DeepSeek出圈，AI模型开启终端侧「范式转移」

Core Viewpoint - The emergence of DeepSeek's distilled models is significantly enhancing the capabilities of edge AI, allowing for the deployment of larger models on resource-constrained devices, thus marking a pivotal shift towards edge AI as the new user interface (UI) for end-users [2][6][10]. Group 1: DeepSeek and Edge AI Development - DeepSeek's distilled models enable devices that previously could only deploy 7 billion parameter models to achieve performance equivalent to 14 billion parameter models, thus elevating edge AI capabilities [2]. - The transition of AI inference to the edge is driven by the need for real-time processing, privacy, and efficiency, as highlighted by Qualcomm's leadership in this domain [2][10]. - The performance gap between distilled models and large models is narrowing, with over 75% of large-scale AI models expected to have parameters below 100 billion by 2024 [8]. Group 2: AI as the New User Interface - AI agents are set to become the primary interface for user interaction, handling various tasks seamlessly in the background, thus simplifying user experiences [10][12]. - The ability of AI agents to manage complex tasks with unclear instructions enhances user experience, making interactions smoother and more intuitive [12]. Group 3: Hardware and Software Collaboration - Qualcomm's strategy involves creating a comprehensive solution that integrates hardware, software, and ecosystem collaboration to overcome limitations in computing power and application ecosystems for edge AI [14]. - The Snapdragon 8 Gen 2 platform showcases significant performance improvements, with a 45% increase in CPU performance and a 40% reduction in GPU power consumption, facilitating advanced AI applications [15]. - Qualcomm's AI software stack simplifies model deployment and enhances performance, allowing developers to create applications without needing to understand hardware intricacies [16]. Group 4: AI Hub and Developer Ecosystem - Qualcomm AI Hub supports a wide range of models and facilitates easy deployment and optimization for developers, significantly shortening development cycles [19]. - Over 1,500 companies, including major players like Meta and OpenAI, are utilizing Qualcomm AI Hub, indicating its growing influence in the AI ecosystem [17]. Group 5: Multi-Device Coverage - The Snapdragon 8 Gen 2 platform enables AI agents to function as native applications on smartphones, enhancing user interaction through personalized features [21]. - Qualcomm's solutions extend beyond smartphones to PCs, automotive, and IoT devices, providing a comprehensive platform for developers to create scalable AI applications across multiple devices [23].