Group 1 - The core viewpoint of the report emphasizes the expansion of AI themes from a single algorithm to diversified scenarios, including computing power, C-end applications, autonomous driving, robotics, and AI for Science, indicating a transition from the "technology trigger" phase to a "steady recovery" and "production maturity" phase in the computing power sector [1][22][24] - The report highlights significant performance growth in domestic optical communication manufacturers, with NewEase's net profit expected to increase by over 380% year-on-year in Q1 2025, demonstrating the realization of performance in the computing power sector [1][22][24] - The report suggests that downstream applications of AI, such as AI agents, autonomous driving, and AI for Science, are likely to accelerate into maturity stages, showcasing practical commercial value [2][23] Group 2 - The report identifies a shift in institutional holdings from high-risk preferences to value growth in the computing power sector, with notable reductions in holdings of companies like Zhongji Xuchuang and Tianfu Communication [2][24] - The "pendulum effect" in the computing power sector is discussed, indicating a rebalancing of fundamentals and market chips, where lower-valued stocks may attract capital due to technological breakthroughs or order increases [3][24] - The report indicates that the optical module sector is entering a golden period of value growth, driven by product upgrades, connection value expansion, and a stable market structure [4][25][27] Group 3 - The report recommends focusing on key companies in the computing power supply chain, particularly in the optical module industry, such as Zhongji Xuchuang and NewEase, as well as smaller firms like Tianfu Communication and Shijia Photon [7][29] - It emphasizes the importance of technological barriers and high order visibility in the core segments of the computing power sector [3][24] - The report provides a detailed overview of the optical module industry, highlighting the demand for 800G optical modules in North America, projected to reach 20 million units by 2025 [4][27]

在数智化转型加速演进的浪潮中，算力自主可控逐渐成为产业发展的核心命题。记者6月11日从爱簿智 能获悉，该团队近期研制出面向边缘场景的E300 AI计算模组并推向市场。 据悉，该成果实现了50TOPS（每秒50万亿次整数运算）的国产算力突破，无需庞大、昂贵的服务器， 能将320亿参数（简称"32B"）大模型轻量化地部署在本地（大模型本地部署，通常指将大模型装在本 地的设备上，不用联网、不依靠远程服务器），解决了边缘AI场景中算力不足和延迟高的问题，有望 加速AI为教育、能源、医疗、金融、交通等数字经济重点行业智能化转型赋能的进程。 记者：孙奇茹 如遇作品内容、版权等问题，请在相关文章刊发之日起30日内与本网联系。版权侵权联系电话：010-85202353 来源：北京日报客户端 据了解，E300支持DeepSeek、Qwen、Llama等主流开源大模型的大模型边缘推理，最高能够将320亿参 数的大语言模型部署在本地，并支持上百种视觉模型的推理，通过将数据本地处理与私有化部署来保护 数据隐私与系统安全，从而满足工业、能源、水利、气象、政务、教育等关键领域对高安全性、高可靠 性AI应用的要求。 所谓边缘AI，通俗理解 ...