记者2月12日从浙江新型研发机构之江实验室了解到，"三体计算星座"已实现星间组网突破，通过在轨协同完成了10个人工智能模型与应用的部署与验 证，探索了深空探测、智慧城市建设、自然资源普查等场景的太空计算创新应用。 记者了解到，已部署的各模型已成功执行了多次在轨任务。人工智能模型在太空的应用，还在推动太空科研范式的变革。以天文学研究为例，"三体计算 星座"的两颗卫星搭载了宇宙X射线偏振探测器，通过部署天基天文时域模型，实现对伽马射线暴（GRB）的在轨快速判定和分类。在极大减少下传数据 和数据处理时间的同时，保持了高达99%的事件识别准确率。 "三体计算星座"是之江实验室协同全球合作伙伴共同打造的千星规模太空计算基础设施，通过"计算上天、星间互联、模型上天"，推动太空科研范式变 革。2025年5月14日，"三体计算星座"首发12颗计算卫星成功发射入轨。 之江实验室计算星座总体部技术总师李超说，三体计算星座首发任务已实现将模型或应用从地面灵活部署至星座任意卫星节点的能力。用户自定义的模型 或应用经地面数字孪生系统验证，就可以及时通过天基分布式操作系统进行在轨部署与更新。此外，团队近期还实现了六颗卫星在轨建链，意味着 ...

Core Insights - The article discusses the successful deployment and testing of the "Three-body Computing Constellation," which focuses on water environment monitoring and has achieved significant advancements in space-based artificial intelligence applications [1][2]. Group 1: Technological Achievements - The Three-body Computing Constellation has successfully completed a full-link capability verification involving satellite payload operation, on-orbit data processing, inter-satellite collaboration, and model computation [1]. - A total of 10 artificial intelligence models have been deployed and validated in orbit, including the largest parameter models globally, such as the 8 billion parameter space-based remote sensing model and the 8 billion parameter space-based astronomical time-domain model [2]. - The overall on-orbit computing power of the constellation has reached 5 petaflops (5POPS), capable of supporting the deployment and inference of models with up to 140 billion parameters, making it the largest space computing constellation in the world [5]. Group 2: Applications and Innovations - The deployed AI models have successfully executed multiple on-orbit tasks, including infrastructure surveys in challenging weather conditions, demonstrating the practical applications of space-based AI [2][3]. - The astronomical research capabilities have been enhanced through the deployment of a space-based astronomical time-domain model, which significantly reduces data transmission volume and processing time while maintaining a high event identification accuracy of 99% [3]. - The integration of a space-based distributed operating system has allowed for unified management and scheduling of computing resources, enhancing the efficiency of constellation operations [4]. Group 3: Future Prospects - The Three-body Computing Constellation aims to drive innovation in space research paradigms, particularly in areas such as deep space exploration and smart city development, by leveraging advanced AI capabilities [1][2]. - The successful inter-satellite networking and ground-satellite integration mark a significant step towards breaking down barriers between satellite networks and terrestrial internet [3].

记者2月12日从浙江新型研发机构之江实验室了解到，"三体计算星座"已实现星间组网突破，通过在轨 协同完成了10个人工智能模型与应用的部署与验证，探索了深空探测、智慧城市建设、自然资源普查等 场景的太空计算创新应用。 专家表示，把模型送上天，是推动人工智能在太空应用与发展的关键。截至目前，之江实验室团队已实 现10个人工智能模型和应用的在轨部署。其中，80亿参数天基遥感模型和80亿参数天基天文时域模型， 是全球在轨运行的参数规模最大的模型之一，有6个模型与算法实现了在轨更新部署。 记者了解到，已部署的各模型已成功执行了多次在轨任务。人工智能模型在太空的应用，还在推动太空 科研范式的变革。以天文学研究为例，"三体计算星座"的两颗卫星搭载了宇宙X射线偏振探测器，通过 部署天基天文时域模型，实现对伽马射线暴（GRB）的在轨快速判定和分类。在极大减少下传数据和 数据处理时间的同时，保持了高达99%的事件识别准确率。 之江实验室计算星座总体部技术总师李超说，三体计算星座首发任务已实现将模型或应用从地面灵活部 署至星座任意卫星节点的能力。用户自定义的模型或应用经地面数字孪生系统验证，就可以及时通过天 基分布式操作系统进行在 ...

来源：科技日报 科技日报记者 江耘 通讯员 陈航 2月初，三体计算星座开展了面向水环境监测的三星协同在轨智能处理试验，通过地表要素提取模型在 水面结冰的情况下提取出了关键水体，验证了"卫星载荷工作—在轨数据处理—星间协同传输—在轨模 型计算—任务结果下传"的全链路能力。12日，记者从浙江的之江实验室获悉，三体计算星座实现了星 间组网突破，通过在轨协同完成了10个人工智能模型与应用的部署与验证，探索了深空探测、智慧城市 建设、自然资源普查等场景的太空计算创新应用。 三体计算星座是之江实验室协同全球合作伙伴共同打造的千星规模太空计算基础设施，通过"计算上 天、星间互联、模型上天"，推动太空科研范式变革，助力"航天+人工智能"创新发展。2025年5月14 日，三体计算星座首发12颗计算卫星成功发射入轨。 之江实验室计算星座总体部技术总师李超介绍，经过九个月的在轨测试，三体计算星座首发任务已形成 组网、计算、模型部署以及科学载荷在轨验证等四大核心能力。 目前，三体计算星座首发任务实现了所有在轨计算节点的协同运行，整体在轨算力达5P OPS，最大可 支持1400亿参数模型在轨部署与推理，是目前全球算力规模最大的太空计算 ...

"人工智能不能因为缺失算力而缺席太空。"正如中国工程院院士、之江实验室主任王坚所言，这是实验 室打造三体计算星座的初衷。模型的部署与应用正是基于三体计算星座组网与计算能力的突破。首发12 颗计算卫星均搭载了之江实验室研发的星载计算单元，单星最高算力可达744TOPS，即每秒744万亿次 运算。 李超表示，已部署的各模型成功执行多次在轨任务。2024年11月10日，团队对我国西北某地189平方公 里城区开展了体育场、桥梁等基础设施普查试验，通过天基遥感模型，在大雪覆盖条件下，自动识别桥 梁、田径场等设施。 人工智能模型用于太空，还在推动太空科研范式的变革。以天文学研究为例，三体计算星座2颗卫星搭 载了宇宙X射线偏振探测器，通过部署天基天文时域模型，实现对伽马射线暴（GRB）的在轨快速判定 和分类。 该模型将下传数据量从每天几百兆字节减少至几十千字节，仅为传统观测方式的万分之一，处理时间也 从数小时缩短至数秒，且保持了高达99%的事件识别准确率，为在轨天文观测提供了全新技术支撑。 算力上天后，团队展开星地组网、星间组网测试。三体计算星座首发任务实现了所有载荷、地面站的IP 化，打破了卫星网络与地面互联网的壁垒。团 ...

Core Viewpoint - SpaceX proposes the "Orbital Data Center System" to deploy up to 1 million satellites in low Earth orbit, aiming to provide computational power for AI models using solar energy, which is seen as a more economical and energy-efficient solution compared to traditional ground-based data centers [1][3][4]. Group 1: SpaceX's Proposal - SpaceX's application to the FCC highlights the rapid growth in demand for AI, machine learning, and edge computing, which exceeds the processing capabilities of ground infrastructure [3][4]. - The projected global data center electricity demand is expected to more than double by 2035, reaching approximately 1200-1700 terawatt-hours (TWh), potentially accounting for up to 4% of total global electricity consumption [3][4]. - The orbital data centers will operate at altitudes between 500 to 2000 kilometers, utilizing solar energy to minimize operational costs and environmental impact compared to terrestrial data centers [3][4]. Group 2: Technical Aspects - SpaceX plans to design various satellite hardware versions to optimize operational efficiency, with each satellite equipped with heat dissipation panels and a lifespan of five years [4]. - The new satellite constellation will feature high-speed laser links for data transmission, connecting to both the proposed system and existing Starlink satellites [4]. - SpaceX's reusable rocket technology, particularly the Starship, is expected to significantly reduce launch costs, enabling the deployment of substantial amounts of computational power in space [4][6]. Group 3: Competitive Landscape - Other companies, such as Google and Nvidia-supported Starcloud, are also exploring space-based computing solutions, with Google planning to launch its TPU satellites by 2027 and Starcloud having already sent satellites equipped with H100 chips into orbit [7][8]. - China's initiatives include the launch of 12 computing satellites by Zhijiang Laboratory, which will form a network capable of 1000 Peta Operations Per Second (POPS) [10]. - Beijing's plan for a space data center aims to establish a large-scale system with over 1 gigawatt (GW) of power, designed to host a million servers and provide AI computing capabilities in space [11].

Core Viewpoint - The development of China's commercial space industry is at a critical juncture, with significant opportunities and challenges ahead, particularly in achieving reusable rocket technology and scaling up production efficiency [1][7]. Group 1: Technological Innovation - The demand for large-scale rocket launches is driven by the extensive satellite deployment plans, necessitating improvements in rocket cost-effectiveness and the development of reusable technologies [3]. - Blue Arrow Aerospace's Zhuque-2 rocket became the world's first liquid oxygen-methane rocket to reach orbit, marking a significant step towards reusable rocket technology [3]. - Companies like Tianbing Technology and Galaxy Aerospace are making strides in key technologies, such as multi-satellite launches and advanced satellite designs [3]. Group 2: Efficiency Enhancement - Experts estimate that over the next nine years, China needs to launch more than 25,000 satellites, with a total of over 120,000 satellites in the next twelve years [4]. - The industry consensus is to produce satellites in bulk, akin to automobile manufacturing, to meet the high demand efficiently [5]. - Innovations in production lines have significantly reduced the assembly time for satellites, with some companies achieving a turnaround of 20-25 days for satellite assembly [5]. Group 3: Ecological Support - The establishment of a comprehensive commercial space industry chain at the Dongfang Spaceport in Shandong enhances collaboration and reduces costs [6]. - Beijing is highlighted as a hub for commercial space enterprises, hosting a majority of the country's rocket companies and facilitating a closed-loop from rocket manufacturing to satellite utilization [6]. Group 4: Development Challenges - Many commercial space companies are struggling to achieve profitability, relying heavily on financing and facing challenges in market expansion [7]. - The high cost of rocket launches remains a significant barrier to the industry's growth, with the engine and rocket body costs comprising a large portion of total expenses [8]. - The need for a regulatory framework that supports frequent launches is emphasized, as current resources are limited, constraining launch opportunities [10]. Group 5: Market Demand and Application - The true demand for commercial space lies in the application of satellite services, such as satellite internet and remote sensing, which are essential for sustainable development [11]. - The government is actively promoting the expansion of satellite service applications to stimulate market demand and support the commercial space sector [11][12]. - Recent policy initiatives aim to enhance the commercial space industry's development, encouraging private sector participation and innovation [13].

Core Insights - Space computing is emerging as a new frontier in global technological competition and commercial aerospace development, serving as a new engine for high-quality digital economic growth [1][2] - The transition from terrestrial computing to space computing requires a systematic update in concepts and models, rather than merely relocating computing infrastructure to space [1][2] Group 1: Transition from Terrestrial to Space Computing - The core value of space computing lies in overcoming the limitations of terrestrial computing in energy supply and global coverage, utilizing resources collected by satellites in space [1][2] - The China Academy of Information and Communications Technology (CAICT) suggests a phased approach to developing space computing, starting with "space-to-space computing" focused on remote sensing satellite needs [2] Group 2: AI and Space Data Value Extraction - The "Three-body Computing Constellation" project led by Zhijiang Laboratory aims to deploy AI models in space to directly process and extract value from space data, addressing the issue that approximately 90% of space data is currently unprocessed [3] - The project has achieved a computing capacity of 744T OPS per satellite, with an overall in-orbit capacity of 5P OPS, and plans to scale to 100 satellites by 2027 [2][3] Group 3: Industry Collaboration and Technological Challenges - Industry collaboration is accelerating, with companies like Haohan Deep and Yiwei Aerospace signing agreements to establish joint laboratories focusing on high-performance computing and integrated communication technologies [4] - The space computing industry faces challenges, including the need for efficient thermal management of satellites, the development of radiation-resistant chips, and the establishment of a mature inter-satellite communication system [4][5] Group 4: Commercialization and Application Challenges - The cost of deploying computing capabilities in space is significantly higher than on the ground due to the harsh conditions satellites must endure, leading to expensive single-unit costs [4][5] - Establishing a clear application framework for space computing is essential to create value, moving from merely having computing power in space to effectively utilizing it [5]

Group 1 - The core business model of the "Three-Body Computing Constellation" integrates "space communication links + on-orbit computing centers + intelligent data platforms," addressing enterprise-level cross-border data transmission and real-time computing scheduling needs [1][2][3] - The project aims to launch 2,800 computing satellites, achieving a total computing power of 1,000 Peta Operations Per Second (POPS) and inter-satellite laser communication link speeds of 100 Gbps [3] - The estimated annual revenue upon completion of the network is approximately 27 billion yuan, with three main business segments: dedicated network transmission (16.5 billion yuan/year), computing leasing (5.1 billion yuan/year), and data services (5.4 billion yuan/year) [3] Group 2 - The "Three-Body Computing Constellation" represents a paradigm shift from "communication transmission" to "on-orbit intelligent computing," enabling real-time processing of space data and addressing energy supply and deployment flexibility issues for next-generation AI clusters [1][2] - The model includes three dimensions: space communication and star cable plans, space computing leasing services, and intelligent data services, which collectively enhance data transmission efficiency and provide edge computing capabilities [2][3] - The project aligns strategically with national initiatives such as the "integration of space and ground networks" and the "East Data West Computing" project, positioning itself as a core hub in the space economy ecosystem [3]

Core Viewpoint - The report from Galaxy Securities highlights the long-term growth potential of satellite internet across various industries, recommending attention to three main lines in the satellite internet industry chain: potential operators of space computing power, satellite communication equipment providers, and satellite application service providers [1] Group 1: Strategic Cooperation - PuTian Technology and Changsha Nonferrous Institute have signed a strategic cooperation framework agreement to combine PuTian's technological advantages in 5G, satellite internet, data elements, and artificial intelligence with Changsha's expertise in mining and metallurgy [2] - This collaboration aims to provide advanced and comprehensive digital solutions for industry clients, serving as a practical model for upgrading traditional industries by integrating new communication technologies with specific industry know-how [2] Group 2: Accelerated Application Deployment - The new generation of communication technology, represented by satellite internet, is rapidly penetrating various industries, with a focus on three main lines of application [3] - The first line involves the acceleration of direct satellite connections for consumer and automotive sectors, with approximately 40 direct satellite phones expected to be launched by China Telecom and partners by the end of 2025, and an estimated cumulative shipment of over 25 million units [3] - The second line is the nationwide commercial trial of satellite IoT, initiated by the Ministry of Industry and Information Technology, which aims to establish a foundation for large-scale applications through phased, multi-scenario trials [3] - The third line emphasizes the evolution of space computing power, transitioning satellites from mere connectivity to also providing computational capabilities, which can alleviate bandwidth bottlenecks and create value for time-sensitive industries [4] Group 3: Key Developments in Space Computing - The "Three-Body Computing Constellation," in which PuTian Technology is deeply involved, has entered the networking phase, with plans for a launch of "one rocket, twelve satellites" by May 2025, featuring a single satellite computing power of up to 744 TOPS and inter-satellite laser communication rates of 100 Gbps [4] - This constellation aims to serve as an application accelerator by providing on-orbit computing and instant response capabilities, exploring business models such as inter-satellite link leasing and computing power leasing for various industries [4]