刘涛披露，公司智算云业务Q3账单收入达7.8亿元，同比增速近120%，主要受益于行业内具身智能场景 及AI推理场景爆发。其中，小米金山生态的贡献显著，2025年上半年相关收入占年度关联交易总额的 40%。除了小米金山生态这个基本盘，金山云的外部客户用量也在齐增。(格隆汇) 金山云(3896.HK)大涨超9%，最高触及7.05港元。消息面上，在上周举办的金山云年度Tech Talk上，金 山云高级副总裁刘涛宣布，智算平台金山云星流已完成从资源管理平台向一站式AI训推全流程平台的 战略升级。据悉，金山云星流训推平台提供从模型开发、训练到推理的完整生命周期管理，具备开发、 训练、推理和数据处理四大模块能力，通过降低多模块协同复杂度，能实现"开箱即用"的AI开发体验。 ...

Investment Rating - The investment rating for the company is "Buy" with a current price of 30.44 HKD and a target value of 50.25 HKD [4]. Core Insights - The report anticipates a stabilization in gaming revenue, steady performance in office software, and a rebound in profits for Q4. The company is expected to see a 10% quarter-on-quarter revenue increase in Q4 2025, despite a year-on-year decline of 5% [8][10]. - The gaming segment is projected to experience a year-on-year revenue decline of 32% and a quarter-on-quarter decline of 2% in Q4 2025, with new game "Goose Duck" exceeding expectations [8][10]. - The office software segment is expected to grow by 19% year-on-year in Q4 2025, driven by government projects and the enhancement of WPS 365 as a global AI collaboration platform [8][10]. - The cloud segment is forecasted to grow by 20% year-on-year in Q4 2025, with continued high capital expenditures expected in 2026 [8][10]. Financial Forecasts - Revenue projections for the company are as follows (in million RMB): - 2023A: 8,534 - 2024A: 10,318 - 2025E: 9,730 - 2026E: 11,571 - 2027E: 13,416 - The growth rates are expected to be 11.7% for 2023, 20.9% for 2024, -5.7% for 2025, 18.9% for 2026, and 15.9% for 2027 [3][10]. - The EBITDA is projected to be 1,268 million RMB in 2023, increasing to 3,511 million RMB by 2027 [3][10]. - The net profit attributable to shareholders is expected to rise from 483 million RMB in 2023 to 2,081 million RMB in 2027, with significant growth in 2024 [3][10]. Business Segment Analysis - Gaming revenue is expected to be 3,977 million RMB in 2023, with a decline to 3,768 million RMB in 2025, followed by a recovery to 4,239 million RMB in 2026 [9][10]. - Office revenue is projected to grow from 4,557 million RMB in 2023 to 5,962 million RMB in 2025, and further to 7,332 million RMB in 2026 [9][10]. - The cloud segment is anticipated to generate significant growth, with revenue expected to reach 12,215 million RMB by 2026 [9][10]. Valuation and Investment Recommendation - The report suggests a valuation of 50.25 HKD per share based on a combination of gaming, office, and cloud business valuations, applying a PE ratio of 13x for gaming, 35x for office, and 3x PS for cloud [12][16]. - The overall recommendation remains "Buy" based on the expected performance and growth potential of the company [12].

[Table_Page] 跟踪研究|软件与服务 证券研究报告 [Table_Title] 【广发计算机&海外】金山云（03896.HK） 前瞻：预计25Q4 AI云高增长，26年延续高资本开支 [Table_Summary] 核心观点： 盈利预测：备注：报告中汇率为 1 港元=0.893 人民币；货币若无特别说明均为人民币。 | [Table_ 单位:Finance] 人民币百万元 | 2023A | 2024A | 2025E | 2026E | 2027E | | --- | --- | --- | --- | --- | --- | | 营业收入（百万元） | 7,047 | 7,785 | 9,481 | 12,215 | 14,687 | | 增长率 （ % ） | -13.8% | 10.5% | 21.8% | 28.8% | 20.2% | | 经调整EBITDA（百万元） | -265 | 639 | 2,292 | 4,072 | 5,657 | | 经调整净利润（百万元） | -1,291 | -825 | -567 | -519 | -387 | | EPS（元/股） | -0.61 ...

金山云高级副总裁刘涛介绍道，在平台效率方面，金山云星流训推平台提供从模型开发、训练到推理的 完整生命周期管理，具备开发、训练、推理和数据处理四大模块能力，通过降低多模块协同复杂度，能 实现"开箱即用"的AI开发体验；作为面向机器人开发与落地的全链路云原生平台，金山云星流机器人平 台深度融合数据采集、存储、标注、模型开发、训练、部署与仿真等核心环节，打造具身场景专属的数 据、模型、仿真一体化引擎；面向大模型应用开发者和企业用户，金山云星流平台模型API服务提供高 可用、易集成的模型调用与管理能力，覆盖模型调用的全生命周期。 艾媒咨询CEO兼首席分析师张毅对《华夏时报》记者表示，金山云进行场景细分的战略意义在于，这是 其从单纯提供通用算力向深度定制行业解决方案的关键战略升级。"通过针对大模型训练、具身智能等 高价值、高增长场景提供精准匹配的解决方案，金山云不仅能有效提升算力资源的利用率和交付效率、 优化成本结构，更能以此增强客户粘性，构筑起差异化的竞争壁垒，从而在激烈的市场竞争中确立长期 优势。" 正是因为不断对业务进行优化，金山云这两年的业绩提升获得助益。财报显示，金山云2025年第三季度 总收入为24.78亿 ...

11月22日，金山云早盘股价一度涨近3%触及6.26港元，截止发稿，金山云上涨0.65%报6.15港元，市值 278.7亿港元，成交额1.32亿港元；隔夜美股市场，金山云上涨4.67%报11.87美元。 此前，金山云宣布公司在智算云领域的新突破，其智算平台金山云星流已完成从资源管理平台向一站式 AI训推全流程平台的战略升级。金山云星流训推平台提供从模型开发、训练到推理的完整生命周期管 理，具备开发、训练、推理和数据处理四大模块能力，通过降低多模块协同复杂度，能实现"开箱即 用"的AI开发体验。 此前财报数据显示，得益于在智算云领域的前瞻性布局和技术能力建设，智算云已 经成为金山云业务的"新底色"，金山云2025年Q3营收同比增长31%，其中智算云账单收入实现近120% 的高速增长，在公有云收入占比达45%。 责任编辑：栎树 港股频道更多独家策划、专家专栏，免费查阅>> 国信通院预计，2045年后我国在用人形机器人超过1亿台，进入各行业领域，整机市场规模可达约10万 亿元级别。金山云作为AI概念股表现抢眼，频频走高，港股多次上涨约20%。分析人士指出，凭借在智 算云领域的前瞻性布局和技术能力建设，金山云有望持续 ...

Core Insights - Kingsoft Cloud's intelligent computing platform, StarryFlow, has completed a strategic upgrade from a resource management platform to a one-stop AI training and inference platform [1][3] - The upgraded StarryFlow platform creates a full-loop ecosystem that includes heterogeneous resource scheduling, self-healing for training task failures, support for robotic industry applications, and commercialization of model API services [1][3] Platform Efficiency - The StarryFlow training and inference platform offers complete lifecycle management from model development to inference, featuring four key capabilities: development, training, inference, and data processing [1][3] - By reducing the complexity of multi-module collaboration, the platform enables an "out-of-the-box" AI development experience [1][3] - The self-developed GPU self-healing technology, combined with task observability design, allows for real-time monitoring of hardware health and task progress, automatically triggering fault migration and task rescheduling to minimize computing interruptions and ensure stable long-term training task operations [1][3] Robotic Platform Integration - The StarryFlow robotic platform deeply integrates core processes such as data collection, storage, annotation, model development, training, deployment, and simulation, creating a unified engine for data, models, and simulation tailored to specific scenarios [2][4] - The model API service on the StarryFlow platform provides high availability and easy integration for model invocation and management, covering the entire lifecycle of model usage [2][4] - This service supports high-concurrency inference and multi-model management, enabling users to efficiently access various model resources, thus facilitating the implementation of large model applications [2][4] - Currently, the StarryFlow platform's model ecosystem supports nearly 40 different models, including DeepSeek, Xiaomi MiMo, Qwen3, and Kimi, allowing clients to access multiple models through a one-stop interface while focusing on AI business innovation and value creation [2][4]

Group 1 - Kingsoft Cloud (KC.US) shares rose 3.17% to $11.70 in pre-market trading [1] - The company announced a strategic upgrade of its AI training and deployment platform, Kingsoft Cloud Star Stream, transitioning from a resource management platform to a one-stop AI training and deployment solution [1] - Analysts believe that the AI application sector is on the verge of explosive growth, and Kingsoft Cloud has been enhancing its full-stack AI capabilities in preparation for this surge [1] Group 2 - The company has laid out a roadmap for its AI development, with plans for infrastructure in 2023, platformization and serverless architecture in 2024, and a comprehensive AI training and deployment platform by 2025 [1] - Kingsoft Cloud is expected to benefit from the growth in AI development, potentially unlocking new avenues for revenue growth [1]

Core Insights - KSYun, a well-known cloud service provider, has upgraded its AI training and deployment platform, KSYun Starflow, to a one-stop solution to meet the increasing demand for intelligent computing driven by the AI application boom [1] Industry Overview - According to Forrester Research, the total addressable market (TAM) for AI in China's new cloud sector is projected to reach 212 billion yuan by 2025 and exceed 410 billion yuan by 2028 [1] - Companies are increasingly focused on the efficiency and usability of AI applications [1] Company Developments - KSYun Starflow platform features four key modules: development, training, inference, and data processing, which simplifies the complexity of multi-module collaboration and provides an "out-of-the-box" AI development experience [1] - The platform incorporates self-developed GPU fault recovery technology and task observability design, enabling real-time monitoring of hardware health and task progress, which reduces the risk of computing interruptions and ensures stable operation of long-duration training tasks [1] - Currently, the KSYun Starflow platform supports nearly 40 models, including DeepSeek and Xiaomi MiMo, enhancing the efficiency and convenience of AI deployment [1]

Core Viewpoint - Kingsoft Cloud (KC.US) has announced a strategic upgrade of its StarStream platform, evolving it into a one-stop AI training and deployment platform [1] Group 1: Platform Upgrade - The upgraded StarStream platform creates a full-link closed loop from heterogeneous resource scheduling, training task self-healing, to robotic industry application support and model API service commercialization [1] - The platform deeply integrates core aspects such as data collection, storage, annotation, model development, training, deployment, and simulation [1] Group 2: Industry Context - The platform achieves a full-link closed loop in the field of embodied intelligent data engineering, efficiently serving core needs such as model training and simulation application scenario analysis [1] - The current market for embodied intelligence is thriving, with the China Academy of Information and Communications Technology predicting that by 2045, there will be over 100 million humanoid robots in use across various industries, with the overall market scale potentially reaching approximately 10 trillion yuan [1]

Core Insights - The AI industry is entering a new cycle driven by "cloud + AI," with generative AI accelerating large-scale implementation and shifting the focus from model training to inference, leading to exponential growth in the inference market [1] - McKinsey predicts that the global AI inference market will reach $150 billion by 2028, with a compound annual growth rate (CAGR) exceeding 40% from 2025 to 2028 [1] - The intelligent computing infrastructure is expected to experience unprecedented market growth as it supports the explosion of inference demand [1] Company Developments - Kingsoft Cloud has upgraded its Starry Flow platform from a resource management platform to a one-stop AI training and inference platform, creating a full-cycle closed loop from heterogeneous resource scheduling to commercializing model API services [1][2] - The Starry Flow platform offers complete lifecycle management for model development, training, and inference, featuring four key modules that enhance platform efficiency and reduce complexity for AI development [2] - The platform's self-developed GPU fault recovery technology ensures real-time monitoring of hardware health and task progress, minimizing computational interruptions and ensuring stable long-term training task operations [2] Industry Trends - Kingsoft Cloud's Starry Flow platform integrates data collection, storage, annotation, model development, training, deployment, and simulation, creating a comprehensive engine for embodied intelligence data engineering [3] - The platform's model API service supports high concurrency inference and multi-model management, facilitating efficient access to various model resources for large model application developers and enterprise users [3] - The AI cloud market in China is projected to reach 22.3 billion yuan in the first half of 2025, with generative AI driving a 148% growth, expected to reach 193 billion yuan by 2030 [4] Ecosystem Advantages - Kingsoft Cloud's strategic positioning in the intelligent computing cloud sector has established it as a key player, providing quality services to ecosystem enterprises and enhancing rapid response capabilities for intelligent computing demand [4] - The Starry Flow platform currently supports nearly 40 different models, allowing customers to efficiently access multiple models through a one-stop service, thereby focusing on AI business innovation and value creation [4]