Soochow Securities Co.,Ltd. released a research report stating that major internet and cloud computing companies are accelerating their efforts to build a hardware ecosystem featuring end-cloud collaboration, thereby strengthening the hardware foundation for their AI transformation. Both cloud computing firms and internet giants are continuously increasing their focus on the device side by establishing a closed-loop hardware ecosystem that integrates end devices with cloud services, solidifying the underlying hardware support for their comprehensive shift towards AI. From an investment perspective, closely monitoring the strategic moves of major global tech companies in device-side deployment and identifying enterprises deeply embedded in their supply chains, especially those that have successfully become core hardware suppliers, will allow investors to fully benefit from industry growth dividends and identify clear, substantial investment opportunities. The main viewpoints of Soochow Securities are as follows:

AI application iteration is driving a continuous increase in demand for device-side hardware. The push for higher computing power on the device side is reshaping the existing market landscape for traditional mobile phones and PCs. Industry leaders need to rely on AI software demands to drive hardware innovation to consolidate their positions. The implementation of AI applications depends on device-side hardware support, and their rapid development continually elevates hardware requirements. Examples such as the Doubao phone form factor and the Mac Mini popularized by Openclaw are landmark cases of AI terminal deployment. AI applications impose specific demands on the computing power and efficiency of device-side hardware, prompting accelerated upgrades of traditional mobile phone and PC chips towards higher performance and driving ongoing innovations in manufacturing processes and architectural design. PCs and mobile phones, as primary user interfaces, serve as the first point of contact for large models moving from computing centers to the physical world, reaching both consumer and business end-users. They are also the largest carriers for device-side AI. This arena has attracted major cloud providers to cross over and layout, with emerging players further reshaping the competitive dynamics. Companies that capture key device-side access points and actively adapt to new AI applications to redefine PC and mobile phone chip products will gain a competitive edge in the AI race. Although industry giants possess significant technical barriers in device-side chips, meeting core requirements for low power consumption and high-end computing power, they must still keep pace with the times, using software models to drive hardware product innovation to maintain their leading positions.

The automotive scenario represents an ideal practice ground for device-side AI implementation. The iterative upgrade of automotive chips and the construction of a domestic ecosystem present significant development opportunities. Vehicles naturally incorporate the ultra-high computing power chips required for autonomous driving, human-machine interfaces, and connectivity control chips for IoT. Furthermore, the large onboard battery power supply can partially mitigate device-side power consumption limitations, making it an optimal scenario for device-side AI hardware application. Nvidia's proposed world model for autonomous driving further emphasizes the demand for device-side computing power. Automotive chips are mainly divided into two categories: cockpit chips and autonomous driving chips. For cockpit chips, domestic chips are catching up and substituting strongly, with IoT chips iterating upwards, and autonomous driving and mobile phone chips attempting to enter this space. Technologically, they are evolving towards greater intelligence, with hardware supporting the development of software ecosystems for mobile-vehicle connectivity. Autonomous driving chips are undergoing a continuous computing power leap from L2 to L4 levels, transitioning from perceptual intelligence to world models. Simultaneously, domestic chips are achieving breakthroughs through strategies like "autonomous driving democratization" and comprehensive multi-price-point layouts. By closely collaborating with automotive manufacturers and adapting software ecosystems, they are redefining the industrial model for automotive intelligence. Meanwhile, cockpit chips and autonomous driving chips are gradually converging towards an ultimate single-chip form factor, a process that will still take time. Domestic chips, through deep cooperation with new energy vehicle makers, accompanying the global expansion of domestic new energy models, and leveraging software capabilities to build their own ecosystem barriers, are poised for core development opportunities.

The IoT market is currently the largest blue ocean market and represents a core opportunity for domestic substitution. IoT covers diverse scenarios such as wearables, smart homes, and industrial applications, requiring not only hardware technical capabilities but also customized solutions and software ecosystems tailored to specific scenarios and terminals. Domestic chips, leveraging China's vast consumer electronics industry base, have broad opportunities for cooperative development. Among these, AI glasses remain a promising, yet unproven, high-quality device-side scenario. Whether as an extension of mobile phones or exploring alternatives to phones, the industry continues to seek optimal solutions. Embodied intelligence has the potential for smooth technology transfer to the IoT and autonomous driving sectors, with its application scenarios becoming clearer. These emerging, not-yet-fully-defined terminal AI scenarios present significant development opportunities for domestic IoT chips.

The development trend for device-side AI is clear and certain; memory price increases are merely short-term disruptions. The extension of AI cloud computing power and AI applications into the physical world is an inevitable phase of technological and industrial evolution. Device-side systems must possess AI and connectivity capabilities to effectively meet the implementation demands of cloud computing and applications. The core rationale for device-side AI includes privacy protection, security, low latency, and preliminary multimodal processing capabilities enabled by on-device computing. Memory prices began rising in Q2 2025, entering a phase of comprehensive acceleration in the second half of the year, temporarily overshadowing the underlying upward trend in the device-side industry's fundamentals. Looking ahead to 2026, if pressure from memory market price increases gradually eases, or if relevant companies effectively mitigate cost pressures through their own measures, the fundamental improvements brought by AI-enabled device-side hardware are expected to become fully apparent. Short-term cyclical disruptions cannot obscure the long-term technological development trend of device-side AI.

Relevant companies in the industry chain include: Device-side AI SoC chips: Allwinner Technology, Rockchip, SigmaStar Technology, BES, Espressif Systems, Horizon Robotics, Black Sesame Technologies, etc.; Device-side memory chips: GigaDevice, etc.; Consumer electronics terminals and supply chain: Luxshare Precision Industry, Goertek, Lingyi iTech, Dongshan Precision, Ugreen Group, etc.; End-cloud ecosystem: Alibaba Group, etc.

Risk warnings: Domestic demand for device-side computing power interconnectivity and end-cloud collaboration construction may fall short of expectations; The recovery cycle for replacement of consumer terminals like mobile phones and PCs may be slower than expected; The progress of domestic substitution for device-side AI chips and mass production may lag; Industry competition may intensify; Adjustments in domestic data security and compliance policies could introduce operational uncertainties.