2025年算力调度平台行业 优化计算资源，支撑AI应用 ◼ 研究背景 随着人工智能技术的迅猛发展，全球 范围内对算力的需求呈现指数级增长， 因此需要算力调度来实现跨地域、跨 平台的算力资源整合与优化调度。 ◼ 研究目标 概览标签：算力、算力调度 Computing Power Scheduling Platform Industry 計算力スケジューリングプラットフォーム業界 报告提供的任何内容（包括但不限于数据、文本、图表、图像等）均系头豹研究院独有的高度机密性文件（在报告中另 行标明出处者除外）。 ，任何人不得以任何方式擅自复制、再造、传播、出版、引用、改 编、汇编本报告内容，若有违反上述约定的行为发生，头豹研究院保留采取法律措施、追究相关人员责任的权利。头豹 研究院开展的所有商业活动均使用"头豹研究院"或"头豹"的商号、商标，头豹研究院无任何前述名称之外的其他分支机构 ，也未授权或聘用其他任何第三方代表头豹研究院开展商业活动。 头豹研究院 1 01 异构算力调度面临的挑战： ◆ 异构算力调度面临多重核心挑战：资源异 构性与软件环境碎片化显著增加调度复杂 性；跨架构任务迁移成本高导致效率低下； 缺乏统一调 ...

Group 1 - The KubeCon+CloudNativeCon China 2025 conference was held in Hong Kong, focusing on the integration of cloud-native technologies and AI, featuring global open-source developers and industry leaders [1] - Huawei participated significantly in the conference, delivering three keynote speeches and over ten technical sessions, showcasing breakthroughs in the field of cloud-native and AI integration [1][3] - Open-source is recognized as a key driver for innovation, enabling faster development cycles and creating exponential value for the next technological revolution [1] Group 2 - Huawei's Chief Open Source Liaison Officer, Ren Xudong, discussed the evolution of computing clusters in the AI era, emphasizing the need for unified management of computing resources to support major AI frameworks [3] - Bilibili, with over 100 million monthly active users, has developed a cloud-native AI platform using Volcano and Karmada to optimize workload scheduling and enhance resource utilization [4][6] - iFlytek achieved significant improvements in AI model training efficiency by utilizing Volcano, increasing GPU utilization by over 40% and reducing fault recovery time by 70% [7][9] Group 3 - Karmada is highlighted as a multi-cloud management engine that addresses challenges in deploying AI tasks across multiple clusters, enhancing the reliability of large-scale data platforms [10][12] - KubeEdge, as a graduate project of CNCF, is recognized for its capabilities in edge computing, providing unified management and real-time advantages in various industry applications [14] - Kmesh, a cloud-native traffic management engine, utilizes eBPF technology to improve performance and reliability in handling large-scale traffic applications [16] Group 4 - openGemini, a high-performance time-series database, is designed for various industries, including IoT and aerospace, and aims to reduce business costs through efficient data storage and querying [18] - openFuyao focuses on building a diverse computing ecosystem to address the complexities of AI workloads, promoting effective resource management and scheduling [20] - Huawei continues to collaborate with global enterprises and developers to drive innovation in AI and cloud-native technologies, contributing to the intelligent transformation of various industries [20]

Core Insights - The KADC2025 event highlighted the growth of the Kunpeng and Ascend computing ecosystems, with over 6.65 million developers and more than 8,800 partners contributing to 23,900 solution certifications as of May 2025 [5][9] - The event focused on deep technical discussions and hands-on coding experiences, aiming to enhance developers' understanding and mastery of Kunpeng and Ascend tools to accelerate AI innovation [6][7] - Huawei is committed to building a robust and user-friendly computing foundation to support developers and partners in the AI era, emphasizing the importance of ease of use, flexibility, and scalability in AI platforms and tools [7][9] Event Highlights - KADC2025 featured two summits, over 40 technical forums, and a 2,000 square meter exhibition area showcasing collaborative innovations and industry applications [6] - The event included a series of new technologies, tools, and platforms, such as the Kunpeng AI+ solution and the openFuyao software community, aimed at enhancing AI application development [7] - The event also showcased practical applications of AI in various industries, including manufacturing and education, demonstrating significant efficiency improvements and cost reductions [10][11] Industry Trends - According to Gartner, by 2027, 80% of Chinese enterprises are expected to deploy multi-model generative AI locally, and by 2029, 60% will integrate AI into products and services, driving revenue growth [9] - The integration of AI into manufacturing processes is accelerating, with companies like Great Wall Motors demonstrating advanced AI applications in production efficiency [11] Developer Engagement - Huawei plans to implement incentive programs to stimulate innovation across the entire technology development and commercial application chain, focusing on talent cultivation through various initiatives [12][13] - The Ascend AI Innovation Competition has successfully engaged thousands of developers globally, fostering creative solutions that drive industry transformation [13]

大会期间，华为正式发布多样化算力集群软件开源社区 openFuyao，华为、华鲲振宇、某国有大行、中国移动、联通数科、博云科技等六家单位作为筹备 委员会成员出席发布仪式。华鲲振宇副总裁赵彦钧入选筹备委员会成员，AI 开发部部长徐洋入选技术委员会，深度参与社区长期发展与技术路线规划。 5月23日，以 "心怀挚爱，共绽光芒" 为主题的鲲鹏昇腾开发者大会 2025 在北京中关村国际创新中心成功举办。华鲲振宇作为鲲鹏&昇腾生态唯一战略级 伙伴，向开发者全景呈现AI全栈解决方案，携手华为、伙伴及开发者共推国产化算力自主创新、共建多样化算力集群软件生态。 自研元启AI开发平台开放体验 在展区内，华鲲振宇独立展台开放AI全栈解决方案体验，开发者现场深度体验自研元启AI开发赋能平台，通过零代码构建智能体、自动化生成知识图 谱、图形化流程编排等功能，开发者可以直观感受像搭积木一样搭建个人/企业智能体，有效解决AI应用开发门槛高的痛点，助力企业让大模型真正运行 起来。 作为鲲鹏唯一战略级整机合作伙伴，华鲲振宇算力创新产品——"卡多多"AT9508 G3推理服务器受邀亮相鲲鹏展区，基于鲲鹏创新架构，支持生态内最高 规格的10卡双宽 ...

Investment Rating - The report maintains a "Positive" investment rating for the communication industry [11]. Core Insights - The communication sector experienced a decline of 1.85% in the 21st week of 2025, ranking 26th among primary industries in the Yangtze River region. Since the beginning of 2025, the sector has decreased by 4.71%, also ranking 26th [2][5]. - Recent releases of AI models, such as Google's Gemini 2.5 and Anthropic's Claude 4, continue to drive demand in data center communications, indicating a sustained positive outlook for AI and computing power [2][9]. - The current valuation of the telecommunications business segment remains low, suggesting a high cost-performance ratio for overall industry allocation [2]. Summary by Sections Market Performance - In the 21st week of 2025, the communication sector saw a 1.85% decline, with a year-to-date drop of 4.71%, both placing it 26th among primary industries [2][5]. - Among companies with a market capitalization above 8 billion, the top three gainers were Dongtu Technology (+14.5%), Haige Communication (+12.8%), and Mengwang Technology (+11.7%). The largest decliners were Lian Te Technology (-10.5%), Taicheng Light (-10.2%), and Cambridge Technology (-8.9%) [5]. AI and Computing Power Developments - Google's annual developer conference introduced the Gemini 2.5 series, enhancing multi-modal capabilities and platform integration, which is expected to bolster the AI ecosystem [6]. - Anthropic launched the Claude 4 series, focusing on coding capabilities and tool usage, with the Claude Code programming assistant now operational [7]. - Huawei's KADC 2025 emphasized the integration of AI and computing power, showcasing advancements in their software and hardware platforms [8]. Investment Recommendations - The report recommends focusing on several key players: - Telecom Operators: China Mobile, China Telecom, China Unicom [9]. - Optical Modules: Zhongji Xuchuang, Xinyi Sheng, Tianfu Communication, with a focus on Taicheng Light and Shijia Photon [9]. - Domestic Computing Power: Fenghuo Communication, Huafeng Technology, Yingweike, Runze Technology, Guangxun Technology, ZTE, and Unisplendour, with attention to Kehua Data and Aofei Data [9]. - IoT: Heertai, Tuobang, Yiyuan Communication, Meige Intelligent, and Aojie Technology [9]. - Satellite Applications: Haige Communication and Huace Navigation, with a recommendation to monitor Chengchang Technology [9].

Group 1 - The core viewpoint of the articles highlights Huawei's commitment to continuous investment in core technology innovation and system architecture innovation to build an open and prosperous computing industry ecosystem [1] - As of May 2025, Huawei's Kunpeng and Ascend processors have developed over 6.65 million developers and more than 8,800 partners, completing over 23,900 solution certifications [1] - The rapid development of AI technology is accelerating industry intelligence, making the usability, flexibility, and scalability of AI platforms and tools essential for AI development, deployment, and operation [1] Group 2 - Huawei officially launched the Kunpeng AI+ solution for the general computing field and released a diversified computing cluster software open-source community called openFuyao [1] - The Ascend supernode architecture breaks through traditional cluster interconnection bottlenecks, achieving efficient collaboration between nodes and allowing computing clusters to operate as a powerful single computer [2] - The architecture supports the industry's largest scale 384-card high-speed bus interconnection system, which can enhance training performance by three times, significantly improving overall computing efficiency for large-scale AI tasks [2]

Core Insights - The KADC2025 conference showcased a series of new technologies, tools, and platforms aimed at enabling partners and developers for efficient development [1] - Huawei's ecological strategy emphasizes "hardware openness, software open-source, enabling partners, and talent development," focusing on core technology and system architecture innovation [1] - As of May 2025, over 6.65 million developers and more than 8,800 partners have been engaged, with 23,900 solutions certified [1] Group 1 - The Kunpeng AI+ solution was officially launched, along with the open-source community openFuyao for diverse computing cluster software [1] - The AI Core segment introduced the Kunpeng+xPU inference solution, compatible with Ascend and other domestic GPUs, covering scenarios from data centers to edge computing [1] - AI Infra components were released, enhancing data processing and retrieval capabilities with improved performance and security [1] Group 2 - AI application capabilities were presented, including business orchestration, deployment, and scheduling, aimed at building end-to-end AI solutions for enterprises [2] - The Kunpeng RAG solution 1.0 was launched, creating an open ecosystem for RAG business orchestration, AI microservices, deployment, scheduling, security, and storage [2] - The openFuyao project focuses on "cloud-native + AI," providing diverse computing interconnect management and scheduling capabilities [2] Group 3 - Ascend introduced the CATLASS operator template library, MindIE Motor inference service acceleration library, and inference microservice MIS, enhancing the ease of operator development and application deployment [2] - Continuous upgrades to the layered open CANN capabilities and the MindSpeed RL reinforcement learning suite were announced, improving model training and inference efficiency [2]

Core Insights - The Kunpeng Ascend Developer Conference 2025 (KADC2025) was held in Beijing, where a series of new technologies, tools, and platforms were launched [1] - Kunpeng officially introduced the Kunpeng AI+ solution and open-sourced the diverse computing cluster software community openFuyao, aiming to build an excellent engine for the AI era [1] - Ascend released the CATLASS operator template library, MindIE Motor inference service acceleration library, inference microservice MIS, and continued to upgrade the layered open capabilities of CANN, MindSpeed RL reinforcement learning suite, and large-scale expert parallel inference solutions [1] Company Strategy - Huawei's ICT Marketing Department Head Zhou Jun emphasized the company's commitment to an ecosystem strategy of "hardware openness, software open-source, enabling partners, and developing talent" [1] - The company continues to invest in core technology innovation and system architecture innovation, collaborating with partners and developers to create an open and prosperous computing industry ecosystem [1] Developer and Partnership Growth - As of May 2025, Kunpeng and Ascend have developed over 6.65 million developers and more than 8,800 partners, completing over 23,900 solution certifications [1] - Huawei is focused on building a solid and user-friendly computing foundation to empower developers and partners, accelerating industry intelligence [1]