Zhitong Financial APP learned that Guotai Junan released a research report stating that the hardware forms of terminals such as AIPC and AI mobile phones are ready-made, and the landing speed of AI terminals is mainly determined by the capabilities and willingness of the AI model. Taking GPT-4o and Gemini as typical examples, their displayjellycrushsagaIt provides emotional natural voice interactions and recognition of the real world. Take Windows Copilot as an example, which demonstrates the management of systems and files. The capabilities of AI models have reached a critical point for end-to-side applications. In addition, Qualcomm and Apple are expected to further lead the era of AI terminals, and the supply chain will benefit deeply. Recommended target Lixun Precision (002475jellycrushsaga.SZ), Pengding Holdings (002938.SZ); beneficiary target Shihua Technology (688093.SH).
Guotai Junan's main views are as follows:
AI has reached a critical point.
The hardware forms of terminals such as AIPC and AI mobile phones are all ready-made, and the landing speed of AI terminals is mainly determined by the capabilities and willingness of the AI model. Typical by GPT-4o and Gemini, it demonstrates emotional natural speech interactions and recognition of the real world. Take Windows Copilot as an example, which demonstrates the management of systems and files. The capabilities of AI models have reached a critical point for end-to-side applications. AI models have been focusing on large-scale computing power hardware capital expenditures for one and a half years, and the method of computing power reasoning token charging has recently entered the stage of price war. AI models have the willingness to promote end-to-end, sell upgraded hardware to consumers, and find brands to collect licensing fees.
Terminal AI is gradually transitioning from APP level to system level.
Most of the early AI end-side products were APP-based AI, and most of them were single-task scene-based AI. But Windows Copilot's Recall or Game Assistant demonstrates the potential of system-level AI, and Apple's Ferret UI model focuses on AI-based human-computer interaction. End-to-side focuses on end-to-side matters, privacy, personality and low cost are the starting point, system-level cross-task interaction is the pain point, and personal assistant digital personality is the ideal; end-to-side and cloud do not conflict, and there is a clear distinction.
Qualcomm and Apple are expected to further lead the era of AI terminals.
Apple has the necessary conditions for end-to-side implementation such as user base, application ecosystem, chip support, iOS iteration, large-model business cooperation, and cloud computing power guarantee. Its subsequent AI terminals will first define product for the industry, and subsequent Iterative upgrades will further drive the wave of machine replacements. At the same time, Qualcomm is expected to benefit deeply based on its core capabilities in AI terminal processing chips. The mobile phone side is 8Gen 4, the PC side is Windows on Arm, and the low power consumption of the wearable/IoT side will be fully upgraded. Coupled with the advanced process upgrades of TSMC, Qualcomm will become the NVDA in the era of AI end-side implementation.
Risk warning: The risk of AI mobile phones not penetrating as expected; the uncertainty of Sino-US trade frictions.