他把当前在未来十年的阶段性，形容为"即将进入到人工智能革命高潮的前夜"。 在中关村举办的2025人工智能+大会，中国AI未来十年的关键"进度条"正在变得清晰。 大会间隙，人工智能百人会高级顾问——智谱董事长刘德兵与面壁智能联合创始人兼首席科学家、清华大学副教授刘知远接受了智东西的独家 采访。两位长期深耕一线的实践者，从基础模型到智能体演进，分享了他们对未来十年的观察与思考。 在谈到基础模型竞争时，刘德兵并不回避现实：在开源成为主流、结果可公开验证的当下，模型能力的差距会被迅速放大——"在一线开源模 型做到90分的情况下，再训一个85分的模型就没多少竞争力。" 他同时强调，坚持做难而正确的事情很重要，哪怕投入巨大，因为"基础模型决定了整个AI产业发展的上限"。他认为，未来的关键变量将更 多来自开源生态的成熟、行业场景的深度落地，以及AI逐渐成为"全民能力"所带来的广泛参与。 在刘知远看来，2025年的一个显著拐点是"AI+编程"，这一能力正在成为软件生产力的重要支撑。 对于大模型如何迈向智能体，他强调的不是堆叠更多知识，而是让模型具备"在指定工作岗位上自主学习的成长能力"，像大学毕业生一样，通 过真实任务的反馈 ...

证 券 研 究 报 告 计算机行业深度研究报告 推荐（维持） 海外 AI 应用：从大模型到各领域落地 行业研究 计算机 2025 年 11 月 18 日 华创证券研究所 证券分析师：吴鸣远 邮箱：wumingyuan@hcyjs.com 执业编号：S0360523040001 联系人：杨玖祎 邮箱：yangjiuyi@hcyjs.com 行业基本数据 | | | 占比% | | --- | --- | --- | | 股票家数(只) | 338 | 0.04 | | 总市值(亿元) | 60,550.68 | 4.98 | | 流通市值(亿元) | 54,914.11 | 5.60 | 相对指数表现 | % | 1M | 6M | 12M | | --- | --- | --- | --- | | 绝对表现 | -5.9% | 27.2% | 31.4% | | 相对表现 | -7.4% | 5.8% | 15.4% | -10% 7% 23% 39% 24/11 25/01 25/04 25/06 25/09 25/11 2024-11-18~2025-11-14 计算机 沪深300 相关研究报告 《计算机 ...

Group 1 - The article emphasizes the importance of selecting valuable research reports and highlights the need for timely insights to avoid missing investment opportunities [1] - The focus is on the recovery of the lithium battery industry chain, with a specific mention of the stock price increase of approximately 30% for Jia Yuan Technology since September 12 [1] Group 2 - A rare investment opportunity in the AI and programming sector is identified, with projected profits of 300 million yuan by 2027 and an expected increase in net profit of up to 185.84% by 2025 [2] - The Chinese government is promoting the integration of "vehicle-road-cloud" systems, with pilot projects in 20 cities and significant funding for roadside equipment and operational services, indicating a broad long-term market potential [2]

Investment Rating - The report assigns an investment rating of "Buy (First Coverage)" with a current price of 64.70 CNY and a target price of 101.82 CNY over a 12-month period [1][5]. Core Insights - The report highlights that the IDE (Integrated Development Environment) business is expected to be the core growth driver for the company, with significant revenue growth projected from 2025 to 2027 [5][33]. - The company is strategically positioned in the AI and programming sector, leveraging AI technologies to enhance its IDE offerings [5][27]. - The global software development tools market is projected to grow from 6.517 billion USD in 2025 to 14.680 billion USD by 2031, with a CAGR of 14.5% [5][28]. Company Overview - The report details that Jiangsu Zhaoyi Information Technology Co., Ltd. was established in May 2008 and went public on the Shanghai Stock Exchange in December 2019 [13][19]. - The company’s main business segments include IDE business, core firmware for cloud computing devices, and cloud services [13][33]. Financial Performance - The company’s revenue is projected to grow from 414.33 million CNY in 2025 to 827.17 million CNY in 2027, with corresponding net profits expected to rise from 93.87 million CNY to 289.99 million CNY during the same period [6][5]. - The report provides a detailed financial forecast, indicating a significant increase in both revenue and net profit over the next three years [6][5]. Business Analysis - The IDE business is identified as a key area for future growth, with products like SnapDevelop and EazyDevelop being central to the company's strategy [5][36]. - The company has a dual strategy of "IDE+AI" and "AI+IDE," focusing on enhancing development efficiency through AI integration [5][36]. - The report notes that the company has a strong management team with a stable background, which is crucial for executing its growth strategy [5][23]. Industry Analysis - The IDE industry is undergoing significant transformation due to advancements in AI, cloud computing, and low-code technologies, which are expected to drive future growth [27][31]. - The report emphasizes the scarcity of domestic companies with mature IDE development capabilities, positioning the company as a rare player in the market [27][28].

Core Insights - The article discusses the tumultuous events surrounding Prem Qu Nair, a key employee at Windsurf, following the company's failed acquisition by OpenAI and the subsequent fallout with its founders [1][12]. Group 1: Company Events - Windsurf's former CEO and founder Varun Mohan fled, leading to instability within the company and the failure of the acquisition by OpenAI, which was valued at $30 billion [5][10]. - Following the acquisition failure, Google acquired a portion of Windsurf's core team for $24 billion, leaving many employees abandoned by the founders [12][18]. - Cognition, a former competitor, later acquired the remaining parts of Windsurf, ensuring that all employees' rights were protected [15][16]. Group 2: Employee Decisions - Prem Qu Nair, as the second employee of Windsurf, initially received a job offer from Google that was valid for less than a day, which required him to forfeit all his existing shares [20][23]. - Ultimately, Nair chose to join Cognition instead of accepting Google's offer, expressing excitement about returning to a passionate work environment [21][24]. - Nair's decision resulted in a significant financial loss, as he only received 1% of the value of his shares accumulated over three and a half years at Windsurf [23][24]. Group 3: Industry Context - The article highlights a broader trend in Silicon Valley, where tech giants are aggressively competing for talent, offering exorbitant salaries to attract employees from startups [24]. - Meta's aggressive recruitment strategy is noted, with reports of offers reaching up to $100 million for AI researchers, indicating a fierce talent war in the industry [25][28].

Core Insights - The article discusses the tumultuous events surrounding the AI startup Windsurf, including the departure of its CEO and core team, leading to significant financial losses for remaining employees [2][4][11] - It highlights the competitive landscape in Silicon Valley, where tech giants like Google and Meta are aggressively recruiting talent from startups, often with lucrative offers [20][23][24] Group 1: Windsurf's Situation - Prem Qu Nair, the second employee of Windsurf, lost nearly all his equity when the CEO and core team left, resulting in him receiving only 1% of the original value of his shares [3][16] - The acquisition talks between OpenAI and Windsurf for $3 billion fell through, leading to the departure of key personnel to Google, which subsequently acquired the team for $2.4 billion [9][11] - Following the turmoil, Cognition, a former competitor, acquired the remaining parts of Windsurf, ensuring that employee rights were protected [14][15] Group 2: Talent Wars in Silicon Valley - Google made aggressive recruitment moves, including offering "same-day expiration" job offers to attract talent, which reflects the high-stakes competition for skilled workers in the AI sector [21][23] - Meta has been reported to offer up to $100 million in compensation packages to lure AI researchers from OpenAI, indicating the fierce competition among tech giants [24][25] - The talent acquisition battle began in June, with Meta acquiring a significant stake in Scale AI, further intensifying the competition for AI talent [27]

Market Overview - On July 16, AI application and programming concept stocks experienced a collective surge, with companies like Puyuan Information and Tiandi Online hitting the daily limit, and others like Baoland and Foxit Software seeing significant gains [1]. Events - Domestic AI company MiniMax has submitted an IPO application in Hong Kong [2]. - Elon Musk's AI company launched a "companion" feature based on the Grok 4 model, available to SuperGrok subscribers for $30 per month [2]. - Amazon's AWS department officially launched an AI programming tool named Kiro on July 15 [2]. - Google announced the acquisition of the founders and core employees of AI programming startup Windsurf for $2.4 billion on July 11 [2]. Institutional Insights - In June, MiniMax launched several AI products, including new reasoning models and video generation models, with the MiniMax-M1 model being open-sourced on the first day of the launch week [3]. - The AI companion market is projected to grow significantly, with ARK Invest estimating a market expansion of 2300-5000 times, increasing from $30 million to between $70 billion and $150 billion by 2025, with a compound annual growth rate (CAGR) of 200%-236% [3]. - The AI programming tools market is expected to grow from $6.21 billion in 2024 to $18.2 billion by 2029, with a CAGR of 24.0% [3]. - AI programming can reduce development time by 5-10 times and lower development costs to 10% of current levels, leading to structural changes in the software industry [3]. - Integrated Development Environments (IDEs) are becoming increasingly important as the core entry point for AI programming [3]. Additional Insights - IDEs and AI programming tools are complementary and promote mutual development [4]. - AI programming needs to integrate better with the overall programming environment to maximize its effectiveness, with IDEs being essential [4]. - The trend of democratizing large models is pushing IDE vendors to integrate with AI, solidifying their core position in the programming field [4]. - The AI application sector is on the verge of rapid growth, with significant cost reductions and increasing penetration rates, indicating a blue ocean market with vast opportunities [4]. - The application layer is expected to dominate revenue and cash flow in the AI sector, similar to traditional applications [4].

Investment Rating - The report assigns an "Outperform" rating to Xiaomi Group, Lenovo Group, BYD Electronics, and Sunny Optical, while a "Neutral" rating is given to SMIC and ASMPT [1]. Core Insights - Generative AI is expected to accelerate a software productivity revolution, pushing the global software market to exceed $2 trillion by 2030, with enterprise software and cloud services projected to grow at CAGRs of 12.3% and 20.7% respectively [4][12]. - The AI + Coding sector is becoming a key area for generative AI applications, with major players including OpenAI Codex and GitHub Copilot leading the market [5][32]. - AI is transforming software development processes, enhancing efficiency and reducing development costs significantly, with expectations of a 90% reduction in project development cycles and labor costs [22][38]. Market Overview and Growth Drivers - The global software market is projected to grow from approximately $737 billion in 2024 to over $2 trillion by 2030, driven by digital transformation, remote work, and cloud computing [17][48]. - North America holds the largest market share at 44%, followed by Europe at 28% and Asia-Pacific at 22% [17][48]. - The AI programming tools market is expected to grow from $6.21 billion in 2024 to $18.2 billion by 2029, with a CAGR of 24% [20]. Technological Breakthroughs and Applications - AI + Coding tools are evolving from simple code completion to full-process development assistants, significantly improving development efficiency [12][29]. - The core technology relies on large language models (LLMs) that understand and generate code, with applications in code generation, testing automation, and project management [24][32]. - The introduction of generative AI is expected to democratize software development, allowing non-professionals to participate through low-code/no-code platforms [32][52]. Hardware Foundation and Cost Structure - The cost structure of AI programming tools is primarily driven by hardware costs, which account for about 70% of total costs, with significant reductions in inference costs due to model optimization [34][37]. - Innovations in CPU and GPU technologies are reshaping the computational ecosystem, with RISC-V architecture gaining traction for its flexibility [37][38]. Industry Impact and Future Outlook - AI + Coding is anticipated to reduce development time by 5-10 times and lower enterprise development costs to 10% of current levels, fundamentally reshaping the software industry ecosystem [38][40]. - By 2028, it is predicted that 75% of enterprise software engineers will use AI programming tools regularly, a significant increase from less than 10% in early 2023 [38][40]. Investment Opportunities and Strategic Recommendations - Three main investment themes are identified: platform-based AI programming tools and IDE providers, vertical industry AI solution providers, and IT outsourcing firms transitioning to AI-driven models [40][44]. - Companies are encouraged to strategically adopt AI programming tools, restructure development processes, and establish governance frameworks for AI-generated code [44].

Core Viewpoint - The introduction of DeepSeek V3 0324 significantly enhances programming capabilities, making AI programming accessible to the general public, thus promoting technological equality [1][2][3]. Group 1: AI Programming Accessibility - AI programming has not been widely adopted among novice users due to previous limitations, but with DeepSeek V3 0324, ordinary individuals can easily leverage AI programming [2][6]. - The ability to program using AI is crucial for enhancing the general public's AI skills, as programming is currently the most effective application of AI as a productivity tool [3][7]. Group 2: Practical Applications of AI Programming - The transition from searching for tools to creating tools signifies a paradigm shift in problem-solving, with AI programming enabling users to generate solutions directly [11]. - Examples of practical applications include creating visual content for social media, replacing traditional presentation tools like PPT with visual web pages, and generating personal websites or resumes in HTML format [28][42][43]. Group 3: Installation and Usage of Trae - The article outlines the installation process for Trae, including downloading the software, initializing settings, and switching to the latest V3 0324 model for optimal performance [13][14][20]. - Users are guided on how to utilize AI programming to enhance efficiency in daily tasks, such as creating visually appealing web pages and automating design processes [23][29]. Group 4: Limitations and Future Prospects - Despite the strong front-end capabilities of DeepSeek V3, there are notable limitations, including issues with backend functionality and occasional system failures [67]. - Anticipation exists for future versions of DeepSeek that may address current shortcomings, suggesting a positive outlook for the evolution of AI programming tools [68].