

From the perspective of shipments in the past three years, according to GFK, the iPhone shipments from 2021 to 2023 are 235 million, 230 million, and 231 million units respectively, remaining stable overall. In the first quarter of 2024, iPhone shipments are expected to decrease by 9.6% year-on-year to 50 million units. We expect that the shipments in the second quarter of 2024 will be relatively stable, with a possible year-on-year increase of 10%-15% to 45-48 million units. On an annual basis, we expect a slight decrease in shipments. Looking ahead, we believe that consumers' upgrade demand comes from premium experience + hardware exclusivity. Currently, Apple's cross-app AI experience has achieved a "good experience" (the next step is to integrate more third-party apps); and the second step of "hardware exclusivity" is also being realized. Apple has stated that the implementation of Apple Intelligence is limited to models equipped with A17pro, M1/M2/M3/M4 series chips, and subsequent iterative products. Among the existing iPhone models, only the iPhone 15 Pro and iPhone 15 Pro Max support Apple Intelligence. We believe that this requirement is highly related to the operating memory specifications of terminal products. We estimate that the terminal using Apple Intelligence must have at least 8GB of operating memory. Specifically, the memory configuration of iPhone 14 Pro/Pro Max is 6GB, while iPhone 15 Pro/Pro Max has upgraded to 8GB of DRAM, so the previous models of iPhone 15 Pro/Pro Max cannot use Apple Intelligence; Similarly, devices equipped with the M1 chip such as MacBook Air (M1, 2020), MacBook Pro (M1, 2020), iPad Air (M1, 2022), iPad Pro (M1, 2021), all start with 8GB of DRAM, thus all can utilize Apple Intelligence.

Risk Factors:

Lower-than-expected consumer willingness to upgrade devices; increase in terminal hardware costs affecting demand; changes in the international industrial environment and escalating trade frictions; slower-than-expected commercialization progress of AI; slower-than-expected hardware computing power upgrades; slower-than-expected iteration speed of large models.

Investment Strategy:

Considering that Apple's AI functions are limited to only a few models, coupled with the expectation that future iPhone models will continue to innovate in computing power, large model applications, heat dissipation, optics, etc., we believe that the iPhone is likely to experience strong upgrade momentum and enter an upward trend in shipments