嬰夢ギ | · 洞见AI新机遇 食品饮料行业 AI转型自皮书 w 2025食品饮料行业数智化转型领先实践 « Preamble 当下，全球食品饮料行业正经历一场由技术革命与消费变革交 织的深度重构。人口结构变迁、消费行为分层分化、需求个性 化升级、供应链不确定性加剧，叠加气候与资源约束，传统生 产模式和增长逻辑面临前所未有的挑战。 与此同时，以人工智能为核心的新一代数字技术，正穿透从田 间到餐桌的全产业链条，为行业打开效率跃迁与价值重塑的想 象空间。以大语言模型为代表的生成式AI技术取得了突破性进 展，从最初简单的文本生成，到如今能够生成高质量的图像、 音频、视频等多模态内容，其应用范围不断拓展，潜力也愈发 凸显。如何借助AI技术实现转型升级，已成为食品饮料行业发 展的关键命题。 本白皮书旨在为食品饮料行业的AI转型提供一份前沿的、真实 的、本土化的落地指南。我们深知，在AI转型的道路上，企业 面临着诸多困惑和难题，需要有一份具有针对性和可操作性的 经验总结来引领方向。 第一章深入洞察了食品饮料行业的变化趋势，并简述了AI技术 的演变路径、大语言模型的技术原理，以及现阶段应用大模型 技术的主流路径，和未来通用 ...

Industry Overview - GPU dominates the AI chip market with a 90% share in China's accelerator chip market in H1 2023, while non-GPU chips hold 10% [2] - The market size of China's accelerator chips is expected to grow from over 500,000 units in H1 2023 to over 900,000 units in H1 2024, with GPU share decreasing to 80% and non-GPU share increasing to 20% [2] - Domestic AI chip brands account for 10% of the market in H1 2023, expected to grow to 20% in H1 2024 [2] AI Chip Development - Google's TPU development timeline shows continuous advancements from TPU v1 in 2015 to TPU v5p in 2024, focusing on cost-effective inference and AI model training [4] - The evolution of TPUs has enabled large-scale training and inference capabilities, supporting AI model development [4] Data Center and Computing Power - China's data center capacity exceeds 8.3 million standard racks by June 2024, with computing power reaching 246 EFLOPS (FP32), growing at over 65% YoY [6] - Intelligent computing centers are rapidly expanding, with a focus on high power density (20-100kW per rack) and energy efficiency [5] - The AI server market in China grew from RMB 14.9 billion in 2020 to RMB 69.2 billion in 2023, projected to reach RMB 143.3 billion by 2028 [14] Cloud and AI Integration - China's cloud computing market reached RMB 616.5 billion in 2023, growing at 35.5% YoY, with expectations to reach RMB 2.14 trillion by 2027 [18] - Cloud providers are increasing AI computing power investments to meet the demands of large model training and inference [33] - AI cloud services offer advantages such as pay-as-you-go pricing, low-latency inference, and flexible resource deployment [21] Key Players and Innovations - Zhonghao Xinying, a domestic AI chip company, has developed the "Chana" TPU architecture, which outperforms international GPU chips by 1.5x in AI/ML scenarios [30] - The company's "Taize" AI computing cluster supports over 1,024 chips with high-speed interconnects, achieving performance tens of times higher than traditional GPU clusters [30] - Qingyun Technology has partnered with the National Supercomputing Center in Jinan to optimize computing resource management, achieving efficient scheduling of over 3,000P computing power [26] Market Trends and Applications - The global AI application market is booming, with downloads increasing by 26% YoY to 2.2 billion in the first eight months of 2024, and revenue growing by 51% to $2 billion [64] - AI applications are expanding across various sectors, including autonomous driving, smart cities, and industrial optimization [67] - The automotive industry is integrating AI models into vehicles, with companies like NIO, XPeng, and Li Auto developing their own AI platforms [86] Green Computing and Sustainability - China's data centers are focusing on reducing PUE (Power Usage Effectiveness) to meet national and local energy efficiency standards [90] - AI-driven energy optimization techniques are being implemented to improve resource allocation and cooling efficiency in data centers [91] - The integration of renewable energy sources and advanced cooling technologies is becoming a key focus for sustainable computing [100]

2 0 2 4 人工智能开源大模型生态研究 开源为先 场景突破 出品机构：甲子光年智库 研究指导：宋涛 报告撰写：努尔麦麦提·买合木提（小麦） 发布时间：2024.06（初版） 更新时间：2024年6月 Part 01 发展人工智能产业的重要性与新机遇 目 录 Part 02 人工智能大模型的开源生态体系分析 CONTENTS ...

Industry Overview - The AI Agent industry is at an early stage, with significant potential for growth and exploration in product and service models [14] - AI Agents are evolving from Copilot models to more autonomous systems, with 2024 being a pivotal year for this transition [14] - The distinction between Copilot and AI Agent lies in the ability for autonomous planning and execution, with AI Agents requiring minimal human intervention [15] Technological Advancements - Large Language Models (LLMs) have revolutionized AI Agents by providing enhanced understanding and generalization capabilities, enabling better handling of multiple tasks and contexts [5][6] - AI Agents are seen as an evolution of prompt engineering, with a focus on increasing autonomy and reducing human intervention [9] - Memory and planning are critical academic concepts for AI Agents, though their full potential in commercial products is yet to be realized [12] Market Dynamics - The AI Agent market is poised for a significant boom, with a shift from Copilot models to more autonomous AI Agent systems [14] - The market is currently in a state of confusion between academic and commercial concepts, with some Copilot products being considered as AI Agent models [14] - The ultimate form of AI Agents will require only initial user instructions and feedback, with no need for human intervention during the process [15] Applications and Use Cases - AI Agents can be built by ordinary individuals, making them accessible for personalized applications in the AIGC field [21] - In enterprise settings, AI Agents can collaborate with employees to enhance productivity and bring direct business value [24] - AI Agents are expected to bring about significant changes in application software forms and business models, becoming super entry points for vertical applications [36] Competitive Landscape - The AI Agent field is a crucial arena for realizing AI value, with various types of enterprises leveraging their unique strengths to enter the market [66] - Chinese AI Agent market has seen a surge in participants, including internet giants, generative AI companies, enterprise SaaS providers, startups, and 3C hardware companies [69] - The diversity of AI Agents in China is expected to further enrich, with a potential explosion in the number of enterprises [69] Key Players and Innovations - Baidu's Wenxin Intelligent Agent Platform aims to lower development barriers and provide equal opportunities for developers to create, operate, and benefit from intelligent agents [71][72] - ByteDance's Coze platform allows for rapid, low-threshold development of personal Chatbots, with integration into various channels like Doubao, Feishu, and WeChat [74][75] - Alibaba Cloud's DingTalk has launched an AI Assistant Market (AI Agent Store), significantly reducing the threshold for creation and attracting a wide range of users [79][80] Future Prospects - AI Agents are expected to become new digital productivity tools, with the combination of AIGC and workflow capabilities being key to future success [39] - The integration of AI Agents into workflows can lead to more efficient and intelligent decision-making and automation services [39] - AI Agents will play a crucial role in the accumulation and efficient reuse of enterprise knowledge assets, driving innovation in knowledge management [41]