而要支撑这种转变，端侧设备必须满足一系列严苛条件：运行百亿参数级基座模型、加载企业专有知识 库、支持多智能体协作、处理超长上下文，并保障数据不出域。这些需求共同指向一个核心命题：端侧 AI如何真正落地？ 在近期由MINISFORUM铭凡与AMD联合举办的AI产品体验会上，多位与会业内人士认为，端侧AI要真 正"落地"，不能仅靠算力堆砌，还需在硬件架构、应用场景和生态协同上形成完整闭环。 2025年，人工智能正经历一场结构性迁移——大模型能力不再局限于云端数据中心，而是加速向终端设 备下沉。这一趋势被AMD大中华区市场营销副总裁纪朝晖称为"AI智能体元年"的开端：AI正从对话式 助手演变为具备任务执行能力的生产力工具。 算力下沉：端侧设备如何承载百亿参数大模型？ 过去两年，尽管"端侧AI"概念火热，但实际部署仍面临三重障碍：算力不足、成本过高、生态割裂。传 统消费级PC或笔记本受限于显存容量与内存带宽，难以承载主流开源大模型（如Llama 3 70B、 DeepSeek-R1 70B等）。而企业若选择专用AI服务器，则需承担数十万元硬件投入、专用机房部署及持 续运维成本。云服务虽提供弹性算力，却在数据隐私、响应 ...

Core Insights - Alibaba has officially established a "Robotics and Embodied Intelligence Group," marking a strategic shift towards becoming a core player in the embodied intelligence sector [1][2] - The move aligns with Alibaba's broader strategy to transition from being a passive investor to an active participant in the AI and robotics landscape, as highlighted by CEO Wu Yongming's endorsement at the Cloud Summit [2][3] - The competition in the embodied intelligence space is intensifying, with major players like Tesla, SoftBank, and Google DeepMind also making significant advancements [2][4] Alibaba's Strategic Moves - Alibaba's recent actions are part of a two-year strategic transformation aimed at deepening its involvement in embodied intelligence [2][10] - The establishment of the new group signifies a shift from a broad investment strategy to a focused self-research approach, integrating its AI capabilities with hardware [10][11] - The company has made several investments in robotics firms over the past two years, emphasizing the importance of practical applications in the robotics sector [6][10] Industry Context - On the same day as Alibaba's announcement, SoftBank revealed its acquisition of ABB's robotics division for nearly $5.4 billion, indicating a significant move towards integrating AI with robotics [4][5] - SoftBank's long-term strategy in AI and robotics has culminated in this acquisition, which provides a mature and profitable industrial manufacturing capability [5][6] - The simultaneous actions of Alibaba and SoftBank highlight a consensus among industry leaders that integrating AI with physical robotics presents a vast market opportunity [5][6] Technical Framework - Alibaba's approach aligns with the "one brain, multiple forms" concept, which utilizes a universal model to drive various robotic forms [11][12] - The integration of NVIDIA's simulation tools with Alibaba's AI models aims to create a unified training and testing environment for different robotic forms [12][14] - Alibaba's extensive data ecosystem, derived from its various business operations, provides a unique advantage in training AI models and reducing costs associated with data collection [14][16] Challenges Ahead - Both Alibaba and SoftBank face significant challenges in bridging the gap between AI software and hardware, which is crucial for successful implementation [15][16] - The complexity of integrating diverse hardware architectures and communication protocols poses a major hurdle for Alibaba's strategy [15][16] - The high costs associated with advanced hardware and data collection present additional barriers to commercializing AI-driven robotics [15][16]