关于资本市场做好金融"五篇大文章"，证监会出台重磅举措 2月7日，中国证监会发布实施《关于资本市场做好金融"五篇大文章"的实施意见》，提出18条政策举 措。《实施意见》聚焦支持新质生产力发展，突出深化资本市场投融资综合改革，增强制度包容性、适 应性，推动要素资源向科技创新、先进制造、绿色低碳、普惠民生等重大战略、重点领域、薄弱环节集 聚。《实施意见》提到，更大力度支持新产业新业态新技术领域突破关键核心技术的科技型企业发行上 市；扩大保险资金开展长期股票投资试点范围。 证监会新闻发言人周小舟就《关于资本市场做好金融"五篇大文章"的实施意见》答记者问表示，支持优 质科技型企业发行上市，优化科技型上市公司并购重组、股权激励等制度。打通社保、保险、理财等中 长期资金入市卡点堵点，加快推进数字化、智能化赋能资本市场。增强资本市场制度的包容性、适应 性，加大投资端改革力度。 吴清：保护投资者合法权益是证券监管的首要任务 金融监管总局：开展保险资金投资黄金业务试点 国家金融监管总局发布《关于开展保险资金投资黄金业务试点的通知》，自《通知》发布之日起开展保 险资金投资黄金业务试点。在试点内容方面，《通知》界定了试点投资黄金的 ...

中国人工智能（AI）企业深度求索（DeepSeek）发布的开源大模型DeepSeek- R1持续引发海内外热议。据DeepSeek称，该模型榜单测试成绩与美国行业巨头 OpenAI于2024年12月正式上线的o1模型相似，在某些方面甚至表现更好，但是， 其推理定价仅约为后者的五十分之一至二十五分之一。此前，DeepSeek发布的开源 大模型DeepSeek-V3已凭借极低的大模型训练价格受到国内外AI产业界热议。对一 家名不见经传的中国AI企业来说，取得这样的成绩是了不起的。国外同行对 DeepSeek异军突起也不吝赞美之词。在刚刚过去的春节假期里，DeepSeek爆 火"出圈"，不少国人感到欢欣鼓舞。 值得注意的是，从DeepSeek的优异表现中可以看到，在AI大模型开闭源之争 中，开源模型正在赶超闭源模型。图灵奖得主、主导Meta AI研究的首席科学家杨立 2 1 财新观察｜AI竞争以开源促开放 CAIXIN| 听文章 0:00 / 5:36 昆（Yann LeCun）对此解读称："DeepSeek得益于开源研究和开源项目（例如 PyTorch和Llama）。他们在他人工作的基础上提出了新想法并进行了构建 ...

Summary of the Conference Call Transcript Industry Overview - The report focuses on the **Memory Market**, particularly the impact of **DeepSeek's R1 model** on the demand for AI hardware in data centers and the overall memory sector dynamics [2][51]. Key Points and Arguments 1. **Market Dynamics**: - The memory sector has experienced an **11% decline** in average share prices over the last two weeks, compared to a **9% decline** in the SOX index [2]. - Concerns about a slowdown in GPU demand are prevalent, but the report emphasizes the growth potential from **edge-AI migration** and the **AI ASIC market** [2]. 2. **Demand Projections**: - The report anticipates a **25% to 40% adoption rate** for edge AI in PCs and smartphones by 2026, with a potential **220%+ upside** in edge-AI bit demand, translating to a **20% increase** in global DRAM Total Addressable Market (TAM) [2][10]. - The inference demand for HBM is expected to be limited, with more upside potential for conventional DRAM due to the low attach rate of HBM for inference ASIC chips [2]. 3. **Supply and Demand Analysis**: - A **10% decline** in HBM demand could lead to a **2.5% surplus** in ex-HBM bit supply, indicating that the HBM market remains tight despite concerns of oversupply [9][18]. - The report suggests that **20% weaker demand** could result in a significant oversupply scenario, necessitating a reallocation of HBM wafer inputs [18]. 4. **CSP Capex Outlook**: - Major Cloud Service Providers (CSPs) like Microsoft and Alphabet reported strong growth in their datacenter divisions, with capex guidance raised to **US$80 billion** and **US$75 billion** respectively for FY25 [51]. - Despite concerns about AI infrastructure spending, CSPs are expected to continue investing heavily in data clusters [51]. 5. **Competitive Landscape**: - The competition among memory makers is intensifying with the introduction of **GDDR7** and **LPDDR6**, which promise higher speeds and improved efficiency [25]. - Samsung, SK Hynix, and Micron are all expected to begin mass production of GDDR7 memory modules in **1Q25**, highlighting the competitive nature of the market [25]. 6. **Investment Recommendations**: - The report recommends investors to focus on the **NVDA-centric supply chain** and highlights **SK Hynix** as a preferred pick in the Asian memory space [2]. - Investors are advised to consider edge AI proxies and key players in the Apple supply chain for potential growth opportunities [55]. Other Important Insights - The report notes that **ASICs** are expected to become a larger driver of HBM demand in the long term, although short-term visibility remains stronger for merchant GPUs [36]. - The memory sector's performance has been lackluster, with an average decline of **10%** over the past three months, attributed to macro risks and the impact of DeepSeek [51]. - The report anticipates volatility in memory stock prices in the near term as investors adjust their earnings growth expectations [55]. This summary encapsulates the critical insights and projections regarding the memory market and its dynamics, particularly in relation to AI and edge computing trends.

Summary of Key Points from the Conference Call Industry Overview - The focus is on the **China AI ecosystem**, particularly the rise of **DeepSeek** as a significant milestone in AI development, indicating competitive capabilities with global solutions [2][10][11]. Core Insights - **DeepSeek's Development**: It is not an isolated success but part of a broader evolution within China's AI ecosystem, which has been developing over the past two years [1][11]. - **Customization Needs**: DeepSeek requires more customization and higher servicing, making it potentially less scalable outside of China’s unique enterprise landscape [3][12]. - **Shift in Focus**: The China AI ecosystem is transitioning from foundational model development to application-focused development, emphasizing open-source solutions [4][17]. - **Competitive Landscape**: The AI market in China is becoming increasingly competitive, with a rise in cottage industries replicating models at lower costs [4][19]. Winners and Losers - **Winners**: Companies that can effectively monetize through application development and avoid price wars will thrive. Key sectors include cloud providers with full-stack capabilities and enterprise software with strong subscription models [4][18]. - **Losers**: Companies that merely repackage compute resources or model developers without differentiation are expected to struggle [4][18][22]. Stock Recommendations - **Positive Outlook**: - **Kingdee**: Strong mid-enterprise platform with potential for AI monetization [5][24]. - **Kingsoft Office**: Notable for its AI monetization track record across consumer and enterprise products [5][24]. - **TCOM and TME**: Positive setup but with a focus on cost optimization and long-term EPS growth [5][24]. - **Negative Outlook**: - **Baidu**: Expected to lose market share in search and face challenges in AI cloud competitiveness [5][28]. Market Dynamics - **Monetization Challenges**: Not all applications can effectively monetize their AI capabilities due to rapid copying and consumer expectations [5][23]. - **Unique Content Advantage**: Companies with unique content or strong industry verticals are better positioned to leverage AI for differentiation [5][23]. Investment Implications - **Cloud Providers**: Favorable for those with full-stack capabilities that can support open-source development and provide value-added services [5][24]. - **Software Companies**: Subscription-based models targeting consumers and SMEs are likely to see better performance compared to traditional ERP systems [5][24]. - **Model Developers**: The competitive landscape is expected to intensify, leading to pricing pressures and potential margin squeezes for less differentiated players [5][24]. Conclusion - The China AI ecosystem is evolving rapidly, with significant implications for investment strategies. Companies that can adapt to the changing landscape and focus on application development, customization, and unique content will likely emerge as winners in this competitive environment [4][19][24].

Summary of Key Points from the Equity Research Report Industry and Company Overview - **Industry**: AI and Software - **Company**: DeepSeek, a Chinese AI startup known for its reasoning model DeepSeek-R1, which competes with OpenAI's models Core Insights and Arguments 1. **DeepSeek-R1 Performance**: DeepSeek-R1 is a reasoning model that generates a Chain of Thought (CoT) to enhance response accuracy, showing robust performance at a significantly lower cost compared to OpenAI's o1 model, with API output costs being 96% cheaper [3][4][22] 2. **AI Application Monetization**: The introduction of DeepSeek-R1 is expected to accelerate AI application monetization, particularly in logical reasoning, which has been a limitation for previous domestic models. This model's low token costs will reduce AI training and inference costs, enhancing R&D efficiency for software companies [4][24][28] 3. **Focus on AI Agents**: The monetization of AI agents is gaining traction, with several US companies leading the way. Chinese software companies are also launching localized AI agents, which are expected to contribute significantly to revenues in the second half of 2025 [5][29] 4. **Investment Recommendation**: A "Buy" rating is maintained for Sangfor Technologies, which is well-positioned to benefit from the growing AI application market due to its hyper-converged infrastructure business [6][36] Additional Important Insights 1. **DeepSeek's Engineering Innovations**: The model utilizes pure reinforcement learning and incorporates cold-start data, which enhances its reasoning capabilities and reduces training costs to approximately USD 6 million compared to OpenAI's USD 500 million [12][22][21] 2. **Developer Adoption**: Since its release, DeepSeek-R1 has gained significant traction among developers, surpassing Llama 3 in GitHub stars, indicating a growing interest and adoption in the developer community [27] 3. **Market Positioning**: DeepSeek's model addresses critical industry pain points, such as high costs and insufficient reasoning capabilities in existing models, making it more accessible for developers and enhancing the potential for diverse AI applications [24][25][28] 4. **AI Agents in China**: Various Chinese companies, including Kingsoft Office and DBAPP, are integrating DeepSeek's models into their AI products, indicating a trend towards the adoption of advanced AI solutions in the domestic market [31][34] Conclusion The report highlights the significant advancements made by DeepSeek in the AI sector, particularly with the launch of DeepSeek-R1, which is set to transform AI application monetization and enhance the capabilities of software companies in China. The ongoing development and adoption of AI agents further emphasize the potential for growth in this industry.

中新社北京2月8日电 (记者 刘育英)记者8日从中国工信部获悉，中国三家基础电信企业均全面接入 DeepSeek开源大模型，实现在多场景、多产品中应用。 工信部称，2025年春节期间，中国基础电信企业积极利用AI开展创新服务，融合其在5G、云平台、大 数据等方面能力，多领域布局，拓展AI应用场景，丰富AI应用供给。 中国三家基础电信企业均全面接入DeepSeek开源大模型，实现在多场景、多产品中应用，针对热门的 DeepSeek-R1模型提供专属算力方案和配套环境，助力国产大模型性能释放。 工信部称，2025年春节期间，中国基础电信企业积极利用AI开展创新服务，融合其在5G、云平台、大 数据等方面能力，多领域布局，拓展AI应用场景，丰富AI应用供给。 中国三家基础电信企业均全面接入DeepSeek开源大模型，实现在多场景、多产品中应用，针对热门的 DeepSeek-R1模型提供专属算力方案和配套环境，助力国产大模型性能释放。 DeepSeek是杭州人工智能公司深度求索官方推出的AI助手，是继ChatGPT之后的又一现象级人工智能产 品。2月8日，移动互联网商业智能服务平台QuestMobile数据显示，Dee ...

Summary of the Conference Call Company and Industry Overview - The discussion centers around **DeepSake**, a Chinese AI company established in July 2023, focusing on achieving General Artificial Intelligence (AGI) [1][2] - DeepSake's recent developments, particularly the release of its **R1 model**, are compared to leading models from OpenAI, highlighting its competitive edge in performance and cost [2][3] Key Points and Arguments - **Cost Efficiency**: - DeepSake's V3 model training cost is **$557.6 million**, significantly lower than OpenAI's GPT-3, which costs around **$100 million** [2][4] - The R1 model also demonstrates low training costs, with **2.2 USD per million tokens** compared to OpenAI's **60 USD** [4][5] - **Performance Comparison**: - The R1 model matches OpenAI's latest offerings in mathematical reasoning, coding, and natural language processing tasks [3][4] - DeepSake's models are noted for their high performance with significantly lower resource consumption, using **2.8 million GPU hours** compared to competitors' **30.8 million GPU hours** [4][5] - **Market Impact**: - The launch of DeepSake's models has led to a surge in daily active users, exceeding **20 million** within 20 days of the app's launch [3] - The company is gaining traction among both domestic and international tech firms, indicating a shift in the AI landscape [3][6] - **Investment Opportunities**: - Investors are encouraged to explore sectors benefiting from DeepSake's advancements, particularly in **computer software, media, and communication services** [7][8] - The **A-share market** is expected to see significant impacts, especially in B2B applications of AI [7][8] Additional Important Insights - **Open Source Advantage**: - DeepSake's models are open-source, allowing users to leverage and build upon them, which is seen as a competitive advantage over proprietary models [5][6] - **Sector Focus**: - Key sectors to watch include **software development, gaming, digital media, and communication services**, which are expected to benefit from the advancements in AI technology [8][9] - **Investment Strategy**: - Investors are advised to diversify their portfolios and not concentrate too heavily on a single sector, especially given the volatility in tech stocks [10][11] - The importance of understanding the underlying assets in ETFs and mutual funds is emphasized to mitigate risks associated with high volatility in the tech sector [10][11] - **Future Projections**: - The domestic AI smartphone market is projected to reach **118 million units**, with a year-on-year growth of **59.8%**, while global AI PC shipments may reach **114 million units**, growing by **165.5%** [18][19] - **Risks and Considerations**: - Investors should remain cautious of the inherent risks in the tech sector, particularly regarding the maturity of technology and the potential for market fluctuations [24][25] - The need for a balanced approach to investment, considering both growth and value opportunities, is highlighted [28][29] Conclusion - The emergence of DeepSake represents a significant shift in the AI landscape, with implications for investment strategies in the tech sector. Investors are encouraged to stay informed about market trends, sector performance, and the potential for new opportunities arising from advancements in AI technology [32][33]

更多资料加入知识星球：水木调研纪要 关注公众号：水木纪要 对话知行科技：DeepSeek 如何赋能高阶智驾？ 摘要 知行科技的成立背景和主要产品有哪些？ 知行科技成立于 2016 年，并在 2018 年建立了自己的生产基地。公司主要产品 包括自驾一体机和预控制器，提供软硬件一体化交付。早在 2022 年，公司基于 Super Atation 产品实现了极氪 001 和其他吉利高端车型的量产落地。2022 年 至 2023 年间，这些高价值、高阶智驾属性的产品实现了年产 10 万台以上的交 付，带来了良好的现金流。在 2023 年 12 月，知行科技成为首家在港交所上市 的中国大陆自动驾驶公司，并在同年 10 月于欧洲设立研发中心，以服务欧洲高 Q&A • 知行科技 Operation 产品已在极氪和吉利高端车型上量产落地，2022-2023 年交付超 10 万台，带来良好现金流。2023 年 12 月成为首家在港交所上市 的大陆自动驾驶公司，并在欧洲设立研发中心，服务欧洲高端市场。 更多一手调研纪要和海外投行研报数据加V：shuinu9870 更多一手调研纪要和海外投行研报数据加V：shuinu9870 更多 ...

更多资料加入知识星球：水木调研纪要 关注公众号：水木纪要 产业端看智驾：DeepSeek 对智能驾驶产业影响 摘要 • 知行科技专注于量产落地的自动驾驶解决方案，已与吉利、长城等超过 20 家客户达成合作，定点超过 60 万车型，其中 30 多款实现量产，产品线覆 盖软硬一体及仓架一体，全栈自研智能一体机及域控制器。 更多一手调研纪要和海外投行研报数据加V：shuinu9870 更多一手调研纪要和海外投行研报数据加V：shuinu9870 更多一手调研纪要和海外投行研报数据加V：shuinu9870 更多一手调研纪要和海外投行研报数据加V：shuinu9870 更多一手调研纪要和海外投行研报数据加V：shuinu9870 更多一手调研纪要和海外投行研报数据加V：shuinu9870 更多一手调研纪要和海外投行研报数据加V：shuinu9870 更多一手调研纪要和海外投行研报数据加V：shuinu9870 更多一手调研纪要和海外投行研报数据加V：shuinu9870 Q&A 请介绍一下 Deep Six 模型在智能驾驶领域的应用及其影响。 Deep Six 模型相较于传统大模型，采用了强化学习的训练方式，使得模 ...