Core Viewpoint - The company expects to achieve revenue from large model-related businesses of approximately RMB 600 million to RMB 620 million for the fiscal year ending December 31, 2025, representing a year-on-year growth of about 1057% to 1095% compared to RMB 51.87 million for the fiscal year ending December 31, 2024 [1][5]. Group 1 - The significant growth in large model-related business revenue is attributed to the company's sustained industry-leading core technology capabilities and accelerated commercialization processes [1][5]. - The company has developed the "Shanhai" series of large models, which includes a complete matrix layout comprising the "Shanhai" large language model, "Shanhai" multimodal large model, and "Shanhai" specialized industry large model [1][5]. - The company has established sustainable systemic advantages in key dimensions such as model architecture design, underlying algorithm capabilities, engineering delivery, and industry adaptability [1][5]. Group 2 - The company has created various intelligent agents based on a unified intelligent platform, covering typical business scenarios such as smart customer service, smart marketing, and document review [1][5]. - The company's large model products are continuously being implemented and replicated in serious application scenarios with high professionalism and complexity, such as healthcare, medical insurance, and transportation [1][5]. - The increasing customer recognition and accelerated commercialization processes are driving rapid growth in the company's large model-related business revenue during this period [1][5].

Core Insights - The enterprise-level AI application industry is transitioning from a technology exploration phase to a large-scale application phase, driven by advancements in large language models [1][14] - Key challenges in scaling AI applications include the need for systematic, end-to-end implementation capabilities rather than relying solely on technological breakthroughs [1][23] - AI Agents are becoming the core vehicle for enterprise-level AI applications, facilitating deep integration with business processes [1][29] Application Layer - AI Agents are central to the implementation of enterprise-level AI applications, breaking down tasks into smaller units and integrating with business processes through various methods [1][29] - The focus is on enhancing efficiency in processes, amplifying knowledge, and innovating value through AI applications [17][27] Supporting Layer - A data-centric approach is essential for model selection, emphasizing the construction of a robust data foundation and a data security system tailored for AI [1][41] - High-quality datasets are critical for AI development, enabling effective model training and application [41][42] Infrastructure Layer - The evolution of AI computing infrastructure is moving towards a heterogeneous model, highlighting the importance of deep collaboration between software and hardware in the context of domestic alternatives [1][50][53] - AI infrastructure is crucial for optimizing the performance and cost-effectiveness of AI applications [53] Organizational Layer - Leadership commitment and top-level design are vital for driving AI transformation within organizations, alongside the need for role upgrades among employees [1][56][60] - Employees must transition from traditional roles to AI collaborators, requiring new skills to effectively integrate AI into business processes [60] Vendor Landscape - The enterprise-level AI application market consists of four main categories: application software, technical services and solutions, cloud services, and AI model providers, creating a dynamic competitive landscape [2][65] - Established companies leverage their industry expertise to extend AI applications, while startups focus on specific scenarios to complement existing systems [65][66] Development Trends - Future trends include the evolution of large models from single architectures to multi-architecture iterations, deep integration of AI into business processes, and the emergence of AI-native applications [2][8] - AI is expected to reshape research processes and enhance competitive advantages for enterprises [2][8] Financing and Investment - Over 50% of AI financing events are concentrated in the application layer, with AI in healthcare emerging as a popular investment area [12][14] Challenges in Scaling - Key bottlenecks in scaling AI applications include weak data foundations, lack of quantifiable business value, and a shortage of talent with both technical and business insights [23][27]

Core Viewpoint - The company expects to achieve revenue from large model-related businesses of approximately RMB 600 million to RMB 620 million for the fiscal year ending December 31, 2025, representing a year-on-year growth of about 1057% to 1095% compared to RMB 51.87 million for the fiscal year ending December 31, 2024 [1] Group 1 - The significant growth in large model-related business revenue is attributed to the company's leading core technology capabilities and accelerated commercialization processes [1] - The company has developed a complete model matrix, including the "Shan Hai" series of large models, which encompasses the "Shan Hai" large language model, "Shan Hai" multimodal large model, and "Shan Hai" specialized industry large model [1] - The company has established sustainable systemic advantages in key dimensions such as model architecture design, underlying algorithm capabilities, engineering delivery, and industry adaptability [1] Group 2 - The company has created various intelligent agents based on a unified intelligent platform, covering typical business scenarios such as smart customer service, smart marketing, and document review [1] - The company's advanced and mature technology and product system have led to successful implementation and replication of large model products in serious application scenarios with high professionalism and complexity, such as healthcare, medical insurance, and transportation [1] - The increasing customer recognition and accelerated commercialization process have driven rapid growth in the company's large model-related business revenue during this period [1]

Core Viewpoint - The company expects to achieve revenue from large model-related businesses of approximately RMB 600 million to RMB 620 million for the fiscal year ending December 31, 2025, representing a year-on-year growth of about 1057% to 1095% compared to RMB 51.87 million for the fiscal year ending December 31, 2024 [1] Group 1 - The significant growth in large model-related business revenue is attributed to the company's leading core technology capabilities and accelerated commercialization processes [1] - The company has developed a complete model matrix, including the "Shan Hai" large language model, "Shan Hai" multimodal large model, and "Shan Hai" specialized industry large model, addressing both general and industry-specific application needs [1] - The company has established sustainable systemic advantages in key dimensions such as model architecture design, underlying algorithm capabilities, engineering delivery, and industry adaptability [1] Group 2 - The company has created various intelligent agents based on a unified intelligent platform, covering typical business scenarios such as smart customer service, smart marketing, and document review [1] - The company's large model products are continuously being implemented and replicated in serious application scenarios with high professionalism and complexity, such as healthcare, medical insurance, and transportation [1] - The increasing customer recognition and accelerated commercialization process have driven rapid growth in the company's large model-related business revenue during this period [1]

香港交易及結算所有限公司及香港聯合交易所有限公司對本公告的內容概不負責，對其準確性或完整性亦不發表 任何聲明，並明確表示概不就因本公告全部或任何部分內容而產生或因倚賴該等內容而引致的任何損失承擔任何 責任。 （股份代號：9678） 2025年度大模型相關業務收入業績預告 本公告乃由雲知聲智能科技股份有限公司（「本公司」）根據香港聯合交易所有限公司證券 上市規則（「上市規則」）第13.09條及第13.10B條及《證券及期貨條例》（香港法例第571章） 第XIVA部項下內幕消息條文（定義見上市規則）作出。 一、大模型相關業務收入預計情況 UNISOUND AI TECHNOLOGY CO., LTD. 雲知聲智能科技股份有限公司 （於中華人民共和國註冊成立的股份有限公司） 1 三、風險提示 本公告所載大模型相關業務收入數據僅反映本公司部分業務表現，並不代表本公 司整體收入水平。本期間內預計本公司錄得整體收入合計約為人民幣11.8億元至 人民幣12.4億元，相較截至二零二四年十二月三十一日止年度錄得整體收入人民 幣9.4億元，預計同比增長26%至32%。本期間內本公司大模型相關業務收入佔整體 收入比例預計約為48% ...

Core Insights - The article discusses the rise of embodied intelligence technology in various sectors, highlighting its role in enhancing new productive forces in industries and public services [1] - Liu Liang, a key figure in the field, emphasizes the importance of collaboration between educational institutions and industries to foster talent and technological breakthroughs [2] Group 1: Technological Development - Embodied intelligence is transitioning from fixed programming to autonomous understanding and planning, which is crucial for adapting to complex scenarios [9] - The core advantage of robots lies in their cost-effectiveness over time, continuous operation without interruption, and precision in high-stakes environments [6][7] - The focus on practical applications of robots, particularly in logistics and manufacturing, is becoming more mature, while humanoid robots are still in the developmental phase [5] Group 2: Educational Institutions' Role - Universities are pivotal in the development of the robotics industry, responsible for both talent cultivation and technological innovation [8] - Collaborative efforts between universities and industries, such as joint laboratories and mentorship programs, are enhancing the alignment of educational outcomes with industry needs [10] - The establishment of a knowledge system and the promotion of research outcomes into practical applications are essential for driving innovation [2][10] Group 3: Industry Collaboration and Ecosystem - Beijing is building a complete ecosystem for embodied intelligence, fostering interaction between universities, industry chains, and application scenarios [2] - The integration of educational institutions with local industrial parks is facilitating the transition from research and development to commercialization [2][10] - Current applications of humanoid robots in public spaces serve as educational tools, helping to familiarize the public with robotic technologies [11] Group 4: Future Prospects - The ongoing advancements in embodied intelligence are expected to lead to broader applications across various industries in the capital [3] - Participation in robotics competitions is seen as beneficial for technological refinement, public awareness, and industry networking [12]

Core Viewpoint - The focus of the Chengdu Municipal Political Consultative Conference is on the development of the digital economy, particularly in the areas of large models and computing power, highlighting the need for a more centralized and robust digital industry in Chengdu [1][7]. Group 1: Digital Economy and AI Applications - The construction of vertical model intelligent agents is seen as a key direction for industry development, particularly in the architectural design sector, where AI can significantly reduce costs and improve efficiency [3][4]. - The current cost of training a senior architectural designer is approximately 500,000 yuan per year, and AI systems can generate multiple design options in the same timeframe, leading to substantial cost savings [3]. - Vertical model intelligent agents require extensive training data and powerful computing resources, as they need to be deeply optimized for specific industries like architecture [4][5]. Group 2: Computing Power and Infrastructure - There is an urgent need for computing power to support the training of intelligent agents, especially as demand in vertical fields like industrial internet increases [5]. - The National Supercomputing Center in Chengdu has established a computing power system that supports over 30 fields, including AI and architectural design, since its operation began in September 2020 [5][9]. - The newly launched "Rongshu·AI Service Platform" by Chengdu Data Group is designed to provide essential computing power for training vertical model intelligent agents [5]. Group 3: Industry Structure and Talent Attraction - Chengdu's digital industry is characterized by a lack of leading enterprises, particularly those valued at over 100 billion yuan, resulting in a fragmented industry structure [7]. - To attract and retain high-quality talent, there is a call for increased investment in cultural and sports facilities in emerging areas like the Tianfu New Area, which are essential for enhancing the quality of life for residents [8]. - The development of public cultural and sports facilities is crucial for meeting the growing demand from the increasing population of young professionals in the region [8].

Core Viewpoint - The article discusses the financial dynamics and business models of two AI companies, Minimax and Zhipu, which both went public with valuations around $60 billion but are facing significant losses due to high operational costs and investments in model training [5][8]. Group 1: Revenue and Expenditure Dynamics - Both Minimax and Zhipu are characterized as "short and agile" companies with fewer than 1,000 employees, rapidly iterating products and achieving annual revenues approaching $100 million within a few years [9]. - Despite rapid revenue growth, expenditures for both companies are approximately ten times their current income, with Minimax's spending being over five times its revenue in the first nine months of 2025 [11]. - The article raises questions about whether increasing revenue will lead to a narrowing of losses or exacerbate them, indicating a potential scale inefficiency in their business models [14]. Group 2: Role of Computational Power - The article emphasizes the critical role of computational power in the business model of AI companies, noting that training costs are a significant portion of total expenditures, often exceeding 50% [21][24]. - For Minimax, the revenue generated in 2024 is only 65% of the training costs incurred in 2023, while Zhipu's coverage is even lower at 30% by mid-2025 [25][26]. - The companies rely heavily on third-party cloud services for computational power, which contributes to their high operational costs [20]. Group 3: Human Resource Investment - Both companies have a high percentage of R&D personnel, with Minimax's monthly salary per employee reaching up to 160,000 RMB, indicating a focus on talent density rather than sheer headcount [16][18]. - The overall salary expenditure for Minimax is around $10 million annually, which is about 90% of its revenue, reflecting a high investment in skilled labor [18]. Group 4: Business Model Challenges - The article highlights a fundamental contradiction in the business model: while revenue is increasing, it is not sufficient to cover the rising costs of model training and operational expenses, leading to significant losses [30][34]. - The companies are caught in a cycle where they must continuously invest in new models to remain competitive, requiring additional financing that often exceeds their revenue [35]. - The potential for a sustainable business model hinges on the ability to amortize training costs over a longer period, which is currently not feasible due to the rapid pace of model iteration [30][37]. Group 5: Competitive Landscape - The competitive landscape is described as a capital-intensive game, where companies must secure financing to survive, with only a few players likely to emerge as long-term leaders [39][44]. - The article notes that many smaller companies are struggling to compete against larger firms and open-source models, leading to a consolidation in the market [43].

Core Viewpoint - The appointment of Yin Qi as the chairman of Jiyue Xingchen marks a significant strategic move in the AI industry, aiming to enhance the company's competitive edge and commercial viability in the rapidly evolving large model sector [1][2][8]. Group 1: Leadership Changes - Yin Qi has been appointed as the chairman of Jiyue Xingchen, responsible for overall strategic direction and technology [1]. - Yin Qi is also the chairman of Qianli Technology, bringing extensive experience in AI and automotive integration [1][3]. - The new management team includes CEO Jiang Daxin, Chief Scientist Zhang Xiangyu, and CTO Zhu Yibo, indicating a strong leadership structure [1]. Group 2: Financing and Market Position - Jiyue Xingchen completed over 5 billion RMB in B+ round financing, setting a record for single financing in China's large model sector in the past 12 months [1][2]. - The company is transitioning from a financing race to a phase of value realization, with increasing market attention on the "AI Six Little Tigers" [1][8]. Group 3: Strategic Focus - The company aims to become one of the best in the foundational model field in China, focusing on the integration of AI and terminal applications [5][6]. - Jiyue Xingchen has developed three generations of foundational models and has partnered with major smartphone brands and automotive companies to implement AI functionalities [6][7]. Group 4: Commercialization Efforts - Yin Qi's dual role is expected to enhance the commercialization of AI technologies, with a focus on strategic direction and specific projects [7]. - The company is working on a complete chain from foundational models to productization and market implementation, particularly in automotive and mobile sectors [7][8]. Group 5: Future Outlook - Jiyue Xingchen's revenue is projected to reach 1 billion RMB in 2025, indicating strong growth potential [8]. - The company remains optimistic about future financing opportunities but has not disclosed specific IPO plans [9][10].

Group 1 - The domestic computing power is in a very favorable position, with a shift in focus towards high-performance and cost-effective chips due to changing industry demands [1][5] - The third-generation inference GPU chip, S3, was launched by Xiwang, aiming to reduce the cost of one million tokens to one cent, reflecting the industry's transition from training to inference [3] - By 2030, it is expected that inference chips will account for 80% of the company's resource allocation, indicating a strategic focus on optimizing inference capabilities [3] Group 2 - The integrated training and inference chips face challenges such as high costs, unstable supply, and complex deployment, highlighting the need for a reasonable computing power to memory access ratio [4] - The "memory wall" has become a significant bottleneck in chip performance, as the speed of computing unit enhancements outpaces memory bandwidth improvements, particularly in inference chips [4] - Companies like DeepSeek are driving innovation across the entire technology chain, from model architecture to inference systems, aiming to reduce dependency on NVIDIA's CUDA ecosystem [4] Group 3 - The reduction of costs in AI applications significantly boosts the number of applications in the market, with the domestic computing power positioned advantageously to capitalize on this trend [5]