芯原股份 20260226 摘要 公司通过 IP 授权与一站式定制服务，帮助客户降低 Capex 固定成本和 研发投入，解决运营成本问题。Fabless 企业通常面临研发费用压力， 而公司通过减少客户自建研发资源投入，支持客户在更多产品线中维持 投入效率。 IP 授权收入由 IP license（知识产权授权使用费）和 royalty（特许权 使用费）构成，约占公司总收入的 1/3。一站式芯片定制服务约占 2/3，包括芯片设计服务 NRE 和芯片量产服务（TMT），客户在不同阶 段支付不同费用。 公司全球员工超过 2000 人，研发人员占比 89%，硕士及以上学历超过 88%。公司采用全球化运营模式，设有 9 个研发中心和 11 个销售办事 处，研发人员主要在国内，但超过 30%的收入来自国外市场。 2025 年营业收入同比增长 35%至 31 亿元，其中量产业务增长 73%， 占比提升，NRE 增长 20%，IP 业务保持增长。数据处理领域收入同比 增长 95%，占比提升至 34%。 一站式芯片定制服务的收费环节如何划分，主要包含哪些收入项？ 一站式芯片定制服务约占公司总收入的 2/3，主要由芯片设计服务 ...

大/小/微模型赋能先进制造： 实 践与思考 Large/Small/Miero Al Models for Manufacturing (Al4M): App licatiansandlnsights 宋学官 大连理工大学机 械工程学院 2 一、 Al4M的背景意义 二、 Al4M的基础知识 三、 Al4M的研究进展 四、 Al4M的案例展示 五、 Al4M的瓶颈所在 六、 Al4M的科学问题 七、 Al4M的发展方向 八、 思 考与总结 汇 报 提纲 一、Al4M的背景意义 二、AI4M的基础知识 三、AJ4M的研究进展 四、AI4M的案例溪示 五、A14M的范颈所在 六、AJ4M的科学问题 七、AI4M的发展方向 八、思考与总结 Al4M的背景意义 5 口先进制造是指采用高新技术和先进设备来改善制造业过程和生产效率的统称，是 衡量一个国家科技发展水平的重要标志，关乎国民经济发展和国防安全建设。 Al4M的背景意义 6 口《中国制造2025》:加快推进制造业转型升级，到2035年整体达到世界制造强国中等水平 □2022年10月，美国发布《国家先进制造业战略》,先进制造业是美国经济和国家安全引擎 美国：Ind ...

Core Viewpoint - The debate on Artificial General Intelligence (AGI) is polarized, with one perspective arguing that AGI will not become a reality due to physical and computational limitations, while the opposing view suggests that AGI may already be achieved or is on the verge of realization [2][4][10]. Group 1: AGI Debate - Tim Dettmers argues that AGI is constrained by physical limits such as memory transfer, bandwidth, and latency, leading to a slowdown in computational growth [10][39]. - Dan Fu counters that the potential of current hardware has not been fully realized, suggesting that significant improvements in computational efficiency are still possible [12][45]. - Both researchers converge on the definition of AGI, emphasizing its impact on changing work processes rather than merely its cognitive capabilities [14][15]. Group 2: Computational Potential - Dan Fu estimates that the theoretical available computational power could increase by nearly 90 times through hardware advancements, system optimizations, and larger clusters [13][46]. - Current models are often based on outdated hardware, and the industry has yet to fully leverage the capabilities of new hardware [49][50]. - The discussion highlights the importance of optimizing hardware utilization, with current effective utilization rates being significantly lower than potential [45][46]. Group 3: Role of Agents - The emergence of code agents is seen as a transformative development, significantly enhancing productivity in programming tasks [20][62]. - Both researchers agree that agents can handle a majority of coding tasks, allowing human experts to focus on oversight and quality control [21][66]. - The ability to effectively use agents is becoming a critical skill in the industry, with those who adapt likely to thrive [68][70]. Group 4: Future Directions in AI - The future of AI is expected to see a diversification of hardware and a shift towards specialized models, with new architectures emerging beyond the dominant Transformer model [23][25]. - Chinese AI teams are recognized for their innovative approaches and practical focus on real-world applications, contrasting with the more centralized technological routes in the U.S. [26][56]. - The potential for AI to revolutionize various sectors, including healthcare and automation, is acknowledged, with significant advancements anticipated in the coming years [57][58].

Core Insights - Google has made significant advancements in the field of AI with the release of T5Gemma 2 and FunctionGemma, focusing on small models that can operate efficiently on edge devices [1][3][37] Group 1: T5Gemma 2 Overview - T5Gemma 2 is part of the Gemma 3 family and emphasizes architectural efficiency and multimodal capabilities, distinguishing itself from larger models like Gemini [3][4] - The model is available in three sizes: 270M, 1B, and 4B parameters, showcasing its versatility [5] - T5Gemma 2 outperforms corresponding models in the Gemma 3 series across various benchmarks, particularly in code, reasoning, and multilingual tasks [9][11] Group 2: FunctionGemma Overview - FunctionGemma is designed for function calling optimization, allowing it to run on mobile devices and browsers, making it suitable for applications like voice assistants and home automation [7][40] - The model has 270M parameters and is optimized for specific tasks, demonstrating that smaller models can achieve high performance in targeted areas [44][46] - FunctionGemma aims to transition AI from a conversational interface to an active agent capable of executing tasks and interacting with software interfaces [43][56] Group 3: Architectural Innovations - T5Gemma 2 represents a return to the encoder-decoder architecture, which is seen as a modernized revival of classical Transformer models, contrasting with the dominant decoder-only models like GPT [14][30] - The model's architecture allows for better handling of "hallucination" issues and provides inherent advantages in multimodal tasks [32][34] - Google employs a technique called "model adaptation" to efficiently train T5Gemma 2, leveraging existing models to reduce computational costs [36] Group 4: Strategic Implications - The release of these models reflects Google's strategic positioning in the AI landscape, particularly in mobile computing and edge AI, as it seeks to maintain control over the Android ecosystem [52][64] - FunctionGemma's design philosophy aims to democratize AI capabilities across various applications, making advanced functionalities accessible to developers without significant infrastructure costs [64] - By establishing a standard protocol for AI interactions with applications, Google is enhancing its competitive edge in the mobile AI market [57][58]

Government Initiatives - The Ministry of Industry and Information Technology (MIIT) has revised the "Management Measures for Public Service Platforms for Industrial Technology," effective from December 5, 2025, focusing on key industries such as equipment, petrochemicals, steel, and artificial intelligence [1][1] - The National Development and Reform Commission, along with other ministries, has issued opinions to strengthen the construction of data element disciplines and digital talent teams, aiming to support the development of a digital economy and society [1][1] - The Ministry of Ecology and Environment has released guidelines for the construction of a product carbon footprint factor database to support the establishment of a carbon footprint management system [1][1] - MIIT is seeking public opinions on the "Comprehensive Standardization System Construction Guide for the Metaverse Industry (2026 Edition)," aiming to establish over 50 national and industry standards by 2030 [1][1] Local Actions - Shandong Province is promoting the metaverse as a new economic growth point, supporting cities like Jinan and Qingdao in building future industry pilot zones [1][1] - Jiangsu Province has established a Metaverse Standardization Technical Committee in Nanjing to fill the gap in the standardization system within the province [1][1] Industry Developments - The GPU leader, Moore Threads, has officially listed on the STAR Market, becoming the first domestic GPU stock, with a market capitalization of 305.5 billion yuan and an opening surge of 468.78% [3][3] - Google has integrated AI simultaneous translation into all its headphones and launched an experimental browser named "Disco," aiming to redefine web browsing experiences [3][3] Academic Insights - Academician Zhang Yaqin predicts that the future of large models will not exceed ten, emphasizing the integration of information, physical, and biological intelligence [4][4] - Academician Tan Jianrong stresses the importance of small models as the foundation for large models, advocating for a shift towards precision small models and industry-specific intelligent agents [4][4] Technology and Applications - The Ministry of Industry and Information Technology has granted approval for China's first batch of L3-level conditional autonomous driving vehicles, marking a significant step towards commercialization [6][6] - Mathematician Terence Tao and his team have solved the 50-year-old Erdős 1026 problem in just 48 hours using AI tools, showcasing the potential of AI in solving complex mathematical challenges [6][6]

Core Insights - The importance of both large and small AI models is emphasized, with a warning that without small models, the implementation of artificial intelligence becomes challenging [2][3] - Knowledge engineering is identified as a core technology for achieving artificial intelligence, alongside models, computing power, and algorithms [4] Group 1 - The need to focus on large models while not neglecting small models is highlighted, indicating a balanced approach is necessary for AI development [2][3] - Knowledge is categorized into qualitative and quantitative types, with models representing quantitative knowledge [4] - Large models require significant computing power for training on diverse data, underscoring the necessity of substantial computational resources behind big data and models [4]

Core Insights - The article highlights the success of NVIDIA's small model, NVARC, which achieved a top score of 27.64% in the ARC-AGI 2 competition, outperforming GPT-5 Pro, which scored 18.3% [2][4] - NVARC's cost per task is only $0.20, significantly lower than GPT-5 Pro's cost of over $7, making it a cost-effective solution [4] - The key innovation of NVARC lies in its zero pre-training deep learning method, avoiding biases and data dependencies associated with large-scale pre-trained models [5] Performance and Methodology - ARC-AGI 2 is a challenging test that assesses a model's ability to acquire new skills beyond its training data, eliminating overlap with public training datasets [6] - NVIDIA's strategy involves moving complex reasoning tasks to an offline synthetic data pipeline, allowing for the training of smaller models that can run quickly during evaluation [9][10] - The NVARC team utilized a large-scale synthetic dataset, creating over 3.2 million augmented samples through a structured pipeline that ensures data quality [18][19] Technical Innovations - The NVARC model is based on an improved ARChitects method, utilizing a small parameter model, Qwen3-4B, and simplifying puzzle understanding through dialog templates [19] - Key to NVARC's success was the implementation of Test-Time Fine-Tuning (TTFT) and LoRA fine-tuning techniques, allowing the model to adapt quickly to new rules for each task [21] - The decoding phase was optimized with batch processing to address non-deterministic outcomes, and eight data augmentation operations were unified to evaluate candidate solutions [22][23] Strategic Implications - The article emphasizes that small models, when optimized for specific tasks, can perform competitively against larger models, highlighting their advantages in cost, speed, adaptability, and domain focus [25] - The success of NVARC suggests that the right methodologies applied in the right contexts can yield significant value, challenging the notion that larger models are always superior [25]

Core Insights - The article highlights the transformation of a small e-commerce service company, Mosquito Club, founded by Wang Junhua in 2014, into a significant player in the digital economy by 2025 [1][2]. Group 1: Company Development - Mosquito Club was established during the rise of e-commerce in China, particularly in Zhejiang, with a focus on providing hands-on training for merchants on how to operate their stores [3]. - The company successfully pivoted to the live-streaming sector early on, capitalizing on the trend initiated by Alibaba, and produced several top streamers within three years [3][4]. - In 2021, Mosquito Club evolved into Wantong Group, which encompasses multiple branches, including public relations and brand consulting, reflecting its growth from serving e-commerce sellers to partnering with global brands [4]. Group 2: Industry Insights - Wang Junhua emphasizes the importance of "small data" over "big data," arguing that while big data shows sales figures, small data reveals critical user insights [7]. - The company adopts a cautious yet open approach to new technologies, exploring virtual anchor technology while recognizing the limitations of AI compared to human flexibility [7]. - The success of Mosquito Club is attributed to confidence and talent, with Wang noting the supportive environment in Zhejiang that fosters innovation and collaboration [7].

Core Insights - The key opportunity in AI entrepreneurship lies in developing specialized intelligent agents rather than competing in the race for larger models [1][2][8] - The intelligent agent market is projected to grow from $5.1 billion to $69.1 billion by 2032, indicating a significant growth potential [4] - Entrepreneurs should focus on solving practical, measurable problems in various industries, rather than pursuing general artificial intelligence (AGI) [10][12] Group 1: Intelligent Agents - Intelligent agents break down tasks into smaller, manageable sub-tasks, enhancing execution and adaptability [4] - The advantage of intelligent agents is their specialization, allowing them to outperform expensive standard models in specific domains [8] - The current AI landscape requires entrepreneurs to build trust in their AI applications, as the technology itself is widely accessible [20][23] Group 2: Market Trends - The market for small models is expected to grow from $930 million in 2022 to $5.45 billion by 2032, with edge computing projected to reach $378 billion by 2028 [13][15] - Edge computing enables applications that handle sensitive data locally, enhancing privacy and reducing costs [16][17] - The military applications of AI are emerging as a significant area of investment, creating opportunities for dual-use technologies [26][27] Group 3: Entrepreneurial Strategies - Entrepreneurs should target industries with repetitive, data-intensive tasks that consume significant human resources [12] - Utilizing open-source models can reduce costs and accelerate product development, allowing startups to operate with lower burn rates [12][28] - The focus should be on building reliable and transparent AI systems to maintain a competitive edge in the market [22][23]

Core Insights - Deutsche Bank's report emphasizes that companies are still in the early stages of developing their AI transformation roadmaps after attending the 2025 New York AI Leaders Summit [1] Group 1: Investment Return and Data Readiness - There is a lack of consensus on measuring return on investment (ROI), with business leaders defining their own key metrics [2] - Data readiness remains critical, with management's understanding of data and its storage locations being fundamental issues [2][3] - Only 10%-20% of total time is reportedly spent on training models, indicating that the quality of models heavily depends on input data [2] Group 2: AI Implementation and Governance - Approximately 80% of clients are still in the phase of optimizing existing business processes, while 20% are more willing to experiment [2] - Regulatory and governance policies are seen as barriers to the speed of AI adoption across enterprises [3] - Many leaders believe that maintaining human involvement in agent processes is crucial for reasonableness checks [2][3] Group 3: Preferences and Strategies - There is an increasing preference for small language models (SLMs) over large language models (LLMs) due to better control and efficiency [3] - Multi-cloud strategies appear to be the preferred approach, with leaders favoring a "best of breed" method [3] - Low-risk and repetitive workloads are identified as the first areas to leverage AI, with significant opportunities for value creation in backend functions [3] Group 4: Survey Insights - 73% of participants believe their organizations are making uneven progress in AI application journeys, with only 18% of systems in production and 9% in early pilot stages [4] - 70% prioritize balancing AI innovation with security, while the rest distribute focus among rapid deployment, risk management, and regulatory compliance [4] - The biggest barrier to creating seamless AI-driven customer experiences is legacy system integration (56%), followed by unclear ROI (33%) and data silos (11%) [4]