+852-2978-2930 | allen.k.chang@gs.com Goldman Sachs (Asia) L.L.C. Verena Jeng +852-2978-1681 | verena.jeng@gs.com Goldman Sachs (Asia) L.L.C. 更多资料加入知识星球：水木调研纪要 关注公众号：水木纪要 17 February 2025 | 10:12AM HKT iFlytek (002230.SZ): DeepSeek on iFlytek open platform; Spark X1, Deep reasoning model launched; Neutral iFlytek launched its first deep reasoning model Spark X1 in Jan 2025 and upgraded version of Spark LLM (Large Language Model) V4.0 Turbo post the launch of Spark V4.0 Turbo in Oct 2024 (link). Management h ...

更多资料加入知识星球：水木调研纪要 关注公众号：水木纪要 17 February 2025 | 8:21AM HKT Navigating China Internet: Top 400 app tracker: Rise of AI applications with DeepSeek launches; softer eComm trend in Jan on earlier CNY China's top 400 mobile apps maintained the prior month's momentum with total time spent up +4% yoy in January 2025 (vs. +5% in Dec 2024), with healthy WeChat time spent growth (+4% yoy) and robust Douyin app engagement performance (main app +20% yoy/Lite +17% yoy), slower eCommerce time spent -9% yoy on an earlier ...

证券时报网讯，2月20日，佛山石门中学举行"石门星辰"发布仪式，宣布本土化部署DeepSeek 671B完整 版，成为全国首个实现此技术的高中。 该项目由石门中学校友企业晨曦科技提供技术支持和战略投入，得到了南海区教育局及多方代表的高度 认可。 DeepSeek 671B系统采用本地服务器架构，确保数据安全与低时延，覆盖教学全场景，提供智能教学助 手、深度教学研究等多模块服务，为师生带来本地化、精准化、智能化的教学支持。 这一里程碑事件标志着人工智能技术与基础教育场景的深度整合，为教育数字化转型提供了创新范本。 ...

依托辽宁省电子政务外网，实现"国芯、国算、国模、国盾"四位一体全栈国产化AI能力落地，实现从底 层硬件到上层模型的全方位自主可控，推动人工智能技术在政务服务领域的深度应用，有效提升政府服 务效能和资源利用水平。 证券时报网讯，据"辽宁发布"消息，2月18日，辽宁省数据局（省营商局）在辽宁省政务云成功部署 DeepSeek-R1多版本模型。 辽宁省政务基础设施，在原有国产化芯片、国产算力云平台、国产安全体系基础上，升级部署国产大模 型。 ...

Summary of Newcap's Conference Call Company Overview - Newcap was founded in 2000 by nine teachers and one official from Zhengzhou University, focusing on campus information technology in higher education. The company has expanded its services from cafeteria payment systems to various campus management solutions, serving approximately 1,420 higher education institutions, capturing about 45% of the domestic market share [2][3]. Core Business and Market Position - Newcap has developed a comprehensive smart campus management architecture, including business, data, and technology platforms, serving around 200 universities such as Xi'an University of Technology and Fudan University [2][3]. - The "Vision Perfect Campus" platform has over 10 million authenticated users, with daily active users (DAU) ranging from 260,000 to 300,000 and monthly active users (MAU) between 700,000 and 800,000, providing services like payment and course selection [2][4]. Financial Performance - Due to the impact of the pandemic, Newcap's profit dropped to 108 million yuan in 2023. However, the company maintained stable performance through operational optimization and network expansion [2][5]. - For 2024, Newcap's performance is expected to remain stable, serving over 1,400 institutions, with more than 60% being "985" universities and over 50% being "211" universities [2][21]. AI and Product Development - Newcap is actively exploring AI applications, having launched the digital assistant "Xiao Mei" for intelligent Q&A and service provision. The company is also integrating AI technology into existing products while developing new ones based on original value creation [2][6]. - The company has achieved over 90% accuracy in localized AI Q&A projects, significantly higher than the industry average, and is expanding its product offerings to include various management systems [2][7]. Technological Advancements - Newcap collaborates with various partners, including Huawei and Alibaba, to enhance its AI capabilities and has established AI research labs with multiple universities [4][9]. - The company is currently training its long-sequence reasoning model and upgrading its data insight models to improve internal R&D efficiency [9]. Project Updates - Newcap has launched a fully localized project in Hunan, providing intelligent Q&A, business processing, and data analysis services. A new project in Shanghai is in the delivery phase, utilizing the DeepSeek model for performance testing [10][11]. Market Outlook - The company anticipates a positive market response in 2025, with many projects from 2024 expected to materialize, supported by AI technology [21][22]. - The overall acceptance of AI technology in higher education is high, with institutions eager to adopt new technologies to enhance efficiency [17][19]. Investment and Future Expectations - Newcap has maintained a stable development trajectory since its IPO in 2011, positioning itself as a leader in the education information technology sector. The company aims to continue delivering reliable services to over 10 million university students [22].

Investment Rating - The report does not explicitly state an investment rating for the industry or specific companies involved Core Insights - The market is currently influenced by DeepSeek, which has short-term impacts on the MV chain, while AI agents and applications benefit from high investment in AI infrastructure that has been maintained for nearly two years [3] - The demand for GPU computing power is expected to shift towards more cost-effective AI computing solutions, with domestic options becoming viable alternatives [3][5] - The performance of NVIDIA's supply chain partners is mixed, with high profit margins but challenges in bargaining power [5] - The cancellation of the supercapacitor tray in the GB300 MV link 72 reference design may affect CSP adoption willingness, leading to stock price fluctuations for Musashi and domestic Jianghai [6] Summary by Sections Market Changes and Core Views - The market has been affected by DeepSeek, leading to a temporary setback in the MV chain, while AI applications continue to thrive [3] - High-performance GPUs like NVIDIA's Blackwell and future Grace are still focused on cutting-edge large model fields [3] Supercapacitor Industry Developments - The supercapacitor combines the advantages of traditional capacitors and batteries, offering high power density and energy density, making it suitable for applications in automotive and data centers [3][7] - Musashi Precision Industry, through its subsidiary, is expanding supercapacitor production to meet the demand for peak shaving applications in data centers [3][6] Applications in Data Centers - Supercapacitors are primarily used in data centers for peak shaving and backup power, addressing GPU power fluctuations and enhancing overall system efficiency [10][12] - The integration of supercapacitors in AI supercomputing data centers helps stabilize power supply during training and inference phases, thus improving GPU performance [14] Technological Advantages - Musashi's hybrid supercapacitor (HSC) technology offers superior energy and power density compared to traditional double-layer capacitors, with significant advantages in high-pressure and high-temperature environments [16] Market Outlook - The demand for supercapacitors in data centers is expected to grow, with limited suppliers currently in the market, indicating a concentrated competitive landscape [20] - In optimistic scenarios, each cabinet may require 250 to 300 supercapacitors, translating to a value range of approximately $15,000 to $55,000 [21]

羊城晚报记者 王丹阳 "DeepSeek一周速成硕士论文。" "DeepSeek教你35岁女性如何保持20岁状态。" "评论'777'，现在给你内测机会抢鲜体验DeepSeek红利，让生意更智能更高效。" 蛇年新春，当深度求索DeepSeek在数据垄断突围之战中，以"东方智慧"的低成本高性能技术令全球为之 瞩目后，一边是创新者低调发力技术更新，另一边则是借当红流量密码、围绕AI的"焦虑收割"的流量生 意在国内高调开卖。 卖课件、卖培训、卖"万物皆可DeepSeek"的相关服务……"画皮""套壳"卖卖卖的剧本并不陌生，只是死 灰复燃。 最近一次仅是一年前Sora出现时，"知识二道贩子"就借"不学就慢人一步"的AI焦虑，赚足一波，鼓了腰 包。 尽管两周前DeepSeek 在官方通告中强调，"DeepSeek官方网页端与官方正版 App 内不包含任何广告和 付费项目。目前除 DeepSeek官方用户交流微信群外，我们从未在国内其他平台设立任何群组，一切声 称与 DeepSeek官方群组有关的收费行为均系假冒，请大家仔细辨别，避免财产损失。" 但这些披着AI+ DeepSeek外衣的"速食生意"依旧卖得热火朝天。究竟 ...

记者 刘怡鹤 2月18日，OpenAI的一则意见征求再度透露了其尝试开源的新动向。上周，百度也接连宣布了文心一言 免费、开源的消息。这背后无疑是DeepSeek强大的开源模型带来的压力。 开源模型意味着更低的使用成本、更灵活的部署和开发。自DeepSeek-V3、R1模型爆火以来，各行各业 的公司都在积极接入DeepSeek，而近日微信接入DeepSeek-R1，更让这一开源大模型有了几何级增长的 渗透潜力。 DeepSeek掀起开源浪潮 走开源路线的DeepSeek发布了详细的论文和技术文档，供全球开发者和研究人员使用。这种开源模式 不仅降低了AI技术的使用门槛，也迅速扩大了其全球影响力。 "DeepSeek最大的特点就是开源和开放。它证明了开源不一定比闭源差，而这一点连科技巨头Meta都没 有做到。过去无论是大厂还是创业公司，模型开源的部分还是有限的，特别是最新模型的开源。"国盛 证券研究所所长助理、计算机行业首席刘高畅对记者说。 凭借性能媲美OpenAI o1，但价格仅为其三十分之一的推理模型DeepSeek-R1，DeepSeek一定程度上冲 击了OpenAI的领先地位。长期以来，OpenAI因为不" ...

一石激起千层浪。短短一个月时间，从科技大厂到政务部门、从汽车企业到手机厂商，各行各业都在积 极拥抱DeepSeek。 2月16日，百度搜索和文心智能体平台宣布将全面接入DeepSeek。搜索用户可免费使用DeepSeek和文心 大模型深度搜索功能，文心智能体平台的开发者也将随时调用DeepSeek模型创建并调优智能体。 百度并非第一家接入DeepSeek的搜索平台。近日，部分微信用户发现，微信搜索已经上线"AI搜索"功 能，并接入DeepSeek-R1提供的"深度思考"服务。腾讯方面表示，微信搜一搜在调用混元大模型丰富AI 搜索的同时，已正式灰度测试接入DeepSeek。被灰度到的用户，可在对话框顶部搜索入口，看到"AI搜 索"字样，点击进入后，可免费使用DeepSeek-R1满血版模型，获得更多元化的搜索体验。 同在2月16日，深圳市宣布，基于政务云环境面向全市各区各部门正式提供DeepSeek模型应用服务，实 现了基于DeepSeek的人工智能政务应用一体化赋能升级。而此前，广州、呼和浩特、赣州、无锡等地 政务部门也已宣布部署应用DeepSeek，推动政务服务智能化，涵盖公文写作、民意速办、城市治理等 领 ...

证券时报网讯，DeepSeek金融朋友圈持续扩大。 近日，记者了解到，多家消费金融公司正在推进接入DeepSeek大模型的工作，将大模型应用于产品营 销、风险控制、客户服务、贷后管理等多个场景，进一步为消费金融领域的智能化建设探索创新解决方 案。 业内人士表示，人工智能技术在金融领域的巨大发展前景，为塑造产业新业态打开大门。 同时，人工智能大模型在应用过程中也可能带来数据安全问题，如何确保金融数据的安全存储、传输、 处理，防止数据泄露和滥用，这是包括消费金融公司在内的各类金融机构需要面对的挑战。 ...