云天励飞-U（688343.SH）深度报告 算法芯片化助力，全产业链发展打造推理 AI 龙头 2025 年 09 月 30 日 ➢ 公司 AI 产品以消费级、企业级、行业级三方向发展。公司以"算法芯片化" 作为底层优势，在消费级、企业级、行业级三种场景部署产品和服务。企业级场 景主要为 AI 推理芯片与相关产品、算力服务及 IP 授权服务等，其中公司自研的 DeepEdge 系列通过"算力积木"架构可提供 8-128T 算力，已经在智慧交通、 机器人、家庭主机、推理服务器等领域与头部企业合作并实现商用，16.07 亿元 的 AI 算力服务订单已经验收并正常回款。消费级场景主要包括噜咔博士 AI 拍学 机、AI 眼镜等 AI 原生产品和 AI 耳机、AI 手表等可穿戴设备模组等 AI 赋能产 品，并实现了快速增长。行业级注重项目质量，考核目标从销售额转向利润端。 ➢ Deepseek 开源模型，推动"算法-芯片协同进化"。Deepseek R1 通过关 键设计的奖励系统、训练模板、自进化现象，以及用于强化学习的核心算法 GRPO，用低算力的 GPU 训练出了媲美 OpenAI o1 的模型，证实了在 AI 行业 ...

Investment Rating - The report initiates coverage with a "Buy" rating for the company [9] Core Views - The company reported better-than-expected revenue in the first half of 2025, with a revenue of 646 million yuan, representing a year-on-year increase of 123.10% [3] - The gross margin improved to 28.53%, up by 12.77 percentage points year-on-year, while the net profit attributable to shareholders was a loss of 206 million yuan, reducing the loss by 104 million yuan year-on-year [3] - The growth in revenue was primarily driven by AI toys and computing power services, with significant contributions from consumer and enterprise-level business segments [4] Company Overview - The latest closing price of the stock is 88.44 yuan, with a total market capitalization of 31.7 billion yuan [2] - The company has a total share capital of 359 million shares, with 264 million shares in circulation [2] - The asset-liability ratio stands at 19.8% [2] Financial Performance - For the second quarter of 2025, the company achieved a revenue of 382 million yuan, reflecting a year-on-year growth of 99.84% [3] - The company’s revenue is projected to grow significantly over the next few years, with forecasts of 1.352 billion yuan, 1.798 billion yuan, and 2.265 billion yuan for 2025, 2026, and 2027 respectively, indicating growth rates of 47.33%, 33.01%, and 25.98% [7][11] - The net profit attributable to shareholders is expected to improve from a loss of 322 million yuan in 2025 to a profit of 52 million yuan by 2027 [7][11] Business Segments - The company’s revenue is derived from three main business segments: consumer-level, enterprise-level, and industry-level applications, with respective revenues of 269 million yuan, 298 million yuan, and 76 million yuan in the first half of 2025 [4] - The consumer segment includes AI-native products and AI-enabled products, with significant sales of AI toys [4] - The enterprise segment focuses on providing AI training and inference computing power services, with ongoing collaborations with well-known companies in various sectors [4][5]

Core Viewpoint - The company reported its 2025 half-year results, exceeding revenue expectations with significant growth in both revenue and gross margin, while still facing net losses [1]. Financial Performance - For the first half of 2025, the company achieved revenue of 646 million, a year-on-year increase of 123.10%, with a gross margin of 28.53%, up by 12.77 percentage points [1]. - In Q2 2025, revenue reached 382 million, reflecting a year-on-year growth of 99.84%, with a gross margin of 27.88%, an increase of 12.6 percentage points [1]. - The net profit attributable to shareholders was -206 million, a reduction in loss by 104 million year-on-year, while the non-recurring net profit was -235 million, reducing loss by 111 million year-on-year [1]. Revenue Breakdown by Business Segment - The company's revenue is primarily derived from enterprise-level, consumer-level, and industry-level business segments, with H1 2025 revenues of 269 million, 298 million, and 76 million respectively [2]. - Consumer-level products include various AI-native and AI-enabled products, with over 70,000 units shipped by the time of the mid-year report, and new AI pet products expected to launch in Q3 2025 [2]. - Enterprise-level offerings include a range of products such as neural network processors, AI inference chips, and edge computing boxes, providing AI training and inference services to clients [2]. Industry Solutions - The company provides integrated hardware and software solutions for system integrators and public service scenarios, focusing on project quality and economic benefits to enhance revenue and profit quality in the industry [3]. Product Development - The DeepEdge10 chip platform fully supports the DeepSeek model series, with successful deployment in a city in South China for government applications [4]. - The company is developing multiple chip models to cater to diverse application scenarios, including the DeepVerse and DeepXBot series [4]. Profit Forecast and Investment Recommendation - Revenue projections for 2025-2027 are 1.352 billion, 1.798 billion, and 2.265 billion, with year-on-year growth rates of 47.33%, 33.01%, and 25.98% respectively [5]. - The net profit attributable to shareholders is forecasted to be -322 million, -171 million, and 52 million, with year-on-year growth rates of 44.31%, 47.01%, and 130.52% respectively [5]. - The company is initiating coverage with an "overweight" rating [5].

Core Viewpoint - The company reported a significant increase in revenue for the first half of 2025, driven by sales growth in both consumer and enterprise scenarios, despite continuing net losses [2][3]. Financial Performance - Revenue for the first half of 2025 reached 645,787,334.97 CNY, representing a year-on-year increase of 123.1% compared to 289,462,659.77 CNY in the same period last year [3]. - The net profit attributable to shareholders was -20,593.82 million CNY, an improvement of 10,380.84 million CNY from the previous year's loss [2][3]. - The net cash flow from operating activities was 39,532,687.45 CNY, a significant increase of 59,345.54 million CNY compared to the previous year [6]. Profitability Metrics - The basic earnings per share for the first half of 2025 was -0.58 CNY, an improvement from -0.87 CNY in the same period last year [5]. - The weighted average return on equity increased to -5.29%, up by 2 percentage points from -7.29% [5]. - Research and development expenses accounted for 31.36% of revenue, a decrease of 31.18 percentage points from the previous year, indicating that revenue growth outpaced R&D growth [5][6]. Business Segments - In enterprise scenarios, the company developed a series of products including neural network processor IP, AI inference chips, and edge computing boxes, which are widely used in various applications such as cameras, robots, and smart computing centers [7]. - In consumer scenarios, the company launched AI-native products like the AI learning machine and plush toys, leveraging AI model capabilities [7]. - In industry scenarios, the company provided integrated hardware and software solutions for system integrators and public service applications, focusing on urban governance and smart transportation [7].

Core Viewpoint - The white paper discusses the AI toy industry, covering its current development status, consumer insights, technological support, best practices, and future trends. It highlights the rapid growth of the AI toy market, driven by favorable government policies and technological advancements, with a significant market presence on platforms like JD.com [1][8][12]. Group 1: Current Development Status - AI toys represent a fusion of AI and traditional toys, utilizing large models and various technologies for intelligent interaction, with 2025 seen as a pivotal year for the industry [1][24]. - The global and Chinese AI toy markets are experiencing rapid growth, with JD.com leading the online market [1][50]. - The industry chain features a technology-intensive upstream, a manufacturing-focused midstream, and a diverse downstream channel structure [1][51]. Group 2: Consumer Insights - AI toys differ significantly from traditional toys in interaction methods, providing emotional companionship and educational value [1][59]. - The consumer demographic has expanded from children to all age groups, with varying motivations and usage scenarios, such as Z-generation parents focusing on education and companionship, while single adults seek emotional support [1][66][70]. Group 3: Technological Support - The technology framework consists of a software layer and a hardware layer, with advancements in real-time communication and sensor modules [1][12]. - The integration of cognitive interaction systems is a notable trend, with JD.com's JoyInside technology offering multimodal interaction advantages [1][12]. Group 4: Best Practices - Successful products like LING Universe and companies like LuKa Doctor have demonstrated strong sales and marketing performance, leveraging technological and resource integration [1][12][14]. Group 5: Future Trends - The commercialization of the industry is in its early stages, with potential for diverse revenue channels in both B2B and B2C sectors [2][12]. - The AI toy market is projected to reach over 100 billion yuan by 2030, with a compound annual growth rate (CAGR) exceeding 70% in China [1][49][50].

Core Viewpoint - Yuntian Lifei is advancing towards a dual listing on the A-share and Hong Kong stock markets after successfully landing on the Sci-Tech Innovation Board in 2023, aiming to accelerate its international capital layout [2][3]. Group 1: Company Overview - Founded in 2014, Yuntian Lifei focuses on the research, design, and commercialization of AI inference chips, offering products and services across enterprise, consumer, and industry applications [3]. - The company has established a complete closed loop from AI inference infrastructure construction to product design, development, and commercialization, enabling rapid migration of AI across diverse scenarios [3]. Group 2: Product Offerings - In enterprise applications, Yuntian Lifei provides AI inference chips, computing power services, and IP licensing, primarily for SoC development, large model services, and service robots [4]. - For consumer applications, the company has developed AI-native products such as AI learning machines, AI glasses, and AI-enabled toys, leveraging advancements in AI technology [5]. - In industry applications, Yuntian Lifei offers integrated hardware and software solutions for smart public transport, smart city governance, and emergency response, addressing complex real-world challenges [5]. Group 3: Market Position - According to a report by Zhaoshang Consulting, Yuntian Lifei ranks among the top three providers of AI inference chip-related products and services in China for 2024, and is also among the top two in the NPU-driven AI inference chip sector [5]. Group 4: Financial Performance - Yuntian Lifei's revenue for 2022, 2023, and 2024 is reported as 546 million, 506 million, and 917 million respectively, with a year-on-year revenue growth of over 168% to 264 million in Q1 of this year [6]. - The market size for AI inference chip-related products and services in China is projected to grow from 11.3 billion in 2020 to 162.6 billion in 2024, with a compound annual growth rate of 94.9%, and is expected to reach 1,383 billion by 2029 at a 53.4% CAGR [6]. Group 5: Strategic Initiatives - Yuntian Lifei plans to increase investment in AI inference chips, focusing on edge computing, cloud-based large model inference, and embodied intelligence [7]. - The company has achieved significant milestones in edge computing, with multiple demonstration projects in high-density video intelligent analysis and small-scale edge large model inference [7]. - In cloud inference acceleration, Yuntian Lifei has launched a reasoning acceleration card based on domestic technology, utilizing a "computing power building block" architecture for modular expansion and task parallelism [8]. - The company has established partnerships with over ten robotics manufacturers, deploying edge chip products in service robots, anticipating embodied intelligence to become a new growth engine in the robotics industry within 3-5 years [8].

Core Insights - The 21st Cultural Expo showcased various AI applications, including AI volunteers, piano-playing robots, and AI-powered toys, highlighting the integration of culture and technology [1][2] - The event featured an AI exhibition area for the first time, with over 60 leading AI companies participating, indicating a significant shift towards digitalization in cultural heritage [1][4] Group 1: AI Applications in Cultural Heritage - Tencent's "Remarkable Oracle Bone Script" project utilizes AI and gamification to lower the barriers for public understanding of ancient scripts, promoting cultural heritage transmission [2] - The AI translation robot and interactive AI exhibits allowed real-time cross-language communication and engaging educational experiences for visitors [2][3] Group 2: New Business Models and Opportunities - The integration of AI in cultural industries is creating new business models, such as AI-driven cultural experiences and smart devices that enhance cultural consumption [4] - Traditional culture is transitioning towards digitalization and IP development, driven by advancements in AI and new technologies [4] Group 3: Policy Support for AI in Cultural Industries - Recent policies in Guangdong province emphasize the application of AI in cultural industries, offering financial incentives for high-quality productions utilizing advanced technologies [5][6]

Core Insights - AI technology is increasingly penetrating the hardware sector, particularly in the realm of AI companionship products aimed at children, elderly, and women, with a strong focus on children's products [1][2] - The primary competition in AI children's companionship products revolves around the integration of education, as these products often serve dual purposes of companionship and educational support [1][2][3] Group 1: Market Dynamics - The AI companionship market for children is characterized by a diverse range of products, including AI toys, smartwatches, and educational devices, with notable brands like FoloToy achieving significant sales growth [2][3] - FoloToy aims to enhance emotional companionship through advanced technology, such as voice simulation, while still embedding educational themes in its products [2][3][6] - The market is expected to face consolidation, with predictions of a "brand death wave" as competition intensifies and only a few brands may survive [7][10][11] Group 2: Target Audience - The target audience for AI companionship products is distinct, comprising both the children who use the products and the parents who make purchasing decisions, necessitating a balance between the two groups' needs [6][9] - Children's developmental stages align well with the capabilities of current AI products, making them more receptive to these technologies, while parents are increasingly open to introducing AI to their children [9][10] Group 3: Competitive Barriers - The core competitive barrier in the AI children's companionship market is technology, particularly in areas such as multi-modal emotional expression and intelligent interaction capabilities [10][11] - The ability to adapt to advanced models and enhance user experience through ongoing data support is crucial for product differentiation and success in the market [10][11] Group 4: Future Trends - The industry is at a crossroads, with the potential for a shift from companionship-focused products to those emphasizing educational value, driven by market demands and consumer expectations [11][12] - The outcome of this transition will determine whether the market will prioritize "companionship" or "education" as the primary value proposition for AI children's products [12]