Huachuang Securities: AI terminal innovation may lead a new round of consumer electronics innovation, driving the increase in value across multiple hardware segments.

Huachuang Securities released a research report stating that the current momentum of AI terminal product categories is strong. By 2025, sales of AI glasses are expected to explode, and the penetration rate of generative AI smartphones continues to rise, indicating huge potential for industry development. The trend of increasing penetration rate of AI on the terminal side is clear, and innovation in AI terminals may lead a new round of consumer electronics innovation, driving value enhancement in segments such as chips, storage, cooling, batteries, and RF. Companies related to the industry chain may benefit greatly from this trend. The report recommends focusing on related targets in the AI terminal industry chain. Huachuang Securities' main points are as follows: The rapid evolution of AI large models requires terminals, and the industry has entered the hardware layout stage. Large models are deeply integrated into various life and work scenarios, with the number of users and token calls for top models rapidly increasing. Technologically, they have surpassed humans in various benchmark tests such as image classification and natural language reasoning, and the growth in model size has also led to the emergence of "emergent abilities" to understand human instructions. Lightweight models and improved hardware efficiency have significantly lowered the deployment threshold for AI on the terminal side, reducing the cost of inference and breaking the dependence on cloud computing power, thus promoting the large-scale application of AI technology. New terminal products such as AIPC, AI glasses, and AI smartphones are emerging, with models like ByteDance's Douyin AI phone and Ola's Friend AI headset becoming benchmarks for AI on the terminal side, enabling local intelligent interaction and backend automation. Furthermore, large model manufacturers have made hardware layout a core strategy, with tech giants like OpenAI, Google, Alibaba, and Tencent integrating software and hardware to create a more dynamic ecosystem. OpenAI's acquisition of IO Company has accelerated industry iteration. Currently, the momentum of AI terminal product categories is strong, with sales of AI glasses expected to explode by 2025 and the penetration rate of generative AI smartphones continuously increasing, indicating significant industry growth potential. Reflecting on the upgrade process from 4G to 5G, innovative terminals lead to an increase in the value of hardware in multiple segments. 5G has completely surpassed the user experience of 4G with features such as network slicing architecture and three core application scenarios of "high bandwidth, low latency, and wide connectivity." The improvement in user experience has led to continuous growth in penetration rate. According to Counterpoint Research data, by the end of Q4 2023, global shipments of 5G smartphones had exceeded 2 billion units, achieving this milestone in less than 5 years; compared to that, it took close to 6 years for 4G networks to reach a similar penetration level. The report analyzed the hardware changes in representative models during the increase in 5G penetration rate, finding that the upgrade to 5G has driven improvements in segments such as RF front-end, baseband, and cooling, leading to growth in the value of related segments. Specifically: 1) In the RF front-end segment, the number of 5G frequency bands has significantly increased, driving growth in the demand for filters, power amplifiers, and switches, and the evolution of RF devices from "discrete devices" to "modular" components; 2) In the baseband chip segment, the increase in 5G downlink rates and signal processing requirements lead to higher computing power demands for baseband chips, which are continuing to evolve from the common 7nm/10nm process nodes of the 4G era to 5nm/4nm nodes; 3) In the cooling segment, as the computing power and communication capabilities of SoCs increase, chip power consumption and heat generation levels rise significantly. Combined with the need for 5G phones to be compatible with multiple frequency bands, an increase in antenna quantity, and limited internal space, efficient cooling devices such as heat pipes and vapor chambers have seen accelerated adoption. Using the example of 5G hardware upgrades, AI terminals are expected to drive an increase in the value of hardware in multiple segments. AI on the terminal side enhances the user experience and may lead to rapid growth in shipments. Canalys predicts that by 2025, the penetration rate of AI smartphones will reach 34%. The proliferation of RF and other components in 5G era is driven by the "connection dividend" of the 5G era. Similarly, innovation in AI terminals is expected to replicate this upgrade path, leading to a restructuring of value in segments such as chips, storage, cooling, batteries, and RF. Specifically: 1) In the chip segment, System-on-Chip (SoC) designs have shifted from traditional CPU/GPU dual-core to a heterogeneous computing system with "CPU+GPU+NPU" clearly defined roles, highlighting the significant role of NPU in AI on the terminal side; 2) In the memory segment, deploying large models locally will require significant local memory storage, as seen with the LLaMA model with 7 billion parameters requiring a minimum of 3.9G memory even in a 4-bit scenario, driving upgrades in phone memory; 3) In the battery segment, AI on the terminal side increases overall power consumption, leading to continuous upgrades in battery specifications to maintain smartphone battery life, with technologies like silicon-carbon anodes and solid-state batteries seeing increased adoption rates, potentially leading to a continuous increase in battery energy density; 4) In the cooling segment, as chip computing power increases, the requirements for cooling will also rise. However, designs must also meet requirements for weight, thickness, and overall smartphone design, encouraging the adoption of technologies like graphene and vapor chambers; 5) In the RF front-end segment, in scenarios involving complex reasoning, token generation, or calls for large models, some calculations still need to be completed through cloud-edge collaboration, placing higher demands on network throughput and latency stability. Technologies like WiFi 7 and 5G-A may become future upgrade directions. Risk Warning: The industry's innovation progress may not meet expectations, and consumer electronics demand may fall short of expectations.