21世纪经济报道新质生产力研究院综合报道 早上好，新的一天又开始了。在过去的24小时内，科技行业发生了哪些有意思的事情？来跟21tech一起 看看吧。 【巨头风向标】 （原标题：小米人事大调整；海光信息终止与中科曙光资产重组丨新鲜早科技） 小米人事大调整 谷歌与XREAL联合发布Project Aura，为Gemini装上"眼睛" 12月9日，谷歌首次公布了 Project Aura。作为硬件产品，Project Aura被称为"实体 Agent"，该项目旨在 展示AI（特别是Gemini模型）如何深度集成到AR眼镜中，实现与真实物理世界的无缝交互。值得关注 的是，Project Aura的核心硬件研发几乎全部由中国团队完成，中国公司XREAL提供光学、芯片与空间 算法能力。XREAL创始人兼CEO徐驰表示："Agent不应仅仅是软件，更应该是一个硬件化的Agent，这 样才能最大范围地释放AI的能力。眼镜形态所带来的增量数据，是AI通向AGI的必经之路。" 阿里加码千问C端 记者12月9日获悉，阿里已成立千问C端事业群，由阿里巴巴集团副总裁吴嘉负责。据悉，该事业群由 原智能信息与智能互联两个事业群合并重组而 ...

本报记者 梁傲男 12月9日，在杭州群核信息技术有限公司（以下简称"群核科技"）2025酷+大会上，群核科技宣布将系统 性持续开放其底层空间智能相关能力，助力技术在千行百业的加速落地。 办方供图 当前，群核科技正从3D空间软件提供商逐步向空间智能基础设施提供商升级，公司使命从帮助人在三 维空间实现想象力进一步拓展至帮助具身智能、AIGC等智能体更智能。 会上，群核科技正式发布了Aholo空间智能开放平台。该平台整合了群核科技过去14年在空间重建、空 间生成、空间编辑与空间理解方面沉淀的核心能力，并将以底层模型和工具的形式逐步对外开放，以赋 能更广泛的行业应用与创新生态。 目前Aholo平台已启动内测。在该平台上，开发者和相关领域企业可通过API或SDK的方式自由调取上 述能力。用户可通过图片、视频或全景图等多模态输入，在平台上快速构建高保真的全息3D空间。 活动现场，浙江华策影视股份有限公司（以下简称"华策影视"）与群核科技达成战略合作，共同推进空 间智能技术在影视内容创作与制作领域的前沿应用与深度融合。华策影视将借助Aholo开放平台，围绕 虚拟片场生成与影视场景重建以及产业生态协同共建等方向展开深度合作， ...

Core Insights - Google officially launched Project Aura, an XR hardware reference design, in collaboration with Chinese AR hardware company XREAL, marking a significant advancement in Android XR operating system architecture [1][3] - Project Aura integrates Gemini AI with spatial awareness capabilities, transitioning AI from a "screen-based intelligence" to a "physical world intelligence" [1][3] Group 1: Project Aura Overview - Project Aura is positioned as the closest realization of the ideal Android XR hardware, enabling Gemini AI to construct environmental semantic maps and understand user intentions in real-time [3][5] - The strategic goal of Android XR is to create an open and unified extended reality platform, embedding AI into the real world [3][5] Group 2: Technical Innovations - Project Aura features a 70° optical field of view (FOV), the largest practical limit for consumer AR, enhancing the natural integration of digital content with the physical environment [5][6] - The X1S spatial computing chip, developed by XREAL, is optimized for spatial AI, providing a high-efficiency inference link from sensor input to semantic output [6][8] - Gemini AI is deeply integrated into the Android XR operating system, functioning as an OS-level service rather than a standalone application, enabling continuous and context-aware interactions [6][8] Group 3: Manufacturing and Future Outlook - Project Aura's core technology relies heavily on Chinese manufacturing, with key components like the X-Prism optical module and X1S chip developed and produced in China [8] - The global supply chain for Project Aura is based in the Yangtze River Delta, facilitating rapid hardware iteration cycles [8] - Project Aura is set to be commercially available by 2026, potentially transforming the XR industry from "display devices" to "spatial intelligent terminals" [8]

已启动港股IPO的"杭州六小龙"之一的群核科技，宣布要做空间智能的"卖水人"！ 12月9日，在群核科技2025酷+大会上，群核科技宣布将系统性持续开放其底层空间智能相关能力，助 力技术加速落地。 群核科技联合创始人兼CEO陈航说："当AI加速迈向三维世界，行业最需要的是底层技术的'卖水人'。 群核致力于提供'可落地'的空间智能能力——不仅实现可视化层面的真实感，更要输出结构化的空间数 据。同时，在算力受限的现实条件下，通过生成、仿真等多元技术组合，更高效连接物理世界与数字世 界。" 群核科技在会上发布战略全景图，系统展示了公司整体布局。当前，群核科技正从3D空间软件提供商 逐步向空间智能基础设施提供商升级，公司使命从帮助人在三维空间实现想象力进一步拓展至帮助具身 智能、AIGC等智能体更智能。 Aholo空间智能开放平台上线 近两年，AI在文本、图像和视频领域突飞猛进，但还迟迟无法满足物理世界的真实需求，空间智能成 为了AI走向三维世界的关键钥匙。而空间智能的加速落地，急需更广泛、更开放的底层能力提供方， 以推动其从"实验室技术"走向规模化产业应用。 会上，群核科技发布了Aholo空间智能开放平台。该平台整合 ...

近两年，AI在文本、图像和视频领域突飞猛进，但还迟迟无法满足物理世界的真实需求，空间智能成为了AI走向三维世界的关键钥匙。而空间智能的加速落地，急 需更广泛、更开放的底层能力提供方，以推动其从"实验室技术"走向规模化产业应用。 会上，群核科技发布了Aholo空间智能开放平台。该平台整合了群核科技过去14年在空间重建、空间生成、空间编辑与空间理解方面沉淀的核心能力，并将以底层 模型和工具的形式逐步对外开放，以赋能更广泛的行业应用与创新生态。 已启动港股IPO的"杭州六小龙"之一的群核科技，宣布要做空间智能的"卖水人"！ 12月9日，在群核科技2025酷+大会上，群核科技宣布将系统性持续开放其底层空间智能相关能力，助力技术加速落地。 群核科技联合创始人兼CEO陈航说："当AI加速迈向三维世界，行业最需要的是底层技术的'卖水人'。群核致力于提供'可落地'的空间智能能力——不仅实现可视化 层面的真实感，更要输出结构化的空间数据。同时，在算力受限的现实条件下，通过生成、仿真等多元技术组合，更高效连接物理世界与数字世界。" 群核科技在会上发布战略全景图，系统展示了公司整体布局。当前，群核科技正从3D空间软件提供商逐步向空 ...

群核科技今日首次发布空间智能开放平台Aholo，公司CEO陈航表示，"当AI开始迈向三维世界，需要空 间智能的'卖水人'。"记者现场获悉，Aholo将陆续开放空间重建、空间生成、空间编辑、空间理解等群 核底层空间智能能力，最终可应用于3D空间创作、影视短剧、工业孪生、机器人仿真等领域。（智通 财经） ...

Core Insights - The partnership between Yingtong Communications and Montong Intelligent marks a strategic collaboration aimed at developing AI glasses solutions, focusing on the emerging spatial intelligent interaction market [1][2][3] - Montong Intelligent, established in 2022, specializes in AI+AR glasses solutions and has rapidly gained market traction, partnering with major global brands [2][3] - The collaboration is structured as a progressive model involving a framework agreement and project agreements, emphasizing mutual strengths to tackle industry challenges [2][3] Company Overview - Yingtong Communications is a significant player in the smart terminal sector, involved in product innovation, manufacturing, and market expansion, supported by a robust supply chain and quality control standards [1] - Montong Intelligent's core technologies include FuseSpaceAI and SpaceAIOS, with a team sourced from leading companies, enhancing its competitive edge in the AI glasses market [2] Industry Context - The AI glasses market is currently characterized by intense competition, with challenges such as technology fragmentation and homogenized user experiences needing to be addressed [3] - The collaboration is seen as a positive signal for industry integration, combining Yingtong's industrial foundation with Montong's technological agility to accelerate market development [3] - A technical meeting is scheduled to facilitate deep technical alignment between the two companies, focusing on product definition, technology validation, and cost analysis [3]

Core Viewpoint - The company, Wuyishijie, has updated its prospectus for the third time, aiming for a public offering after previously attempting to list on the STAR Market and the Beijing Stock Exchange. It has received approval from the China Securities Regulatory Commission [1][3]. Financial Performance - In the first half of 2025, Wuyishijie reported revenue of 53.82 million RMB, a year-on-year increase of 62%. However, the adjusted net loss expanded by 11.4% to 67.18 million RMB [1][4]. - Revenue has shown consistent growth from 169.977 million RMB in 2022 to 256.302 million RMB in 2023, and 287.362 million RMB in 2024. The adjusted net loss decreased from 132 million RMB in 2022 to 68.08 million RMB in 2023, and further to 43.07 million RMB in 2024, before reversing in 2025 [4][5]. - The gross margin has declined from 65% in 2022 to 41.1% in the first half of 2025, a drop of 8.9 percentage points compared to the same period in 2024 [4][5]. Cost Structure - Sales costs increased significantly, with 2025 first-half sales costs rising by 90.7% to 31.69 million RMB compared to the previous year, leading to a sales cost-to-revenue ratio of 58.9% [4][5]. - The cost of purchasing software accounted for 60.5% of sales costs in the first half of 2025, an increase of 41.8 percentage points from the same period in 2024 [5][6]. Research and Development - R&D expenses in the first half of 2025 were 30.14 million RMB, showing a slight increase from 28.59 million RMB in the same period of 2024. However, R&D spending has decreased in proportion compared to sales and administrative expenses [5][6]. - The company plans to hire 50-100 R&D personnel over the next three years, with the current R&D team size decreasing from 130 in January 2024 to 114 in the first half of 2025 [5][6]. Business Segments - The digital twin platform 51Aes remains the primary revenue driver, contributing approximately 81.4% of total revenue. Other platforms, such as 51Sim and 51Earth, contributed 14.8% and 3.9% respectively in the first half of 2025 [1][10]. - The company has emphasized new products in spatial intelligence and embodied intelligence, including the Clonova platform, although this product was not mentioned in the latest prospectus [11][12]. Legal Matters - Wuyishijie settled a legal dispute by paying 7 million RMB to a service provider, which had previously claimed 23 million RMB for contract violations [6][9].

规模与渗透率为物理 AI 提供"落地土壤" 虽然从单纯的语言模型，拓展至能够在语义、物理、几何与动态等多重复杂世界中进行理解、推理、生 成与交互的能力看似遥远，但在智能化技术飞速发展的当下，"物理AI"却已经在中国具备了初步的落 地"土壤"。 （文/观察者网 张家栋 编辑/高莘） 当智能电动车的硬件与软件竞争日益白热化，造车新势力正在把赌注押向新的战略高地——"物理AI"。 继小鹏汽车在今年科技日上正式宣布其"物理AI"战略后，理想也正在尝试将"物理AI"运用在其商业化生 态当中。尽管在具体的呈现方式上各有不同，但其背后折射出的，是中国智能电动车产业已经从"技术 配置竞争"进入"跨终端智能平台竞争"的趋势。 李飞飞曾经主导的Marble空间智能项目 World Labs 无独有偶，被业内誉为"AI教母"的斯坦福大学教授李飞飞也在近日公开表示，当前以大型语言模型为代 表的AI虽然擅长处理抽象知识，却如同在"黑暗中行走"，缺乏对物理世界的真实理解。而空间智能，即 人类用以理解、导航并与三维世界交互的底层能力，将是实现机器真正智能的关键突破。"要实现空间 智能，我们需要的远不止LLM（大语言模型）那样的体系，而是更 ...

编译：Haozhen、Gemini 如今 LLM 的语言理解与生成能力已展现出惊人的广泛适用性，但随着 LLM 的发展，一个事实越 发凸显：仅靠语言，仍不足以支撑真正的智能。 从更本质的角度看，人类处理世界的方式从来不只依赖文字，而是通过视觉、空间感知、物理直觉 与行动能力等共同构成完整的认知体系。语言只是对三维世界的"有损压缩"：它记录结论，却省略 过程；它表达结构，却隐藏动态。而真正的智能，源于不断与世界互动、不断在空间中推理和行动 的能力。 正因如此，构建能够"理解并操作世界"的空间智能（Spatial Intelligence）与世界模型（World Models）成为继 LLM 之后的关键方向。 2024 年，李飞飞、Justin Johnson 等学者创立了 World Labs，今年 11 月推出了 Marble 这个 3D 世界 生成模型。团队尝试突破模型"只懂文本"的限制，让模型具备在三维环境中定位、推理、模拟、生 成甚至执行任务的能力。这不仅意味着新的技术路线，也意味着新的 AI 价值尺度：从语言走向世 界、从描述走向交互、从静态认知走向动态智能。 本文整理了李飞飞和 Justin Joh ...