来源：新浪基金∞工作室 9月26日，数据港涨5.3%，截至发稿，报41.31元/股，成交13.45亿元，换手率4.73%，总市值296.76亿 元。 风险提示：市场有风险，投资需谨慎。本文为AI大模型自动发布，任何在本文出现的信息（包括但不 限于个股、评论、预测、图表、指标、理论、任何形式的表述等）均只作为参考，不构成个人投资建 议。 数据显示，汇添富基金旗下1只基金位居数据港十大流通股东。中证上海国企ETF（510810）二季度增 持150.75万股，持有股数336.75万股，占流通股的比例为0.47%。根据测算，今日浮盈赚取约700.43万 元。 中证上海国企ETF（510810）成立日期2016年7月28日，最新规模79.42亿。今年以来收益9.41%，同类 排名3568/4220；近一年收益34.72%，同类排名2658/3824；成立以来收益6.14%。 中证上海国企ETF（510810）基金经理为吴振翔。 截至发稿，吴振翔累计任职时间15年237天，现任基金资产总规模184.78亿元，任职期间最佳基金回报 189.86%， 任职期间最差基金回报-31.53%。 资料显示，上海数据港股份有限公司位于 ...

资金流动层面，科技巨头的经营现金流虽然尚能覆盖资本支出，但两者的差距正在缩小。若资本支出继 续超越内生现金流，未来的投资将更依赖于外部融资。这种情况下，AI基础设施建设的财务健康性会 受到挑战，尤其是在风险投资市场陷入窘境的背景下。 巴克莱强调，AI资本支出的放缓不仅影响股市，也可能对更广泛的美国经济产生重大影响。仅在2025 年上半年，数据中心相关投资对美国GDP增长贡献了约1个百分点。如果美国经济因劳动力市场或其他 因素遭遇衰退，而AI投资同步放缓，这种负面共振可能对股市造成比单一行业变动更严重的打击。 分析显示，从2023到2025年间，AI相关股票的涨幅大部分由估值扩张推动。一旦增长故事受到质疑， 估值可能迅速回调。巴克莱的压力测试结果表明，如果数据中心资本支出下降20%，标普500指数的盈 利将受到一定拖累，但估值的下降幅度则更为显著。 近期市场传闻中，人工智能（AI）资本支出放缓被视为可能成为美股的重大隐患。巴克莱股票策略团 队的分析指出，如果数据中心的资本支出在未来两年内减少20%，标普500指数将面临3%至4%的盈利下 行压力，更为严重的是估值可能下降10%至13%。这表明，数据中心资本支出的 ...

Core Viewpoint - The stock price of Shandong High-Speed Holdings Group (0412.HK) experienced a dramatic drop of 76.02% on September 19, 2025, following scrutiny from the Hong Kong Securities and Futures Commission regarding its highly concentrated shareholding structure [2][3]. Group 1: Stock Performance and Market Reaction - Following the initial drop, the stock continued to decline, with a further decrease of 4.39% on September 25, marking the second consecutive day of losses [3]. - The concentrated shareholding structure, where the top two shareholders hold a combined 68.46% and public ownership is only 7.54%, contributes to significant stock price volatility and liquidity issues [4][5]. Group 2: Shareholding Structure and Risks - The major shareholders include Shandong High-Speed Group (43.37%) and JS High Speed Limited (25.09%), with 20 other shareholders holding 24% [5]. - A significant portion of shares (15.78%, approximately 950 million shares) held by a leading public fund has been pledged to banks, creating a risk of forced selling if the stock price continues to decline [6]. Group 3: Company Financials and Business Model - Shandong High-Speed Holdings reported a 506% increase in net profit to 476 million yuan in the first half of 2025, despite an 11.64% decline in revenue to 2.503 billion yuan [8]. - The company operates in investment and financing, focusing on the integration of green energy and computing power, and has announced a share buyback plan of up to $100 million [8]. Group 4: Market Sentiment and Investor Behavior - Investor sentiment is mixed, with some looking to buy the dip while others caution against the risks associated with concentrated shareholding [9]. - The stock's price movements are increasingly detached from fundamental analysis, raising concerns about the long-term value realization and the need for a more optimized shareholding structure [9].

海南自由贸易港政策宣介暨 "向数图强"招商推介会在杭州举办 海南日报杭州9月25日电(海南日报全媒体记者 刘梦晓)9月25日，借助第四届全球数字贸易博览会 平台，海南自由贸易港政策宣介暨"向数图强"招商推介会在浙江杭州举办，旨在向全球展示全岛封关后 的政策红利与投资机遇，吸引数字经济头部企业、国际资本及技术资源落地海南。超400家企业参会， 覆盖电子商务、人工智能、AI算力、数据中心、智能制造等领域。 省委常委、宣传部部长王斌出席推介会并致辞。 近年来，海南自由贸易港数字经济蓬勃发展，全省数字经济年产值已超1000亿元。当前，海南正在 聚焦"五向图强"加快培育发展新质生产力，积极构建具有特色和优势的现代化产业体系，其中之一就 是"向数图强"。海南不仅从制度设计上聚焦促进企业发展，更从服务角度出发，充分利用地理区位优 势、国际海缆基础设施优势和数据安全有序流动制度优势，在游戏出海、来数加工等领域开展了成功的 技术验证和实践探索，形成了可复制、可推广的经验。 推介会现场，海南省商务厅、省财政厅、省委网信办、省工业和信息化厅有关负责人介绍海南"向 数图强"相关工作情况，澄迈、陵水、昌江、保亭、海口复兴城互联网信息产 ...

Core Viewpoint - Barclays' stock strategy team indicates that a 20% decline in data center capital expenditures over the next two years could lead to a 3-4% earnings pressure on the S&P 500 index, with a more severe potential valuation drop of 10-13% [4][24]. Group 1: Risks Identified - The report identifies three major risks that could trigger a crisis: 1. **Efficiency Risk**: Rapid improvements in AI model efficiency may lead to overbuilding of computing facilities, reminiscent of the "dark fiber" tragedy during the dot-com bubble [5][11]. 2. **Physical Limitation Risk**: Increasing electricity shortages are becoming a hard constraint on data center construction, potentially cooling capital expenditures [6][12]. 3. **Liquidity Risk**: If capital expenditures continue to outpace cash flow generation, financing pressures may arise, exacerbated by dwindling venture capital [7][18]. Group 2: AI Investment Landscape - Despite the potential slowdown in data center capital expenditures, the foundation for AI investment remains solid, with a projected annual growth of 30% in capital expenditures [8][9]. - The demand for computing power continues to exceed supply, driven by the proliferation of advanced reasoning models and AI agents [9][11]. Group 3: Economic Impact - Data center-related investments are projected to contribute approximately 1 percentage point to the 1.4% GDP growth in the first half of 2025 [19]. - If the U.S. economy enters a recession due to other factors, the simultaneous slowdown in AI capital expenditures could act as an "accelerator," worsening the situation [21][22]. Group 4: Valuation and Earnings Impact - The impact on earnings (EPS) from a 20% decline in data center capital expenditures is expected to be relatively mild, with a 3-4% drag on the S&P 500 index's EPS for the fiscal year 2026 [24]. - However, the valuation (P/E) impact is projected to be severe, leading to a 10-13% compression in the overall S&P 500 index [25]. - For industries directly benefiting from AI infrastructure, the average P/E compression could reach 15-20% [27].

尽管AI投资主题基础依然稳固，但巴克莱指出数据中心资本支出放缓可能成为美股最大的系统性风险。 据硬AI，9月25日巴克莱股票策略团队发表研报，指出如果数据中心资本支出在未来两年下降20%，标普500指数将面临3-4%的盈利下行 压力，更严重的是估值可能下跌10-13%。 报告指出了可能引爆这场危机的三大潜在"巨雷"： 技术与效率风险：AI模型效率的飞速提升可能导致现有计算设施被"过度建设"，重演科网泡沫时期的"暗光纤"悲剧。 物理限制风险：日益严重的电力短缺正成为数据中心建设无法逾越的"硬墙"，可能强制性地给资本支出降温。 资金流动性风险：当资本支出增长开始超越现金流创造能力时，融资压力和枯竭的VC资本可能成为压垮骆驼的最后一根稻草。 AI热潮根基稳固，但并非无懈可击 首先，报告肯定了AI投资主题的坚实基础。 即便是在一个预计每年增长30%的数十万亿美元资本支出预测下，市场对算力的需求仍然远超供应。 高级推理模型和AI代理的普及进一步推高了需求天花板。数据显示，标普1500指数中，有十分之一的公司在财报中提及AI带来的效率提 升。 报告将当前超级巨头的资本支出/销售额占比（约25%）与科网泡沫时期的电信公司（ ...

"我们也将在完善AIDC产业生态、打造循环经济示范区、攻关行业大规模存储难题等方面展开深度合 作。这次合作也进一步巩固了公司在国内算力产业发展中的地位。"上述工作人员说道。 此外，紫光股份有限公司（以下简称"紫光股份"）也在不断加码智算中心建设。 当前，智算中心建设呈现出快速发展态势。"众多国家和地区都在积极推动相关项目的落地实施，以满 足日益增长的AI研究和应用需求。"盘古智库（北京）信息咨询有限公司高级研究员余丰慧对《证券日 报》记者表示。 据甬兴证券研报，在多种因素共同驱动下，预计2028年中国智算中心市场投资规模有望达2886亿元。 面对持续攀升的市场需求，智算中心加速建设，相关上市公司积极布局。 北京秦淮数据有限公司（以下简称"秦淮数据"）工作人员对《证券日报》记者表示，秦淮数据在今年8 月份与宁夏回族自治区中卫市人民政府在北京签署战略合作框架协议。根据协议，秦淮数据计划投资 240亿元，在中卫市建设3座零碳超大规模智算园区，项目建成后的园区总容量可达1.2GW。 9月24日至26日，2025云栖大会在杭州云栖小镇举行。期间，阿里巴巴集团董事兼首席执行官吴泳铭表 示，公司正积极推进3800亿元的AI ...

Group 1 - Alibaba Group is actively advancing its AI infrastructure construction with an investment of 380 billion yuan and plans for larger investments [1] - The competition among technology companies in AI infrastructure is intensifying, benefiting the development of related industries such as data center construction, chip manufacturing, and software development [1] - The proportion of intelligent computing power in China's total computing power is expected to increase from 3% in 2016 to 35% by 2025 [1] Group 2 - The investment scale of China's intelligent computing center market is expected to reach 288.6 billion yuan by 2028 [2] - Beijing Qinhuai Data Co., Ltd. plans to invest 24 billion yuan to build three zero-carbon super-large intelligent computing parks in Zhongwei City, with a total capacity of 1.2 GW [3] - Unisplendour Corporation is expanding its intelligent computing center construction and has established strategic cooperation with several provinces, with seven Turing Towns currently built or under construction [4] Group 3 - Unisplendour has launched a high-density liquid cooling cabinet solution that supports 64 high-performance AI accelerator cards or thousand-core CPUs, meeting the stringent energy consumption requirements of high-computing clusters [4]

Core Viewpoint - The article discusses the growing importance of liquid cooling technology in data centers due to increasing power consumption and heat generation from AI server chips, highlighting the shift from traditional air cooling to liquid cooling solutions. Group 1: Liquid Cooling Technology Types - Liquid cooling technology is categorized into two main types: cold plate and immersion cooling, with cold plate cooling holding a 91% market share in 2022 due to its maturity and compatibility [3][31]. - Immersion cooling, while having the highest heat dissipation capability, accounts for about 8% of the market and is represented by companies like Shuguang Data and Guangdong He Yi [4][31]. - Cold plate cooling primarily uses water-based coolants, offering advantages such as low cost and fast heat dissipation [5]. Group 2: Cooling Liquid Material Systems - The main types of cooling liquids include water-based, oil-based, and fluorinated coolants [7]. - Water-based coolants are mainly used in single-phase cold plate cooling, supplied by companies like Compton and Unification [8]. - Oil-based coolants are used in single-phase immersion cooling, with suppliers like China Petroleum and Unification [8]. - Fluorinated coolants are crucial for high reliability and extreme physical conditions, with companies like Juhua and Dongyue Group achieving large-scale production [8]. Group 3: Microchannel Liquid Cooling Breakthroughs - Microchannel cold plate technology is becoming a focus for AI chip manufacturers, with companies like Nvidia and Microsoft pushing for MLCP standards [9]. - Domestic companies such as Highlan and Feirongda are gaining certifications and expanding their market share in this area [9]. Group 4: Market Size and Price Logic - By 2028, the global AI data center installation capacity is expected to reach 59 GW, corresponding to a cooling liquid demand exceeding 89,000 tons [10]. - The Chinese market for liquid-cooled data centers is projected to grow at a compound annual growth rate (CAGR) of 51.4%, with the industry scale expected to exceed 100 billion yuan by 2027 [10]. Group 5: Leading Projects - Major companies like Alibaba, Shuguang, Huawei, and Lenovo are implementing liquid-cooled data center demonstration projects, with Alibaba achieving a PUE of 1.12 in its 5A green liquid-cooled data center [14][15]. - Shuguang has realized the world's first large-scale commercial application of immersion phase change liquid cooling [15]. - Huawei's cloud data center in Ulanqab utilizes indirect evaporative cooling, achieving a PUE of 1.15 [16]. Group 6: Mainstream Cooling Pathways - The article outlines the mainstream cooling pathways, including water-based, oil-based, and fluorinated cooling liquids, emphasizing the rapid growth of the liquid cooling market [18][21].

中国日报9月25日陵水电（记者 王卓琼 陈博文 马萌萌）在海南省陵水黎族自治县的清水湾30米至40米深的海底，一座海底机房正安静地运转着，这标志着 全球首个商用海底数据中心的成功运营，为我国"陆数海算"开启零碳算力新模式。 海南省提出了海洋经济发展的"十四五"规划，明确提出要促进海底数据中心研发及建设。而北京海兰云海底数据中心项目正是由海南省国资委牵头引进，落 地陵水的海洋新基建类项目。 项目自2022年启动，分三期进行，计划完成100个数据舱的建设。建成后，将为"东数西算"提供重要补充，同时向东南亚等"一带一路"国家延伸，为海南数 字经济发展提供支撑。 目前项目一期于2023年11月建设完成，今年2月海南海底智算中心集群启用。 公司目前聚焦海上风电融合型海底数据中心研发与建设。 随着数据公司需要时延要求越来越严格，物理距离不能过长，大部分发达城市都聚焦在沿海地区，科技公司也聚焦在这些发达城市。相较于陆地数据中心， 海底数据中心海底数据中心由岸站、海底光电复合缆、海底分电站及海底数据舱组成，使用海水自然冷却，具有PUE值低、能效高、模块化建设、 快捷部 署（90天）、布放海底节约土地以及高可靠性免维护等优点。 ...