Group 1 - The emergence of generative AI has significantly increased the demand for computing power, transitioning from large models to intelligent agents and embodied intelligence [2] - The global AI server market is projected to grow from $125.1 billion in 2024 to $158.7 billion in 2025, reaching $222.7 billion by 2028, with generative AI servers' market share increasing from 29.6% in 2025 to 37.7% in 2028 [3] - In China, the intelligent computing power is expected to reach 1,037.3 EFLOPS by 2025 and 2,781.9 EFLOPS by 2028, with a compound annual growth rate (CAGR) of 46.2% from 2023 to 2028 [3] Group 2 - The trend of cross-domain and cross-cluster mixed training of large models is emerging, supported by advancements in computing network infrastructure [4] - The "East Data West Computing" initiative has seen over 43.5 billion yuan invested, with a total investment exceeding 200 billion yuan, improving network latency and energy efficiency [4] - The construction of computing networks is evolving towards AI-driven and distributed models, with a focus on multi-node and multi-mode collaboration [10] Group 3 - Companies face challenges in cross-cluster mixed training, particularly in integrating different computing service providers and ensuring effective communication protocols [5] - The shift in user demand from training to inference computing power is evident, indicating a transition from a scale-driven to an efficiency-driven industry [6][8] - The service model is evolving from traditional Infrastructure as a Service (IaaS) to Model as a Service (MaaS), focusing on industry-specific solutions [7] Group 4 - The increasing demand for computing power necessitates a reevaluation of self-built computing infrastructure, which may not be cost-effective for many companies [8] - Companies are increasingly opting for computing platforms to manage workloads, raising the bar for service providers to develop efficient scheduling platforms [9] - The construction of computing networks is crucial for driving innovation across various industries, with a focus on AI and distributed computing [9] Group 5 - The rise in computing demand also raises concerns about energy consumption in data centers, with AI data center capacity expected to grow at a CAGR of 40.5% by 2027 [12] - Innovative cooling technologies and strategic data center locations are being explored to reduce energy consumption [12][13] - The integration of AI technologies is enhancing the operational efficiency of data centers, leading to a shift towards fully automated "dark" data centers [15]

Group 1: Company Developments - ByteDance's Seed team launched the general-purpose robot model GR-3, which can efficiently fine-tune with minimal human data, allowing for quick and low-cost adaptation to new tasks and object recognition [1] - Shanghai Artificial Intelligence Laboratory introduced the DeepLink technology solution for large-scale cross-domain mixed training, enabling interconnection and mixed training of large models across multiple intelligent computing centers over distances of up to 1500 kilometers [1] Group 2: Industry Performance - The AI theme index (930713) experienced a slight decline of 0.16% as of July 23, 2025, with mixed performance among constituent stocks [2] - The top-performing stocks included Qianqian Yuyou (002555) with a rise of 2.93%, Tongfang Co., Ltd. (600100) up by 2.74%, and Stone Technology (688169) increasing by 2.18% [2] - Conversely, Guangxun Technology (002281) led the decline with a drop of 2.18%, followed by Xinyi Sheng (300502) down 1.64%, and Deepin Technology (300454) down 1.46% [2] - The AI ETF (512930) also saw a decrease of 0.28%, with the latest price at 1.45 yuan [2] - The AI theme index comprises 50 listed companies involved in providing foundational resources, technology, and application support for artificial intelligence [2] Group 3: Index Composition - As of June 30, 2025, the top ten weighted stocks in the AI theme index (930713) accounted for 52.8% of the index, including companies like Zhongji Xuchuang (300308) and Hikvision (002415) [3] - The AI ETF (512930) has several off-market connections, including the Ping An CSI AI Theme ETF Initiated Connection A (023384) and others [3]

Core Viewpoint - The Shanghai Artificial Intelligence Laboratory has launched the DeepLink technology solution for large-scale cross-domain mixed training, enabling stable mixed training of models with hundreds of billions of parameters across long distances, such as connecting computing centers over 1500 kilometers between Shanghai and Jinan [1] Group 1 - The DeepLink technology supports cross-domain mixed training of large models, facilitating interconnectivity among computing centers in Beijing, Shanghai, and Guizhou [1] - The DeepLink open computing system has been deeply integrated into various intelligent computing platforms, including China Unicom, China Telecom, SenseTime, and China Electronics Technology Group [1] - The operational model of DeepLink is characterized by "1 platform + N types of chips + X regions," ensuring stable performance across different locations [1]

Market Overview - The AI ETF from Huaxia (589010) closed up 0.39% on July 18, with leading stocks like Foxit Software rising 4.99% and Hehe Information up 4.24% [1] - The Robot ETF (562500) remained flat, showing balanced market forces, with Zhongdali De leading at 5.04% and Dongjie Intelligent up 3.74% [1] - The trading volume reached 896 million yuan, with a turnover rate of 5.75%, indicating a stable market sentiment [1] - The Robot ETF's scale hit a new high, increasing by 730 million yuan to reach 15.529 billion yuan, surpassing the total scale of similar funds [1] Key Developments - UBTECH Technology won a robot equipment procurement project worth 90.5115 million yuan, setting a record for the largest contract amount for humanoid robots globally [2] - UBTECH also launched the world's first humanoid robot Walker S2, which supports autonomous battery swapping, with plans to deliver 500 units for smart manufacturing within the year [2] - The Shanghai Artificial Intelligence Laboratory released the DeepLink technology solution for large-scale cross-domain mixed training, marking a breakthrough in inter-provincial connectivity for large models [2] Industry Insights - Dongfang Securities believes that domestic humanoid robots are accelerating breakthroughs and may lead the global industry landscape [3] - The humanoid robot industry has become a national strategy in China, with expected policy support strengthening and core component localization narrowing the R&D gap [3] - The long-term investment value of domestic humanoid robot manufacturers and OEMs is viewed positively, with manufacturers benefiting from algorithm and brand positioning in high-growth scenarios [3]

1.【DeepLink超大规模跨域混训技术方案发布】据上海人工智能实验室官微消息，上海人工智能实验 室7月19日发布DeepLink超大规模跨域混训技术方案，并已完成多个项目落地，支持千公里多智算中 心跨域长稳混训千亿参数大模型，标志着超大规模智算跨省互联实现新突破。（科创板日报） 2. 【马斯克：将开发专门针对儿童的Baby Grok】马斯克在X上表示，他旗下的人工智能初创公司 xAI将开发一款专门针对儿童的应用程序，并将其命名为Baby Grok。本月早些时候，xAI推出了聊天 机器人Grok 4，这距离其上一次迭代仅过去了几个月。更新之前，Grok 4还在X上分享了多条反犹太 言论，引起了广泛谴责。报道称，马斯克未透露"Baby Grok"的具体功能细节，仅表示该应用将为儿 童提供"友好型内容"。（新浪财经） 3.【DuckDuckGo推出AI图像过滤功能，提升搜索体验】DuckDuckGo推出新功能，允许用户在搜索 结果中过滤AI生成的图像，该功能源于用户的反馈需求。用户在"图片"标签页中可通过下拉菜单选择 是否显示AI生成内容，也可在搜索设置中直接启用"隐藏AI生成图像"选项。（DoNews） 此外 ...