机器之心报道 编辑：Panda、泽南 AI 行业很多人相信，我们正在或已经进入所谓的「AI 下半场」。在这一轮 AI 的浪潮中，硬件的竞争早已不再是单纯的算力比拼，而是一场围绕软件、开发者与 生态的「护城河」之战。当国产 AI 生态的转型成为科技领域的时代呼声，华为昇腾及其异构计算架构 CANN 正站在了这场变革的聚光灯下。 2025 年 8 月 5 日，华为轮值董事长徐直军宣布， 昇腾硬件使能的 CANN 将全面开源 开 放 ，并承诺在 12 月 30 日 前完成。 从「砖」到「大厦」 这并非一次简单的技术路线调整，而是国产 AI 基础设施在生态战略上一次深刻的自我革命。它标志着昇腾正试图打破过去由硬件厂商自上而下主导的封闭模式， 转而拥抱一种更开放、更依赖社区共建、也更不确定的未来。这不仅是对行业领导者 CUDA 所建立的强大壁垒发起的挑战，更是对自身的一次考验：从「可用」 到「首选」的道路，关键不再仅仅是技术的迭代，更在于 能否真正构建起一个让开发者愿意参与进来的繁荣生态 。 为了探寻 CANN 开源的真实意义，以及它将为整个生态带来怎样的机遇与挑战，我们与三位身处不同位置的核心开发者与观察者进行了深度 ...

Core Insights - Huawei unexpectedly revealed its future chip roadmap during the Huawei Connect 2025 event, showcasing several new chips including the Ascend 950, 960, and 970 series for AI computing, as well as the Kunpeng 950 and 960 processors for general computing [1][3][10] Group 1: Chip Developments - The Ascend 950 series chips will support low-precision data formats and achieve computing power of 1P and 2P, enhancing training efficiency and inference throughput [3][10] - The Ascend 960 is planned to double the performance of the Ascend 950 and will support Huawei's self-developed HiF4 data format, set to launch in Q4 2027 [7] - The Ascend 970 will further enhance specifications compared to the Ascend 960, with plans for release in Q4 2028 [7] Group 2: Supernode and Cluster Innovations - Huawei introduced the Atlas 950 supernode, which will consist of 8192 Ascend 950DT chips, achieving FP8 computing power of 8E FLOPS and FP4 computing power of 16E FLOPS, set to launch in Q4 2026 [11][13] - The Atlas 960 supernode, planned for Q4 2027, will be based on 15488 Ascend 960 chips, with FP8 computing power reaching 30E FLOPS and FP4 computing power reaching 60E FLOPS [13] - The Atlas 950 SuperCluster will consist of 64 Atlas 950 supernodes, achieving FP8 computing power of 524 EFLOPS, making it the world's strongest computing cluster [18] Group 3: Software and Ecosystem Development - Huawei aims to develop a robust software ecosystem to complement its hardware, with the CANN deep learning framework and MindSpore framework serving as alternatives to NVIDIA's CUDA [21][22] - The company plans to open-source its CANN compiler and virtual instruction set interface by the end of 2025, along with the Mind series application tools [22][24] - Huawei's strategy emphasizes hardware evolution through existing chip technology while fostering an open-source ecosystem to address challenges posed by U.S. sanctions [24]

大会同步发布龙岗区第二批"城市+AI"应用场景清单，涵盖21个领域共424个场景。龙岗区城投与华为 签署合作协议，共建"在地+云端"算力资源池。 龙岗区发布AI CITY标杆城区解决方案，以昇腾算力为底座，打通政务与公共服务数据，聚焦城市管 理、民生服务与产业赋能。主干道通行效率提升18%，社区医院AI辅助诊断准确率超92%。"4T数字生 活空间"以四大数字权利为核心，为市民提供T级存储、网络、公共资源与智能算力支持，构建"云-网- 边-端"一体化服务底座，实现从"城市治理"向"市民服务"转型。 昇腾超节点突破传统集群通信瓶颈，训练性能达传统节点3倍，支撑千亿级大模型训练。CANN全量开 源，推动AI框架深度适配，算子自动生成技术大幅提升开发效率。未来两年计划培育200万昇腾开发 者，夯实人才基础。 大模型训练性能提升3倍 【深圳商报讯】（记者 马小晗）近日，"昇腾超节点暨CANN生态合作大会"在深圳华为坂田基地召开。 大会以"智算新纪元，生态创未来"为主题，集中发布昇腾超节点算力方案、CANN开源生态成果、AI CITY标杆城区解决方案及"4T数字生活空间"。 ...

人的现实生活如何映射到数字世界？怎样让居民快速享受数字技术红利？拥有私人定制AI助手会给生 活带来什么改变？这些问题，很快将在深圳市龙岗区找到答案。 9月15日，昇腾超节点暨CANN生态合作大会在深圳龙岗华为坂田基地举办。会上，龙岗区政务服务和 数据管理局发布AI CITY标杆城区解决方案，明确以昇腾算力为底座，打通政务数据、公共服务数据与 市民需求的"最后一公里"。此外，"龙岗数字生活空间4T for Home"项目也在本次大会上正式推出。依托 4T底座，龙岗将构建政务办事、数字家园、龙岗家医、虚拟社区、本地生活、信息资讯、学习成长、 数据账户八大场景，全面覆盖数字生活的各个方面。 "数字大脑"升级换代 政务服务更聪明、响应更快 过去，政务云更多解决"存"和"查"的问题；随着技术发展，政务云对"算"和"智"的需求愈发迫切。 目前，龙岗已形成政务AI"1+1+4+N"应用体系，具体包括一个政务智能体百千万共创平台、一个统一的 全国产大模型基座、4个核心领域及N个应用场景。其中，4个核心领域涵盖市民和企业服务、民意速 办、政务办公、城市治理。 借助AI技术，居民诉求响应速度大幅提升——这得益于"一网统管"平台通 ...

Market Overview - US stock indices collectively rose, with the Dow Jones up 0.11%, S&P 500 up 0.47%, and Nasdaq up 0.94%, marking new closing highs for S&P 500 and Nasdaq [1] - Major tech stocks saw gains, with Google rising over 4% to become the fourth US-listed company with a market value exceeding $3 trillion, while Tesla and Oracle increased by over 3% [1] - Chinese concept stocks mostly rose, with the Nasdaq Golden Dragon China Index up 0.87%, and notable gains from Li Auto (over 6%), Bilibili (over 5%), NIO (over 4%), and XPeng (over 2%) [1] Industry Insights - Huawei is set to hold a conference on September 19, 2025, to fully open-source its CANN architecture, aiming to disrupt Nvidia's dominance in the AI development ecosystem [2] - The open-source CANN platform allows developers to freely combine computing modules, promoting the development of domestic AI applications [2] - Nvidia is pushing suppliers to develop MLCP technology, which is 3-5 times the cost of existing cooling solutions, to manage the high power consumption of its next-generation AI platforms [3] - The cooling technology industry is shifting from "technical breakthroughs" to "scale applications," driven by the exponential increase in AI computing power [3] - Hesai Technology announced a laser radar order worth over $40 million with a leading US Robotaxi company, set to be delivered by the end of 2026 [4] - The cost of laser radar has significantly decreased, enhancing the sales volume of related manufacturers and improving profit margins [5] Stock Movements - Several companies announced plans for share reductions by major shareholders, including Hai Li Wind Power, Beiwai Technology, and others, indicating potential shifts in ownership dynamics [6][7] - Shanghai Pharmaceuticals' controlling shareholder plans to increase its stake in the company by acquiring up to 74 million H-shares over the next 12 months [8]

Core Insights - Huawei showcased breakthroughs in general computing, large model technology, and AI applications at the "2025 Huawei Computing Scenario Release Conference" [1] - The launch of the "xPN Pioneer Action" indicates Huawei's commitment to support partners in driving digital transformation across various industries [1] Open Source Strategy - Huawei announced the full open-source of CANN (Compute Architecture for Neural Networks) and Mind series application tools, enabling developers to innovate more effectively [3][4] - The open-source strategy aims to expand the ecosystem and accelerate the AI industry's development by allowing partners to access foundational frameworks for performance tuning and toolchain development [6] Product Development and Ecosystem - Huawei's Ascend components and solutions cover all AI scenarios, with a focus on enhancing product competitiveness and ecosystem richness by 2025 [4] - The number of Ascend partners is expected to exceed 100 by 2025, with over 100 products developed, while the Kunpeng partner program will also see similar growth [8] Market Support and Collaboration - Huawei is investing 8 million to support partners in expanding their business, providing comprehensive assistance in research, marketing, and supply chain management [10] - The company aims for a 350% growth in the xPN partner business by 2025, driven by customer demand and support capabilities [12] Future Outlook - Huawei plans to enhance its product matrix and cover a wide range of computing scenarios, focusing on low to high computing power solutions [12] - The strategy emphasizes collaboration with partners to ensure that the benefits of computing power are accessible across various industries [13]

Core Insights - Huawei showcased its latest achievements in general computing, large model technology, and AI application innovation at the "2025 Huawei Computing Scenario Release Conference and Beijing xPN Partner Conference" [1] - The launch of the "xPN Pioneer Action" aims to support partners comprehensively in research, marketing, supply, and service, focusing on core technologies to create competitive products and solutions for various industries [5] Group 1: Business Performance - In 2024, Huawei's operations remained stable, with a 25% year-on-year growth in the government and enterprise business in China, and an 80% growth in the computing industry [3] - The Kunpeng and Ascend platforms played a crucial role in various fields and accelerated the implementation of AI technology [3] Group 2: Partner Strategy - Huawei plans to increase the number of diamond and silver partners by 25% and 22% respectively in 2024, while launching APN and KPN programs to deepen cooperation [3] - The number of certified APN partners has reached nearly 30, and KPN partners exceed 20, indicating a significant improvement in both quantity and quality of partnerships [3] Group 3: Product Development - Huawei's Ascend division aims to cover all AI scenarios with nine types of component products, enhancing product competitiveness and ecosystem richness [4] - The Ascend hardware and CANN software will work together as dual engines to empower various industries, with a complete open-source version of A2 expected by the end of 2025 [4] Group 4: Innovation and Ecosystem - The Kunpeng platform has seen a tenfold growth in computing power compared to x86, establishing itself as the preferred choice for autonomous innovation [4] - Huawei has collaborated with over 50 partners to develop more than 70 products by 2025, fostering a robust KPN ecosystem and supporting innovation in the computing industry [4]

Investment Rating - The report maintains an "Outperform" rating for the industry [1][4]. Core Insights - The AI sector is witnessing significant advancements, with the release of GPT-5 showing performance improvements, although it did not meet expectations for a generational leap. The overall performance gap with competitors remains limited [3][41]. - Domestic software and hardware are accelerating development, with foundational software growth at 13.8% in the first half of 2025, indicating a recovery trend and potential for further acceleration in the second half of the year [3][19]. - The capital expenditure of major US tech firms reached a record high of $99.973 billion in Q2 2025, reflecting a 77% year-on-year increase, driven by AI business growth [4][62]. Summary by Sections Industry Data - In the first half of 2025, the software industry revenue reached 7.06 trillion yuan, growing by 11.9% year-on-year, with a profit total of 858.1 billion yuan, up 12.0% [13][14]. - The IC design sector showed the highest growth at 18.8%, while cloud and big data services grew by 12.1% [18][19]. AI Developments - The number of AI-related project bids reached 574 in July 2025, with disclosed amounts totaling approximately 1.335 billion yuan, marking year-on-year increases of 422% and 540% respectively [3][79]. - Major models in the first tier include OpenAI's GPT-5, xAI's Grok 4, and Alibaba's Qwen3-235B, with the latter being recognized as the strongest open-source model globally [33][73]. Domestic Market Trends - The domestic AI chip manufacturers are experiencing accelerated growth, with the open-source CANN aiding in the establishment of a robust AI ecosystem [3][19]. - The report highlights the increasing preference for domestic software and hardware solutions amid international uncertainties [4][19]. Capital Expenditure Trends - The capital expenditures of six major US tech firms reached a new high, indicating a strong investment trend in AI and cloud infrastructure [4][62].

Investment Rating - The report maintains an "Overweight" rating for the computer industry, expecting the industry index to outperform the market index by over 10% in the next six months [1]. Core Insights - The release of GPT-5 by OpenAI marks a significant advancement in AI capabilities, particularly in programming and creative writing, which is expected to drive deeper AI applications and increase demand for computing power [2][26]. - The computer sector has shown strong performance, with a cumulative increase of 3.17% over the past two weeks, outperforming the CSI 300 index by 0.77 percentage points, and a year-to-date increase of 16.67%, surpassing the CSI 300 index by 10.61 percentage points [9][19]. - The current price-to-earnings (PE) ratio for the SW computer sector stands at 56.04, placing it in the 91.48% percentile for the past five years and 82.35% for the past ten years, indicating a relatively high valuation [19]. Summary by Sections 1. Industry Performance Review - The SW computer sector has increased by 3.17% in the last two weeks, ranking 13th among 31 primary industries [9]. - Year-to-date, the sector has outperformed the CSI 300 index by 10.61 percentage points [9]. 2. Valuation Situation - As of August 14, 2025, the SW computer sector's PE TTM is 56.04, indicating a high valuation compared to historical data [19]. 3. Industry News - OpenAI launched GPT-5, enhancing capabilities in coding and creative writing, which is expected to facilitate deeper AI applications [20]. - Google DeepMind introduced Genie 3, a model capable of generating interactive 3D environments [20]. - Huawei announced the full open-source of its CANN platform, aiming to accelerate innovation in AI applications [20]. 4. Company Announcements - Hikvision reported a revenue of 41.818 billion yuan for the first half of 2025, a year-on-year increase of 1.48% [23]. - Inspur Information launched a super node AI server capable of running trillion-parameter models, indicating advancements in AI infrastructure [20]. 5. Weekly Perspective - The report emphasizes the potential of GPT-5 to drive AI applications from shallow interactions to deep autonomous decision-making, suggesting a sustained demand for computing power [26]. 6. Recommended Stocks - The report suggests focusing on companies like GuoDianYunTong, ShenZhouData, and Inspur Information, which are well-positioned to benefit from the growing demand for AI and computing power [27].

Core Viewpoint - The traditional Scaling Law for AI models is facing significant bottlenecks, particularly in China, where infrastructure investment is lagging behind the US, leading to challenges in AI inference performance and commercial viability [1][4][9]. Group 1: AI Inference Challenges - AI inference has become a critical area, with current demand for inference computing power exceeding that for training, as evidenced by GPT-5's API call volume exceeding 20 billion calls per minute [4][6]. - Chinese enterprises face a "push not moving," "push slow," and "push expensive" dilemma, with domestic models outputting less than 60 tokens per second compared to over 200 tokens per second for foreign models [7][9]. - The increasing complexity of AI applications, such as long text processing and multi-turn dialogues, has intensified the demand for improved inference performance [1][4][6]. Group 2: Huawei's UCM Technology - Huawei has introduced the Unified Cache Manager (UCM), a breakthrough technology designed to enhance AI inference performance by optimizing memory management and overcoming HBM capacity limitations [1][11]. - UCM employs a tiered caching strategy that allows for the efficient storage and retrieval of KV Cache data, significantly reducing inference latency and costs [10][11][18]. - The technology has demonstrated substantial improvements in inference speed, with a reported 125-fold increase in processing speed for specific applications in collaboration with China UnionPay [19][21]. Group 3: Industry Implications and Future Prospects - The introduction of UCM is seen as a pivotal move for the Chinese AI industry, potentially leading to a positive cycle of user growth, increased investment, and rapid technological iteration [18][24]. - Huawei's open-source approach to UCM aims to foster collaboration within the AI ecosystem, allowing various stakeholders to integrate and enhance their frameworks [28]. - The technology is expected to be applicable across various industries, addressing the challenges posed by the increasing volume of data and the need for efficient inference solutions [23][24].