16日~22日这周，中国模型的周调用量进一步冲高至5.16万亿Token，三周大涨127%，而同期美国模型调用量跌至2.7万亿Token。与此同时，全球调用量 排名前五的模型中，中国模型占据四席，这股强大的增长动能，并非依赖单一爆款产品，而是中国AI厂商集群式崛起。 Token是AI模型处理文本的最小单位。相比用户数，Token调用量是更能真实反映AI模型使用强度、用户粘性及商业价值的关键指标。 每经记者｜宋欣悦 每经编辑｜肖勇 王嘉琦 高涵 2月，中国AI的模型调用量爆发式增长，首次超过美国。 全球最大的AI模型API聚合平台OpenRouter数据显示，9日~15日这周，中国模型以4.12万亿Token的调用量，首次超过同期美国模型的2.94万亿Token。 中国模型厂商，正凭借快速迭代和成本优势占领全球市场，国产算力需求正经历指数级增长。 榜单洗牌：中国Token调用量首超美国，四款大模型霸榜 OpenRouter平台，汇聚了全球数百种大语言模型，拥有超过500万开发者用户，是目前全球最大的AI模型API聚合平台。因此，其API调用量数据被视为 洞察全球AI应用落地趋势最真实的"晴雨表"，因为它直接反映 ...

Core Viewpoint - Nebius (NBIS) has been rated as a Buy due to strong demand for AI infrastructure, significant hyperscaler contracts, and strategic capacity expansion, leading to a share price increase of approximately 15% since the last coverage [1]. Group 1: Company Performance - The share price of Nebius (NBIS) has increased by around 15% following the positive outlook on its demand and contracts [1]. Group 2: Analyst Background - The analyst has a strong background in finance and risk analysis, holding an MSc in Applied Risk Management and having worked in various roles across leading firms [1]. - The analyst's expertise includes financial analysis, risk management, data analysis using SQL, Python, and machine learning tools [1]. - The focus is on translating complex financial data into actionable insights for investors, emphasizing data-driven analysis and long-term value creation [1].

Core Viewpoint - Robo.ai Inc. has successfully completed the initial collection and delivery of intelligent data in the Middle East, marking a significant step in the development of its artificial intelligence machine economy platform [1] Group 1: Company Developments - Robo.ai Inc. executed a data collection services agreement with DaBoss.AI Inc. earlier this month [1] - The completion of data collection is crucial as the artificial intelligence sector shifts from large language models to embodied AI [1] Group 2: Industry Context - The acquisition of high-quality, real-world physical interaction data is identified as a structural challenge for the artificial intelligence industry [1]

2026 年 02 月 26 日 ledi/229 相关研究 证券分析师 林起贤 A0230519060002 linqx@swsresearch.com 袁伟嘉 A0230519080013 yuanwj@swsresearch.com 任梦妮 A0230521100005 renmn@swsresearch.com 夏嘉励 A0230522090001 xiajl@swsresearch.com 赵航 A0230522100002 zhaohang@swsresearch.com 黄俊儒 A0230525070008 huangjr@swsresearch.com 李国盛 A0230521080003 ligs@swsresearch.com 研究反馈 张淇元 A0230124080001 zhangqy@swsresearch.com 联系人 张淇元 A0230124080001 zhangqy@swsresearch.com Seedance2.0 突破, Al 视频 格局及产业链机遇 ——GenAl 系列报告之 69 暨 Al 应用深度之四 本期投资提示: 请务必仔细阅读正文之后的各项信息披露与声明 ...

今天分享的是：量子位；2025年度AI十大趋势报告 报告共计：48页 2025年度AI十大趋势报告核心内容总结 量子位智库2025年度AI十大趋势报告从基础设施、模型进化、应用版图、中国发展四大维度，梳理了当下AI产业的核心发展趋势，展现了AI技术从底层基 建到上层应用的全链路进化图景，也凸显了中国在全球AI发展中的独特价值与地位。 在基础设施层面，算力与芯片成为产业发展核心基石。算力基建化成为显著特征，全球超大规模数据中心建设提速，各国均将算力布局提升至战略高度，太 空计算等新型算力方案也进入探索阶段，算力成为智能产业的核心引擎；芯片则向AI原生方向转型，GPU仍是训练核心，NPU成为端侧标配，ASIC/FPGA 迎来增长，中国正加速AI芯片国产替代，实现了国产软硬件协同完成千亿级参数大模型训练的突破。 模型进化领域，技术创新围绕效率与落地展开。预训练架构成为大模型格局的关键，MoE架构、线性/稀疏注意力等技术突破了Transformer的效率瓶颈，模 型蒸馏、强化学习等技术则推动大模型能力向端侧下沉；大模型落地进入推理时代，多模态深度推理、自适应推理成为研发重点，推理需求倒逼框架与硬件 的协同创新；同时， ...

The Telegraph is among a number of major media outlets in new coalition aimed at protecting publishers from tech firms’ copyright abuses - Geoff Pugh News industry rivals have joined forces to stop AI giants cashing in on original journalism without contributing to its costs. The Telegraph, the BBC, Sky News, The Guardian and The Financial Times have formed a united front to push for OpenAI, Google and others to pay to use their publications as training data for their technology. In an open letter the ...

Core Viewpoint - Shanghai-based AI startup Jieyue Xingchen is considering an IPO on the Hong Kong Stock Exchange, aiming to raise approximately $500 million (around 3.9 billion HKD) following a recent 5 billion RMB B+ round financing [2][3]. Group 1: IPO Plans and Market Context - Jieyue Xingchen is in discussions with potential advisors regarding the equity sale for its IPO, which could occur as early as this year, although details on the issuance scale and timing remain uncertain [2]. - The company aims to become the third domestic large model enterprise to list on the Hong Kong Stock Exchange, following the successful IPOs of Zhiyuan and MiniMax, which saw their market values soar to over 320 billion HKD and 300 billion HKD, respectively [3]. - The recent surge in capital interest in the AI sector has been marked by significant IPOs, with companies like TianShu ZhiXin also achieving substantial market valuations shortly after listing [3]. Group 2: Recent Financing and Investment - Jieyue Xingchen completed a record-breaking B+ round financing of over 5 billion RMB on January 26, 2026, marking the highest single financing amount in the Chinese large model sector over the past year [4][6]. - Notable investors in this financing round include Shanghai Guotou, China Life Equity, and Tencent, with existing shareholders like Qiming Venture Partners also participating [4][6]. - The company has a history of successful fundraising, including a previous B round and an angel round, with continuous support from key investors [5][6]. Group 3: Leadership and Technological Advancements - The appointment of AI veteran Yin Qi as chairman is expected to enhance Jieyue Xingchen's strategic direction and commercialization efforts, leveraging his experience from previous ventures [7]. - The company has launched multiple significant AI models, including the Step-1 and Step-2 models, and is actively developing the next generation of models, focusing on language and multi-modal capabilities [7]. - Jieyue Xingchen has released over 30 models, emphasizing its commitment to advancing AI technology and integrating it with terminal systems [7].

Core Viewpoint - The Jiangsu Province Intellectual Property Bureau has officially released the "Jiangsu Province Artificial Intelligence and Intellectual Property Dual Empowerment Action Plan (2026-2030)", marking it as the first national plan focusing on the dual empowerment of artificial intelligence and intellectual property. The plan aims to enhance innovation protection in the AI industry and improve intellectual property service efficiency through 23 specific measures across seven areas, fostering a synergistic effect of "1+1>2" between AI and intellectual property [1][15]. Group 1: Empowering AI Industry Development through Intellectual Property - Strengthening the guarantee of industrial development factors by exploring data intellectual property protection rules and formulating compliance guidelines for new fields and business models [3][17]. - Accelerating the layout of key technology patents in the industry, with a target of increasing effective invention patents in the AI field by no less than 20% [3][18]. - Promoting the transformation of industrial innovation achievements by matching technology supply and demand efficiently and supporting the establishment of intellectual property alliances [4][20]. Group 2: AI Empowering Intellectual Property Creation - Utilizing AI semantic analysis and knowledge graphs to create industry technology maps, helping to identify development trends and optimize intellectual property layout [6][25]. - Implementing intelligent assistance in patent applications to enhance the quality of applications by identifying defects and assessing the reasonableness of protection scope [7][26]. - Establishing a comprehensive management system for the entire lifecycle of intellectual property to improve management precision and efficiency [8][28]. Group 3: AI Empowering Intellectual Property Utilization - Achieving precise matching of technology supply and demand through AI technologies, facilitating the transfer and transformation of patent technology achievements [8][57]. - Developing a multi-dimensional patent value evaluation model to assess the economic value of patents based on various factors [8][31]. - Creating an enterprise innovation capability evaluation model that integrates diverse data to assess R&D capabilities [8][34]. Group 4: AI Empowering Intellectual Property Protection - Constructing an intelligent model for analyzing intellectual property infringement to enhance the efficiency and quality of case handling [9][86]. - Establishing a monitoring system for online infringement behaviors to ensure rapid detection and response [9][89]. - Building a comprehensive overseas intellectual property service platform to proactively manage risks associated with overseas operations [9][40]. Group 5: AI Empowering Intellectual Property Services - Providing comprehensive public services through AI technologies to streamline intellectual property policy consultations and administrative processes [10][90]. - Developing an intelligent legal service platform to enhance accessibility to intellectual property legal services [10][91]. - Creating a tailored training system for intellectual property professionals to meet industry needs and foster talent development [10][92]. Group 6: AI Empowering Intellectual Property Governance - Establishing a decision support system for government management, integrating various data sources to support macro decision-making [12][93]. - Implementing intelligent analysis models for evaluating major economic and technological activities to provide decision-making references [12][94]. Group 7: Building a Collaborative Ecosystem for AI and Intellectual Property - Creating a trusted data space for intellectual property by integrating global data and ensuring data security [13][95]. - Developing a project database for AI applications in intellectual property, supporting the creation of leading models and tools [13][96]. - Strengthening service and talent support by expanding patent pre-examination fields and cultivating specialized service teams [13][98].

CloudBot的爆红与并购，不是Agent创业的终点，而是从概念走向落地的起点。大模型不会包揽一切，中小团队也不必绝望。 出品 | 竞合人工智能 作者 | 竞合 排版 | 洋洋 近期硅谷与国内AI圈被两只"龙虾"搅动：开源项目CloudBot（OpenClaw，社区称龙虾）以本地执行、自主任务闭环迅速走红；随后传出创始人加入 OpenAI、项目获资助并转入基金会运作。几乎同期，聚焦通用Agent的Manus被Meta以高价收购，成为AI应用层最受瞩目的两起并购。一时间，行业共识 似乎被简化为：大模型通吃一切、独立Agent创业空间被压到极致、被收购是唯一结局。 真相远非如此。AI Agent不是大模型的"皮肤"，中小创业者也并非只能在巨头缝隙里求生。本文以CloudBot为样本，拆解Agent创业的真实护城河、国内复 刻可能性、大厂挤压下的生存逻辑，并给出2026年可落地的细分赛道与终局判断。 01 CloudBot的爆红： 不是套壳，是范式转移 先澄清核心事实：CloudBot并非简单调用大模型API的聊天机器人，它的核心价值在于本地自托管、设备执行权、持久记忆、多模型插拔、社交入口交互 五大能力。用户一 ...

Core Viewpoint - The rising reliance of children on AI companion bots for guidance instead of trusted adults raises concerns about their safety and the potential exploitation by companies behind these bots [1][2][3]. Group 1: Concerns Raised by Governor Shapiro - Governor Shapiro expressed empathy for children using AI bots and highlighted the risk of these companies taking advantage of vulnerable users [2][3]. - He proposed measures such as age verification, parental consent, and restrictions on the creation of sexually explicit or violent content involving children to enhance online safety [2][3]. Group 2: Impact of AI Companion Bots - The increasing use of AI companion bots has sparked concerns regarding their effects on mental health and social relationships, with warnings that these apps could lead to users becoming mentally "manipulable" [4]. - The preference for virtual relationships over real-life interactions has been noted, indicating a shift in social dynamics due to AI technology [4]. Group 3: Regulatory Call - The call for regulation by Governor Shapiro reflects a growing awareness of the need to protect vulnerable users from potential harm associated with AI technologies [5].