上 市 公 司 互联网电商 财务数据及盈利预测 | | FY24 | FY25 | FY26E | FY27E | FY28E | | --- | --- | --- | --- | --- | --- | | 营业收入（百万元） | 941,168 | 996,347 | 1,023,358 | 1,118,672 | 1,215,861 | | 同比增长率（%） | 8.3 | 5.9 | 2.7 | 9.3 | 8.7 | | Non-GAAP 归普净利润（百万元） | 158,359 | 158,393 | 95,827 | 122,157 | 157,095 | | 同比增长率（%） | 10.0 | 0.0 | (39.5) | 27.5 | 28.6 | | 调整后每股收益（元） | 62.11 | 65.59 | 39.69 | 50.59 | 65.06 | | ROE (%) | 16 | 16 | 8 | 8 | 9 | | 市盈率（倍） | 17 | 16 | 27 | 21 | 16 | | 市净率（倍） | 2.7 | 2.5 | 1.7 | 1.5 | 1.4 | 2026 年 ...

9.11和9.9谁更大？ 去年，这道小学生难度的数学题难倒了一众海内外AI大模型。当时第一财经记者测试了12个大模型，其中阿里通义千问、百度文心一言、Minimax和腾讯 元宝答对了，但ChatGPT-4o、豆包、kimi都错了，错法各有不同。 今日，当第一财经记者提问阿里巴巴最新公测的千问APP："数字9.9和9.11谁大？"时，这次千问却答错了，称："9.11更大"。 有意思的是，在随后的问题拆解中，千问称这是一个常见的思维陷阱，又经过分析后给出了9.9大于9.11的结论。 此前有AI技术从业人士对第一财经记者表示，大模型本质上还是一个语言模型，它从语言数据中学习的是统计相关性，而这使它不擅长做规则学习，从 而不擅长归纳推理。 尽管在常识性问题上出现失误，但在技术积累、生态布局和全球拓展方面，包括阿里巴巴的Qwen模型等在内的中国大模型已具备不容忽视的全球竞争 力。截至目前，Qwen系列模型的全球下载量已突破6亿次。 不久前，爱彼迎CEO Brian Chesky曾表态称公司已经很大程度上依赖阿里巴巴的Qwen模型，它非常好，速度也很快，而且很便宜，"我们也会用OpenAI的 最新模型，但在实际生产中通常 ...

Investment Rating - The industry investment rating is optimistic (maintained) [1] Core Views - The report emphasizes the significant advancements in AI infrastructure, particularly through the launch of Baidu's Kunlun chip and Alibaba's Qianwen app, indicating a strong focus on AI development and domestic computing power [11][13] - The report highlights the expected explosive growth in the domestic computing power industry, driven by major internet companies like Baidu and Alibaba investing heavily in AI applications and infrastructure [13][14] Summary by Sections Section 1: AI Development and Infrastructure - Baidu's Kunlun chip has introduced new models, including Kunlun M100 and M300, aimed at optimizing large-scale inference and training for multimodal models, with continuous product releases planned over the next five years [12] - Alibaba has rebranded its flagship AI app to Qianwen, aligning its features closer to OpenAI's ChatGPT, and is investing 380 billion yuan in AI infrastructure over three years [13] Section 2: Communication Data Tracking - As of August 2025, China has 4.646 million 5G base stations, with a net increase of 395,000 stations from the end of 2024 [23] - The number of 5G mobile phone users reached 1.154 billion, showing a year-on-year growth of 19.46% [23] - 5G mobile phone shipments in August 2025 were 19.992 million units, accounting for 88.4% of total shipments, with a slight year-on-year increase of 1.2% [23] Section 3: Operator Performance - In the first half of 2025, China Mobile's cloud revenue reached 56.1 billion yuan, growing by 11.3% year-on-year, while China Telecom's Tianyi Cloud revenue was 57.3 billion yuan, up by 3.8% [39] - The average revenue per user (ARPU) for China Mobile remained stable at 48.0 yuan, while China Telecom's ARPU slightly decreased by 0.6% to 46.0 yuan [39]

Core Viewpoint - The launch of the video call feature by Doubao represents a significant technological breakthrough, marking a shift in AI's ability to perceive and understand the real world, which could lead to new business opportunities and models [1][20]. Group 1: Technological Advancements - The transition from traditional image recognition to semantic understanding is crucial, where AI not only identifies objects but also comprehends their context and relationships [3][4]. - The new capabilities of Doubao are based on a multimodal large model that integrates visual and linguistic understanding, allowing for real-time interaction and contextual comprehension [4][5]. - The emergence of "embodied intelligence" signifies that AI can perceive the external world through cameras and respond intelligently, moving beyond being a passive tool [5][6]. Group 2: Potential Applications - In office settings, AI can analyze visual data, such as K-line charts, and provide insights, leading to a new service model termed "Screen as a Service" [12][13]. - In education, AI can assist students by analyzing problems directly from their screens and providing solutions verbally [14]. - In healthcare, AI can enhance efficiency by quickly analyzing CT scans and identifying abnormalities, significantly reducing the time required for diagnosis [16]. - Retail experiences will evolve, allowing consumers to instantly access product information and make purchases through visual recognition, streamlining the buying process [17][18]. - Manufacturing processes will benefit from AI's ability to monitor production lines for defects, enhancing quality control and operational efficiency [19][20]. Group 3: Competitive Landscape - Doubao, backed by ByteDance, is positioned to leverage user feedback and data-driven product innovation, potentially expanding its capabilities to other platforms like Douyin and AI glasses [21][22]. - Other companies, such as Quark and Tencent, are also developing AI visual functionalities, but their approaches differ, with Quark focusing on information retrieval and Tencent's product facing user retention challenges [22][24]. - The long-term success of Doubao will depend on the quality and diversity of data for training, as well as its integration into various industries like education, healthcare, and manufacturing [25][26].