对话Arm邹挺：2026年物理AI加速，芯片将有这些新进展

Core Insights - The AI industry is rapidly evolving, with a focus on "physical AI" expected to dominate applications by 2026, driven by advancements in modularity and energy efficiency in computing [1][2] - Arm predicts a new era of intelligent computing in 2026, emphasizing the seamless interconnection of cloud, physical terminals, and edge AI environments [1] - The development of a robust software ecosystem and flexible heterogeneous computing infrastructure is crucial for the AI industry to address fragmentation issues in hardware and software [1][4] Group 1: Physical AI Development - "Physical AI" is recognized as a key application area, particularly in embodied intelligence and autonomous driving, although significant time is still needed for large-scale deployment [2][3] - Arm's analysis indicates that breakthroughs in multimodal models and efficient training will enable the large-scale deployment of physical AI systems, transforming various industries such as healthcare, manufacturing, and transportation [2] - The emergence of general computing platforms for automotive and robotic automation is anticipated, enhancing economies of scale and accelerating the development of physical AI systems [2][3] Group 2: Technical Challenges and Solutions - The industry faces challenges in the evolution of world models and VLA (visual-language-action) models, both of which are critical for the implementation of physical AI [2][3] - Arm has established a "Physical AI" division to integrate its automotive, robotics, and autonomous device businesses, aiming to create a real-time closed-loop AI solution that emphasizes power efficiency and reliability [3][4] - Arm's layered solution includes hardware, software, and system-level optimizations to enhance energy efficiency and support the deployment of numerous devices [4] Group 3: AI in Mobile Devices - Arm is a key player in the current AI smartphone trend, with high-end phones expected to run large models with 30 billion parameters by 2025 without internet connectivity [5] - Advances in model compression and architecture design are enabling the development of small language models (SLMs) that maintain computational capabilities while being easier to deploy on edge devices [5][6] - The introduction of Arm Mali GPUs with dedicated neural acceleration technology in smartphones is set to enhance mobile AI capabilities significantly by 2026 [5] Group 4: XR Devices and Market Trends - The XR (extended reality) market is evolving, with AR (augmented reality) expected to be the future focus despite challenges faced in 2025 [7][8] - The integration of AR and VR devices in various work scenarios is anticipated, driven by advancements in lightweight design and battery life [7][8] - Challenges for XR devices include balancing computational power with energy efficiency, meeting stringent design specifications, and ensuring low latency for real-time interactions [8][9] Group 5: AI Chip Market Evolution - The demand for AI chips is evolving, with a focus on specialized accelerators like ASICs and NPUs, which are suited for specific applications [9][10] - Arm is enhancing NPU capabilities through heterogeneous architecture collaboration and comprehensive software ecosystem support [10][11] - The trend towards system-level collaborative design for custom chips is reshaping chip performance, with major cloud service providers leading this transformation [11][12]