Core Viewpoint - Hangzhou has made a significant investment in AI company Zhipu, completing a strategic financing round exceeding 1 billion RMB to promote technological innovation and ecosystem development of its domestic GLM model [1][4]. Group 1: Investment and Company Formation - Zhipu has established Zhejiang Zhipu Huazhang Technology Co., Ltd. in Hangzhou in 2023, and another company, Zhejiang Zhipu Xinpin Technology Co., Ltd., was formed on February 24, 2023, with a registered capital of 450 million RMB [1][4]. - The main investors in Zhipu's recent financing round include Hangzhou Urban Investment Industry Fund and Shangcheng Capital, both of which are state-owned enterprises [4]. Group 2: Zhipu's Position in the AI Industry - Zhipu is recognized as one of the "Six Little Tigers" in AI, being the first among them to surpass a valuation of 20 billion RMB [6]. - Zhipu is the only domestic company that fully benchmarks against OpenAI, with a comprehensive layout across various model types including foundational, dialogue, code, multi-modal, reasoning models, and AI agents [7]. Group 3: Technological Advancements and Commercialization - Zhipu is leading in the development of AI agents, with a focus on Agentic LLMs, which are expected to be a major technological breakthrough by 2025 [7][8]. - The company has established a new ecosystem for model services, supporting over 700,000 enterprises and application developers, and has achieved over 100% growth in commercial revenue for 2024 compared to 2023 [9][10]. - Zhipu's MaaS platform has seen a 30-fold increase in API annual revenue and a 150-fold increase in daily token consumption [10].

Investment Rating - The report maintains an "Overweight" rating for the technology sector and the computer industry [6]. Core Insights - DeepSeek has significantly reduced inference costs, achieving a theoretical daily revenue of $562,027 against a cost of $87,072, indicating a profit margin of 545% if 15% of tokens are paid [2][4]. - The optimization of the DeepSeek-V3/R1 inference system focuses on higher throughput and lower latency through Expert Parallelism (EP) [3]. - The pricing difference in inference services between domestic and international models reflects the constraints in external computing power supply, with DeepSeek offering a more cost-effective solution [4]. Summary by Sections Investment Rating - The report recommends "Buy" for Inspur Information (浪潮信息) with a target price of 61.41 CNY, reflecting a strong growth outlook in the AI server market [9][14]. Cost and Revenue Analysis - DeepSeek's inference system operates at a peak node utilization of 278 nodes, with an average of 226.75 nodes, and a GPU rental cost assumed at $2 per hour [2]. - The average processing cost per million tokens is $0.11, while the R1 model pricing is significantly lower than competitors like OpenAI [2][4]. Technical Optimization - The DeepSeek-V3/R1 system employs a pre-fill and decode architecture to enhance parallel computation across nodes, aiming for reduced latency and improved performance [3]. Market Dynamics - The report highlights that domestic AI model providers are optimizing hardware performance under supply constraints, which may lead to increased market share in global applications [4][5].

Summary of Conference Call Notes Industry Overview - The discussion revolves around the AI computing power industry, specifically focusing on the DeepSeek V3/R1 inference system and its implications for major cloud service providers like Alibaba Cloud, Huawei Cloud, and Tencent Cloud [1][2][3]. Key Points and Arguments - **DeepSeek V3/R1 System Performance**: The DeepSeek V3/R1 inference system has a daily average cost of $87,072, assuming a GPU rental cost of $2 per hour. The theoretical daily revenue can reach $562,027, resulting in an extraordinary profit margin of 545% [1]. - **Engineering Optimization**: The system optimizes limited computing power cards to significantly increase throughput and reduce latency. This includes three major open-source technologies: Expert Parallel (EP), Computation Communication Overlap, and Distributed File System 3FS [1]. - **Impact of Open Source Technologies**: The open-sourcing of these technologies has led to significant breakthroughs in the AI computing power supply chain, allowing large companies to enhance their Return on Invested Capital (ROIC) while enabling smaller teams and startups to access valuable data insights more easily [2]. - **Profitability Assumptions**: The 545% profit margin is based on ideal conditions, including full capacity and a 100% API payment rate, without considering redundancy issues in cloud services. Under a 3:1 redundancy assumption and varying payment rates, cloud providers could achieve profitability ranging from 4.3% to 93% [2]. - **Increased API Call Volume**: The introduction of DeepSeek has led to a 3-4 times increase in token calls on Tencent Cloud, with similar trends observed on Alibaba Cloud, indicating a substantial rise in demand for AI services [3]. - **Demand vs. Efficiency**: The growth in demand for AI services is significantly outpacing the improvements in efficiency, suggesting a robust market for AI computing power in the medium to long term [4]. Additional Important Insights - **Market Dynamics**: The discussion highlights the competitive landscape among major cloud service providers, emphasizing the need for them to increase their payment rates to improve profitability [2]. - **Jevons Paradox**: The anticipated increase in API call volume will likely drive up the demand for chips and servers, aligning with Jevons' Law, which states that as technology improves efficiency, overall consumption can increase [3].

Core Viewpoint - The article discusses the rising significance of "distillation" technology in the AI sector, particularly how companies like OpenAI, Microsoft, and Meta are leveraging it to reduce costs and enhance accessibility to advanced AI capabilities, while also highlighting the competitive threat posed by startups like DeepSeek [1][2]. Group 1: Distillation Technology - Distillation technology allows a large language model (the "teacher model") to generate predictive data, which is then used to train a smaller, more efficient "student model," enabling rapid knowledge transfer [2]. - This technology has recently gained traction, with industry experts believing it will serve as a "cost-reduction and efficiency-enhancement" tool for AI startups, allowing them to build efficient AI applications without relying on extensive computational resources [2][5]. - The operational costs of training and maintaining large models like GPT-4 and Google's Gemini are estimated to be in the hundreds of millions of dollars, making distillation a valuable method for developers and businesses to access core capabilities at a lower cost [2][3]. Group 2: Industry Impact and Competition - Microsoft has implemented this strategy by distilling GPT-4 into a smaller language model, Phi, to facilitate commercialization [3]. - OpenAI is concerned that DeepSeek may be extracting information from its models to train competitive products, which could violate service terms, although DeepSeek has not responded to these allegations [3][7]. - The rise of distillation technology poses challenges to the business models of AI giants, as lower computational costs lead to reduced revenue from distilled models, prompting companies like OpenAI to charge lower fees for their use [6]. Group 3: Performance Trade-offs - While distillation significantly reduces operational costs, it may also lead to a decrease in the model's generalization ability, meaning distilled models might excel in specific tasks but perform poorly in others [5]. - Experts suggest that for many businesses, distilled models are sufficient for everyday applications like customer service chatbots, which can run efficiently on smaller devices [5][6]. Group 4: Open Source and Competitive Landscape - The widespread application of distillation is seen as a victory for open-source AI, allowing developers to innovate freely using open-source systems [7]. - However, the competitive landscape is becoming more complex, as companies can quickly catch up using distillation technology, raising questions about the sustainability of first-mover advantages in the rapidly evolving AI market [8].

Core Insights - The article highlights significant investment and growth opportunities for Chinese companies expanding overseas, particularly in emerging markets and through innovative strategies. Group 1: Investment and Financial Highlights - Mixue Ice Cream's IPO subscription amount exceeded HKD 1.77 trillion, setting a record in the Hong Kong stock market [6] - TikTok plans to invest USD 8.8 billion in Thailand over the next five years [4] - Alibaba announced an investment of over CNY 380 billion in cloud and AI hardware infrastructure over the next three years, marking the largest investment in this sector by a private Chinese company [5] Group 2: Company Expansion and Market Strategies - Stone Technology expects a significant increase in overseas revenue in 2024, driven by optimized sales structures and refined channel layouts [12] - Chery Automobile submitted its IPO application to the Hong Kong Stock Exchange, aiming to expand its product range and enhance its global market presence [8] - Xiaomi plans to increase its R&D investment to CNY 300 billion by 2025, with a focus on AI and related businesses [11] Group 3: Industry Trends and Market Dynamics - The 2024 Chinese mobile phone export volume is projected to grow for the first time in eight years, reaching 814 million units, a 1.5% increase [18] - The global fashion shopping website SHEIN has the highest traffic among clothing and fashion sites, indicating strong consumer interest [5] - The report from SNE Research shows CATL maintaining a 41% market share in the global energy storage market, marking a 5% increase from the previous year [12] Group 4: Strategic Collaborations and Partnerships - Alibaba International Station has partnered with Maersk to enhance logistics for small and medium enterprises, aiming to reduce costs by 10% [5] - Wanglaoji has entered the Saudi market through a partnership with Adook International Holdings to promote its products [12] - Dingdong Maicai has formed a strategic partnership with Lee Kum Kee to develop new products for the Hong Kong market [13]

Market Performance - The main board led the gains with notable increases from companies such as Shida Group (+10.09%), Yanhua Intelligent (+10.01%), and Zhichun Technology (+10.00%) [1] - The ChiNext board saw significant growth with GQY Video (+20.06%), Kaiwang Technology (+20.01%), and Hongjing Technology (+20.01%) leading the charge [1] - The Sci-Tech Innovation board was led by Huicheng Co. (+7.84%), Shihua Technology (+3.08%), and Yongxin Zhicheng (+0.76%) [1] - Active sub-industries included SW Education Publishing (-1.58%) and SW Semiconductor Materials (-2.52%) [1] Domestic News - A partnership was announced between Wuliangcai Glasses of Bailian Group and AR technology company Rokid, unveiling two new AR glasses products [1] - TSMC is considering a strategic investment in Korean chip design startup FuriosaAI, while Meta is reportedly looking to diversify its data center chip portfolio through a potential acquisition of FuriosaAI [1] - Xiamen Silan Jihong Semiconductor's 8-inch silicon carbide (SiC) power device chip manufacturing line project has officially topped out after eight months of construction [1] - DeepSeek revealed core technology details and commercialization data for its DeepSeek-V3/R1 inference system, boasting a theoretical cost profit margin of 545% [1] Company Announcements - ZTE Corporation reported a total operating revenue of 121.299 billion yuan for 2024, a decrease of 2.38% year-on-year, with a net profit attributable to shareholders of 8.425 billion yuan, down 9.66% [2] - Liyang Chip announced that its wholly-owned subsidiary received government subsidies totaling 4.3896 million yuan [2] - Haiguang Information reported a total operating revenue of 9.162 billion yuan for 2024, reflecting a year-on-year growth of 52.4%, with a net profit of 1.931 billion yuan, up 52.87% [2] - Cambrian Technology released its 2024 annual performance report, showing an operating revenue of 1.174 billion yuan, a year-on-year increase of 65.56%, but a net loss of 444 million yuan [2] Overseas News - TrendForce's latest research indicates that global DRAM industry revenue is expected to exceed 28 billion USD in Q4 2024, a quarter-on-quarter increase of 9.9%, driven by rising contract prices for server DDR5 and concentrated shipments of HBM [3] - Micron announced it has begun shipping samples of its 1γ sixth-generation (10nm) DRAM node DDR5 memory designed for next-generation CPUs to ecosystem partners and select customers [3] - Former President Trump proposed a 25% tariff on goods from Mexico and Canada, effective March 4, along with an additional 10% tariff on imports from China [3] - CINNO Research reported that global TFT-LCD and AMOLED panel capacity is projected to reach 409 million square meters in 2024, a year-on-year increase of 2.5%, with further growth of 2.3% expected in 2025 [3]

Core Insights - DeepSeek V3/R1 inference system shows a theoretical daily income of $562,027 against a daily cost of $87,072, resulting in a profit margin of 545% [1] - Actual profit margins are expected to be significantly lower due to factors such as lower pricing for V3, limited monetization of services, and discounts during off-peak hours [2] Profitability Analysis - The theoretical calculations assume full load operation at R1 pricing, but real-world conditions may not allow for such efficiency [3] - Daily average token calls include 6,080 million input tokens and 1,680 million output tokens, leading to a total of 7,760 million tokens called daily [3] - Estimated daily income from V3 is approximately 665,600 yuan, while R1 generates about 1,996,800 yuan, totaling around 2,662,400 yuan in daily API income [3] Technological Advancements - DeepSeek employs a mixture of experts (MoE) model to optimize throughput and reduce latency, utilizing parallel processing across multiple GPUs [5] - The system implements a dual-batch overlapping strategy to minimize communication costs and enhance overall throughput [6] - Load balancing mechanisms are in place to ensure even distribution of computational tasks across GPUs, preventing bottlenecks [7] Infrastructure and Resource Management - A distributed file system (3FS) is utilized for efficient data transfer between computers without CPU intervention, enhancing throughput and reducing latency [8] - The introduction of DualPipe allows for complete overlap of forward and backward computation-communication phases, minimizing pipeline stalls [8] - The use of redundant experts in the expert-parallel load balancer dynamically allocates input to less loaded expert replicas during inference [8] Market Implications - DeepSeek's open-source approach is seen as a significant opportunity for domestic cloud and AI applications, reducing reliance on GPUs and breaking monopolies in the industry [4] - The advancements in DeepSeek's technology are expected to create favorable conditions for large cloud providers and applications in the domestic market [4]

嘉宾 丨 肖朝军、傅天予 整理 丨 程曼祺 上周，DeepSeek、Kimi 都放出了新的大模型架构改进和优化成果，分别是 NSA、MoBA。二者都聚焦对大 模型中 "注意力机制" 的改进。 o 1 、 R 1 等 推 理 模 型 的 出 现，给 了 长 文 本 新 课 题 。 注意力机制是当前大语言模型（LLM）的核心机制。2017 年 6 月那篇开启大语言模型革命的 Transformer 八 子论文，标题就是：Attention Is All You Need（注意力就是你所需要的一切）。 而优化 Attention 的计算效率和效果，又能帮助解决 AI 学界和业界都非常关心的一个问题，就是长文本（long context）。 不管是要一次输入一整本书，让模型能帮我们提炼、理解；还是在生成现在 o1、R1 这类模型需要的长思维 链；又或者是希望模型未来能有越来越长的 "记忆"，这都需要长文本能力的支持。 这期节目我们邀请了两位做过 Attention 机制改进的 AI 研究者做嘉宾。 一位是清华计算机系自然语言处理实验室的博士生肖朝军，他是 InfLLM 注意力机制改进的一作，导师是清华 计算机系副教授 ...

Investment Rating - The report maintains an "Increase" rating for the media sector [6]. Core Viewpoints - The media sector experienced a decline of 8.06% during the week of February 24-28, 2025, influenced by market conditions. The outlook for 2025 is optimistic, focusing on AI applications and mergers and acquisitions, particularly in state-owned enterprises [1][10]. - The release of "Nezha 2" has further boosted the popularity of domestic IPs, highlighting significant opportunities in the IP monetization value chain, including trendy toys and film content [1]. - The publishing and gaming sectors are expected to benefit from tax relief policies, with the publishing industry projected to see high growth in 2025 [1]. Summary by Sections Market Overview - The media sector's performance was notably poor, ranking among the bottom three sectors, with a decline of 8.06% [10]. - The top-performing sectors included steel, building materials, and real estate, while the computer and communication sectors also faced significant declines [10]. Subsector Insights - Key focus areas include: 1. Resource integration expectations: Companies like China Vision Media, Guoxin Culture, and others are highlighted [2]. 2. AI applications: Companies such as Aofei Entertainment and Tom Cat are noted for their potential [2]. 3. Gaming: Strong recommendations for companies like Shenzhou Taiyue and Kaixin Network [2]. 4. State-owned enterprises: Companies like Ciweng Media and Anhui New Media are emphasized [2]. 5. Education: Companies like Xueda Education and Action Education are mentioned [2]. 6. Hong Kong stocks: Notable mentions include Tencent Holdings and Pop Mart [2]. Key Events Review - The release of GPT-4.5 by OpenAI, which boasts over ten times the computational efficiency of GPT-4, is a significant development in AI technology [21]. - DeepSeek's open-source initiatives, including the release of various codebases, are aimed at enhancing data access and model training efficiency [21]. - Alibaba's launch of the video generation model Wan 2.1 showcases advancements in video technology, particularly in generating synchronized movements and text within videos [21]. Subsector Data Tracking - The gaming sector is seeing a variety of new game releases, with popular titles currently available for pre-order [23]. - The domestic film market's total box office for the week was approximately 431 million yuan, with "Nezha: The Devil's Child" leading the box office [24][26]. - The top-rated series and variety shows reflect strong viewer engagement, with "Difficult to Please" and "Mars Intelligence Agency Season 7" leading in viewership [27][28].

Core Insights - DeepSeek claims a theoretical profit margin of 545% based on its online service's "cost-profit ratio" [1] - The company estimates a potential daily revenue of $562,027 if all usage is billed at R1 pricing, although actual revenue is significantly lower due to various factors [2] - DeepSeek's technology has gained attention by outperforming OpenAI's ChatGPT in the Apple App Store, ranking 6th in the productivity category [3] Financial Projections - The estimated daily revenue of DeepSeek is $562,027 based on R1 pricing for its V3 and R1 models [1] - The cost of leasing GPUs is reported to be $87,072, indicating a substantial gap between potential revenue and actual costs [2] - DeepSeek acknowledges that its actual income is "significantly lower" than projected due to discounts and limited commercialization of services [2] Market Position - DeepSeek's new model has been noted for matching OpenAI's performance in certain benchmarks while having a lower development cost [2] - The company faces challenges due to U.S. trade restrictions that limit access to advanced chips for Chinese firms, impacting its market potential [2] - The technology has disrupted traditional players in the AI space, evidenced by its rise in app store rankings [3]