10月12日，在2025移动合作伙伴大会"算力网络联合创新"分论坛上，中国移动携产业合作伙伴重磅发布GSE 2.5版本协议标准，并同步启动"国芯国连"智算开放互联GSE生态攻坚计划，标志着 随着智能计算规模持续跃升，超万卡乃至十万卡级集群正成为全球大模型竞争的重要基础设施。中国移动积极发挥产业链"链长"作用，以"技术标准+产业生态"双轮驱动，携手产业界共同构建开 本次发布的GSE 2.5版本标准，在GSE 2.0版本基础上实现多项关键技术增强。GSE2.5版本全面支持精简报文设计、RDMA全操作类型乱序接收机制（覆盖Wirte\Read\Atomic\Send），并 作为协同推进GSE技术商用部署的关键举措，"国芯国连"生态攻坚计划同步启动。该计划以汇聚产业力量、构建开放完善的GSE生态体系为目标，联合芯片、设备厂商等产业链伙伴制定了里程碑 中国移动未来将与产业各方持续开展技术攻关、芯片研发、生态共建，共同加速"国芯国连"智算集群的产业化进程，筑牢我国算力基础设施的自主可控根基，全面提升中国智算产业的全球竞争力 ...

今天分享的是：2025智算产业发展研究报告 报告共计：47页 2025智算产业全景扫描：全球竞速、中国突围与未来五大趋势 人工智能正以不可逆转之势重塑全球经济格局，据行业预测，其十年内将推动全球经济规模增长15%，2035年更将为我国GDP 贡献11万亿元。2025年作为AI规模化应用的关键元年，智算产业已成为驱动数字经济增长的核心引擎，从政策布局到技术突 破，从市场需求到区域发展，都呈现出全新的变革态势。 全球范围内，智算产业政策竞赛已全面打响。欧美国家凭借雄厚技术积累，持续加码AI基础设施战略布局。美国启动"星际之 门"项目，计划4年内投资5000亿美元提升AI基础设施能力，同时通过多份政策文件强化AI供应链主权管控；欧盟则推出"投资人 工智能"计划，筹集2000亿欧元用于AI投资，并设立200亿欧元基金建设人工智能超级工厂，力求将人才与产业优势转化为AI领 域的全球领导力。日韩也在加速追赶，日本计划到2030年前投入超650亿美元支持AI与半导体产业，韩国则推进"国家AI计算中 心"建设，目标建成配备3万张高性能GPU、算力达1 EFLOPS的超级计算平台，缩小与头部梯队的差距。 我国智算产业发展则走出 ...

Investment Rating - The report does not explicitly state an investment rating for the industry Core Insights - The rapid development of large model technology is fundamentally transforming the global computing paradigm, with a shift from traditional CPU cloud computing to GPU-centric intelligent computing [3] - The cloud service industry's value chain is being reshaped, with profound changes in the business logic, commercial models, and competitive factors across IaaS, PaaS, and SaaS layers [3] - The industry is transitioning from a "compute-first" approach to building model ecosystems and commercializing applications, with a significant application explosion expected around 2025 [16][19] Summary by Sections Industry Overview - The intelligent computing industry has formed a four-layer structure: hardware, cloud, model, and application, currently transitioning from "compute-first" to "model ecosystem building" and "application commercialization" [10][16] - The industry value transfer trend indicates a shift from heavy investment in infrastructure to a focus on model ecosystem construction and the commercialization of industry-specific agents [19] IaaS Layer Opportunities - The core value of the IaaS layer lies in selling GPU computing power, with high-end card retention and vertical industry capability construction being key to capturing high-value clients [30][33] - Major clients include internet giants, government enterprises, and AI-driven emerging industries, all of which have significant demand for GPU computing power [31] - The growth path for IaaS involves selling bundled solutions of IaaS, PaaS, and MaaS to industry clients, enhancing customer stickiness and increasing average transaction value [36] PaaS Layer Opportunities - MaaS is becoming the core revenue entry point for cloud vendors, with those possessing self-research capabilities needing to adopt a dual strategy of open and closed source to lock in customers [41][42] - The success of MaaS hinges on the ability to provide a rich model marketplace and a comprehensive toolchain for clients, driving resource consumption in IaaS and PaaS layers [42][44] SaaS Layer Opportunities - The SaaS market is shifting towards deep vertical integration, where the ability to build industry-specific products is crucial for customer retention [52][55] - Large cloud vendors face challenges from smaller firms that leverage low prices and localized services to capture market segments that larger companies cannot reach [55] - Effective strategies for entering the SaaS market include "soft-hard integration" and building Agent platforms to transform from direct competitors to value integrators and ecosystem enablers [57][59]

Group 1 - The company, Junxin Co., is focusing on the integration of waste incineration with emerging fields such as digital industry and intelligent manufacturing [2] - A strategic cooperation agreement has been signed with Changsha Digital Group Co., aimed at developing "waste-free cities," environmental digital upgrades, green electricity, and collaborative development in the intelligent computing industry [2] - The company will disclose any significant developments related to this cooperation in accordance with regulations [2]

根据协议，三方将聚焦智算产业核心环节，整合资源优势形成发展合力：扬州大数据集团将依托数据技术与资源 优势，构建智算中心的数据底座与安全屏障；扬州广陵新城投资发展集团将发挥产业载体运营与生态培育优势， 为智算中心提供落地支撑与场景对接，推动智算能力与制造、文旅等本地特色场景快速适配。 其中，作为领先的AI基础设施及智算云提供商，九章云极DataCanvas将充分发挥在普惠算力领域的技术优势，通 过创新研发的九章智算云平台及"按需配置、即时响应"的弹性供给机制，解决传统AI算力成本高企、利用率低、 供需错配等痛点。锚定扬州市特色产业核心需求，九章云极将为普惠智算中心建设提供全维度支撑——覆盖从底 层算力调度到上层大模型微调应用的全栈AI解决方案，同步输出稳定可靠的普惠算力服务能力，为当地"AI+"场 景的深度应用落地筑牢根基。 此次三方合作不仅是政企协同共筑智算生态的典范，更标志着扬州在构建"算力+算法+数据"新型基础设施方面迈 出关键一步。随着普惠算力服务的广泛渗透，将有力支撑扬州市特色产业体系的数字化转型，为长三角数字经济 一体化发展注入新动能，助力打造全国智算赋能产业升级的标杆样板。 近日，由扬州市国资委、 ...

Core Insights - The launch of the OISA 2.0 protocol at the 2025 China Computing Power Conference signifies a significant step forward in China's intelligent computing industry, focusing on technological innovation and ecosystem collaboration [1][3] - The OISA (Omni-Directional Intelligent Sensing Interconnection Architecture) protocol aims to create an efficient, intelligent, flexible, and open AI chip interconnection system, enhancing the performance of super-node computing clusters [3] Industry Developments - The OISA 2.0 protocol supports up to 1024 AI chips, with bandwidth exceeding TB/s and interconnection latency reduced to hundreds of nanoseconds, facilitating data-intensive AI applications such as large model training and inference [3] - The collaboration between China Mobile and partners like Moore Threads emphasizes the importance of domestic GPU innovation and the establishment of an open, collaborative ecosystem in the AI and computing sectors [3] Company Contributions - Moore Threads has actively participated in the formulation of the OISA standard, showcasing its core technological capabilities in AI and computing, and ensuring compatibility with mainstream AI frameworks and applications [3] - As OISA 2.0 is implemented, Moore Threads plans to integrate the latest technological requirements into its product iterations, enhancing product advancement and optimizing architecture to support higher levels of independent innovation and integrated development in China's GPU ecosystem [3]

Core Viewpoint - Guizhou Province, specifically Guiyang, has initiated local legislation in the field of computing power, providing a legal framework for the orderly development of the computing power industry, with the "Guiyang Computing Power Industry Promotion Regulations" taking effect on August 1 [1] Group 1: Legislative Framework - The "Guiyang Computing Power Industry Promotion Regulations" is the first local regulation to define the concept of "computing power," covering the entire value chain including infrastructure construction, software and hardware R&D, technological innovation, service applications, and security [1] - The regulations consist of five chapters and thirty-two articles, mandating that governments at the county level and above take measures to promote the collaborative development of the computing power industry and related sectors [1] Group 2: Industry Development - Guizhou's computing power scale is projected to exceed 55 Eflops in 2024, doubling from 2023, with intelligent computing accounting for over 90% of this capacity [2] - The province has issued 177 computing power vouchers nationwide, successfully surpassing the 10 billion yuan mark in the computing power industry scale [2]

Core Viewpoint - The international academic seminar on high-level multilingual models highlighted the urgent need to address the digital divide affecting low-resource languages, as global mainstream models inadequately support these languages, posing a risk of marginalization for small language countries [1][5]. Group 1: Expert Opinions - Professor Vlado Delić from Serbia emphasized that Serbian language representation in general models is less than 0.1% in token usage, significantly lower than Slovenian, advocating for national-level models that reflect cultural identity to avoid translation errors in critical fields like healthcare and law [3]. - Avner Algom, founder of the Israeli Human Language Technology Association, stated that language services should not be designed solely for major languages, advocating for technological equity for small languages [5]. - Nipat Jongsawat from Thailand highlighted that language sovereignty is a strategic necessity for nations, not merely a choice [5]. Group 2: Technological Developments - Liu Cong, director of iFlytek Research Institute, reported that their multilingual model, Spark X1, expanded from supporting 81 languages in October 2022 to over 130 languages by July 2023, aiming to provide a comprehensive multilingual model and applications [5]. - The Spark X1 model reportedly outperforms GPT-4.1 in key languages such as Arabic, German, French, Korean, and Japanese, while the speech synthesis model supports 55 languages with leading industry performance [5]. - Zhang Xiao, deputy general manager of iFlytek's intelligent computing division, identified challenges in the rapid iteration of computing power and low efficiency in existing resources, proposing a five-element solution combining computing power, algorithms, data, applications, and ecosystem [7]. Group 3: Data Quality and Collaboration - Gábor Prószéky, director of the Hungarian Linguistics Research Center, stressed that data quality is more critical than quantity for building reliable large language models, noting the unique challenges posed by the Hungarian language's complex morphology [7]. - His team has developed the PULI model family, collaborating with Chinese AI counterparts to create a complete closed loop from training to application [7].

Group 1: Core Insights - The event held by DataCanvas in Shanghai showcased the company's latest achievements in AI computing power accessibility, providing practical solutions for the high-quality development of the AI industry [1][2] - Shanghai is actively implementing the national "East Data West Computing" strategy, introducing new industrial policies to support the development of the intelligent computing industry, aiming for a transformation towards self-controllable and green efficiency by 2025 [3][4] - DataCanvas has launched its new product "New Start" at the event, which focuses on commercializing intelligent agent technology and aims to create a comprehensive solution for enterprise-level intelligent agents [4][5] Group 2: Technological Innovations - DataCanvas's Alaya NeW Cloud has successfully overcome performance bottlenecks, achieving "second-level generation of millions of tokens," aimed at providing intelligent computing infrastructure services for global AI innovation enterprises [4][6] - The company has introduced a unique "1-degree computing power" measurement standard to standardize the supply and elastic scheduling of computing resources [4][6] Group 3: Market Strategy and Applications - DataCanvas has tailored computing power and channel policies for the Shanghai market, establishing a service system that supports flexible scheduling from single cards to clusters, addressing the differentiated computing power needs of key sectors like finance and healthcare [6][7] - The collaboration with various industries has demonstrated significant results, such as the partnership with Wuxi Yinlv Shandong AI Technology Co., which has transformed music creation through AI technology, producing millions of works with minimal sample learning [7][8] Group 4: Industry Impact - The establishment and development of DataCanvas in Shanghai not only accelerates the local intelligent computing industry but also provides valuable experience for the national intelligent computing industry construction [7] - The distributed computing network built by Alaya NeW Cloud aims to optimize computing resources across regions, making AI computing power a widely accessible social resource, thus supporting the high-quality development of China's digital economy [7][8]

Core Viewpoint - The 2025 DEMO WORLD Enterprise Innovation and Investment Conference will be held in Shanghai Songjiang, focusing on the integration of technological and industrial innovation, particularly in the intelligent computing service sector [1] Group 1: Event Overview - The conference will feature a special session on "Intelligent Computing Services & Shanghai Songjiang Economic Development Zone," aimed at showcasing the vitality of the intelligent computing industry and fostering cross-industry collaboration [1] - The event is organized by Chuangyebang, in collaboration with various local government entities, to promote innovation and investment opportunities in the region [1] Group 2: Industry Development - The intelligent computing service sector is one of the seven key industries in Songjiang's "2+7" modernization industrial system, with plans to establish a core supply source for intelligent computing in the Yangtze River Delta by around 2030 [1] - Songjiang aims to support the intelligent computing industry through nine initiatives, including financial support of up to 200 million yuan for qualifying enterprises [1] Group 3: Economic Performance - Songjiang Economic Development Zone ranks 16th among 233 national-level development zones in China and 2nd in Shanghai, housing 108 high-tech enterprises and 47 specialized and innovative companies [2] - The intelligent computing capacity in the Songjiang Economic Development Zone reaches 31,300 PetaFLOPS, positioning it among the leaders in the Yangtze River Delta [2] Group 4: Conference Activities - The conference will include presentations on the latest industry policies and investment environments, as well as deep sharing from leading enterprises and their partners [3] - A "New Quality Discussion" segment will feature over 10 intelligent computing service companies sharing their latest technologies and products, followed by a roundtable discussion on investment trends [4]