Core Viewpoint - Despite potential threats, Nvidia's position remains strong in the AI chip market, with significant demand for its products continuing from major tech companies [10]. Group 1: Financial Performance and Market Position - Nvidia is expected to report fourth-quarter revenue of $38.16 billion, with a gross margin exceeding 70%, indicating strong pricing power [2][3]. - The company's earnings per share (EPS) is projected at $0.85, with historical performance showing that Nvidia typically exceeds EPS expectations by 3-5% [4]. - The data center business accounts for over 75% of Nvidia's total sales, making it crucial for the company's growth trajectory [4]. Group 2: Competitive Landscape - The emergence of cost-effective AI training models, such as DeepSeek's R-1, raises concerns about pricing pressure on Nvidia's products [2][9]. - DeepSeek claims to have developed its AI model at a cost of only $5.6 million, which has sparked skepticism regarding the feasibility of such low-cost AI training [5][9]. - Despite the competitive threat posed by DeepSeek, leading tech companies continue to order Nvidia's H20 GPUs, indicating sustained demand [10]. Group 3: Future Outlook - The upcoming earnings report will be critical, particularly the forward guidance for Q1 2025, which will influence market sentiment [4]. - If Nvidia raises its guidance and continues to exceed expectations, the AI-driven growth momentum is likely to persist [4]. - The launch of the H200 GPU in 2025 is expected to further solidify Nvidia's leadership in AI acceleration [10].

Quantitative Models and Construction Methods Model Name: DeepSeek-R1 - **Model Construction Idea**: The DeepSeek-R1 model leverages a mixture of experts (MoE) architecture and dynamic routing technology to reduce inference costs while maintaining high performance[16] - **Model Construction Process**: - **Mixture of Experts (MoE)**: Integrates multiple "expert" models to enhance overall model performance. A gating network determines which expert(s) should handle specific inputs[27] - **Group Relative Policy Optimization (GRPO)**: Eliminates the need for a separate critic model in reinforcement learning, reducing training costs by using group scores to estimate the baseline[31] - **Self-evolution Process**: The model improves its reasoning capabilities through reinforcement learning, exhibiting complex behaviors like reflection and exploration of alternative approaches[39][41] - **Cold Start**: Introduces high-quality long CoT data to stabilize the model during the initial training phase[42] - **Model Evaluation**: The model demonstrates significant cost efficiency and high performance, making it a groundbreaking development in AI applications[16][43] Model Name: DeepSeek-V2 - **Model Construction Idea**: The DeepSeek-V2 model is a powerful MoE language model designed with innovative architectures like Multi-head Latent Attention (MLA)[23] - **Model Construction Process**: - **Multi-head Latent Attention (MLA)**: Improves performance over traditional Multi-head Attention (MHA) by reducing KV cache, enhancing inference efficiency[25] - **Mixture of Experts (MoE)**: Similar to DeepSeek-R1, it uses a gating network to activate specific experts based on input, optimizing resource usage and performance[27] - **Model Evaluation**: The model shows advantages in performance, training cost, and inference efficiency, making it a strong, economical, and efficient language model[23][27] Model Name: DeepSeek-V3 - **Model Construction Idea**: The DeepSeek-V3 model aims to enhance open-source model performance and push towards general artificial intelligence[33] - **Model Construction Process**: - **Multi-Token Prediction (MTP)**: Enhances model performance by predicting multiple future tokens at each position, increasing training signal density[34] - **FP8 Mixed Precision Training**: Improves computational efficiency and reduces memory usage while maintaining model accuracy by using lower precision data types[36] - **Model Evaluation**: The model effectively balances computational efficiency and performance, making it suitable for large-scale model training[33][36] Model Backtesting Results - **DeepSeek-R1**: Demonstrates significant cost efficiency, achieving performance comparable to ChatGPT-01 with much lower training costs[43] - **DeepSeek-V2**: Shows superior performance and efficiency in training and inference compared to traditional models[23][27] - **DeepSeek-V3**: Achieves high computational efficiency and maintains model accuracy, making it effective for large-scale training[33][36] Quantitative Factors and Construction Methods Factor Name: Scaling Laws - **Factor Construction Idea**: Describes the predictable relationship between model performance and the scale of model parameters, training data, and computational resources[21] - **Factor Construction Process**: - **Scaling Laws**: As model parameters, training data, and computational resources increase, model performance improves in a predictable manner[21] - **Data Quality**: High-quality data shifts the optimal allocation strategy towards model expansion[22] - **Factor Evaluation**: Provides a strong guideline for resource planning and model performance optimization[21][22] Factor Backtesting Results - **Scaling Laws**: Demonstrates a predictable improvement in model performance with increased resources, validating the factor's effectiveness in guiding model development[21][22]

Core Viewpoint - DeepSeek's low-cost AI inference model has caused over $1 trillion in market fluctuations globally, outperforming many Western competitors [1][2]. Group 1: DeepSeek's AI Model - DeepSeek is accelerating the launch of its successor to the R1 model, initially planned for May, now aimed for an earlier release without a specific timeline [1]. - The new R2 model is expected to enhance code generation capabilities and expand to more non-English languages [1]. - The R1 model, utilizing relatively weaker Nvidia chips, competes effectively with high-end AI models developed by major US tech companies that have invested billions [1]. Group 2: Industry Impact and Competition - The release of DeepSeek's R2 model may serve as a pivotal moment in the AI industry, potentially prompting global companies to accelerate their own R&D efforts and challenge the current dominance of a few major players [1]. - The U.S. government may express increased concerns regarding the R2 launch, as it could further motivate Chinese companies to enhance their AI strategies [1]. - Numerous Chinese firms have already indicated plans to integrate DeepSeek's models into their products [1]. Group 3: Company Background - Information about DeepSeek is limited, with its founder Liang Wenfeng having gained billionaire status through the quantitative hedge fund Huanfang Quant [2]. - The company is characterized more as a research laboratory than a traditional profit-driven enterprise, as indicated by insights from former employees and industry professionals [2].

Core Viewpoint - The overall sentiment for 2025 is optimistic, with expectations of significant economic improvement compared to 2024, driven by clear policy shifts and a recovering real estate market [2][67]. Group 1: Economic Outlook - The economic "feel" in 2025 is expected to be better than in 2024, with a clear turning point in policies [67]. - The real estate sector is anticipated to stabilize, particularly in first-tier cities, which are closely tied to market confidence and asset values [16][69]. - The real estate market has seen a significant reduction in inventory, with over 7.5 billion square meters removed in the past three years [15][14]. Group 2: Market Trends - The investment landscape is shifting, with a focus on small-cap technology growth stocks, which are expected to gain traction as market conditions stabilize [18][22]. - Dividend investments are seen as having absolute returns but lacking relative returns in a more optimistic market environment [25]. - The semiconductor industry is viewed as a high-risk area due to high valuations and significant stock price volatility [82][83]. Group 3: Case Studies - The case of "胖东来" (Pang Dong Lai) illustrates a customer-centric business model that emphasizes trust and collaboration with suppliers and employees, which is seen as a sustainable competitive advantage [6][30]. - The analysis of the U.S. internet development provides insights into the current AI landscape, questioning whether companies like NVIDIA will face challenges similar to those experienced by Cisco in the past [8][46]. Group 4: Sector Analysis - The public utility sector is undergoing significant changes due to the rise of renewable energy, which is expected to alter profit models and market positions [91]. - The consumer sector remains challenging, with a need for substantial policy support to revitalize growth [93][95]. - The Hong Kong market is being approached with caution, focusing on high-quality growth stocks that can withstand international scrutiny [102][104].

Core Insights - Cohere has experienced significant growth, with annualized revenue reaching $70 million, more than tripling since March of the previous year [1] - The company is considering allowing the sale of employee shares, which may provide investors insight into its valuation in the competitive enterprise AI software market [2] - Following the potential share sale, a major Series E funding round is anticipated by some investors [2]

Core Viewpoint - Nvidia's stock price surge, which once made it the highest-valued company globally, has stagnated as investors become cautious about further investments, recognizing that the adoption of AI computing will not be a straightforward path and will not solely depend on Nvidia's technology [1]. Group 1: Nvidia's Growth Factors and Challenges - Nvidia's most profitable product is the Hopper H100, an enhanced version of its graphics processing unit (GPU), which is set to be replaced by the Blackwell series [3]. - The Blackwell design is reported to be 2.5 times more effective in training AI compared to Hopper, featuring a high number of transistors that cannot be produced as a single unit using traditional methods [4]. - Nvidia has historically invested in the market since its founding in 1993, betting on the capability of its chips to be valuable beyond gaming applications [3][4]. Group 2: Nvidia's Market Position - Nvidia currently controls approximately 90% of the data center GPU market, with competitors like Amazon, Google Cloud, and Microsoft attempting to develop their own chips [7]. - Despite efforts from competitors, such as AMD and Intel, to develop their own chips, these attempts have not significantly weakened Nvidia's dominance [8]. - AMD's new chip is expected to improve sales by 35 times compared to its previous generation, but Nvidia's annual sales in this category exceed $100 billion, highlighting its market strength [12]. Group 3: AI Chip Demand and Future Outlook - Nvidia's CEO has indicated that the company's order volume exceeds its production capacity, with major companies like Microsoft, Amazon, Meta, and Google planning to invest billions in AI and AI-supporting data centers [10]. - Concerns have arisen regarding the sustainability of the AI data center boom, with reports suggesting that Microsoft has canceled some data center capacity leases, raising questions about whether it has overestimated its AI computing needs [10]. - Nvidia's chips are expected to remain crucial even as AI model construction methods evolve, as they require substantial Nvidia GPUs and high-performance networks [12]. Group 4: Competitive Landscape - Intel has struggled to gain traction in the cloud-based AI data center market, with its Falcon Shores chip failing to receive positive feedback from potential customers [13]. - Nvidia's competitive advantage lies not only in hardware performance but also in its CUDA programming language, which allows for efficient programming of GPUs for AI applications [13].

Group 1 - Tesla's stock price dropped over 8%, resulting in a market value loss of approximately $89.2 billion, bringing its total market value below $1 trillion [2][3][4] - Major technology stocks experienced declines, with Nvidia and Google down over 2%, and Microsoft and Meta down over 1% [4] - Chinese concept stocks mostly rose, with the Nasdaq China Golden Dragon Index increasing by 0.58%, driven by significant gains in Li Auto and XPeng [4] Group 2 - Baidu announced a $2.1 billion acquisition of YY Live from JOYY, with plans to invest the returned funds into cloud and AI infrastructure [7][8] - WeChat's Windows 4.0.2 test version now supports receiving red envelopes on PC, although sending red envelopes from PC is not yet available [9][11] - DeepSeek reopened its API recharge service, with updated pricing for token inputs and outputs, following a previous halt due to server resource constraints [12][14]

定量研究 | 定量研究专题报告 证券研究报告 报告日期 2025 年 02 月 23 日 分析师：郑兆磊 S0190520080006 zhengzhaolei@xyzq.com.cn 相关研究 【兴证金工】布局" 硬科技"突围新思路， 科创板全能选手来袭—鹏华上证科创板 综合 ETF 盛大发行中-2025.02.19 【兴证金工】浙江国资创新崛起，杭州概 念领跑市场—浙江国资 ETF 投资价值分 析-2025.02.13 【兴证金工】政策春风助力，把握科创板 块投资机遇-2025.02.12 政策支持不断落地，科技东风带来民企经济新启程— —弘毅远方国证民企领先 100ETF 投资价值分析 投资要点： 风险提示：基金投资有风险，本报告不代表投资建议；基金经理历史业绩不代表未来， 请投资者知悉。 | 一、 | 支持政策不断落地，科技东风带来民企经济新启程 | 4 | | --- | --- | --- | | （一） | 民营企业是中国经济的重要支撑 | 4 | | （二） | 支持政策不断落地，民营企业迎来历史机遇期 | 5 | | （三） | 民营企业在中美科技竞争前沿一线的重要性再度凸显 | 6 | | ...

全面发力。 作者 I 周佳丽 报道 I 投资界PEdaily 将出台促进风投创投高质量发展政策文件； 不辩不解，只有行动。 杭州爆红，放大了深圳的焦虑 2 0 0 6年，26岁的杭州人汪滔从香港科技大学毕业，思前想后回到深圳，带着一个稍显稚 嫩的学生团队开始创业。 十年后，大疆开始称霸全球无人机，当时汪滔曾动情做过总结："我常常想，这一群初出 茅庐的年轻人，不用去阿谀奉承、投机取巧，就可以在踏实做事的埋头苦干当中，达到创 业之巅，这样的故事恐怕只有深圳才可以实现。" 二十年后，伴随De e pSe e k、宇树科技意外爆红，"杭州六小龙"令各地省市掀起一场反思 潮。这当中，"大疆之后再无深圳创新"的声音尤为刺耳。 经历了一个多月的批判，深圳没有更多解释，而是拿出行动——刚刚过去的周日，深圳市 政府新闻办召开"打造最好科技创新生态和人才发展环境"主题发布会上，宣布了一系列 魄力十足的举措： 大学毕业生来深圳找工作可免费住15天，青年创业最高1 0 0万资助； 1 9 7 9年，珠江口咸腥的海风中，宝安县渔民撒下最后一网时，没有人会想到这片荒芜的 滩涂即将迎来一场人类历史上最剧烈的城市嬗变。 那是属于深圳的繁花时 ...

划重点 ① DeepSeek 宣称以 GPT-4十分之一的成本训练模型，引发算力需求崩溃担忧。但多模态模型、 实时推理等"水下需求"逐渐凸显，加之模型迭代加速，反而推高市场对算力的长期需求。 ② 除了历次财报给出的爆炸性数据，英伟达真正的护城河藏在 2.8 亿行 CUDA 代码构筑的生态 壁垒中，它所具备的软件栈优势已形成行业标准，迁移成本极高。即便竞争对手硬件性能接 近，软件生态差距仍难以弥合。 ③ 当前数据证明，尽管 DeepSeek 的突破意义重大，但没有一家科技巨头削减计算和数据中心 方面的资本支出。即将发布的 2024Q4 财报可能再次提振投资者情绪，预计英伟达 2025 营收依 然极其乐观，且 3 月 17 日的 GTC 大会（重点转向 GB300、Rubin 以及机器人等实体 AI 项目） 还会披露新看点。 RockFlow 本文共3875字， 阅读需约15分钟 2025 年 1 月 27 日，美股见证了历史性一幕：英伟达单日暴跌 17%，市值蒸发 5900 亿美元，创美股史上最大单日市值损失纪录。 这场地震的震源，来自一家中国 AI 公司 DeepSeek——它宣称以 GPT-4 十分之一的成 ...