图片来源：Photo illustration by Getty Images 夏日渐去、秋日将至，众多科技界人士开始担忧寒冬的到来。上月末，彭博社专栏作家提出疑问：“人工智能寒冬终于要来了吗？”英国《每日电讯报》则态度更为笃定：“下一轮人工智能寒冬即将来临”。与此同时，社交平台X上关于“人工智能寒冬或将来临”的讨论也甚嚣尘上。 “人工智能寒冬”是人工智能领域从业者用以指代特定时期的称谓：这一时期，大众对“机器能如人类般学习、思考”这一理念的热情渐趋冷却，对人工智能产品、企业及研究的投资也随之枯竭。这一词汇之所以频繁从人工智能评论员的口中说出，实则有其深刻缘由——在长达70年的人工智能研究历史中，我们已经历过数次“寒冬”。倘若如部分人所忧虑的那样，我们即将步入新一轮“寒冬”，那么这至少将是第四次。 近期关于寒冬将至的讨论，根源在于投资者愈发忧虑人工智能技术可能无法达成炒作营造出的预期，且诸多人工智能相关公司估值过高。在最糟糕的情况下，这场人工智能寒冬可能伴随着人工智能概念催生的股市泡沫的破裂，进而对整个经济产生影响。虽然此前也曾经历人工智能炒作周期，但从未有哪次像本轮生成式人工智能热潮这样，吸引投资者投入高达数千亿美元资金。因此，若新一轮“寒冬”到来，其冲击力或将如极地涡旋般猛烈。 近期OpenAI首席执行官萨姆·奥尔特曼（Sam Altman）的言论引发了市场恐慌。他向记者坦言部分风投支持的人工智能初创企业估值严重虚高（当然，OpenAI不在此列，它是史上估值最高的风投支持型初创企业之一）。随后，麻省理工学院发布的研究报告指出95%的人工智能试点项目以失败告终。 回顾过往的人工智能寒冬及其成因，或许能帮助我们判断当前空气中的“寒意”究竟只是一阵短暂的凉风，还是“冰河时代”即将来临的先兆。有时人工智能寒冬是由学术研究揭示特定人工智能技术的局限性引发的；有时则因人工智能技术在现实应用中屡屡受挫；有时两者兼而有之。但历次人工智能寒冬的共同之处在于：当承载厚望的新进展未能兑现炒作所赋予的过高期待时，出资方便会陷入幻灭。 第一轮人工智能炒作周期 冷战初期，美国及其盟国政府便在人工智能研究领域投入了巨额资金。彼时与当下情形一样，美国政府将这项技术视为可能带来战略和军事优势的领域，因此美国国防部提供了大部分人工智能研究经费。 当时，人工智能领域存在两种对立的方法论。其一，是借助硬编码逻辑规则，将输入数据分类为符号，再通过对这些符号进行操控来得出输出结果。依靠此方法，计算机在跳棋、国际象棋领域首次取得重大突破，世界上首批聊天机器人也由此诞生。 另一种方法则基于感知器技术——即当今神经网络的前身，是大致模仿大脑运行机制的人工智能。感知器并非从规则和逻辑出发，而是通过数据学习归纳完成特定任务的规则。美国海军研究办公室为感知器的早期研究提供了大量资金支持，而康奈尔大学神经科学家兼心理学家弗兰克·罗森布拉特（Frank Rosenblatt）是该技术的开创者。美国海军和中央情报局（CIA）均对感知器进行了测试，试图验证其能否对目标进行分类——例如识别敌舰轮廓，或辨别航空侦察照片中的潜在目标。 两大对立阵营都高调宣称，其技术将迅速催生出与人类智力相当甚至超越人类智力的计算机。1958年，罗森布拉特在接受《纽约时报》采访时表示，他研发的感知器很快就能识别人脸并喊出其姓名，距离实现即时语言翻译“仅一步之遥”，最终人工智能系统还将具备自我复制能力并拥有意识。与此同时，麻省理工学院人工智能实验室联合创始人、符号人工智能阵营领军人物马文·明斯基（Marvin Minsky）在1970年接受《生活》杂志采访时宣称：“未来三到八年内，我们将拥有具备普通人类通用智能的机器。” 这正是人工智能寒冬的首要前提：炒作。如今多位人工智能领域知名人士的言论与彼时存在明显的相似之处。今年1月，OpenAI首席执行官萨姆·奥尔特曼在其个人博客中写道：“我们如今笃定——已掌握构建具备传统意义上人类水平的通用人工智能的方法”，并表示OpenAI正日益将重心转向研发超越人类的“超级智能”。他还称，今年“我们可能见证首批人工智能代理'加入劳动力队伍'，并切实改变企业的产出”。Anthropic联合创始人兼首席执行官达里奥·阿莫迪（Dario Amodei）曾预测，具备人类水平的人工智能将于2026年问世。与此同时，谷歌DeepMind联合创始人兼首席执行官戴密斯·哈萨比斯（Demis Hassabis）则表示，在所有认知领域均达到人类水平的人工智能将在未来“五到十年内”诞生。 政府失去信心 但引发人工智能寒冬的，是部分确凿证据表明炒作的愿景无法兑现。第一次寒冬的爆发源于一系列沉重打击：1966年，受美国国家研究委员会（National Research Council）委托的委员会发布了一份关于自然语言处理和机器翻译现状的负面报告，结论是计算机翻译比人工翻译成本更高、速度更慢且准确性更低。该委员会此前已为早期语言人工智能研究投入2000万美元（按如今币值计算，至少相当于2亿美元），随后便停止了所有资金支持。 随后在1969年，明斯基又挥出第二记重拳。这一年，他与人工智能研究者西蒙·派珀特（Seymour Papert）合著的专著对感知器进行了全面批判。在书中，明斯基与派珀特通过数学论证证明：单层感知器（如罗森布拉特1958年高调展示的那种）仅能进行精确的二元分类——换言之，它只能识别物体是黑是白、是圆是方，却无法将事物归入两个以上的类别。 事后证明，明斯基与派珀特的批判存在重大缺陷。尽管多数人将此书视为基于神经网络的人工智能永远无法企及人类智力水平的决定性证据，但他们的论证仅适用于结构简单的单层感知器：输入层由若干接收数据的神经元构成，且所有输入层神经元仅与一个输出层神经元相连。他们很可能刻意忽略了这样一个事实：早在1960年代，部分研究者已开始探索多层感知器——这种感知器在输入层神经元与输出层神经元之间增设了一个由神经元构成的中间“隐藏层”。作为当今“深度学习”技术的真正前身，多层感知器实际上具备将数据归入两个以上类别的能力。但当时训练这种多层神经网络难度极大。而这已无关紧要——损害已然造成。明斯基与派珀特的著作出版后，美国政府对基于神经网络的人工智能方法的资金支持基本终止。 明斯基与派珀特的批判不仅说服了美国国防部的资助机构，还让众多计算机科学家相信神经网络研究已走入死胡同。部分神经网络研究者甚至指责明斯基使该领域的发展倒退了数十年。2006年，曾助力重新点燃神经网络研究热情的研究员特伦斯·谢诺夫斯基（Terry Sjenowski）在一次会议上公开质问明斯基：“你是魔鬼吗？”明斯基无视提问，转而详细阐述他眼中神经网络存在的缺陷。谢诺夫斯基继续追问，恼怒的明斯基大声回应道：“没错，我就是！” 但明斯基代表的符号人工智能，很快也面临资金短缺的困境。同样是在1969年，美国国会强制要求曾为两种人工智能研究方法提供大量资金支持的美国国防部高级研究计划局（Defense Advanced Research Project Agency，DARPA）改变拨款方式。该机构被告知要资助那些具有明确军事应用场景的研究项目，而非更侧重理论探索的“蓝天研究”（指无明确实用目标、纯基础领域的研究）。尽管部分符号人工智能研究符合这一标准，但大多数研究并不符合。 1973年，致命一击降临：英国议会委托剑桥大学数学家詹姆斯·莱特希尔（James Lighthill）对英国人工智能研究现状展开调查。他在结论中指出，在实现与人类智力水平相当这一宏大目标上，人工智能未能显露出任何希望，其推崇的诸多算法虽能解决“玩具级问题”（指简单模拟场景中的问题），却永远无法应对现实世界的复杂性。基于莱特希尔的这一结论，英国政府终止了对人工智能研究的所有资金支持。 尽管莱特希尔的调查仅聚焦于英国的人工智能研究，但美国国防部高级研究计划局以及其他资助人工智能研究的美国机构均注意到了这一结论，这进一步加深了他们对人工智能的怀疑态度。到1974年，美国对人工智能项目的资助额仅为1960年代的零头。人工智能寒冬就此降临，并一直持续到20世纪80年代初。 如今，当研究表明人工智能未能达到预期时，也出现了与第一次人工智能寒冬相似的情形。苹果公司与亚利桑那州立大学近期发表的两篇研究论文，对前沿人工智能模型是否真正具备推理能力提出质疑——这些模型本应通过“思维链”推理如何回应提示词。两篇论文均得出一致结论：这些模型并未像人类理解的推理那样，学习如何将可泛化的逻辑规则和问题解决技巧用于解决新问题，而仅仅是试图将当前问题与训练数据中出现过的问题进行匹配。这些研究或许会成为当代版“明斯基与派珀特批判感知器”的标志性事件。 与此同时，关于当前人工智能模型实际影响的研究正日益增多，这类研究与莱特希尔报告及美国国家研究委员会的报告类似。例如，麻省理工学院的一项研究得出结论，95%的人工智能试点项目未能推动企业营收增长。赛富时（Salesforce）研究人员近期发布的研究发现，当前多数大型语言模型无法准确执行客户关系管理（CRM）任务——这一结论颇具讽刺意味，因为赛富时自身正大力推广人工智能代理，以实现客户关系管理流程自动化。Anthropic的研究表明，其Claude模型无法成功运营自动售货机业务——相较于科技鼓吹者宣称将被人工智能代理“彻底颠覆”的众多业务，这已是相对简单的业务。人工智能研究机构METR的研究还揭示：实际上，相较于不借助人工智能编程助手的情况，使用这类工具的软件开发人员，完成任务的速度降低19%。 但存在部分关键差异。最显著的是，当前的人工智能热潮并不依赖公共资金。尽管包括美国军方在内的政府机构正成为人工智能企业的重要客户，但推动当前热潮的资金几乎完全来自私营领域。自2022年11月ChatGPT推出以来，风险投资机构已向人工智能初创企业投入至少2500亿美元。这还不包括微软、谷歌母公司Alphabet、亚马逊和Meta等大型上市科技公司在自身人工智能项目上的巨额投入。仅今年一年，用于建设人工智能数据中心的支出就高达3500亿美元，预计明年这一数字还会进一步攀升。 此外，与第一次人工智能寒冬时期人工智能系统主要停留在研究实验阶段不同，如今人工智能已在各行业广泛部署。人工智能还成为一项规模庞大的消费技术——仅ChatGPT的周用户量就达7亿——这在以往是从未有过的情况。尽管当今的人工智能似乎仍缺乏人类智能的某些关键要素，但相较于过去的人工智能系统已有显著进步，而且人们确实发现这项技术在大量任务中具有实用价值，这一点毋庸置疑。 第二次人工智能寒冬：企业失去耐心 第一次人工智能寒冬在20世纪80年代初逐渐消退，这主要归功于计算能力的提升和算法技术的改进。这一时期，人工智能领域的炒作主要集中在“专家系统”上——这类计算机程序旨在将特定领域人类专家的知识编码为逻辑规则集，软件根据这些规则执行特定任务。 尽管如此，企业界仍热情高涨，认为专家系统将推动生产力大幅提升。在这轮人工智能炒作周期的鼎盛阶段，近三分之二的《财富》美国500强企业宣称已部署专家系统。到1985年，美国企业在这方面的总投入已超过10亿美元，围绕该技术的完整产业也应运而生，其中大部分得到了风险投资的支持。大部分资金用于研发名为LISP机的专用计算机硬件，这些硬件经过优化可运行专家系统——其中许多系统正是用LISP编程语言编写的。此外，自1983年起，美国国防高级研究计划局通过新推出的“战略计算计划”重新资助人工智能研究，最终向全美多所大学的90余个人工智能项目投入逾1亿美元资金。 尽管专家系统借鉴了符号人工智能研究者开创的诸多方法，但许多计算机科学领域的学者担忧，过高的期望值将再次引发“繁荣-萧条”周期，进而对该领域的发展造成损害。明斯基和人工智能研究学者罗杰·尚克（Roger Schank）在1984年的一场人工智能会议上创造了“人工智能寒冬”这一术语。他们选用这个新词，意在呼应“核冬天”——大规模核战争后可能出现的、不见天日的毁灭性萧条时期。 随后发生的三件事引发了新一轮寒冬。1987年，太阳计算机系统公司（Sun Microsystems）推出新型计算机工作站。这类工作站，以及IBM和苹果推出的性能日益强大的台式机，使得专用LISP机变得不再必要。不到一年时间，LISP机市场便彻底崩塌。许多风险投资机构血本无归，从此对人工智能初创企业避之不及。同年，纽约大学计算机科学家杰克·施瓦茨（Jack Schwartz）出任美国国防部高级研究计划局计算研究部门负责人。他向来对人工智能持否定态度，尤其反对专家系统，随即大幅削减相关经费。 与此同时，企业逐渐发现专家系统的构建与维护成本高昂且难度极大。这类系统还存在“脆弱性”——虽能高效处理高度常规化任务，但遇到稍有异常的情况，就难以应用预设的逻辑规则。此时，系统往往会输出怪异且不准确的结果，甚至直接彻底崩溃。事实证明，要制定出能覆盖所有极端情况的规则，是一项不可能完成的任务。因此到20世纪90年代初，企业开始放弃专家系统。与首次人工智能热潮中科学家和政府资助方对技术产生质疑不同，第二次寒冬的主要推手是企业的失望情绪。 如今人工智能领域的发展，与彼时存在明显的相似之处。例如，微软、Alphabet、亚马逊云科技、埃隆·马斯克的X.ai以及Meta正斥资数千亿美元建设人工智能数据中心。OpenAI正与软银、甲骨文及其他投资者共同推进耗资5000亿美元的“星门计划”数据中心项目。英伟达之所以能凭借4.3万亿美元市值成为全球市值最高的公司，很大程度上是因为其生产的人工智能芯片满足了数据中心的需求。数据中心热潮背后的核心假设之一是：最前沿的人工智能模型，其规模即便不比现有顶尖模型更大，至少也会与之相当。而训练和运行这类规模的模型，需要极其庞大的数据中心支持。 然而与此同时，多家初创企业已找到巧妙方法，成功研发出规模小得多却能模拟大型模型诸多功能的模型，且所需计算资源远少于后者，有些甚至无需使用英伟达生产的专用人工智能芯片，规模小到可在智能手机上运行。若这一趋势持续下去，那些巨型数据中心可能会变得不再必要——就像当年LISP机被证明并非必需品一样。这意味着，投入人工智能基础设施的数千亿美元资金，最终可能沦为“搁浅资产”。 当今的人工智能系统在诸多方面比20世纪80年代的专家系统更强大、更灵活。但企业仍发现其部署过程复杂且成本高昂，投资回报往往难以捉摸。尽管当下的人工智能模型比专家系统更具通用性与韧性，但依旧不可靠，尤其是在处理训练数据未充分覆盖的特殊案例时。它们容易产生幻觉，会笃定地输出错误信息，有时甚至会犯人类绝不会犯的错误。这意味着企业和政府无法将人工智能用于关键任务流程自动化。企业是否会像当年对专家系统那样，对生成式人工智能和大型语言模型失去耐心，目前尚难预料，但这种情况确实存在发生的可能性。 第三次人工智能寒冬：神经网络的兴衰与复兴 20世纪80年代，另一种人工智能方法——神经网络也重新引发关注，这在一定程度上得益于大卫·莱姆哈特（David Rumelhart）、杰弗里·辛顿（Geoffrey Hinton）和罗纳德·威廉姆斯（Ronald Williams）的研究。1986年，他们成功找到了破解自20世纪60年代以来便一直困扰多层感知器的关键难题的方法。他们的创新成果被称为反向传播（backpropagation，简称backprop），这种方法能在每次训练过程中对中间“隐藏层”神经元的输出结果进行修正，从而让整个神经网络实现高效学习。 反向传播算法，再加上性能更强大的计算机，共同推动了神经网络的复兴。很快，研究人员构建的多层神经网络便具备多种能力：能识别信封和支票上的手写字母、分析家谱中人物的亲属关系、识别打印字符并通过语音合成器朗读，甚至能为早期自动驾驶汽车导航，使其保持在高速公路车道内行驶。 这在20世纪80年代末引发了短暂的神经网络热潮。但神经网络也存在显著缺陷：训练过程需要海量数据，而许多任务根本无法获取所需的海量数据；在当时的计算机硬件条件下，训练速度极慢，有时运行过程中会出现迟滞。 这意味着神经网络仍存在大量无法完成的任务。与当初企业争先恐后地采用专家系统不同，如今企业并未急于引入神经网络——因其应用场景似乎极为受限。与此同时，其他统计机器学习技术正快速取得进展，这些技术所需数据量更少、对计算能力要求更低。如此一来，许多人工智能研究者和工程师再次对神经网络失去信心，又一个长达十年的人工智能寒冬来临。 推动第三次寒冬回暖，有两大因素发挥作用：其一，互联网产生了海量数字数据，且获取这些数据变得相对轻松，这解决了20世纪80年代神经网络发展面临的数据瓶颈问题；其二，自2004年起，先是马里兰大学的研究者，随后是微软的研究者，开始尝试使用“专为电子游戏设计的新型计算机芯片”——图形处理器（GPU）——来训练和运行神经网络。图形处理器具备并行执行大量相同运算的能力，而这恰恰契合了神经网络的运算需求。很快，杰弗里·辛顿及其研究生证明：基于海量数据集训练的、在图形处理器上运行的神经网络，能够完成诸如将图像分类为上千种类别等任务——这在20世纪80年代末是不可能实现的任务。现代“深度学习”革命就此拉开序幕。 这场热潮至今仍在持续。最初，对神经网络的训练多以实现单一特定任务为核心目标——下围棋或人脸识别。但2017年谷歌研究人员设计出名为转换器的特殊神经网络，它擅长解析语言序列，这一突破将人工智能的盛夏推向了更深层次。2019年，OpenAI的一项研究让这股热潮再获助力——他们发现，依托海量文本数据完成训练的转换器模型，不仅具备生成高质量文本的能力，还能掌握翻译、摘要等多种语言任务。三年后，基于该模型的神经网络升级版GPT-3.5，成为风靡全球的聊天机器人ChatGPT的核心引擎。 如今ChatGPT推出三年后，人工智能的炒作热度空前高涨。若以过往人工智能寒冬为参照，如今确实出现若干秋日征兆——随风飘落的零星落叶。这究竟是“又一场将让人工智能投资陷入长达一代人冰封期的极寒风暴”的前奏，还是“阳光重现前短暂的寒流”，唯有时间才能给出答案。（财富中文网） 译者：中慧言-王芳 As summer fades into fall, many in the tech world are worried about winter. Late last month, a Bloomberg columnist asked “is the AI winter finally upon us?” British newspaper The Telegraph was more definitive. “The next AI winter is coming,” it declared. Meanwhile, social media platform X was filled with chatter about a possible AI winter. An “AI winter” is what folks in artificial intelligence call a period in which enthusiasm for the idea of machines that can learn and think like people wanes—and investment for AI products, companies, and research dries up. There’s a reason this phrase comes so naturally to the lips of AI pundits: We’ve already lived through several AI winters over the 70-year history of artificial intelligence as a research field. If we’re about to enter another one, as some suspect, it’ll be at least the fourth. The most recent talk of a looming winter has been triggered by growing concerns among investors that AI technology may not live up to the hype surrounding it—and that the valuations of many AI-related companies are far too highl. In a worst case scenario, this AI winter could be accompanied by the popping of an AI-inflated stock market bubble, with reverberations across the entire economy. While there have been AI hype cycles before, they’ve never involved anything close to the multiple hundreds of billions of dollars that investors have sunk into the generative AI boom. And so if there is another AI winter, it could involve polar vortex levels of pain. The markets have been spooked recently by comments from OpenAI CEO Sam Altman, who told reporters he thought some venture-backed AI startups were grossly overvalued (although not OpenAI, of course, which is one of the most highly-valued venture-backed startups of all time). Hot on the heels of Altman’s remarks came a study from MIT that concluded that 95% of AI pilot projects fail. A look at past AI winters, and what caused them, may give us some indication of whether that chill in the air is just a passing breeze or the first hints of an impending Ice Age. Sometimes those AI winters have been brought on by academic research highlighting the limitations of particular AI techniques. Sometimes they have been caused by frustrations getting AI tech to work well in real world applications. Sometimes both factors have been at play. But what previous AI winters all had in common was disillusionment among those footing the bill after promising new advances failed to deliver on the ensuing hype. The first AI hype cycle The U.S. and allied governments lavishly funded artificial intelligence research throughout the early days of the Cold War. Then, as now, Washington saw the technology as potentially conferring a strategic and military advantage, and much of the funding for AI research came from the Pentagon. During this period, there were two competing approaches to AI. One was based on hard-coding logical rules for categorizing inputs into symbols and then for manipulating those symbols to arrive at outputs. This was the method that yielded the first great leaps forward in computers that could play checkers and chess, and also led to the world’s first chatbots. The rival AI method was based on something called a perceptron, which was the forerunner of today’s neural networks, a kind of AI loosely built on a caricature of how the brain works. Rather than starting with rules and logic, a perceptron learned a rule for accomplishing some task from data. The U.S. Office of Naval Research funded much of the early work on perceptrons, which were pioneered by Cornell University neuroscientist and psychologist Frank Rosenblatt. Both the Navy and the CIA tested perceptrons to see if they could classify things like the silhouettes of enemy ships or potential targets in aerial reconnaissance photos. The two competing camps both made hyperbolic claims that their technology would soon deliver computers that equalled or exceeded human intelligence. Rosenblatt told The New York Times in 1958 that his perceptrons would soon be able to recognize individuals and call out their names, that it was “only one more step of development” before they could instantly translate languages, and that eventually the AI systems would self-replicate and become conscious. Meanwhile Marvin Minsky, cofounder of MIT’s AI Lab and a leading figure in the symbolic AI camp, told Life magazine in 1970 that “in three to eight years we will have a machine with the general intelligence of an average human being.” That’s the first prerequisite for an AI winter: hype. And there are clear parallels today in statements made by a number of prominent AI figures. Back in January, OpenAI CEO Sam Altman wrote on his personal blog that “we are now confident we know how to build [human-level artificial general intelligence] as we have traditionally understood it” and that OpenAI was turning increasingly towards building super-human “superintelligence.” He wrote that this year “we may see the first AI agents ‘join the workforce’ and materially change the output of companies.” Dario Amodei, the cofounder and CEO of Anthropic, has said the human-level AI could arrive in 2026. Meanwhile, Demis Hassabis, the cofounder and CEO of Google DeepMind, has said that AI matching humans across all cognitive domains would arrive in the next “five to 10 years.” Government loses faith But what precipitates an AI winter is some definitive evidence this hype cannot be met. For the first AI winter, that evidence came in a succession of blows. In 1966, a committee commissioned by the National Research Council issued a damning report on the state of natural language processing and machine translation. It concluded that computer-based translation was more expensive, slower and less accurate than human translation. The research council, which had provided $20 million towards this early kind of language AI (at least $200 million in today’s dollars), cut off all funding. Then, in 1969, Minsky was responsible for a second punch. That year, he and Seymour Papert, a fellow AI researcher, published a book-length takedown of perceptrons. In the book, Minsky and Papert proved mathematically that a single layer perceptron, like the kind Rosenblatt had shown off to great fanfare in 1958, could only ever make accurate binary classifications—in other words, it could identify if something were black or white, or a circle or a square. But it could not categorize things into more than two buckets. It turned out there was a big problem with Minsky’s and Papert’s critique. While most interpreted the book as definitive proof that neural network-based AI would never come close to human-level intelligence, their proofs applied only to a simple perceptron that had just a single layer: an input layer consisting of several neurons that took in data, all linked to a single output neuron. They had ignored, likely deliberately, that some researchers in the 1960s had already begun experimenting with multilayer perceptrons, which had a middle “hidden” layer of neurons that sat between the input neurons and output neuron. True forerunners of today’s “deep learning,” these multilayer perceptrons could, in fact, classify data into more than two categories. But at the time, training such a multilayer neural network was fiendishly difficult. And it didn’t matter. The damage was done. After the publication of Minsky’s and Papert’s book, U.S. government funding for neural network-based approaches to AI largely ended. Minsky’s and Papert’s attack didn’t just persuade Pentagon funding bodies. It also convinced many computer scientists too that neural networks were a dead end. Some neural network researchers came to blame Minsky for setting back the field by decades. In 2006, Terry Sjenowski, a researcher who helped revive interest in neural networks, stood up at a conference and confronted Minsky, asking him if he were the devil. Minsky ignored the question and began detailing what he saw as the failings of neural networks. Sjenowski persisted in asking Minsky again if he were the devil. Eventually an angry Minsky shouted back: “Yes, I am!” But Minsky’s symbolic AI soon faced a funding drought too. Also in 1969, Congress forced the Defense Advanced Research Project Agency (DARPA), which had been a major funder of both AI approaches, to change its approach to issuing grants. The agency was told to fund research that had clear, applied military applications, instead of more blue-sky research. And while some symbolic AI research fit this rubric, a lot of it did not. The final punch came in 1973, when the U.K. parliament commissioned Cambridge University mathematician James Lighthill to investigate the state of AI research in Britain. His conclusion was that AI had failed to show any promise of fulfilling its grand claims of equaling human intelligence and that many of its favored algorithms, while they might work for toy problems, could never deal with the real world’s complexity. Based on Lighthill’s conclusions, the U.K. government curtailed all funding for A.I. research. Lighthill had only looked at U.K. AI efforts, but DARPA and other U.S. funders of AI research took note of its conclusions, which reinforced their own growing skepticism of AI. By 1974, U.S. funding for AI projects was a fraction of what it had been in the 1960s. Winter had set in—and it would last until the early 1980s. Today, too, there are parallels with this first AI winter when it comes to studies suggesting AI isn’t meeting expectations. Two recent research papers from researchers at Apple and Arizona State University have cast doubt on whether the cutting edge AI models, which are supposed to use a “chain of thought” to reason about how to answer a prompt, are actually engaging in reasoning at all. Both papers conclude that rather than learning to apply generalizable logical rules and problem-solving techniques to new problems—which is what humans would consider reasoning—the models simply try to match a problem to one seen in its training data. These studies could turn out to be the equivalent of Minsky’s and Papert’s attack on perceptrons. Meanwhile, there are also a growing number of studies on the real world impact of today’s AI models that parallel the Lighthill and NRC reports. For instance, there’s that MIT study which concluded 95% of AI pilots are failing to boost corporate revenues. There’s a recent study from researchers at Salesforce that concluded most of today’s large language models (LLMs) cannot accurately perform customer relation management (CRM) tasks—a particularly ironic conclusion since Salesforce itself has been pushing AI agents to automate CRM processes. Anthropic research showed that its Claude model could not successfully run a vending machine business—a relatively simple business compared to many of those that tech boosters say are poised to be “utterly transformed” by the AI agents. There’s also a study from the AI research group METR that showed software developers using an AI coding assistant were actually 19% slower at completing tasks than they were without it. But there are some key differences. Most significantly, today’s AI boom is not dependent on public funding. Although government entities, including the U.S. military, are becoming important customers for AI companies, the money fueling the current boom is almost entirely private. Venture capitalists have invested at least $250 billion into AI startups since ChatGPT debuted in November 2022. And that doesn’t include the vast amount being spent by large, publicly-traded tech companies like Microsoft, Alphabet, Amazon, and Meta on their own AI efforts. An estimated $350 billion is being spent to build out AI data centers this year alone, with even more expected next year. What’s more, unlike in that first AI winter, when AI systems were mostly just research experiments, today AI is being widely deployed across businesses. AI has also become a massive consumer technology—ChatGPT alone is thought to have 700 million weekly users—which was never the case previously. While today’s AI still seems to lack some key aspects of human intelligence, it is a lot better than systems that existed previously and it is hard to argue that people are not finding the technology useful for a good number of tasks. Winter No. 2: Business loses patience That first AI winter thawed in the early 1980s thanks largely to increases in computing power and some improved algorithmic techniques. This time, much of the hype in AI was around “expert systems”. These were computer programs that were designed to encode the knowledge of human experts in a particular domain into a set of logical rules which the software would then apply to accomplish some specific task. Nevertheless, business was enthusiastic, believing expert systems would lead to a productivity boom. At the height of this AI hype cycle, nearly two-thirds of the Fortune 500 said they had deployed expert systems. By 1985, U.S. corporations were collectively spending more than $1 billion on expert systems and an entire industry, much of it backed by venture capital, sprouted up around the technology. Much of it was focused on building specialized computer hardware, called LISP machines, that were optimized to run expert systems, many of which were coded in the programming language LISP. What’s more, starting in 1983, DARPA returned to funding AI research through the new Strategic Computing Initiative, eventually offering over $100 million to more than 90 different AI projects at universities throughout the U.S. Although expert systems drew on many of the methods symbolic AI researchers pioneered, many academic computer scientists were wary that inflated expectations would once again precipitate a boom and bust cycle that would hurt the field. Among them were Minsky and fellow AI researcher Roger Schank who coined the term “AI winter” during an AI conference in 1984. The pair chose the neologism to echo the term “nuclear winter”—the devastating and bleak period without sunlight that would likely follow a major nuclear war. Three things then happened to bring about the next winter. In 1987, a new kind of computer workstation debuted from Sun Microsystems. These workstations, as well as increasingly powerful desktop computers from IBM and Apple, obviated the need for specialized LISP machines. Within a year, the market for LISP machines evaporated. Many venture capitalists lost their shirts—and became wary of ever backing AI-related startups again. That same year, New York University computer scientist Jack Schwartz became head of DARPA’s computing research. He was no fan of AI in general or expert systems in particular, and slashed funding for both. Meanwhile, businesses gradually discovered that expert systems were difficult and expensive to build and maintain. They were also “brittle”—while they could handle highly routinized tasks well, when they encountered slightly unusual cases, they struggled to apply the logical rules they had been given. In such cases, they often produced bizarre and inaccurate outputs, or simply broke down completely. Delineating rules that would apply to every edge case proved an impossible task. As a result, by the early 1990s, companies were starting to abandon expert systems. Unlike in the first AI boom, where scientists and government funders came to question the technology, this second winter was mostly driven much more by business frustration. Again there are some clear echoes in what’s happening with AI today. For instance, hundreds of billions of dollars are being invested in AI datacenters being constructed by Microsoft, Alphabet, Amazon’s AWS, Elon Musk’s X.ai, and Meta. OpenAI is working on its $500 billion Project Stargate data center plan with Softbank, Oracle and other investors. Nvidia has become the world’s most valuable company with a $4.3 trillion market cap largely by catering to this demand for AI chips for data centers. One of the big suppositions behind the big data center boom is that the most cutting edge AI models will be at least as large, if not larger, than the leading models that exist today. Training and running models of this size requires extremely large data centers. But, at the same time, a number of startups have found clever ways to create much smaller models that mimic many of the capabilities of the giant models. These smaller models require far less computing resources—and in some cases don’t even require the kinds of specialized AI chips that Nvidia makes. Some might be small enough to run on a smart phone. If this trend continues, it is possible that those massive data centers won’t be required—just as it turned out LISP machines weren’t necessary. That could mean that hundreds of billions of dollars in AI infrastructure investment winds up stranded. Today’s AI systems are in many ways more capable—and flexible—than the expert systems of the 1980s. But businesses are still finding them complicated and expensive to deploy and their return on investment too often elusive. While more general purpose and less brittle than the expert systems were, today’s AI models remain unreliable, especially when it comes to addressing unusual cases that might not have been well-represented in their training data. They are prone to hallucinations, confidently spewing inaccurate information, and can sometimes make mistakes no human ever would. This means companies and governments cannot use AI to automate mission critical processes. Whether this means companies will lose patience with generative AI and large language models, just as they did with expert systems, remains to be seen. But it could happen. Winter No. 3: The rise and fall (and rise) of neural networks The 1980s also saw renewed interest in the other AI method, neural networks, due in part to the work of David Rumelhart, Geoffrey Hinton and Ronald Williams, who in 1986 figured out a way to overcome a key challenge that had bedeviled multilayered perceptrons since the 1960s. Their innovation was something called backpropagation, or backprop for short, which was a method for correcting the outputs of the middle, hidden layer of neurons during each training pass so that the network as a whole could learn efficiently. Backprop, along with more powerful computers, helped spur a renaissance in neural networks. Soon researchers were building multilayered neural networks that could decipher handwritten letters on envelopes and checks, learn the relationships between people in a family tree, recognize typed characters and read them aloud through a voice synthesizer, and even steer an early self-driving car, keeping it between the lanes of a highway. This led to a short-lived boom in neural networks in the late 1980s. But neural networks had some big drawbacks too. Training them required a lot of data, and for many tasks, the amount of data required just didn’t exist. They also were extremely slow to train and sometimes slow to run on the computer hardware that existed at the time. This meant that there were many things neural networks could still not do. Businesses did not rush to adopt neural networks as they had expert systems because their uses seemed highly circumscribed. Meanwhile, there were other statistical machine learning techniques that used less data and required less computing power that seemed to be making rapid progress. Once again, many AI researchers and engineers wrote off neural networks. Another decade-long AI winter set in. Two things thawed this third winter: the internet created vast amounts of digital data and made accessing it relatively easy. This helped break the data bottleneck that had held neural networks back in the 1980s. Then, starting in 2004, researchers at the University of Maryland and then Microsoft began experimenting with using a new kind of computer chip that had been invented for video games, called a graphics processing unit, to train and run neural networks. GPUs could perform many of the same operations in parallel, which is what neural networks required. Soon, Geoffrey Hinton and his graduate students began demonstrating that neural networks, trained on large datasets and run on GPUs, could do things—like classify images into a thousand different categories—that would have been impossible in the late 1980s. The modern “deep learning” revolution was taking off. That boom has largely continued through today. At first, neural networks were largely trained to do one particular task well—to play Go, or to recognize faces. But the AI summer deepened in 2017, when researchers at Google designed a particular kind of neural network called a Transformer that was good at figuring out language sequences. It was given another boost in 2019 when OpenAI figured out that Transformers trained on large amounts of text could not only write text well, but master many other language tasks, from translation to summarization. Three years later, an updated version of OpenAI’s transformer-based neural network, GPT-3.5, would be used to power the viral chatbot ChatGPT. Now, three years after ChatGPT’s debut, the hype around AI has never been greater. There are certainly a few autumnal signs, a falling leaf carried on the breeze here and there, if past AI winters are any guide. But only time will tell if it is the prelude to another Arctic bomb that will freeze AI investment for a generation, or merely a momentary cold-snap before the sun appears again. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
图片来源：Counter/Getty Images • 分析人士警告称，少数科技巨头（尤其是英伟达）估值过高，已在美股中形成类似历史泡沫时期的集中度风险。 标普500指数周三下跌0.69%。英伟达（Nvidia）跌幅超过两倍，下跌1.95%。此外，由于投资者对依赖融资的政府信誉丧失信心，债市对长期收益率走高反应消极。 整体市场情绪略显紧张。 德意志银行近日发布的一份研究报告更是火上浇油，直指核心问题：“美国股市是否正处在泡沫之中？”根据分析师吉姆·里德（Jim Reid）、亨利·艾伦（Henry Allen）和拉杰谢卡尔·巴塔查里亚（Rajsekhar Bhattacharyya）的观点，答案或许是肯定的。 他们指出，英伟达是造成泡沫风险问题的重要驱动因素。它的市值非常庞大。是不是过于庞大了？ 三位分析师在报告中写道：“英伟达的市值，如今已经超过除美国、中国、日本和印度之外世界上任何一个国家的整个股票市场的总市值。”这种现象扭曲了美国股市格局，因为英伟达与另外四家公司（微软、Alphabet、苹果和亚马逊）合计占据标普500指数总市值的30%。相比之下，在2000年互联网泡沫时期，标普500前五大公司的集中度还不到这个数字的一半。 下图清晰展示了当前市场对前五大股票的高度依赖程度： 这些股票的估值之高，已使美国股市规模以史无前例的方式远超海外市场。报告指出：“美国股市现在的规模几乎是排名第二的中国的五倍，约为欧洲主要市场的20倍。” 德意志银行团队表示：“这并不意味着一定存在泡沫，但我们似乎正处在未知领域，且市场表现很可能高度依赖少数几家公司。” “值得注意的是，在发达国家股市中，唯有美国可能存在泡沫风险。其他七国集团（G7）当前股市估值与盈利相比仍处于历史平均水平。” 那还能出什么问题呢？ 劳动力市场就是一个隐患。 安永-帕特农（EY-Parthenon）首席经济学家格雷戈里·达科（Gregory Daco）在一份报告中写道：“8月就业报告很可能证实劳动力市场正在明显放缓，原因是，企业领导者正面临终端需求减弱、成本与利率上升以及不确定性加剧等多重挑战，因此持续收紧招聘。” “我们预计就业增长将进一步放缓，8月非农就业人数增幅仅为4万，低于7月的7.3万增幅。失业率预计将小幅上升至4.3%，创2021年10月以来新高。” 请系好安全带，前方将颠簸难行。换句话说，衡量市场波动性的VIX恐慌指数近日持续走高，昨日上涨5.46%。（财富中文网） 译者：刘进龙 审校：汪皓 • U.S. stocks slid as the S&P 500 fell 0.69% and Nvidia dropped 1.95% yesterday. Analysts warn that outsize valuations in a few tech giants (especially Nvidia) have created a concentration risk in U.S. stocks reminiscent of past bubbles. The S&P 500 lost 0.69% yesterday. Nvidia, however, lost more than twice that—down 1.95%. Additionally, the bond market is unhappy with long-term yields going up as investors lose faith in the credibility of governments who want their financing. It’s all looking a bit nervy. That won’t be helped by a research note from Deutsche Bank today, which asks the question, “Are U.S. equities in a bubble?” The answer, according to analysts Jim Reid, Henry Allen, and Rajsekhar Bhattacharyya, is maybe. Nvidia is a big part of the problem, they say. Its market cap is huge. Too huge? “Nvidia’s market cap is now larger than every country’s entire listed stock exchange apart from the U.S., China, Japan, and India,” the trio wrote. That has a distorting effect on U.S. stocks because Nvidia and just four other stocks (Microsoft, Alphabet, Apple, and Amazon) compose 30% of the S&P 500’s entire value. For comparison, the concentration of the top five companies in the S&P during the dotcom bubble of 2000 was less than half that. This chart shows just how weighted toward the top five stocks the market currently is: The valuation of those stocks is so high that the U.S. market now dwarfs foreign markets in a way that it historically did not. “The U.S. is now nearly five times larger than China (in second) and around 20 times larger than Europe’s larger markets,” they said. “This doesn’t automatically mean it’s a bubble, but we appear to be in uncharted territory, and likely means performance heavily depends on a handful of companies,” the Deutsche team said. “We should note that, of [developed country] equity markets, only the U.S. could be considered a bubble risk. Other G7 equity markets currently have historically average valuations vs. earnings.” What could possibly go wrong? The labor market for one thing. The U.S. will publish a new job openings report today (the so-called JOLTS) and a new nonfarm payrolls number on Friday. EY-Parthenon chief economist Gregory Daco forecasted in a note yesterday that he expects Friday’s employment number to be weak: “August’s employment report is likely to confirm that a marked slowdown in labor market conditions is underway, as business leaders—grappling with softer final demand, higher costs and interest rates, and elevated uncertainty—continue to restrain hiring. “We anticipate another step down in job growth, with nonfarm payrolls expected to rise by just 40,000 in August, following a 73,000 increase in July. The unemployment rate is projected to edge higher to 4.3%—its highest level since October 2021.” Buckle up. It’s going to be a bumpy ride. (Or to put it another way, the VIX fear index—which measures volatility—has been elevated in recent days and was up 5.46% yesterday.) 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
阿波罗绘制“通胀高山”示意图。图片来源：阿波罗全球管理公司 从在德意志银行任职，到现任阿波罗全球管理公司（Apollo Global Management）首席经济学家，托尔斯滕·斯洛克多年来始终以精准图表分析在华尔街引发关注。他的市场洞见甚至曾获彭博社专题报道。今年夏天，他更是最早预言股市可能出现人工智能泡沫的权威声音之一。 如今，斯洛克发现，上世纪七八十年代的“通胀高山”与2021年的通胀浪潮之间存在着惊人的相似之处——几乎达到了令人毛骨悚然的“恐怖谷”效应，更令人不安的是，美国经济未来走势正呈现出高度相似的轨迹。在8月31日发布的《Daily Spark》简报中，斯洛克指出关税政策、美元贬值以及联邦公开市场委员会（FOMC）内部关于如何平衡通胀上升与就业放缓之间日益加剧的分歧，正在加大通胀和通胀预期的上行压力。（美国银行研究部门（Bank of America Research）在一份题为《2007年幽灵》的报告中也指出，美联储极少会在通胀上升背景下选择降息。） 斯洛克补充道：“风险正在上升，未来数月可能出现另一座‘通胀高山’。” 预警信号浮现 斯洛克和阿波罗发布的图表将当前美国核心CPI走势与1974—1982年通胀周期进行对比，清晰显示1973—1974年与2021—2022年两轮通胀浪潮存在高度相似性。如箭头所示，第一座“通胀高山”于1970年代出现，之后，第二座于1978年前后再度出现。如果历史重演，美国经济将在2025年秋季再次攀上通胀新高峰。 尽管斯洛克未在简报中明言，但“第一座通胀高山”指的是初始通胀峰值，而“第二座高山”则代表数年后因外部冲击与政策失误导致的更剧烈通胀攀升。 通胀担忧加剧 这并非斯洛克首次发出通胀警告。早在8月下旬，他就指出，杰罗姆·鲍威尔在杰克逊霍尔研讨会上使用“奇妙平衡状态”来形容劳动力市场，表明美联储已意识到关税与移民政策造成的结构性扭曲。斯洛克指出，如果这些因素持续推高通胀，而鲍威尔又迫于白宫压力降息，就可能重蹈上世纪70年代“走走停停"货币政策的覆辙——而这正是第二座通胀高山形成的背景。 这种似曾相识的场景令人想起上世纪70年代，如果美联储过早放松货币政策，通胀可能再度飙升，最终不得不采取鲍威尔前任保罗·沃尔克时代的痛苦纠偏措施——当年沃尔克激进加息，承受了严重的“双重衰退”。 最新通胀数据显示，7月个人消费支出（PCE）物价指数同比上涨2.6%，涨幅与6月持平，且符合经济学家预期。但剔除波动较大的食品和能源类别后，物价同比上涨2.9%，高于6月的2.8%，创下自2月以来新高。《财富》记者伊娃·罗伊特伯格指出，非必需消费品类支出已出现回落迹象。同期，广义消费者价格指数（CPI）增幅为2.7%，低于预期；而生产者价格指数（PPI）则因批发价格上涨3.3%而高于预期。 这些警告出现之际，经济学家们正就2020年代下半程的经济形态展开激烈讨论：究竟会迎来经济衰退，还是会出现伴随斯洛克所述“通胀高山”的滞胀现象？瑞银经过对美国经济硬数据研判，认为衰退风险显著上升，7月风险概率高达93%，不过结合其专有模型对其他条件的分析后，平均衰退风险要低得多。但瑞银仍预测，美国经济前景将趋于“疲软”，这与美国银行研究部门的判断不谋而合。 摩根大通则对7月意外疲软的就业报告深感忧虑，称如此幅度的劳动力需求下滑“无疑是经济衰退预警信号”。同时，穆迪分析公司首席经济学家马克·赞迪8月初曾援引与瑞银相似的硬数据警告称，美国正处在衰退边缘。最近，赞迪将美国经济衰退的概率调整为50%，并表示占美国GDP总量近三分之一的各州要么已陷入衰退，要么正面临衰退风险。斯洛克的分析则提出一个关键问题：当衰退撞上“通胀高山”时，经济将面临何种局面？（财富中文网） 译者：刘进龙 审校：汪皓
近期在特斯拉（Tesla）洛杉矶餐厅开业活动中，人形机器人擎天柱（Optimus）为顾客递上爆米花。图片来源：PATRICK T. FALLON—AFP/Getty Images • 特斯拉首席执行官埃隆·马斯克（Elon Musk）认为，人形机器人是决定特斯拉未来的关键所在。这位全球首富周一在社交媒体发文称，特斯拉约80%的市值最终将来自其自主研发的人形机器人擎天柱。特斯拉周一还发布了“宏图计划”第四篇章，进一步强调了物理人工智能的重要性。 尽管机器人生产屡遇阻碍，但埃隆·马斯克对特斯拉的宏大愿景仍聚焦于“减少对电动汽车的关注、全力押注自主机器人”。 这位特斯拉首席执行官周一宣称，终有一天，公司80%的市值将源自人形机器人擎天柱——该人形机器人于2021年推出，旨在承担人类眼中单调乏味且危险的工厂任务。马斯克对机器人成功前景的预测，是在特斯拉周一公布其“宏图计划”第四篇章之后发布的，该计划概述了特斯拉的未来发展目标。 该计划称：“特斯拉故事的下一篇章将助力打造一个我们尚处于初步构想阶段的世界，其体量之宏大，前所未有。我们正通过研发相关产品与服务，引领人工智能走进物理现实世界。” 为专业机器人植入智能，已成为科技领域领导者的核心关注点之一。英伟达（Nvidia）首席执行官黄仁勋（Jensen Huang）同样将“物理人工智能”视为其四阶段演进过程的终极形态，并于近期推出售价3499美元的机器人“大脑”开发套件，本月起开始发货。有分析师指出，马斯克将人形机器人擎天柱视为特斯拉未来的核心，表明特斯拉也有类似的计划。 Zacks Investment Research高级股票策略师凯文·库克（Kevin Cook）向《财富》杂志表示：“埃隆并非唯一一位洞察到机器人与物理人工智能广阔前景的人。显而易见，15年来他一直在用人工智能系统训练特斯拉汽车，因此将业务拓展至其他自主机器领域，对他而言是顺理成章的事。” 对特斯拉而言，实现战略重心转移绝非易事。人形机器人擎天柱自诞生之初便饱受分析师质疑，此后也不断遭遇阻碍。年初时，马斯克曾预测，特斯拉将在2025年生产数千台人形机器人擎天柱，并表示该项目长期有望创造超10万亿美元营收。然而今年四月，马斯克却向投资者表示，该机器人的生产因中美贸易争端而受阻——美方收紧了对稀土材料的出口管制，而这类材料正是制造特斯拉机器人内置电机的关键。负责特斯拉人形机器人擎天柱项目研发的米兰·科瓦奇（Milan Kovac）也已于6月离职。 随着汽车销量持续大幅下滑——7月欧盟地区交付量锐减40%——特斯拉正全力加速拓展电动汽车生产以外的业务。据该公司报告显示，2025年上半年其全球销量下滑13%，且目前正呈现出连续两年销量下滑的态势。 特斯拉未回应《财富》杂志的置评请求。 人形机器人擎天柱面临的竞争日益激烈 尽管特斯拉尚未兑现其在人形机器人擎天柱项目上的承诺，但人形机器人领域的竞争已愈发激烈。摩根士丹利（Morgan Stanley）今年5月发布的研究报告预测，到2050年，人形机器人市场规模将达到5万亿美元，未来25年内全球投入使用的人形机器人数量或将达到10亿台。美国人形机器人公司Figure AI自2022年成立以来，已筹集逾7亿美元资金，投资方包括杰夫·贝佐斯（Jeff Bezos）旗下的Bezos Expeditions、英特尔资本（Intel Capital）、微软（Microsoft）以及英伟达。路透社2月报道称，这家初创公司拟开展一轮规模达15亿美元的融资，若融资完成，其估值将接近400亿美元。此外，总部位于美国加利福尼亚州的K-Scale Labs已研发出一款人形机器人，其单价仅为9000美元，不到人形机器人擎天柱预计起售价的一半。 “（马斯克）当下在机器人领域面临激烈竞争，”库克（Cook）表示，“目前有数十家小型初创公司在做同类产品，不仅成本更低，还采用开源模式——此外还有Figure这类大型公司参与其中。因此，他面临的挑战相当严峻。” 海外市场的竞争更为激烈。汇丰前海（HSBC Qianhai）8月发布的报告显示，2017年至2024年，中国工业机器人销量近乎翻倍，从15万台增至约30万台。中国还大力推进人形机器人的开源开发，这一举措不仅有助于提升其在该领域的行业声誉，还能降低生产成本。 汇丰前海的报告指出：“作为最先进、最受关注的人形机器人制造商之一，特斯拉发布的生产指引被市场视为衡量人形机器人规模化生产推进程度的基准。不过我们认为，市场可能忽视了中国人形机器人供应商取得的进展——相较于海外同行，其商业化进程正在加速推进。”（财富中文网） 译者：中慧言-王芳
图片来源：Omar Marques/SOPA Images/LightRocket via Getty Images Klarna公司一度是欧洲估值最高的初创公司，但它的遭遇也充分证明，在科创圈多刺激的大起大落都不叫事儿。在经历了长期多次跳票之后，如今Klarna公司的IPO终于姗姗来迟。这家瑞典公司在全球带火了“先买后付”模式，也因此获得了全球95后和00后的青睐。目前该公司的IPO估值在140亿美元上下。这个数字看起来不错，但是已经比它456亿美元时的估值最高点下跌了69%。 2021年，Klarna公司凭借着“先买后付”模式，靠着多轮融资一路高歌猛进，成为欧洲科创界的领军者，在全球金融技术公司中的估值也是仅仅落后于Stripe。虽然现在它的IPO估值只有140亿美元，但是要知道，在此之前，Klarna也走了一段颇不容易“救赎之路”，因为它曾差一点成为又一个创业失败的反面典型。2022年，Klarna的估值一度暴跌85%，降至67亿美元。不过现在Klarna的财务状况已经有所好转，业务也扩展到了广告和消费服务领域，已经迈过了当初“撒币赚吆喝”的阶段，而更像一家运营规范的金融科技平台了。 Klarna背后的投资者包括红杉资本、银湖资本等。目前，该公司已经完成在美IPO申请，预计将于年底前上市。它也将成为近年来欧洲企业在美国规模最大的IPO交易之一。在2021年鼎盛时期，Klarna的估值甚至超过了欧洲的一些银行。但随着利率上涨，以及各国对“先买后付”服务的监管越来越严格，加上投资者对该公司“赔本增长”的质疑，Klarna不得不启动降本裁员，并且在大幅下调估值的基础上拉新融资。 近几个季度，Klarna的业绩出现好转，亏损下降，管理层也将重心从扩张转向“适度增长”和盈利。有分析师表示，其庞大的商户网络和用户的接受度仍是Klarna的核心竞争优势，但是在当前的高利率环境下，Klarna的分期付款模式的可持续性仍是存疑的。Klarna在向监管部门提交的文件中提到，该公司在其成立后的前14年里均实现盈利，但自从2018年进军美国及其他市场后，便未能继续盈利。“2023 年，我们的运营亏损额开始减少，在美国市场的交易利润率也首次扭负为正。”那么，为何Klarna在盈利下滑后，估值出现大幅崩塌？为什么它现在又在上市之际出现了貌以回暖的局面？原因主要有三点： 1. 低利率时代终结 2022年，美联储为抑制通胀而大幅加息，此后科技公司的估值普遍受挫。许多依赖宽松信贷的“泡沫化商业模式”（先买后付模式首当其冲）因融资成本上升而受到冲击。此外，地缘政治动荡与贸易政策的不确定性叠加，导致整体宏观经济波动加剧，给包括Klarna在内的众多企业的投资造成了影响。 与此同时，标普500指数的集中度也变得极高。有时候由所谓的“七巨头”主导，最近更是连特斯拉也滑到了下一档，变成了“六巨头”。另一方面，英伟达公司的市值已突破4万亿美元，仅其一家公司的股价波动，就能影响整个市场走势。在某种程度上，如今的标普500指数更像是“标普10指数”，仅由少数头部公司主导。 2. 消费支出放缓 美国经济的核心引擎是消费，在大多数年费，消费对GDP的贡献都达以了三分之二以上。但是2025年出现了一个有趣的现象，与AI革命相关的数据中心建设对GDP的贡献的增幅，已经超过了消费支出的增幅。这并非是说数据中心建设对GDP的贡献超过了三分之二，而是指这项建设的增速超过了消费支出的增速。而另一方面，当前美国就业市场疲软无力，通胀居高不下，消费支出已经出现了乏力迹象。 阿波罗全球管理公司的首席经济学家托尔斯滕・斯洛科警告称，美国的通胀可能会再次出现2021年的激增势头——而正是2021年的通胀，导致Klarna的估值从最高峰处掉了下来。未来美国可能再次撞上通胀高峰。不过另一方面，由于收入紧张，很多消费者可能会更加依赖先买后付服务。贷款平台LendingTree的数据显示，2024年有14%的美国成年人使用先买后付服务购买食品。到2025年，这一比例已升至25%。也就是说，尽管消费支出放缓，存在通胀上升甚至经济衰退的压力，但这反而可能推动更多人通过先买后付服务来让自己先活下去再说。 3．监管审查趋严 拜登政府时期，Klarna公司遭到了美国消费者金融保护局（CFPB）的审查——相比特朗普执政时期，拜登政府是更倾向于对“先买后付”模式加强监管的。当时，有些议员和州总检察长都敦促CFPB先买后付平台采取更严格的监管措施，因为他们担忧弱势的低收入群体很可能成为这类服务的“目标群体”。 对此，该行业的企业也发起了反击。一个行业组织（成员包括Klarna在内）起诉了CFPB，称其制定的信息披露规则“不可能实现”。截至2025年，CFPB已经降低了对该行业的联邦执法优先级，这意味着监管重心将转向分散化监管，预计将来会有更多由各州自行主导的监管行动和框架出台。 “独角兽” 能否重振旗鼓？ 过去两年，Klarna围绕 “降亏损、拓展广告等周边业务、推动盈利”等目标进行了业务重组。此次IPO考验的是投资者对这类金融科技公司的信心——这些公司曾拥有很高的估值，但现在则需要面对传统融资市场对收益和利润的严格审视。 Klarna的此次IPO，仍将是近10年来欧洲科技行业最重要的IPO交易之一。Klarna能否证明上市后，它能突破“先买后付”服务的一些局限性，实现进一步增长？这些一度大热的金融技术公司能否实现私募市场上的辉煌？投资者对此将密切关注。 在创作本文过程中，《财富》杂志借助AI完成了初稿，经编辑核实了信息准确后发布。（财富中文网） 译者：朴成奎
2023年，美国佛蒙特州多地遭遇洪水袭击。图片来源：John Tully for The Washington Post via Getty Images 根据房地产网站Realtor.com发布的《气候风险报告》，美国超四分之一的家庭住宅（总价值达12.7万亿美元）面临着至少一种“严重或极端的气候风险”，如洪水、飓风和山火等。该报告由经济学家徐佳怡（音译）撰写。该报告详细阐述了日益严峻的气候威胁对房地产市场格局的影响，给业主造成的经济负担，以及对全美保险业造成的负担和威胁。 该报告指出，总的来说，有26%的美国家庭住宅面临严重或极端的气候风险，其中最被联邦政府低估的是洪水风险。由于现在的使用的洪水地图已经过时了，未来30年，美国有近600万套住宅（总价3.4万亿美元）将面临严重洪水威胁，这个数字比美国联邦紧急事务管理局（FEMA）的估计多出了约200万套。光是在迈阿密、纽约、坦帕、洛杉矶和休斯敦等美国主要城市，就有总价数千亿美元的房产存在洪水侵袭的隐患。 实际上，这一比例较2024年的报告还是有所下降的。2024年的同名报告显示，有44%的美国家庭住宅（总价值约22万亿美元）面临严重或极端气候风险。不过Realtor.com的首席经济学家丹妮尔・黑尔对《财富》表示，这两份报告不具备直接可比性。2024年的这份报告涵盖了五种气候风险（洪水、大风、火灾、高温和空气质量），而2025版的报告仅包含了三种气候风险。不过即便仅计算2024年报告中的大风、洪水和山火的风险数据，其涉及的房产总价值仍达到了14.1万亿美元，高于2025版报告的统计结果。 黑尔还指出，Realtor.com在制作这份报告的过程中，与研究公司First Street进行了合作。First Street致力于量化“美国每一套房产”面临的风险，它使用的模型可能每年都会有所不同。她提到，在两份报告之间的这一年，美国发生了多起 “备受关注”的气候事件，比如破坏极大的洛杉矶山火。据《财富》此前报道，这场山火吞噬的房产价值高达1500亿美元。 洪水、飓风和山火高危区域 在洪水和飓风风险方面，最高危的区域是迈阿密—劳德代尔堡—西棕榈滩都市圈。在迈阿密、休斯敦等部分区域，所有住宅均被划入高危类别。就洪水风险而言，以风险住宅占区域内所有住宅的比重来看，新奥尔良州及佛罗里达州的几个大城市的风险比重最高。加利福尼亚州是山火风险最高的州，风险房产价值占全美近40%（约3.4万亿美元），其中洛杉矶和河滨市都是值得关注的高风险区域。除加州外，美国西部的一些城市（如科罗拉多州的科罗拉多斯普林斯、亚利桑那州的图森）的房产也面临着严重的山火威胁。 在一些高风险地区，保费也在快速上涨。以迈阿密为例，业主每年支付的保费平均占房产价值的3.7%，为全美最高水平。目前，洪水保险通常需单独购买，飓风保险的免赔额可能是标准保单的5倍，而能提供山火保险的保险公司相当有限，保费也往往高得让普通人望而却步。据世界经济论坛称，高保费已经让美国房产保险业出现了部分“保险荒漠”。黑尔指出，大多数抵押贷款要求购房者必须购买房产保险，但对于数千万拥有完整产权，且不需要还房贷的美国人来说，他们也可以选择不买房产保险，但风险就只能由自己担着了。 保险保费飙升，灾害事件频发，加之购买相关保险难度加大，不仅影响了人们对居住地的选择，也影响了相关风险地区人们对房价的负担能力。由于相关风险地区的保险越来越难买，不少人可能会选择搬到低风险地区，从而使相关地区的房价出现强劲增长。黑尔表示，Realtor.com发布这份报告已有五年。“人们很容易低估这些风险的烈度”。该公司希望通过这份报告，为购房者提供充足信息，帮助他们做出科学的购房决策。 多少人生活在洪水风险中而不自知？ Realtor.com在报告中指出，First Street所用的风险区域模型，与美国联邦紧急事务管理局的模型在计算风险住宅数量上存在巨大差异，这是因为美国联邦紧急事务管理局的模型“未将大雨和未来气候变化因素纳入考量”。Realtor.com的研究发现，在美国，大约有200套住宅（总价近1万亿美元）有被洪水侵袭的风险，而业主却对此并不知情，因此这些业主很有可能并未购洪水保险。 若是将First Street认定的主要洪水风险区域纳入考量，这一风险缺口可能会更大。以房产价值计算，纽约、洛杉矶和旧金山的风险缺口最大。研究显示，纽约的潜在洪水风险房产总价值达到953亿美元，洛杉矶为656亿美元，旧金山为549亿美元。 目前，美国房地产和保险行业正在努力采取措施，应对这一“定时炸弹”。2024年5月，房利美CEO普里西拉・阿尔莫多瓦在《财富》撰文称，她很认同碧昂斯在她的新歌《YA YA》里的一句歌词：“野火烧毁了他的房子，保险公司不会理赔，房利美也不会。”她还表示，自2021年以来，美国每年平均发生22起造成损失超10亿美元的自然灾害，与上世纪80年代（年均3起）形成了鲜明对比。（财富中文网） 为撰写本报道，《财富》杂志使用生成式人工智能协助完成初稿。在发布前，编辑已核实信息准确性。 译者：朴成奎
图片来源：David Paul Morris—Bloomberg/Getty Images 麦当劳正借助其汉堡和薯条的销售情况来反映美国经济的更大趋势。首席执行官克里斯·肯普钦斯基（Chris Kempczinski）正通过下调超值套餐价格来应对他所称的“双层经济”——即一部分消费者仍在随心所欲地消费，另一部分则开始收紧开支的分化局面。 自2022年通胀潮以来，麦当劳及其快餐竞争对手的套餐价格纷纷突破两位数，因此，多年来他们不得不面对消费者因菜单价格上涨而产生的不满情绪。高收入群体仍保持着对高端产品的消费热情，并且外卖使用频率保持高水平。但首席执行官肯普钦斯基在接受CNBC《Squawk Box》采访时指出，低收入消费者正在收紧开支，快餐对他们而言已不再是日常必需品，而更多成为偶尔消费。他对主持人透露，麦当劳过去一年多来一直在推进“超值套餐之旅”。 “中低收入消费者尤其正在承受巨大压力，”肯普钦斯基对CNBC主持人说，“他们周围很多人在讨论‘经济状况如何，现在怎么样？’而我们看到的情况是，经济实际上正在形成一种双层结构。如果你是年收入超过10万美元的高收入群体，日子还算不错……但中低收入消费者的处境却截然不同。”他指出，麦当劳的中低收入群体客流量下降了两位数，他们要么不吃早餐，要么在家吃饭。 采访中，CNBC主持人还向肯普钦斯基抛出了若干政治话题，包括麦当劳是否契合美国卫生与公共服务部（HHS）部长罗伯特·F·肯尼迪提出的“让美国再次健康”（MAHA）的目标，以及他对小费免税政策的看法。肯普钦斯基表示，他个人支持小费免税，但同时表明这对麦当劳意义不大，因为麦当劳不允许收取小费。他补充说，允许收取小费的餐厅的最低时薪仅为2.13美元，该标准自1991年沿用至今，他称这种模式存在“不公平竞争”，因为你“既想让顾客为你的劳动力买单”，还想享受免税优惠。他呼吁为所有餐厅制定统一的联邦最低工资，并表示麦当劳对提高联邦最低工资持“开放”态度。他还透露，公司正就包括此问题在内的多项议题与白宫保持沟通。 美国现行联邦最低时薪为7.25美元，该标准自2009年7月24日实施后从未调整。这一标准已经逾16年未作调整，这是美国有史以来最低工资标准停滞时间最长的一次。然而，许多州和地方已实施更高的最低工资标准，如哥伦比亚特区最低时薪高达18美元。 2025年，一项名为《提高工资法案》（Raise the Wage Act）的重要新立法提交美国国会审议。该法案计划逐步提高联邦最低工资标准，到2030年达到每小时17美元，并逐步取消对有小费员工、残障员工和青年员工的次级最低工资标准。此外，参议院还提出一项法案，拟自法案通过后次年1月1日起将最低工资提高到每小时15美元。这些立法举措表明，在经历十余年停滞后，联邦层面终于出现了要提高最低工资的势头。 与大萧条时期不同 肯普钦斯基补充说，此轮情况与麦当劳在大萧条时期所经历的不同，“那时候消费者普遍降低消费档次”。因此，麦当劳现在必须采取更具创意的方式来吸引不同消费群体。公司现在推出了改良版5美元套餐，并在核心市场开展更多促销活动，让低收入消费者也能吃得起。广告宣传重点突出了“超值”主题，试图打动那些天天为省钱发愁的节俭家庭。 这一战略凸显了麦当劳的平衡之道。作为少数几家具备足够规模与采购优势、能在不立即重创盈利的前提下降价的全球连锁企业之一，麦当劳得以在小型竞争对手无法施展的地方发力。但在美国经营大部分门店的特许经营商仍心存顾虑，他们担心在工资、房租和保险成本高企的背景下，较低的定价可能会进一步压缩利润空间。尽管如此，肯普钦斯基在接受CNBC采访时仍表示，推行更多超值选择的做法在特许经营商中“几乎获得一致支持”，此结果令主持人颇感意外。 更广阔的零售图景 麦当劳的“双轨”战略反映了美国经济中普遍存在的分化趋势。沃尔玛（Walmart）、塔吉特（Target）等大型零售商也报告了类似情况。Dollar General首席执行官托德·瓦索斯（Todd Vasos）今年3月直言：“许多顾客表示他们的钱只够买基本生活必需品。”而作为富裕消费群体需求风向标的达美航空（Delta Air Lines），尽管因特朗普关税政策带来的不确定性而在2025年下调了业绩预期，但作为美国最赚钱的航空公司，其整体发展态势持续向好。 这种趋势让人联想起疫情期间形成的经济格局：“K型经济”。安永-博智隆（EY-Parthenon）首席经济学家格雷戈里·达科（Gregory Daco）在2023年对《财富》解释称，这意味着中低收入消费者就像“K”字形中向右下方延伸的一笔，走势向下；而高收入群体则持续向上。 麦当劳必须驾驭这种“K型”格局才能最大化消费者价值。这意味着既要守住其数十年来“平价餐首选”的地位，同时也要开拓更高利润机会，以安抚股东。这种平衡策略能否持续，很大程度上取决于美国这种双轨消费经济格局将延续多久。（财富中文网） 译者：刘进龙 审校：汪皓
“将两家公司合并显然并非明智之举，但在我看来，分拆也无法解决问题。”沃伦·巴菲特向美国消费者新闻与商业频道（CNBC）表示。图片来源：Daniel Acker—Bloomberg/Getty Images 卡夫亨氏（Kraft Heinz）——这家由沃伦·巴菲特（Warren Buffett）与巴西私募股权公司3G资本（3G Capital）于2015年联手打造的包装食品巨头，正式宣布将进行分拆。周二这一消息的发布，为巴菲特那笔备受瞩目且惨痛至极的投资画上了句号：当年这场承诺将带来效率提升与行业主导地位的并购案，最终反而导致公司市值蒸发约570亿美元（跌幅达60%）。 消息公布后，卡夫亨氏股价下跌7%。巴菲特旗下伯克希尔-哈撒韦公司（Berkshire Hathaway）仍持有其27.5%的股份，他毫不掩饰自己的感受。 “将两家公司合并显然并非明智之举，但在我看来，分拆也无法解决问题。”他向美国消费者新闻与商业频道表示，并补充称自己对这一决定“深感失望”。 一分为二的分拆计划 该公司宣布，将于2026年底前完成分拆，形成两家上市企业： • 全球风味提升公司（Global Taste Elevation Co.）：专注于经典亨氏产品，包括调味酱、涂抹酱、调味品（亨氏番茄酱、卡夫奶酪通心粉、费城奶油奶酪等均在其列）。 • 北美杂货公司（North American Grocery Co.）：涵盖多个标志性日常食品品牌，如奥斯卡·梅耶（Oscar Mayer）、卡夫切片奶酪（Kraft Singles）、麦斯威尔咖啡（Maxwell House）以及方便午餐盒（Lunchables）。现任首席执行官卡洛斯·艾布拉姆斯-里维（Carlos Abrams-Rivera）将执掌该公司，与此同时，董事会正为全球风味提升公司物色新负责人。 公司执行主席傅玫凯（Miguel Patricio）表示，推动分拆旨在优化资本配置，并在营销层面聚焦战略重点。 他向《华尔街日报》表示：“我们会为每个品牌匹配适配的关注度与资源，充分释放其潜力。” 在经历十年业绩低迷之后，公司决定进行分拆。自2015年卡夫与亨氏合并以来，公司市值已蒸发逾570亿美元，资产减记规模达150亿美元，同时还因消费者转向非加工食品而多次遭受市场抛售的冲击。 巴菲特对这一投资失误向来坦诚。2019年，他承认伯克希尔“为卡夫支付了过高溢价”。此后，这位“奥马哈先知”对该持股计提数十亿美元资产减值，而3G资本则悄然退出，使得伯克希尔成为持股比例最高、损失最为惨重的股东。 分拆能否扭转局面？ 此次分拆给投资者带来更棘手的问题：倘若消费者正抛弃那些未能契合当下健康标准的“传统”食品杂货品牌，那么分拆后的独立品牌又如何能在相同的业务领域表现更好呢？ 分析师、雅虎财经（Yahoo Finance）执行主编布莱恩·索齐（Brian Sozzi）在领英发文指出：“加大营销支持力度绝非包治百病的灵丹妙药。” TD Cowen的罗伯特·莫斯科（Robert Moskow）向《华尔街日报》指出，食品巨头常高估规模效益带来的影响。“食品企业发现，它们在食品杂货店内的广泛影响力未必能带来预期优势。”他如是说。 换言之，对卡夫亨氏进行分拆，或许能解决部分因官僚主义而导致的效率低下问题，但无法改变消费者对热狗或方便午餐盒类加工食品需求持续萎缩的事实。对巴菲特而言，此次分拆为其罕见的投资失误画上了象征性的句点。当“奥马哈先知”准备年底将伯克希尔的管理权移交给格雷格·阿贝尔（Greg Abel）时，卡夫亨氏将成为警示案例：即便是最具标志性的品牌，也难以抵御消费者口味变化带来的冲击。（财富中文网） 译者：中慧言-王芳
今年《财富》世界500强企业的总营收达到41.7万亿美元，创历史峰值，占到全球名义GDP的三分之一以上。同时，这500家大公司的利润总额连续第三年增长，逼近3万亿美元大关，刷新历史第二高位。这份榜单的最大意义，正是在每年喧嚣的盛夏，当社交媒体的碎片信息如蝉鸣不绝之际，让我们能够透过最重要的数字，看清两点：行业的脉动与国家间愈加紧密的经济纽带。 过去一年，镁光灯照耀下的群体毫无疑问是全球的高科技巨擘。在34家上榜的高技术大公司中，美国企业占据了15席：它们的平均营业收入已经远超千亿美元大关，平均净利润更是高达310亿美元。而今年跻身榜单门槛、排在第500位的企业，其年销售收入为322亿美元——也就是说，一家美国高科技巨头所创造的利润，几乎足以再造出一家全新的《财富》世界500强公司。 用我们的话语体系来总结，美国公司实现了高质量发展，尤其是作为链主的头部科技公司。但是，美国的优势显然不是依靠脱钩或者去中国化带来的。在全球科技产业链的版图中，中国公司已经是不可或缺的枢纽。 两年前，在美国硅谷的一个会议上，英伟达的首席执行官黄仁勋对着屏幕上的中国地图说：“失去这个市场，我们没有Plan B。”英伟达2025年第一季度的全球营收达到411亿美元，其中中国市场贡献12.5%，成为其全球布局的关键一环。为巩固这一阵地，黄仁勋在今年7月专程访华，出席中国国际供应链促进博览会，并宣布对华销售H20芯片的许可获批。这不仅是为推销产品，更是为了维护其全球产业链的稳定性。 苹果公司的200家核心供应商中，超过80%的公司在中国设有生产基地，覆盖了从零部件制造到整机组装的全链条。外部分析认为即便是印度、墨西哥的“替代产线”，仍然高度依赖中国供应商的技术与零部件支撑。这种科技巨头与中国产业链的共生关系，是静水流深的现实。 进一步从现实眺望未来，人工智能占据了我们的全部视野。就在今年的《财富》世界500强榜单发布的前一周，我在新加坡参加《财富》人工智能头脑风暴大会。听完一整天的发言，有一个想法挥之不去：人工智能有没有可能“填平”修昔底德陷阱？从技术供给侧来观察：美国为技术优势竖墙，在一定程度上推动了中国拥抱开源生态，而美国则继续背靠大厂的闭源树立壁垒——这两种路径看似对立，其实可以互补并存。 如果从人工智能的应用侧来看：中国在规模化发展垂直场景，例如智慧城市、工业互联网等；美国则依托公有云生态激活了长尾用户的创造力——前者深入应用场景，后者拓开应用广度，共同构成一个完整的“技术-市场正态分布”。 虽然从芯片到算力，中美竞争是主线，但是这种结构性互补，还需不需要如当年雅典和斯巴达那样“一战”生死？会不会到头来变成“一帐”共谋？17世纪的英国诗人约翰·多恩写过一段名言：“没有人是一座孤岛，可以自全；每个人都是大陆的一片，整体的一部分。”今天再读这句话，感觉它深刻揭示了人类命运的相关性。而对我们自己来说，只需要记住一个事实：越开放融入的一方，越难以被淘汰。（财富中文网） 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
今年以来，外卖大战让阿里、美团、京东，三家巨头的利润受到侵蚀，市场格局得以重塑。外卖业务继续胶着的同时，它们已开始押注下一个赛道寻找增量——硬折扣超市。 硬折扣模式下，商品售价较传统超市更为低廉，而与商家让利和处理尾货的软折扣不同，硬折扣是通过优化供应链，减少中间环节，实现可持续的低价供应。较为常见的策略有精简SKU、发展自有品牌等。 盒马CEO严筱磊自2024年上任后，对盒马业态进行了大刀阔斧的调整，将十余种盒马业态，缩减为两种——除盒马鲜生之外，只留下了硬折扣店盒马NB。随着8月底位于上海的最后一家盒马X会员店停止营业，曾作为盒马“三驾马车”之一、意在对标山姆的盒马X会员店业态已画上句号。 经调整后，盒马实现了盈利。根据阿里巴巴2025财年年报，盒马GMV超过750亿元，跻身中国超市榜单第三，首次实现全年经调整EBITA（息税及摊销前利润）转正。距离严筱磊所设定的目标——三年后年GMV达到1000亿元更进一步。 阿里正在加快硬折扣模式的扩张。按照严筱磊的计划，盒马NB在2025财年计划增长到300家店。消息称，截至目前，其门店数已近300家，或已提前完成目标。 8月29日，盒马NB正式更名为“超盒算NB”。当天，首批位于上海、宁波、苏州、南京等10座城市的17家“超盒算NB”新店同步开业。9月，“超盒算 NB”全面上线淘宝闪购，将这一线下业务接入阿里今年着重布局的即时零售。 这一动作也符合阿里CEO吴泳铭上任以来，推行的集团业务间集体作战模式——盒马能借助阿里为即时零售投入的流量，同时，具备性价比优势的超盒算NB也能丰富淘宝闪购的供给。 根据尼尔森IQ数据，2024年中国市场硬折扣市场规模已突破2000亿元，但渗透率仅为8%，具备增长潜力。预计未来十年，中国硬折扣业态年复合增长率将达5.6%，远超大卖场的2.5%。 当前环境下，具备发展潜力的业务对互联网公司而言都是值得抓住的机遇。今年参与外卖大战的主要玩家均未缺席对硬折扣超市的押注。 在盒马NB更名的同一天，美团自营硬折扣超市“快乐猴”首店在杭州开业，与盒马相似，快乐猴的商品包含大量自有品牌。快乐猴相关负责人表示，该品牌的运营具有“科学品控体系和自有供应链的双重保障”，商品主要围绕一日三餐与日常高频需求。 这不是美团首次尝试线下门店。2017年，美团就曾开设线下生鲜超市“掌鱼生鲜”，此举当时被市场解读为对盒马鲜生的效仿。次年，该品牌更名为小象生鲜。这一实体品牌的表现未达预期，一年后关闭了多家店面。之后，美团便将重点回归至他们更熟悉的线上领域。“小象”如今已被美团用于对其自营即时零售业务的命名。 据报道，部分小象超市、美团优选的供应商也参加了快乐猴的招商，美团先前积累的部分供应链和技术经验也将为快乐猴提供支撑。 消息称，美团快乐猴的开店目标为1000家店，今年预计将开出10家左右门店，前期主要分布在北上广，和杭州等一线城市。这些数字暂未得到官方证实。 参照已布局多年的盒马NB当前300家的门店数量，要达成千家实体门店有一定难度。在这方面，新晋外卖巨头京东也有发言权——2018年，京东计划在之后的3到5年开设1000家七鲜超市，而在内外部多重因素作用下，目前其全国门店数仍未突破100家。 京东也于近日切入了硬折扣赛道。8月30日，京东折扣超市在刘强东的家乡江苏宿迁四店齐开。据悉，京东折扣超市采用大店型，门店面积约5000平方米，商品种类超5000款，自有品牌占比约20%。无论门店面积和SKU，与盒马NB相比都更具规模。 刘强东曾强调，京东的业务都围绕供应链展开。硬折扣超市作为以供应链为核心的业务，京东多年积累的能力能够直接用于旗下硬折扣门店的运营。 虽然三家巨头的资金实力已经在外卖大战中得以展现，但线下实体业态的经营仍需它们探索更多新经验。在硬折扣领域，竞争对手远不止彼此熟识的互联网老面孔，还有以ALDI奥乐齐为代表的海外硬折扣品牌，以及国内超市品牌的业务扩充。 进入中国市场已有六年的德国品牌奥乐齐，过去一直未走出上海。而今年4月，奥乐齐进入了江苏，两家门店分别开在苏州和无锡。在此一个月前，盒马NB刚在无锡开出首店。 国内品牌方面，7月25日，知名连锁超市集团物美旗下的硬折扣店“物美超值”开业，首批6家门店落地北京。集团计划于今年年内在京开出25家物美超值门店。（财富中文网） 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
英伟达（NVIDIA）首席执行官黄仁勋（Jensen Huang）。图片来源：GETTY IMAGES • 因投资者对人工智能企业的增长上限提出质疑，科技股8月应声下跌。英伟达、美满科技（Marvell Technology）及超微电脑公司（Super Micro Computer Inc.）上月表现均逊于大盘。这种不确定性可能影响以“七巨头”科技公司为主的标普500指数。 纳斯达克100指数上周五收盘时下跌1.22%，尽管美国市场周一因劳动节假期休市，但该指数的期货合约交易并未休市：今日上午其交易持平，这意味着投资者对周二纽约股市开盘时科技股的表现并不抱太大期望。 尽管整体标准普尔500指数上涨3.56%，但纳斯达克100指数8月收盘时仍下跌0.16%。 整个八月，科技股的表现始终受限于“人工智能是否处于泡沫期”的讨论。麻省理工学院（MIT）的研究表明，95%的企业尚未从其人工智能投资中获得回报。 正如德意志银行（Deutsche Bank）分析师吉姆·里德（Jim Reid）及其团队近日所言：“英伟达（上周五下跌3.32%）是市场疲软态势的主要推手。其股价下跌的原因在于美满科技的前景展望引发市场对数据中心设备需求的质疑，以及中国阿里巴巴推出新人工智能芯片。上周三，英伟达公布的财报虽略好于预期，但数据中心部门营收增速放缓，部分原因在于对华人工智能芯片销售暂停。” 美满科技总部位于加利福尼亚州圣克拉拉，主营半导体芯片业务。该公司与英伟达建立了合作关系。在8月28日举行的2026财年第二季度财报电话会议上，该公司首席执行官马特·墨菲（Matt Murphy）表示：“我们预计第三季度数据中心整体营收将与上一季度持平。”持平意味着没有增长，这一消息导致美满科技股价次日下跌19%。（5月，美满科技以宏观经济不确定性为由取消了投资者日活动。） 在失望情绪出现的前一天，英伟达举行了财报电话会议。该公司报告称数据中心业务营收增长强劲，但仍低于分析师预期。 超微电脑公司（另一家受益于人工智能热潮的芯片制造商）8月初将全年营收预期从2月的400亿美元下调至330亿美元。此外，8月28日，该公司年度报告披露：“我们发现财务报告内部控制存在重大缺陷，若不加以整改，或将对我们及时准确披露财务状况和经营业绩的能力产生不利影响。”此消息一经公布，其股价下跌5.5%，当月累计跌幅达27%。 人工智能股票的波动可能会对更广泛的市场产生影响。七大科技巨头（苹果、亚马逊、谷歌母公司Alphabet、Meta、微软、英伟达和特斯拉）均在人工智能领域投入巨资，目前它们的总市值占标普500指数总市值的34%。（财富中文网） 译者：中慧言-王芳 • Tech stocks declined in August as investors questioned the limits to the growth of AI companies. Nvidia, Marvell Technology, and Super Micro Computer Inc. all underperformed the broader market last month. This uncertainty may impact the S&P 500, which is dominated by the “Magnificent 7” tech giants. The Nasdaq 100 closed down 1.22% on Friday and while U.S. markets are closed today for the Labor Day holiday, futures contracts for the index are not: They’re trading flat this morning, implying that investors are not expecting much from tech stocks once the opening bell rings in New York on Tuesday. The Nasdaq 100 closed down for the month of August (-0.16%) even though the broader S&P 500 was up 3.56%. Tech stocks were dogged all month by discussion about whether AI was in a bubble. And a study by MIT suggested that 95% of companies have yet to see a return on their investment in AI. As Jim Reid and his team of analysts at Deutsche Bank said this morning: “Nvidia (-3.32% on Friday) was a major driver of this softness, losing ground after Marvell Technology’s outlook raised doubts over demand for data-centre equipment and as China’s Alibaba unveiled a new AI Chip. Last Wednesday, Nvidia’s results delivered a modest quarterly beat but saw slowing revenue growth for the data centre division, in part due to a pause in sales of AI chips to China.” Marvell Technology is based in Santa Clara, California, and makes semiconductor chips. It has a partnership with Nvidia. On its fiscal Q2 2026 earnings call on August 28, CEO Matt Murphy said, “We expect overall data center revenue in the third quarter to be flat sequentially.” Flat is not up, and that sent Marvell’s stock down 19% the next day. (In May, Marvell cancelled its investor day presentations, citing macroeconomic uncertainty.) That disappointment came after Nvidia’s earnings call the day before. The company reported robust data center revenue growth, but it was nonetheless below analyst expectations. And then there is Super Micro Computer Inc., another chipmaker buoyed by the AI boom. In early August, it reduced its revenue outlook for the year to $33 billion. Back in February, it had estimated $40 billion. On top of that, on August 28th, the company said in its annual report, “We have identified material weaknesses in our internal control over financial reporting, which could, if not remediated, adversely affect our ability to report our financial condition and results of operations in a timely and accurate manner.” Its stock fell 5.5% after that and was down 27% for the month. Shakiness in AI stocks could have consequences for the broader market. The “Magnificent 7” tech companies (Apple, Amazon, Alphabet, Meta, Microsoft, Nvidia, and Tesla), which have all placed large bets on AI, are currently worth 34% of the entire market cap of the S&P 500. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
图片来源：Bloomberg / Contributor / Getty Images • 亿万富豪、首席执行官贝尼奥夫表示，由人工智能客服接手相关工作后，市值2480亿美元的Salesforce已裁撤约4000个客服岗位。他表示，如今人工智能技术已能处理逾百万次客户对话，自2025年初以来客服成本降低17%，因此，公司“不需要那么多人了”，这一做法与Klarna、微软（Microsoft）类似。而此番言论与他此前声称“人工智能不会引发白领失业潮”的言论截然不同。 ChatGPT问世不过三年。其间，包括马克·贝尼奥夫和黄仁勋在内的多位企业领袖一再坚称，更廉价的人工智能替代不会引发大规模失业。 但现实是，这项技术正导致多家大型企业裁员——其中，Salesforce已裁撤4000个客服岗位，由人工智能客服接手相关工作。 “我重新调整了客服团队规模，”市值2480亿美元的软件公司Salesforce首席执行官马克·贝尼奥夫近日在播客《洛根・巴特利特秀》（The Logan Bartlett Show）中透露，“我已经把员工人数从9000人减到了约5000人，因为我不需要那么多人了。” 他补充说：“如果此番对话发生在一年前，当你致电Salesforce时，全球会有9000名客服人员通过我们的服务云平台为您服务，他们负责管理、创建、读取、更新和删除数据。”而如今，业务量依旧，但“50%由人工智能客服完成，50%由人工客服完成”。 在他看来，这种人机协作模式并非遥不可及的未来。“我完全不觉得这是反乌托邦，”他补充道，“至少对我而言，这就是现实。” Salesforce发言人向《财富》表示：“今年年初我们上线了help.agentforce.com平台。Agentforce让效能大幅提升，促使客服工单量持续下降，我们就不再需要主动补充客服工程师岗位。我们已成功将数百名员工调配到专业服务、销售和客户成功等其他部门。” 数月间态度发生逆转 数月来，这位科技巨头一直在关注客服岗位自动化。他此前曾对《财富》表示，过去六到九个月，人工智能客服已完成逾百万次客户对话。但当时，他还声称不会大规模裁员。 贝尼奥夫坦言：“我一直在观察行业动态，并和多位公司首席执行官交流，问他们大规模裁员时究竟用了哪些人工智能技术？我认为人工智能能增强人的能力，但未必能取代人类。原因就在于，现有的很多技术依然建立在语言模型之上。也许未来会出现更精准的人工智能模型，但至少目前还没到那一步。现在的重点是人机协作。” 尽管客服部门大幅裁员，他依然坚称这对整个公司而言是一个“意义重大”的时刻，并强调人类仍将发挥核心作用。 贝尼奥夫在播客中表示：“我们现已部署全渠道监管系统，用于协调人工智能客服和人工客服的协同工作。这无疑是过去九个月里Salesforce最令人兴奋的进展。” 况且，Salesforce裁撤客服岗位其实早有征兆。这位科技公司首席执行官曾提到，人工智能客服已承担公司30%至50%的工作量，并且客服与销售两类岗位最易被自动化取代。贝尼奥夫称，通过以先进技术取代人力，Salesforce目前已削减17%的客服成本。 不过，自动化浪潮可能并不止于客服和销售两类岗位。贝尼奥夫透露，他正在审视“所有职能部门”，研究如何让公司真正实现智能体化运营。 多家企业裁员，人工智能员工取而代之 首席执行官们曾矢口否认人工智能快速普及会导致大规模裁员，如今却公开宣布以人工智能取代人工的计划。在行业巨头推动岗位自动化的浪潮下，科技企业今年裁员总数已超6.4万人。 7月初，微软宣布将裁员约9000人，这是该公司自2023年以来最大规模的裁员。尽管这家市值3.74万亿美元的公司财务状况良好——上季度净利润同比增长18%，但今年裁员总数已达惊人的1.5万人。然而，并非所有为公司创造利润的员工都能保住饭碗：最新裁员预计将波及销售、客户服务岗位以及Xbox游戏部门。 Meta也加入了这场自动化浪潮，于今年2月裁员3600人。首席执行官马克·扎克伯格甚至表示，人工智能今年内或许就能具备“中级工程师”的编程能力。谷歌（Google）同样毫不避讳地裁撤了Android、Pixel和Chrome等部门的数百个岗位。谈及大规模裁员的原因，这两家硅谷巨头均声称，需要精简人力运营并加大对人工智能的投入。 贝尼奥夫也绝非首位专门在客服岗位上动刀的企业领袖。金融科技公司Klarna的人工智能客服已承担了700名人工客服的工作量。而在受生成式人工智能冲击最严重的职业中，销售代表和客服人员分列第四和第六位。 贝尼奥夫今年早些时候在接受《财富》采访时表示：“我们积累了大量潜在客户线索，根本无法全部跟进。销售人员基本上只能挑选部分线索回访。成千上万乃至数十万条线索始终没有得到跟进。但在智能体时代，这种情况将不复存在。每条线索都能得到及时跟进。”（财富中文网） 译者：刘进龙 审校：汪皓 • Billionaire CEO Marc Benioff said his $248 billion Salesforce has cut about 4,000 customer service roles as AI agents step in to do the work. Just like Klarna and Microsoft, the Silicon Valley mogul revealed he just “needs less heads” now that the technology can handle over a million consumer conversations and has reduced support costs by 17% since around the start of 2025. It’s a change of tune from Benioff’s previous comments that AI wouldn’t lead to a white-collar wipeout. ChatGPT launched only three years ago. Since then, leaders including Marc Benioff and Jensen Huang have been adamant that cheaper alternatives to labor won’t cause mass unemployment. But in reality, the technology is slashing human headcounts at major companies—including Salesforce, which has cut 4,000 of its customer support roles for AI agents to pick up the work. “I was able to rebalance my headcount on my support,” Marc Benioff, CEO of the $248 billion computer software company, recently revealed on the podcast The Logan Bartlett Show. “I’ve reduced it from 9,000 heads to about 5,000, because I need less heads. “If we were having this conversation a year ago and you were calling Salesforce, there would be 9,000 people that you would be interacting with globally on our service cloud, and they would be managing, creating, reading, updating, deleting data,” he added. Today, those same interactions are happening, but “50% are with agents, 50% are with humans.” And he doesn’t see his hybrid AI-human workforce as an otherworldly future. “I don’t think it’s dystopian at all,” he added. “This is reality, at least for me. “At the start of this year we deployed help.agentforce.com. Because of the benefits and efficiencies of Agentforce, we’ve seen the number of support cases we handle decline, and we no longer need to actively backfill support engineer roles,” a Salesforce spokesperson tells Fortune. “We’ve successfully redeployed hundreds of employees into other areas like professional services, sales, and customer success.” A change of tune from months ago The tech titan has shown an interest in automating customer support jobs for some months now, previously telling Fortune that AI agents have completed over a million conversations with customers over the past six to nine months. But at the time, he said that mass layoffs weren’t on the table. “I keep looking around, talking to CEOs, asking, What AI are they using for these big layoffs? I think AI augments people, but I don’t know if it necessarily replaces them,” Benioff revealed. “The reason is because a lot of this is still built on word models. Maybe there’s a future AI model that will be more accurate, but that’s not where we are right now. This is about humans and AI working together.” And while the customer service department gets heavily slashed, he’s still adamant that it’s an “exciting” time for the wider company—and that humans will remain at the core of the function. “There’s also an omni-channel supervisor now that’s helping those agents and those humans work together,” Benioff said on the podcast. “And this is the most exciting thing that’s happened in the last nine months for Salesforce.” Plus, Salesforce’s elimination of support roles in particular should come as no surprise. The tech CEO has said that agents are already doing 30% to 50% of work within the company and that two roles in particular had the potential to be automated by AI agents: support and sales. By pushing humans out in favor of the advanced technology, Benioff said Salesforce has reduced its support cost by 17% so far. But further automation beyond sales and support could be on the cards as the Salesforce boss revealed he’s looking at “every single function” to see how it can become an agentic business. Other companies slashing staff in favor of AI workers CEOs once denied the fact that rapid AI adoption will slash staffers across organizations, but now leaders are openly sharing their plans to replace humans with bots. More than 64,000 have been laid off across the tech sector this year as industry heavyweights lead the charge in job automation. In early July, Microsoft announced that it will cut about 9,000 roles—its largest round of layoffs since 2023. That plan brings the $3.74 trillion company’s total layoffs this year to a whopping 15,000 jobs, despite the company doing well financially; Microsoft posted a 18% year-to-year increase in net income last quarter. However, not all workers get to stick around after bringing home the bacon: The latest round of cuts is expected to hit sales and customer-facing roles, alongside the Xbox gaming division. Meta joined in on the automation push, laying off 3,600 employees in February, and CEO Mark Zuckerberg even said that AI could be “effectively be a sort of mid-level engineer” sometime this year, with the ability to code. Google also wasn’t shy about reducing hundreds of roles across its Android, Pixel, and Chrome sectors. In reasoning about the mass firings, both Silicon Valley giants claimed the need to streamline human operations and invest more in AI. And Benioff is far from being the first leader to cut down specifically customer service jobs; fintech company Klarna’s AI agents are doing the work of 700 customer service employees. And among the professions most impacted by generative AI, sales representatives rank fourth and customer service agents rank sixth. “We have so many leads that we can’t follow up on them all. Salespeople basically cherry-pick what leads they want to call back. Thousands of leads, tens of thousands of leads, hundreds of thousands of leads have never been called back,” Benioff told Fortune earlier this year. “But in the agentic world, there’s no excuse for that. Every lead can be followed up on.” 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
固态电池的起源可追溯至19世纪法拉第对固体电解质的发现，但受限于技术瓶颈，其发展在随后百余年始终缓慢，直至20世纪90年代新型固态电解质的出现让这项技术重回视野。进入21世纪后，在新能源汽车浪潮的推动下，固态电池终于迎来研发与投资的热潮，其从实验室走向量产线的步伐也在逐渐加快。 近期，这场能源存储革命又迎来多项突破性进展。9月2日，亿纬锂能固态电池研究院成都量产基地正式揭牌，其“龙泉二号”全固态电池已成功下线。就在同一天，国轩高科宣布其首条全固态中试线已正式贯通，良品率达90%。另外，孚能科技表示预计今年年底建成硫化物全固态电池中试线，并向战略合作伙伴客户交付60Ah的第一代硫化物全固态电池。 如果说现有的锂离子电池（液态电解质）已经成功推动了电动汽车革命，那么固态电池可能才刚刚翻开新故事的第一章。固态电池根据液体含量可分为半固态、准固态和全固态三种类型，它们的液体含量分别为5-10wt%、0-5wt%、0wt%。最受关注的无疑是全固态电池，其可将电动车续航提升至1000公里以上，且彻底解决液态电池的燃爆风险。我国在2020年首次将全固态电池研发上升至国家层面，之后该领域呈现加速发展态势。根据行业预测，全固态电池将在2027年实现小批量装车，2030年迈入规模量产阶段。 目前，多家企业已制定明确的全固态电池量产时间表。孚能科技称其将于2026-2027年推进小批量量产装车，2030年将实现大规模量产。国轩高科表示已正式启动第一代全固态电池2GWh量产线的设计工作。奇瑞则联合中科院、宁德时代共建“固态电池联合实验室”，在正极材料、电解质、隔膜三大核心领域攻关，预计2026年量产装车后纯电车型续航将突破1000公里，充电时间缩短至10分钟。 尽管技术进展迅速，成本仍是固态电池大规模商业化的主要障碍。有分析机构估算，预计固态电池2026年成本在5700元/kWh左右，这意味着一辆搭载100kWh（即100度电）固态电池的纯电动车仅仅电池的成本就接近60万元，如果再加上关键的三大件和电子电气架构，量产车落地价格可能超过百万。 欣旺达曾公开表示计划在2026年量产全固态电池，并将整体成本控制在2元/Wh（2000元/kWh）以内。但即使这一目标能够实现，100度电的固态电池成本仍将达到20万元级别，导致整车价格轻松超过40万元。 固态电池成本高企的推手，除了新型材料研发成本较高，还有生产工艺与传统液态电池之间存在的显著差异，尤其是干法电极、固态电解质复合、等静压成型等新增环节都推高了生产制造成本。值得关注的是，今年以来，固态电池生产设备需求出现激增：在主要设备供应商中，先导智能上半年新签订单总额达124亿元，同比增长近70%，海目星上半年新增订单约44.21亿元，同比增长117.5%，截至6月末在手订单规模突破100亿元。春江水暖鸭先知，设备供应商显然已经率先感受到固态电池产业化的热度。 材料方面，国泰海通研报指出，硫化物固态电解质凭借更优秀的综合性能有望成为全固态电池的主流选择，其中硫化锂是硫化物固态电解质的核心原材料，潜在的市场空间广阔，吸引了众多企业布局。硫化锂主流制备路线可分为三大类：固相法、液相法、气相法。随着固态电池市场需求逐渐兴起，前瞻布局硫化锂的企业有望受益。 而与固态电池的成本相呼应，其应用层面也在梯度渗透。最初全固态电池将主要应用于eVTOL飞行器、人形机器人等高附加值领域，因这些领域对能量密度和安全性要求极高但对成本相对不敏感。随着产能扩大和成本下降，固态电池逐步渗透到高端电动车市场，最后才会进入大众消费市场（以电动车为例，一般认为是30万元以下价位）。 亿纬锂能最新下线的“龙泉二号”全固态电池就是明确面向人形机器人、低空飞行器及AI高端装备领域，这些应用场景将为全固态电池提供难得的商业化验证机会。中金公司预测，2030年全球固态电池出货量将达到808GWh，其中全固态电池有望于2027年实现技术定型和小规模量产，2030年实现商业化量产、需求量有望超150GWh，届时动力、EVTOL、消费电子分别对应固态装机需求为93、40、23GWh，渗透率分别为3%、40%、15%。 固态电池商业化提速，自然也离不开政策的推动。今年3月，国家层面提出建立全固态电池标准体系。中国汽车工程学会将于9月10-11日在北京召开《固态电池材料评测用模具电池装配方法》等10项固态电池团体标准送审审查会及《硫化物全固态电池硫化氢产气量评价方法》等5项标准项目启动会，将解决硫化物电解质遇水产生硫化氢气体等安全性评估问题，同时也将规范材料评测、电池装配等方法，使行业的发展更加有序和高效，为产业化奠定基础。 从投资角度看，技术突破与商业化进程同步推进的固态电池有望成为长期的黄金赛道。但这场技术革命不会一蹴而就，它更像是一场需要耐心与智慧并重的产业马拉松。未来三到五年，行业或将经历一轮淘汰赛：部分过度依赖融资却缺乏核心技术突破的企业可能面临出清，而真正在材料体系、工艺设备和商业模式上实现突破的企业将逐渐崭露头角。这意味着投资者不仅要关注固态电池的技术进展，更要精准识别各细分领域中的真正领军者。（财富中文网）
图片来源：Getty Images 随着近年来农产品价格暴跌以及总统唐纳德·特朗普发起的贸易战对农业造成的冲击，美国玉米和大豆农户接连发出严峻警告。 上周四，美国国家玉米种植者协会（National Corn Growers Association）警示称：“在投入成本仍接近历史高位之际，农产品价格却大幅下跌，美国农村正遭遇一场经济危机。” 该协会表示，玉米价格自2022年达到峰值以来，已暴跌逾50%，而同期生产成本仅下降3%，意味着每蒲式耳要亏损85美分。该协会进一步强调，明年前景更糟，价格将进一步走低，而成本则持续攀升。 该协会呼吁国会和特朗普政府通过提高乙醇掺混比例、扩大海外市场准入等措施提振需求。 一周前，美国大豆协会（American Soybean Association）致信特朗普，警告称：“美国豆农正站在贸易和金融的悬崖边上。” 该协会敦促特朗普在对华贸易谈判中优先考虑大豆议题，争取让中国承诺大规模采购大豆并取消对美国农产品加征的关税。 信中指出：“长期以来，美国一直是中国客户的首选供应方。然而，由于关税，中国的长期客户已经并将继续转向南美竞争对手来满足需求。而巴西自上次中美贸易战以来大幅增产，足以满足这一需求。” 该协会补充称，美国大豆收获季即将来临，但中国尚未采购任何未来数月交货的美国大豆。 信中强调，中美谈判若迟迟未能达成贸易协议，随着秋收季节日益深入，农民承受的压力将愈发沉重。 与玉米种植者一样，大豆种植者也面临价格暴跌与成本高企的双重压力。大豆价格自2022年触顶以来，已下跌约40%。 美国大豆协会表示：“豆农正承受巨大的财务压力。价格持续下跌的同时，投入成本和设备开支却大幅上涨。美国豆农无法在与最大客户的长期贸易争端中生存下去。” 农场收入与信贷状况持续恶化 美联储最新农场金融状况调查印证了农业经济的惨淡景象。调查显示，收入下降削弱了农民的流动性，从而推高了融资需求。 与此同时，信贷环境持续恶化。芝加哥联储和堪萨斯城联储辖区约30%的受访者表示其偿付率低于去年同期；明尼阿波利斯联储辖区该比例约为40%；而在圣路易斯联储辖区，这一比例更是高达50%。 当然，美国农民也将获得大规模援助。特朗普今年初发动新一轮贸易战后，美国政府和国会议员4月起就开始商讨农户纾困方案。 7月签署的《大而美法案》（One Big Beautiful Bill Act）包含约660亿美元农业专项支出。据美国农场局联合会（American Farm Bureau Federation）数据，其中绝大部分资金（约590亿美元）将用于强化农场安全网。 此外，特朗普谈判达成的其他贸易协议也有望推动亚洲多国加大对美国农产品的采购。 例如，印尼和孟加拉国已承诺按照协议增加采购量。本周有消息人士对路透社透露，越南、菲律宾和泰国可能会增加饲料谷物进口量。 美国大豆出口协会（U.S. Soybean Export Council）东南亚及大洋洲区域总监罗启明（Timothy Loh）在接受路透社采访时表示：“近期的贸易磋商卓有成效，为美国拓宽本地区市场准入提供了机会。” 他补充说：“我们预计东南亚对美国豆粕等农产品的需求将进一步增长。”（财富中文网） 译者：刘进龙 审校：汪皓
图片来源：Getty Images 美国人力资源管理协会（SHRM）最新研究揭示，工作场所中文明行为的缺失，正致使美国企业每日承受高达约21亿美元的损失。该机构梳理粗鲁无礼行为、简短生硬邮件及尖刻互动的报告后发现，工作效率下滑与员工缺勤现象正在侵蚀企业利润。 美国人力资源管理协会的文明指数研究发现，美国员工每天共遭遇2.08亿次“失礼行为”，这一数字在2024年大选期间急剧上升，目前仍接近历史最高水平（较上季度1.98亿次亦有增长）。此类从隐晦轻蔑到公然敌对的持续性不尊重行为，最终转化为代价高昂的缺勤、士气低落和产出减少。 美国人力资源管理协会首席人力资源官吉姆·林克（Jim Link）近期在接受《财富》杂志采访时指出：“我们清楚这一数字，这相当于21亿美元的生产力损失。” 粗鲁无礼行为激增的根源 美国人力资源管理协会表示，办公室不文明行为激增是由更广泛的社会政治紧张局势、疫情引发的压力以及林克所说的“数字勇气”推动的，这个词让人联想到社交媒体时代的“键盘侠”。简而言之，人们在网络上敢于发表面对面交流时绝不敢发表的言论。政治立场、社会议题乃至移民政策分歧正不断在工作场所引发摩擦，根源在于员工们难以妥善应对激烈辩论以及文化分歧。 “数字勇气指的是躲在屏幕这一安全屏障后，对任何话题、任何人畅所欲言的观念。”林克向《财富》杂志解释道，并指出这一现象不仅对美国社区乃至整个社会产生影响，更波及个体及工作场所。“倘若人们在行使数字勇气的权利，那么它可能正在向工作场所、社区和社会渗透或蔓延。我们认为这种情况完全有可能发生。” 美国人力资源管理协会并非唯一研究这一课题的机构。林克指出，杜克大学的“公民对话项目”及迈阿密大学的研究团队也在开展相关工作。不过，美国人力资源管理协会对工作场所中粗鲁无礼行为的解读仍具有独到见解。 对幸福感的实质影响 美国人力资源管理协会的研究发现，办公室不文明行为的影响远远超出了伤害感情的范畴。管理者们报告称，充斥不文明现象的工作场所会导致员工心理安全感降低、团队凝聚力减弱，并在包容性与多样性指标方面表现欠佳——这些因素直接影响企业利润，因此备受首席执行官关注。 林克表示，这可能与美国人力资源管理协会针对工作场所“幸福感”开展的独立研究存在关联，不过他坦言，目前尚未发现二者之间存在因果联系。截至5月，超三分之一的受访员工表示工作带来极大压力。除此之外，员工幸福感整体状况喜忧参半，且已显露出令人担忧的迹象。 林克指出，幸福指数在疫情初期大幅下降，随后在2021年大幅反弹。他表示，他们认为2021年的回升反映了“疫苗带来的喜悦”，总体而言，“基本上67%的受访者表示其幸福状况较疫情前有所恶化，此后大部分时间里评分基本持平。”除此之外，“女性群体评分较低，少数族裔群体评分较低，年轻群体评分也较低。” 文化的重要性 企业领导者必须正视这一问题。美国人力资源管理协会的研究强调了组织文化的重要性：当首席执行官和主管们以身作则并规范文明行为时，信任度与绩效表现将同步提升。美国人力资源管理协会倡导企业明确行为期望、完善友善准则、培训员工掌握主动倾听技巧——促使工作场所对话从争论转变为讨论，而非下达“封口令”或禁止讨论敏感话题。 林克以某封邮件为例说明所谓的“不文明行为”。他向《财富》杂志透露，自己亲阅这封被指存在问题的邮件后，认为其措辞虽直截了当，但“属于正常商务沟通范畴”。诚然，这封邮件的措辞“不够华美”，但“我当时坐在那里想，这到底哪里不文明了？”当员工报告不文明行为时，美国人力资源管理协会会追问具体指控内容。大多数情况不过是邮件措辞简短生硬，或口头交流时态度尖刻。他补充道，所幸肢体暴力事件并不多见。 但他从中获得了一个关键启示：不文明行为“更多与组织文化相关，而非取决于当事人是否有意为之”。他敦促企业要有意识地塑造自身文化，并明确文化期望。美国人力资源管理协会将此称为“文化清晰”。如此一来，不文明行为的界定将更加明晰，或者更不易被随意解读。 “在构建文明行为与文明期望的过程中，文化与领导力同等重要。”他表示。 这并非意味着企业文化本身必然具备文明属性。林克指出，关键在于明确期望。 “当领导者——尤其是首席执行官或高管团队宣告'无论你是否认同，这些就是我们文化的组成部分'时，那么就没有太多解读的余地了。”他说道。（财富中文网） 译者：中慧言-王芳
图片来源：GETTY IMAGES 金融业正经历快速数字化转型，但网络犯罪分子的适应速度同样惊人。银行被迫投入巨资以应对激增的金融欺诈活动。在亚太地区，98%的金融机构不得不扩大合规运营规模，相关成本已飙升至450亿美元以上。这种激增反映出反欺诈策略正转向一体化模式，各国政府和行业正推出有针对性的国家应对措施，以抵御日益复杂的欺诈威胁。 中国香港推出手机版防骗预警系统Scameter，实时提醒用户高风险交易。新加坡实施“共同责任框架”，将诈骗损失责任分配给金融机构和电信运营商，推动反诈措施落地。同样，澳大利亚的“反诈安全协议”则由银行、住房互助协会和信用合作社跨行业合作，旨在提升客户保护标准以应对诈骗。 这些举措均是对日益严峻的区域威胁的有力回应。以东南亚“诈骗基地”为例：这些实体据点由犯罪集团操纵，策划大规模网络诈骗，包括身份欺诈、网络钓鱼、虚假投资和洗钱等犯罪活动。这些犯罪组织伪装成合法企业运作，每年非法获利高达数十亿美元。 推动金融犯罪升级的幕后推手究竟是什么？答案日益清晰：正是人工智能。犯罪网络利用人工智能技术伪造身份，发动大规模钓鱼攻击，并绕过传统安全系统——不仅所需资源更少，作案速度更是达到前所未有的程度。尽管诈骗园区主要集中在亚洲，但金融欺诈威胁已然席卷全球。 正当亚洲犯罪集团频频登上新闻头条之时，亚洲银行正悄然引领一场反欺诈技术变革。与其他地区银行主要将人工智能用于客户服务个性化和呼叫中心支持的银行不同，亚洲银行正通过在欺诈检测、身份验证和反洗钱等领域部署人工智能，对网络犯罪分子展开有力反击。 亚太地区为何能在人工智能反诈防御领域领先全球 亚洲之所以更注重人工智能反诈领域，根本上源于该地区面临的金融犯罪风险。亚洲金融机构始终奋战在打击网络犯罪前线，这种紧迫态势迫使其迅速采用人工智能驱动的防御策略。 金融损失规模令人震惊。仅2024年，亚太地区因欺诈造成的损失就高达约6880亿美元，几乎占全球总损失的三分之二。数字钱包和支付平台在亚洲民众中的快速普及使情况雪上加霜：由于强有力的消费者保护措施未能同步跟进，这种使用习惯为网络犯罪分子提供了可乘之机，也将银行推到了防御最前线。 亚洲银行在采纳ISO 20022新报文标准方面走在前列。该标准让金融机构借助人工智能更精准地识别异常交易，从而降低其暴露在金融犯罪中的风险。 同样的技术，不同的打法 随着人工智能的广泛应用，全球银行业的区域战略重点正呈现明显分化。亚太银行聚焦欺诈防控与安全防护，而欧美金融机构则优先将人工智能用于实现产品个性化和优化客户服务。 据我们的研究显示，英国仅略超半数企业计划通过生成式人工智能提升客户体验。这反映出英国市场竞争异常激烈，用户友好的互动体验是赢得客户忠诚度的关键。美国则将人工智能资源同时投入客户体验和运营自动化，既满足消费者对无缝银行服务的需求，又提升内部效率。 相比之下，58%的亚太银行将人工智能投资重点放在欺诈检测和反洗钱领域，远超全球平均水平。亚太银行面临高风险环境，犯罪网络利用生成式人工智能实施身份欺诈、网络钓鱼和金融诈骗。因此，该地区优先强化网络安全，制定更为明确的安全导向型人工智能战略，将欺诈防御视为核心竞争优势。 值得注意的是，人工智能正在模糊安全与服务之间的界限。随着网络安全威胁日益严峻，客户不仅期望银行保护他们的资金，更期待在不确定时期获得清晰、准确的答复。我们的客户实践显示，人工智能驱动的聊天机器人和身份验证系统能为银行员工调取信息的速度提升30%至40%，这进而带动客户满意度提升：客户对聊天机器人的满意度评分较传统人工客服高出25%。 未来银行业的核心需求 在当今威胁环境下，欺诈检测不能孤立存在，必须嵌入到金融基础设施中。无论是通过澳大利亚“反诈安全协议”等跨行业合作协议，还是借助既能实时验证用户身份又能解答问题的人工智能聊天机器人融合服务与安全功能，亚太地区正在向世界展示，通过人工智能驱动并与运营需求深度契合的一体化系统如何将原始数据转化为可执行的防御措施。 亚太地区的实践表明，金融安全的关键在于主动防御而非被动响应。面对巨额欺诈损失和复杂诈骗网络，亚洲金融机构已迅速将人工智能驱动的反欺诈列为优先事项。相比之下，欧美同行仅将反欺诈视为众多人工智能应用场景之一。随着人工智能驱动的金融犯罪在全球蔓延，这种战略误判将带来严重后果。 人工智能在欺诈中的作用将日益凸显。亚太地区的战略表明，快速行动并将反欺诈嵌入金融基础设施具有重要意义。随着全球威胁升级，世界各地应将亚洲视为榜样，不仅将其视为区域领导者，更应将其作为实现安全无缝金融交易的典范。（财富中文网） Ashish Thapar为NTT DATA亚太区副总裁兼网络安全主管。 《财富》网站评论文章仅代表作者个人观点，不代表《财富》的立场。 译者：刘进龙 审校：汪皓
图片来源：GETTY IMAGES 旧金山科技企业家特雷弗·特雷纳（Trevor Traina）曾就读普林斯顿大学，后获牛津大学和加州大学伯克利分校高等学位。他的儿子罗比（化名）是一名校队运动员，且学业成绩优异，获4.0满分绩点，今年即将步入大学。然而，他对父亲的母校乃至其他任何常春藤盟校都毫无兴趣。罗比最终选择了北卡罗来纳州的维克森林大学（Wake Forest）。 特雷纳表示，儿子做出这一决定的主要原因，是为了避开弥漫在东北部和西海岸名校校园里的激进政治氛围以及令人窒息的政治正确文化。 在他看来，如今的学生们普遍认为这些学校“乏味、爱评判且对白人男生存在偏见”。他补充说，儿子的许多朋友也纷纷选择了更具包容性的南方高校，例如杜克大学（Duke）、范德堡大学（Vanderbilt）和杜兰大学（Tulane）。 这种现象并非个例。最新招生数据显示，美国东北部等地区学生选择南方高校的人数激增。政治因素固然是原因之一，但通过采访家长、学生和校方人员发现，一种新型大学理想正在兴起：校园归属感、经济承受力和文明氛围已成为最重要的择校标准。 人人身着橙色战袍 艾因斯利·马特森（Ainsley Matteson）坦言，她的大学选择让家庭立场出现了分歧——至少在去年某个星期六是这样。那天，在一场关键季后赛橄榄球比赛中，她放弃了对俄亥俄州立大学的长期支持，转而为田纳西大学（University of Tennessee）加油助威。 “在诺克斯维尔，体育让所有人团结在一起，”这位主修供应链管理的大四学生、田纳西大学“新晋粉丝”说。“比赛日只要穿上橙色队服，你就会有一种归属感。” 来自华盛顿特区郊区的高中毕业生卡梅伦·麦克马纳斯（Cameron McManus）同样被南方高校这种强烈的社区氛围吸引，正考虑申请北卡罗来纳大学教堂山分校（UNC Chapel Hill）、克莱姆森大学（Clemson）或南卡罗来纳大学（University of South Carolina）。他之所以对这些学校感兴趣，一方面是因为被抖音和Instagram上展示的体育和兄弟会文化吸引，另一方面是喜欢温暖的气候。 “一年四季都能在户外活动。”他还说，朋友的哥哥姐姐们的亲身经历更让他深信南方高校是“充满活力”的地方。 范德堡大学（Vanderbilt University）正是吸引更多外地学子的高校之一。校长丹尼尔·迪尔迈尔（Daniel Diermeier）表示，学校收到东北部、西海岸（尤其是湾区）的申请量激增。 迪尔迈尔表示，纳什维尔的温和气候和活跃的体育氛围固然是吸引力所在，但潜在学生和家长更看重的是范德堡大学对言论自由的承诺以及在外部政治议题上保持中立。 他说：“我们在与家长的交流中注意到，他们最关心的是校园是否存在意识形态垄断，并希望子女能在多元思想环境中自由成长。” 迪尔迈尔补充说，这类担忧在2023年10月7日之后尤为突出。巴以冲突在在美国校园掀起支持巴勒斯坦的抗议浪潮，出现帐篷营地，甚至迫使哥伦比亚大学等高校取消毕业典礼。 迪尔迈尔表示，当抗议者占领其办公室并袭击一名保安时，他采取了截然不同的处理方式，那就是实施纪律处分并恢复校园秩序。但他同时强调，学校仍然欢迎各种观点。“我们的学生可以探讨最具挑战性的话题，但必须在尊重与文明的氛围中探讨。” 申请量激增50% 华盛顿特区一所公立高中的高年级学生艾迪·罗杰斯（Addie Rogers）表示，她明显感受到身边同学南下求学的意愿日益高涨，而这同样是她的心愿。 她说：“最吸引我的是南方高校特有的校园精神。我上大学，不仅要学习，还要享受大学生活，而这正是南方高校的魅力所在。” 如果罗杰斯最终选择南下求学，她将拥有众多同行者。《华尔街日报》的一份最新报告显示，过去二十年间，就读南方公立高校的北方学生人数增长84%；2018年至2022年间更是激增30%。 与此同时，“通用申请”（Common Application，一种被越来越多高校采用的标准招生流程）最新数据显示，自2019年以来，南方高校申请量增长50%，远超新英格兰和中大西洋地区高校不到30%的增幅。 这在一定程度上反映出进入最顶尖名校的难度空前加大。南方高校申请量激增的另一大原因是，学生申请的学校数量远超以往。 近年来“广撒网式”的申请趋势源自新冠疫情时期。当时，许多高校在招生流程中取消了标准化考试要求。即便如今招生政策回归常态，这一趋势依然延续。 Access Consulting的克里斯塔·贾乔尼（Krista Jajonie）表示，这种“广撒网心态”之所以持续存在，部分原因在于招生办公室不愿直接告知学生——哪怕是完全不符合条件的学生——不要申请他们的课程。申请人数越多，“录取转化率”就越高，而这正是高校之间相互竞争的重要指标。 谈到校园的政治氛围，贾乔尼表示确实有一些家长不愿把孩子送到因巴以冲突而陷入分裂的学校。但对未来的学生而言，南方高校的核心吸引力依然是气候和体育文化。 最后是费用问题——在某些院校仅学费就超过每年7万美元的当下，这已成为许多家庭的首要考量。来自弗吉尼亚州北部的丹妮尔·戴维斯（Danielle Davis）在为儿子挑选潜在大学时，几乎没有考虑校园的政治氛围。 令她震惊的是，本地弗吉尼亚大学（University of Virginia）仅学费就接近每年3.7万美元。最终，他们选择了一所“公立常春藤”院校——佛罗里达大学（University of Florida），总费用（含兄弟会会费）为3.1万美元。她的儿子目前主修金融专业，由于费用相对低廉，即便未来攻读研究生，家庭仍能负担后续费用。（财富中文网） 译者：刘进龙 审校：汪皓 Trevor Traina, a tech entrepreneur from San Francisco, attended Princeton University before pursuing advanced degrees from Oxford and UC Berkeley. His son Robby (not his real name) is a varsity athlete with a 4.0 grade point average who is off to college this year—and wants nothing to do with his father’s alma maters or, for that matter, any other Ivy League school. Robby chose Wake Forest in North Carolina instead. Traina says a big reason for his son’s decision is to avoid a culture of radical politics and stifling political correctness that has come to define the campuses of elite schools in the Northeast and on the West Coast. In Traina’s view, students have come to view these schools as “unfun, judgey and biased against white boys.” He added that many of his son’s friends likewise sought out more welcoming Southern schools like Duke, Vanderbilt and Tulane. They are not alone. Recent admissions data show a surge in students from the Northeast and other regions choosing schools in the South. Politics is not the only reason of course. But interviews with parents, students and university officials suggest the ascendance of a new type of college ideal: A campus where belonging, affordability and civility matter most. Everyone wears orange Ainsley Matteson says her choice of college meant her family became a house divided—or at least it was one Saturday last year when she dropped her lifetime loyalty to Ohio State and rooted for the University of Tennessee during a critical playoff football game. “In Knoxville, sports bring everyone together,” said Matteson, a senior studying supply chain management and Volunteer convert. “If you’re wearing orange on game day, there’s this sense of belonging.” Cameron McManus, a high school senior from the suburbs of Washington, D.C., is also drawn to the idea of a school with a strong sense of community, and has his eye on UNC Chapel Hill, Clemson or the University of South Carolina. His interest has been spurred in part by TikTok and Instagram videos that showcase sports and Greek culture scenes at those schools, and by the promise of warm weather. “You can be outside all months of the year,” he said, adding that stories from friends’ older siblings reinforced his impression that Southern schools are a “vibrant” place to be. One of those schools attracting more students from outside the region is Vanderbilt University. According to Chancellor Daniel Diermeier, the school has seen a surge in applications from the Northeast, West Coast and from the Bay Area in particular. While the Nashville university’s mild climate and lively sports scene are no doubt a draw, Diermeier says prospective students and parents are attracted to Vanderbilt’s commitment to free speech and institutional neutrality on external political issues. “We’ve noticed from conversations with parents that top of mind for them is whether campus will be a place where their son or daughter can thrive without ideological homogeneity,” he said. Diermeier adds these concerns have become especially pronounced since October 7, 2023 when the Hamas massacre of Israelis touched off a regional war, and a wave of pro-Palestine protests on U.S. campuses that produced tent encampments and led schools like Columbia to cancel graduation ceremonies. The Vanderbilt Chancellor says he took a different approach when protestors occupied his office and assaulted a security guard, choosing to mete out discipline and restore order to campus. Diermeier says all views are nonetheless welcome at the school. “Our students explore the most challenging topics but can do so in a climate of respect and civility,” he says. A 50% jump in applications Addie Rogers, a senior at a Washington, DC public high school, says she has noticed a growing desire among her peers to go South for schools, and that it is her aspiration too. “The main thing that appeals to me is the school spirit of Southern schools,” she said. “I don’t want to go to college and focus only on studying. I want to have fun. That’s what Southern schools are all about.” If Rogers does end up traveling south for school, she will have plenty of company. A recent Wall Street Journal report found that the number of Northerners going to Southern public schools has risen 84% over the past two decades, and jumped 30% from 2018 to 2022. Meanwhile, surveys of recent data from the Common Application (a standard admissions process used by a growing number of colleges) shows that applications to colleges in the South are up 50% since 2019. That compares to a rise of less than 30% for schools located in New England and the Mid-Atlantic. Part of this reflects the reality that it is harder than ever to get into the most elite colleges. Another big factor in the surging admissions down South is that students are applying to a far greater number of schools than in the past. This recent effort to cast a very wide net is an outgrowth of the Covid era when many schools dropped standardized tests from their admission process, and has continued even as schools revert to their former practices. According to Krista Jajonie of Access Consulting, this “apply everywhere mentality” has persisted in part because admissions offices are reluctant to ever tell students—even totally unqualified ones—not to apply to their programs since more applications improve the so-called yield rate that schools use a key benchmark against one another. As for the political climate of campuses, Jajonie says she is hearing from parents who don’t want to send their kids to a school riven with conflict over Israel and Palestine. But she says, for prospective students, the prime draw of Southern campuses is the weather and sports culture. Finally, there is the question of cost—a factor that has become an overriding concern for many at a time when some schools cost over $70,000 a year in tuition alone. When Danielle Davis of northern Virginia was exploring potential universities for her son to attend, the issue of campus political culture was hardly top of mind. What concerned her instead was that it would cost nearly $37,000 just for her son to attend the nearby University of Virginia. Instead, they settled on the University of Florida, a “public Ivy” where the total cost was $31,000—all-in, including fraternity dues. Her son is now majoring in finance and, thanks to the relative affordability, the family will have money left if he chooses to pursue graduate school. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
图片来源：Murat Deniz—Getty Images 牛奶从7美元飞涨到14美元，草莓变成奢侈品，人们被迫买加工食品……研究劳动力和农业的经济学家认为，这些都是未来六个月消费者将面临的情况。 然而消费者“根本不知道发生了什么”，得州农工大学布什政府与公共服务学院（Texas A&M’s Bush School of Government）劳动经济学家雷蒙德・罗伯逊告诉《财富》。他曾为美国多家机构提供贸易与劳工政策方面的建议。 罗伯逊说，特朗普政策引发的政治喧嚣分散了选民的注意力，而导致超市物价飙升的真正推手，包括劳动力短缺和关税等正持续产生更严重的影响。驱逐非法移民导致农田里人手减少，农场失去了“绝大多数”由无证工人构成的劳动力。与此同时，对番茄、咖啡、橙汁等日常食品新征收的关税推高进口成本，几乎没有便宜的替代品可供选择。 “影响是明确无疑的，”哈佛肯尼迪学院（Harvard Kennedy School）墨西哥贸易专家、经济学家戈登·汉森告诉《财富》，“相关商品价格面临上行压力。” 白宫未立即回应《财富》的置评请求。 汉森补充说，唯一的问题是农民、批发商和零售商承受的痛苦，最终都会体现在超市货架和消费者的购物车里。 冬季的食品价格寒潮 第一波涨价很可能在今冬袭来。罗伯逊预测，随着库存清空和新合同生效，到明年初农产品价格可能上涨50%至100%。而且与过去几十年不同，以往美国政府会悄悄放松边境执法以保证农田有足够劳动力，如今政治环境下不会再有类似“缓冲”措施。 “就像看着洪水袭来，海啸正在逼近，水位已经上涨了5厘米，”罗伯逊警告说。 劳动力短缺的根源在于，美国出生的工人根本不愿意按外籍无证工人的工资干体力活，罗伯逊解释道。无证工人摘草莓时薪大概18美元，而美国公民去冰淇淋店打工就能赚到同样工资。 要让美国公民去田里干活，“必须支付每小时25—30美元”，罗伯逊说，对大多数农业生产者来说这样的成本难以承受。 劳动力短缺已在实际中显现。在佛罗里达州的多佛，帕克斯代尔农场的马特・帕克告诉The Daily Adda，其家族企业严重依赖专门为外国农业劳动力提供支持的H-2A签证项目填补缺口。 然而经济学家表示，该计划规模太小且程序过于繁琐，不可能解决危机。汉森指出，尽管近年来H-2A计划有所扩张，外籍工人仍只占农场劳动力的“一小部分”。 “这一计划要大规模扩张，达到百万而不是几十万，才能满足美国的需求，”汉森说。 签证每个季度都会到期，农场需要为每个工人重复申请、提供住宿并负担交通费用。 “如想连续五年雇用同一名工人，就得办五次签证，”汉森补充道。 罗伯逊表示同意，但他认为特朗普政府完全可以利用人脸识别等安全技术，将H-2A计划大幅扩容。 “他们就是不做，真让人抓狂，”罗伯逊说。 关税双重夹击 过去在美国作物歉收时，进口还能作为补充，现在也无法缓解问题了。墨西哥可全年种植牛油果、番茄等作物，然而特朗普的关税导致这些农作物更昂贵。 “墨西哥的牛油果产量比美国多得多，”汉森说，“这不是说现在种几棵牛油果树，明年就能收获。” 他还表示，大概六个月后，消费者就会感受到关税冲击。 “消费者不会看到关税对产品价格的全面影响，但至少能感受到50%。” 对消费者而言，驱逐非法移民与关税的“双重夹击”可能很快会改变购物方式。经济学家警告，果蔬和乳制品受到的影响最大，很多家庭将被迫购买更便宜的加工食品。 “随着蔬菜价格一路飙升，人们只能转向高热量的深度加工食品，最终对健康造成负面影响，”罗伯逊说。 在汉森看来，政策制定者唯一能做的就是鼓励“降低关税”。 “原因很简单，”罗伯逊说，“如果能让更多合法的农业工人流入并降低关税，消费者的境况就会变好。其他试图弥补现有政策损害的政策都没有意义。” 类似争论并不是新鲜事。1950年代以来，关税与移民问题一直是美国周期性政治斗争的话题，当前局面只是冲突的“极端表现”。历史经验表明，一旦价格飙升，选民就会迫使立法者采取行动。“现在特朗普还向国会施压，要求对移民保持强硬立场。等到接近中期选举，消费者抗议物价飞涨时，强硬路线就会开始松动，”他解释道。 “政治就是这么回事，”汉森说。（财富中文网） 译者：梁宇 审校：夏林 Milk prices jumping from $7 to $14, strawberries that feel like luxury goods, and a switch to processed food: This is the six-month outlook economists studying labor and agriculture see for consumers. Yet these consumers “don’t have a clue what’s going on,” Raymond Robertson, a labor economist at Texas A&M’s Bush School of Government who has advised U.S. agencies on trade and labor policy, told Fortune. Instead, Robertson said, voters are distracted by the political noise of President Donald Trump’s policies, while the real drivers of grocery sticker shock—labor shortages and tariffs—continue to tighten their grip. Deportations have thinned fields and stripped farms of undocumented workers, who “overwhelmingly” make up the agricultural workforce. At the same time, new tariffs on staples like tomatoes, coffee, and orange juice are pushing up costs on imports, leaving few affordable alternatives. “The impacts are unambiguous,” Gordon Hanson, an economist and expert on Mexican trade at Harvard Kennedy School, told Fortune. “It’s upward pressure on those prices.“ The White House did not immediately respond to Fortune’s request for comment. The only question, Hanson added, is how much of the pain farmers, wholesalers, and retailers can absorb before it lands on the grocery aisles and in consumers’ carts. Winter grocery chill The first wave of grocery-price increases will likely hit this winter. Roberson predicted produce prices could rise 50% to 100% by early next year as inventories clear and new contracts kick in. And unlike past decades, when Washington would quietly ease border enforcement to keep fields staffed, today’s political environment suggests no such check. “This is like when you see a flood coming, the tsunami is coming in, and the water’s gone up two inches,” Robertson warned. The reason for the labor shortage is American-born workers simply do not want to do manual work at the wages typically offered to foreign-born, undocumented workers, Robertson said. Undocumented workers are used to getting paid around $18 an hour to pick strawberries—the type of wage American citizens can get working at an ice cream shop. You would have to pay American citizens “$25 to $30 an hour” to get them in the fields, Robertson said, an unfeasible cost for most agricultural producers. The shortage is already visible on the ground. In Dover, Fla., Matt Parke of Parkesdale Farms told The Daily Adda his family business is leaning heavily on the H-2A visa program—designed particularly to support foreign agricultural labor—to fill the gaps. Economists, however, say the program is too small and too cumbersome to solve the crisis on its own. Hanson noted while H-2A has expanded in recent years, guest workers still account for “a small fraction of the total” farm labor force. “It would have to be much, much larger, in the millions rather than the hundreds of thousands, to meet U.S. demand,” Hanson said. The visas also expire each season, requiring repeated applications, housing, and transportation costs for every worker. “If you want to hire that same worker five years in a row, you have to get five different visas,” Hanson added. Robertson agreed, but thought the Trump administration could easily expand the H-2A program dramatically to meet the capacity, especially given the innovations of facial recognition technology and other security measures. “It blows my mind that they don’t do this,” Robertson said. Tariffs creating a double bind Imports, once a fallback when U.S. crops ran short, can no longer offer relief. Mexico has a structural advantage in crops like avocados and tomatoes—growing the crops year-round—but Trump’s tariffs have made them more expensive by default. “Mexico produces way more avocados than we do,” Hanson said. “It’s not like you can plant new avocado trees and get an additional crop next year.” Hanson also said shoppers will feel the tariffs in about six months. “Consumers are not going to see the full pass-through of the tariffs to product prices, but they’re likely to see at least 50%.” For consumers, the double bind of deportations and tariffs could soon reshape grocery shopping. Economists warn produce and dairy are most exposed, and many families will be forced to trade down to cheaper, processed foods. “As vegetables [prices] keep going up and up, people will just substitute towards these very hot, ultra-processed foods, which ultimately will have adverse effects on their health,” Robertson said. The only thing policymakers could do, in Hanson’s mind, is encourage “lower tariffs.” “It’s simple,” Hanson said. “If we were able to create larger flows of legal farmworkers and lower tariffs, consumers are going to be better off. Any other policy that tries to undo the damages of an existing policy makes no sense.” These fights aren’t new, he said. Tariffs and immigration are topics the U.S. has had periodic political battles about since the 1950s, and today’s environment is a “very intense manifestation” of those conflicts. But history shows that once prices spike, voters force lawmakers’ hands. Now, Trump is pressuring Congress to maintain a hard line on immigration. But then you get closer to midterm elections, and consumers are lashing out against higher prices, and your hard line begins to weaken, he explained. “That’s just kind of how politics work,” Hanson said. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
2023年8 月，加利福尼亚州卡梅尔市，兰博基尼董事长兼CEO斯蒂芬・温克尔曼与一辆Lanzador电动概念车型一同亮相。图片来源：David Paul Morris—Bloomberg/Getty Images • 兰博基尼公司CEO斯蒂芬·温克尔曼近日表示，由于美国与欧盟之间的关税政策仍不稳定，美国的部分富豪已暂缓了购车计划。美国是兰博基尼的最大市场，虽然近年来，兰博基尼在美销量再创新高，而且积压的订单也一定程度上缓冲销售放缓的影响，但温克尔曼仍指出，美国的对欧关税政策仍有可能给兰博基尼造成沉重打击。 特朗普在关税问题上的不确定性，导致整个美国经济都受到了影响。就连兰博基尼的那些富豪客户也未能幸免。 兰博基尼公司CEO温克尔曼近日表示，目前，该公司的部分美国客户正在观望特朗普对欧关税的最终结果。（在美国购买一辆基础版的兰博基尼跑车至少需要40万美元。）虽然美国与欧盟已达成协议，美国对包括汽车在内的部分欧盟产品征收15%的关税，以换取欧盟的多项承诺（包括欧盟从美国采购更多AI芯片和军事装备），但这些客户仍然选择暂缓购车，继续持币观望。 温克尔曼在接受CNBC采访时表示：“有些客户选择持币观望，是想确定这一税率是不是就是最终结果了。也有一部分客户能够接受这一税率，或者表示愿意跟我们谈判协商。” 温克尔曼指出，尽管凭借大量积压订单（现在兰博基尼交付的汽车，均为一两年前就下好的订单），兰博基尼有能力抵御关税带来的部分压力，但美国对欧关税确实也给公司造成了不小的冲击。兰博基尼的所有汽车均在意大利生产，这一点也是该品牌的核心特色之一。尽管美国是兰博基尼的最大市场，但兰博基尼确实无法做到像特朗普建议的那样，将公司生产线搬到美国来规避关税。 “对我们而言，自由贸易才是正确的发展方向。”温克尔曼表示：“我们都知道，这才是我们期待的目标，但是现实情况是，我们必须应对各种复杂局面，而且我们也做好了应对任何复杂挑战的准备。” 兰博基尼隶属于大众汽车旗下的奥迪集团。2024财年，该公司营收首次突破30 亿欧元（约合35亿美元），同比增长16%；营业利润实现两位数增长，达到8.35亿欧元（约合9.74亿美元）。另外，自2023年以来，兰博基尼已经推出了三款全新的插电式混合动力车型。 截至目前，兰博基尼尚未回应《财富》的置评请求。 兰博基尼客户群体的日益多元化，也为公司带来了新的增长动力。随着来自全球各地的年轻富裕客户越来越多，使得兰博基尼车主的平均年龄已降至45岁以下。特别是在亚洲市场，客户的平均年龄仅为30岁左右。兰博基尼车主平均每人拥有5辆汽车。其高端车型Revuelto的车主名下更是平均拥有10辆汽车。 至于兰博基尼客户对关税和可能涨价等问题的态度，温克尔曼表示，要相信这些客户的判断。 他表示：“这些客户能成为百万富翁甚至亿万富翁，这绝非偶然。因此，他们清楚自己在做什么，也明白自己为什么要这么做。”（财富中文网） 译者：朴成奎 • Lamborghini CEO Stephan Winkelmann said some of its moneyed customers are holding back on new car purchases as tariffs remain unstable between the U.S. and EU. The U.S. is the company’s biggest market, and despite recent record results and an order backlog that provides some cushion against a sales slowdown, Winkelmann said tariffs have the potential to deal the company a blow. President Donald Trump’s tariff instability is putting the economy on hold, and even Lamborghini’s deep-pocketed customers aren’t immune. CEO Stephan Winkelmann said some of its U.S. customers, who have at least $400,000 to shell out for a base model, are waiting to see where the Trump administration’s tariff rate for the EU ultimately settles. Its customers are waiting, even though the U.S. reached a deal to impose a 15% tariff on some EU products, including cars, in exchange for several pledges by the EU. This includes buying more AI chips and military equipment from the U.S. “Some are waiting because they want to be sure that this is the final number that is going to be in place,” Winkelmann told CNBC. “Others are fine with it, or we will have negotiations.” Winkelmann said while Lamborghini can resist the pressure of tariffs because of its large backlog of orders—cars delivered today were ordered a year or two ago—the company will still face pressure from U.S. tariffs on the EU. Lamborghini vehicles are made in Italy, and this point is a differentiator for the brand. While the U.S. is its largest market, the car company’s production can’t be moved there, a strategy Trump has suggested for companies to avoid tariffs. “For us, free trade is the right approach,” Winkelmann said. “We all know that one is what we want. The other is the reality, and we have to deal with complexity since we are in business, and we are ready to face whatever comes.” Lamborghini, which is owned by Volkswagen’s Audi Group, reported a 16% year-over-year increase that saw it bring in more than 3 billion euros ($3.5 billion) in revenue for the first time for fiscal 2024. Its operating income also rose by double digits to 835 million euros ($974 million). The company has also launched three new plug-in hybrid models since 2023. Lamborghini did not immediately respond to Fortune’s request for comment. Providing another boost is Lamborghini’s increasingly diverse customer base. Younger wealthy customers from all over the world have pushed the average age of a Lambo owner below 45. In Asia, the average customer is about 30. The average buyer owns five cars, while buyers of its upscale, pricier model, the Revuelto, have 10, on average. As for Lamborghini customers’ behavior on tariffs and possible price increases, Winkelmann said to trust their judgment. “They are maybe millionaires or billionaires for a reason, so they know what they’re doing and why they’re doing things,” he said. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
图片来源：GETTY IMAGES 数周前标普500指数创下历史新高时，笔者曾指出该指数几乎触及里程碑数值——市盈率达30倍。不过，当时我有点“投机取巧”了，正如在那篇文章里提及的：实际市盈率约为29.85倍，只因数值接近整数，四舍五入后便成了30。关键在于，这一数字处于极高水平，然而华尔街分析师和评论员却鲜少提及——他们更倾向于引用基于“明年利润”（这一数据通常被高估）或“营业利润”（剔除利息支出等基本费用）计算得出的更低、更具市场吸引力的市盈率。 但如今，这一数据已载入史册：8月28日下午2点35分，标普500指数攀升至6501点的新高，实际（未经四舍五入的）市盈率达到30倍。该比率基于最关键的指标——过去四个季度依据公认会计准则（GAAP）核算的收益，即实实在在已实现的利润，而非通常过于乐观的预测。近几十年来，唯有1999年第四季度至2022年第一季度科技狂潮席卷市场的十个季度里，大盘股估值曾达到如此高位。（疫情期间及全球金融危机后，市盈率也曾短暂超过30倍，但那只是因为盈利大幅下滑，导致分母数值缩水，使得市盈率被人为推高。） 正如我所指出的，宏观层面危险信号正不断增多。美国劳工统计局最新就业报告显示，7月美国仅新增7.3万个就业岗位，同时大幅下调5月和6月数据，使得过去三个月净增就业岗位总数仅为10.6万个，还不到去年同期增幅的四分之一。海军联邦信贷联盟（Navy Federal Credit Union）首席经济学家希瑟·朗（Heather Long）将这一疲软数据称为“局势逆转点”，直言这表明“劳动力市场正在迅速恶化”。 国内生产总值增长同样令人失望，远低于特朗普政府雄心勃勃的3%的目标。2025年上半年，经济年化增长率仅为1.75%，较去年三、四季度2.7%的平均增速大幅回落。国会预算办公室（CBO）预测2026至2035年经济增速将维持在1.7%-1.8%的低位，远不足以缩减联邦债务——该机构预计联邦债务占国民收入的比例将从今年的100%攀升至2031年的110%。 那么这对当下的投资者意味着什么？市盈率达到30倍，意味着投资者每向标普500成份股投入100美元，仅能获得3美元收益。而就在2022年底，同样投入100美元，还能获得5美元收益。股价飙升并非源于盈利激增——自那以后，盈利增长甚至难以抵消通胀。近年来股价暴涨纯粹源于市盈率飙升，使得股票变得愈发昂贵。这并非预示股市明日、下周或下月必将崩盘，但历史反复证明：当估值攀升至如此高位时，终将回归合理水平。（财富中文网） 译者：中慧言-王芳 A few weeks ago as the S&P 500 hit a new record, this reporter noted that the index virtually hit a landmark reading, a price to earnings ratio of 30. I actually cheated a bit, as I pointed out in the piece: The actual figure was around 29.85, close enough that I rounded it to 30. The point then was, this is a big, big number that you seldom see mentioned by Wall Street analysts or pundits, who’d rather cite a lower, more marketable multiple based on “next year’s” (always over-estimated) profits or “operating earnings” that exclude real charges as basic as interest expense. But now it’s in the record books: At 2:35 PM on August 28, the S&P hit another fresh summit at 6501, and the real, not-rounded-up PE hit 30. That ratio’s based on what matters most, GAAP earnings posted over the last four quarters, profits that really happened as opposed to usually over-rosy predictions. The only span in recent decades when big cap stocks have been this expensive: Ten quarters during the tech frenzy that stretched from Q4 of 1999 to Q1 of 2022. (The PE also briefly exceeded 30 during the pandemic and following the GFC, but only because earnings collapsed, sinking the denominator and skewing the multiple artificially low.) As I noted, on the macro scene, the danger signs are multiplying. The latest employment report from the Bureau of Labor Statistics disclosed that the U.S. added a meager 73,000 jobs in July, and revised the May and June figures radically downward, bringing total net hires for the past three months to just 106,000, less than one fourth the increase for the same period last year. Heather Long, chief economist at Navy Federal Credit Union, described the feeble data as a “game changer” demonstrating that “the labor market is deteriorating quickly.” GDP growth has also proved disappointing, clocking far below the Trump administration’s highly aspirational target of 3%. The economy expanded at an annualized clip of just 1.75% through the first half of 2025, way down from the 2.7% average in Q3 and Q4 of last year. The Congressional Budget Office (CBO) is forecasting tepid expansion of 1.7% to 1.8% from 2026 to 2035, not nearly fast enough to shrink the federal debt that the agency projects will swell from 100% of national income this year to 110% by 2031. So what does that mean for investors now? A 30 PE means you’re getting only $3 in earnings for every $100 you pay for S&P stocks. As recently as late 2022, you were getting $5 for every $100 invested. And the jump in stock prices didn’t occur because earnings soared. Since then, they’ve barely matched inflation. No, the huge ramp in recent years came strictly because PEs jumped, making stocks more and more expensive. That doesn’t mean stocks will crash tomorrow, or next week or next month. But history has proved time and time again that when valuations rise this far into the stratosphere, they are bound to come back to earth eventually. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
图片来源：Abdullah Durmaz—Getty Images 根据LinkedIn最近的一项调查，超过半数的职场人士表示，AI培训让他们感觉仿佛承担了第二份工作。这凸显了在职场自动化项目日益扩张的背景下，员工普遍存在的不满情绪。 多数受访者（51%）认为，AI培训要求的强度和频率过高，已干扰到他们的核心工作职责，并导致职业倦怠。员工们指出，课程内容过于繁杂、截止日期不切实际，以及培训实际效用不够清晰，是引发不满的主要根源。 LinkedIn的数据显示，今年在该平台上谈及“感到压力过大”和“应对变革”的用户数量增长了82%。该公司在报告中写道：“持续加码的AI技能提升压力正在加剧职场人士的不安全感，其中三分之一（33%）承认自己因对AI的理解甚少而感到尴尬，35%表示在工作中谈论AI时会感到紧张，担心暴露自己的认知不足。” 职场影响 在雇主加大投入，推动员工提升技能以适应基于人工智能的新型工作流程之际，这些发现浮出水面。许多职场人士表示，这些培训并未让他们感到更有能力，反而增加了压力并延长了工作时间，而且往往既没有额外补偿，也未带来实际的工作改进。 这种情况带来了切实的后果，也从侧面印证了员工的不安全感并非空穴来风。IgniteTech 首席执行官埃里克·沃恩上月早些时候对《财富》表示，他在员工未能积极响应AI培训后裁减了近80%的人员。而Mindstone的约书亚·沃勒则讲述了一个类似案例：某客户公司的首席执行官要求员工将每周五全部用于AI再培训，如果不能就学习成果作出有建设性的反馈，就被迫离开公司。 调查还发现，在充斥着各类AI相关内容和项目的背景下，职场人士越来越倾向于依赖自己的人脉网络，而非AI工具或搜索引擎，来获得值得信赖的建议和支持，以应对职场变化。约有43%的职场人士表示，“他们的人脉（即认识的人）仍然是职场建议的首要来源”，这一比例高于搜索引擎和AI工具。近三分之二（64%）的职场人士认为，同事的帮助让他们更加快速和自信地做出决策。 对强制性AI培训的日益不满或许只是冰山一角。麻省理工学院（MIT）近期的一项研究发现，95%的企业生成式AI试点项目未能带来任何可量化的投资回报。在企业开支与投资者热情远超实际成果的情况下，这进一步加剧了人们对AI股市泡沫的忧虑。这种现象似乎与低效、举步维艰的AI培训所引发的挫败感密切相关。 麻省理工学院发人深省的研究结论 麻省理工学院的NANDA报告分析了数百个AI部署案例，结果发现，仅有5%实现了营收的快速增长或显著的运营改善。大多数试点项目停滞在测试阶段，或最终被弃用；而大型企业往往需要近一年时间来推动规模化，但成功率极低。报告指出，阻碍的关键不仅在于模型质量，还在于企业整合存在缺陷，以及员工在AI素养方面存在差距。 华尔街及机构投资者已开始敲响警钟，担心创纪录的AI投资未能转化为利润，可能会引发高估科技股的一场痛苦调整。由于担心现实与炒作之间的差距难以为继，部分投资者已开始削减相关持仓，这一幕令人想起过去的科技泡沫。英伟达（Nvidia）的最新财报便折射出市场的紧张情绪：即便营收再创新高，股价仍因投资者抛售下跌了几个百分点。 与职场困境的关联 在企业大举投入AI试点项目和科技股之际，员工们对其商业价值及不断提升技能的要求愈发持怀疑态度。超过半数的职场人士表示AI培训犹如承担了第二份工作，而麻省理工学院的报告则为这一现象提供了新的注解：企业激进推动数字化转型的做法，并未如宣传所言增强员工能力，反而正在消耗他们的精力。 调查结果凸显了技术推广速度与职场人士实际体验之间日益加剧的紧张关系，表明企业或许需要重新审视其AI技能培训的方式，以避免进一步造成员工与公司的疏离。（财富中文网） 为撰写本报道，《财富》杂志使用生成式人工智能协助完成初稿。在发布前，编辑已核实信息准确性。 译者：刘进龙 审校：汪皓 Over half of professionals report that AI trainings feel like a second job, according to a recent LinkedIn survey, highlighting widespread frustration among workers with the proliferation of workplace automation programs. A majority of respondents (51%) find the intensity and frequency of AI training requirements excessive, stating that it’s interfering with their core job responsibilities and contributing to burnout. Employees cited dense training modules, unrealistic deadlines, and a lack of clarity about practical benefits as key sources of dissatisfaction. LinkedIn found an 82% increase in people posting on the platform about feeling overwhelmed and navigating change this year. “The mounting pressure to upskill in AI is fueling insecurity among professionals at work—with a third (33%) admitting they feel embarrassed by how little they understand it, and 35% saying they feel nervous talking about AI at work for fear of sounding uninformed,” LinkedIn wrote. Workplace impact These findings come as employers increase investment in upskilling efforts designed to help staff adapt to new AI-based processes. Instead of feeling empowered, many professionals say these trainings add stress and extend their working hours, often without extra compensation or real improvements to workflow. There are real consequences for this and anecdotal evidence that workers are justified in feeling insecure. IgniteTech CEO Eric Vaughan told Fortune earlier this month that he laid off nearly 80% of his staff after they failed to respond to AI training, while Joshua Wöhle of Mindstone relayed a similar story of a client-CEO who ordered his staff to dedicate all Fridays to AI retraining, and invited them to leave the company if they didn’t report back constructively on their findings. The survey also found that, amid the flood of AI-related content and programs, professionals are increasingly turning to their networks—rather than AI tools or search engines—for trusted advice and support in navigating workplace changes. Some 43% of professionals say “their network, the people they know, is still their No. 1 source for advice at work,” ahead of search engines and AI tools. Nearly two-thirds (64%) of professionals say colleagues are helping them make decisions faster and more confidently. Mounting frustration with mandatory AI trainings may be just the tip of the iceberg. A recent MIT study found that 95% of generative AI pilots at enterprises have failed to deliver any measurable return on investment—fueling growing concerns over an AI stock bubble as corporate spending and investor hype far outweigh results. It seems to be tied to this frustration over ineffective or stumbling AI training efforts. MIT’s sobering findings The MIT NANDA report analyzed hundreds of AI deployments and found only 5% produced rapid revenue acceleration or noticeable operational improvements. The majority of pilots stall in the testing phase or get abandoned, with large companies taking nearly a year to scale projects that rarely succeed. Flawed enterprise integration and a gap in AI literacy—not just model quality—were cited as the main barriers. Wall Street and institutional investors are sounding the alarm, worried that record AI investments aren’t translating to profits and could trigger a painful reckoning for overvalued tech stocks. Some have started trimming exposure, fearing that the gap between reality and hype may be unsustainable, reminiscent of prior tech bubbles. The all-important Nvidia earnings on Wednesday illustrate the jitters, as record revenue still failed to prevent investors taking a few percentage points off the stock. Connections to workforce concerns As companies pour money into AI pilots and tech stocks, employees are increasingly skeptical of both the business value and the constant upskilling requirements. With over half of professionals saying AI trainings feel like a second job, the MIT report adds new context: Companies’ aggressive push for digital transformation is straining workers, not yet augmenting them, as widely billed. The results underscore mounting tension between the pace of technological implementation and the lived experience of professionals, suggesting that companies may need to rethink their approach to AI upskilling to avoid further alienating employees. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
内容衰退？图片来源：Getty Images 美国人观看电视的时长达到了前所未有的高度，但他们愈发倾向于通过流媒体平台而非有线电视观看节目。美国银行研究所（Bank of America Institute）数据显示，今年春季流媒体平台的收视份额正式超越传统“线性”电视，标志着人们消磨闲暇时间的方式发生历史性转变，其规模几乎与线性电视和广播的总和相当。 这种变化更多体现在使用平台的转变，而非需求本身的转变。美国劳工统计局的《美国时间使用调查》显示，美国人每天仍有约5小时用于休闲活动，其中超过一半的时间用于观看电视。 但越来越多的家庭不再频繁切换有线电视频道，转而选择每月付费订阅点播服务。美国银行（BofA）内部支付数据显示，截至7月，流媒体视频和音频支出在所有收入阶层中占比已超10%，自2023年以来，其增速已超过现场活动、影院和主题公园等其他娱乐类别。然而，美国银行高级经济学家大卫·迈克尔·廷斯利（David Michael Tinsley）领导的团队指出，尽管消费者需求不断增长，该行业仍面临新挑战：可能出现“内容衰退”。近年来，内容供应商已减少原创剧集和电影数量，转而深耕为数不多的、制作价值更高的项目。 内容衰退可能以两种形式呈现：一是订购剧集数量减少，强调“质量优于数量”的策略，以维系订户对热门IP的忠诚度；二是必看剧集数量减少，原因是流媒体平台转向低成本、大规模、更接近线性电视的制作模式。早在2023年末，亚马逊工作室（Amazon Studios）前负责人罗伊·普莱斯（Roy Price）在《纽约时报》撰文指出，随着流媒体行业发生根本性变革，精品剧时代已然终结。美国银行分析师警告称，若观众认为缺乏足量的新内容支撑持续上涨的月费，这种趋势可能阻碍行业增长。 用户流失与不平等迹象 美国银行内部信用卡支出数据显示，仍有三分之二的家庭每月流媒体支出低于40美元。但费用正悄然攀升：目前约有六分之一的家庭月流媒体支出超80美元，近10%的家庭月支出超100美元。 与此同时，用户忠诚度并非板上钉钉之事。7月，近五分之一的美国人取消了流媒体订阅服务或开通了订阅服务。这种“流失”表明家庭消费策略灵活：热门新剧或体育赛事上线时选择订阅，热度消退后便弃用平台。倘若内容衰退成为现实，将加速这种动态变化，这可能决定流媒体服务在消费者心中是经济实惠的替代方案，还是与传统有线电视套餐相当的失控开支。流媒体的应对策略似乎是重新包装旧有套餐。 体育、音乐与人工智能——下一个增长前沿 当前流媒体平台正押注两大增长引擎：体育赛事直播与音乐联动。独家体育赛事转播权持续吸引订户，CivicScience调研显示，超三分之一球迷今年秋季愿为观看赛事订阅新服务。橄榄球、篮球及足球赛事的独家转播权，加之音乐与有声读物的优质内容，被视为下一个关键盈利点。女子体育赛事被视为未开发的增长领域，而音乐平台正尝试推出高价套餐，并跨界涉足现场演出领域。 人工智能是关键变量。人工智能能助力主播以更快的速度、更低的成本生成内容与特效，但也降低了新入行者的门槛。长远来看，观众甚至可能借助人工智能工具创作个性化内容，这可能颠覆流媒体公司当前押注的商业模式。 美国人短期内不会减少观看电视的时间。不过，流媒体能否持续吸引观众，更多取决于内容而非技术。若内容衰退加剧，流媒体的主导地位恐将因内容供应减少而面临考验。（财富中文网） 译者：中慧言-王芳 Americans are watching more television than ever, but the screen they turn to increasingly isn’t wired for cable. According to Bank of America Institute, streaming platforms officially overtook traditional “linear” television in viewership share this spring, marking a historic shift in how people spend their downtime. It’s almost as large as linear TV and radio combined. The change is more about delivery than demand. The Bureau of Labor Statistics’ American Time Use Survey shows Americans still devote about five hours a day to leisure, with more than half of that time spent watching TV. But instead of flipping through cable channels, more households are paying monthly for on-demand services. Internal BofA payments data shows streaming video and audio represented more than 10% of spending across all income cohorts as of July, and since 2023, it has outpaced other entertainment categories such as live events, theaters, and theme parks. Yet even as consumer appetite grows, the industry faces a new challenge, writes the BofA team led by senior economist David Michael Tinsley: a possible “content recession.” In recent years, providers have dialed back the sheer volume of original shows and films, shifting instead to fewer projects with higher production values. The content recession could be seen in two ways: either a pullback in the number of shows being commissioned, an emphasis on “quality over quantity” play that may keep subscribers loyal to franchise hits, or a recession in terms of the number of must-watch shows being made as streamers embrace a cheaper, mass production, more linear-TV-like model. As long ago as late 2023, former Amazon Studios chief Roy Price wrote a New York Times opinion piece declaring the era of prestige TV was ending as streaming underwent a fundamental change. BofA analysts caution this could stall growth if viewers feel there isn’t enough fresh material to justify rising monthly bills. Signs of churn and inequality BofA’s internal card-spending data shows two-thirds of households are still paying less than $40 a month for streaming. But the bills are creeping up: Roughly one in six households now spends more than $80 per month, and nearly 10% are shelling out over $100. At the same time, loyalty isn’t a given. Nearly one in five Americans canceled or started a streaming subscription in July. That “churn” means households are nimble, signing up when a hot new series or sporting event arrives, then abandoning platforms when the buzz fades. A content recession materializing could supercharge this dynamic, which could mean the difference between feeling like streaming is an affordable alternative or a runaway expense that rivals old cable bundles. Streaming’s solution to this appears to be reinventing that old bundle in a new form. Sports, music—and AI—as the next frontier For now, streamers are betting big on two growth drivers: live sports and music tie-ins. Exclusive sports rights continue to draw subscribers, with CivicScience finding more than a third of fans are willing to sign up for a new service this fall just to watch games. Exclusive rights to football, basketball, and soccer games, plus premium offerings in music and audiobooks, are seen as the next big moneymakers. Women’s sports are seen as an untapped growth area, while music platforms are experimenting with higher-priced tiers and crossover into live events. The wild card is artificial intelligence. AI could allow streamers to generate content and effects faster and more cheaply, but it also lowers the barriers for new entrants. In the long run, viewers themselves may even use AI tools to create personalized content, potentially disrupting the very model streaming companies are betting on. Americans aren’t giving up screen time anytime soon. But whether streamers can keep those screens tuned in will depend less on technology than on storytelling. If the content recession deepens, streaming’s dominance could be tested by its own shrinking pipeline. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
在中国互联网江湖从二季度掀起的一场史无前例的外卖大战中，美团、淘宝、京东三大巨头的激烈碰撞，不仅让三家公司的利润曲线出现了不同程度的下跌，更让原本稳定的市场格局彻底重塑。伴随着三家公司二季度成绩单的出炉，谁赢谁输也初显轮廓。 根据瑞银最新报告，目前外卖市场格局已从过去的“741”（美团74%、饿了么13%、其他7%）转变为“631”（美团65%、饿了么28%、京东7%）；且三者重合用户数达到3.88亿，同比增长22.8%，这表明用户正在形成多平台使用习惯。而据交银国际测算，仅2025年第二季度，这三家公司在外卖领域的投入就达到250亿元。 这场战事在财务数据上也留下了清晰印记。美团2025年第二季度财报显示，收入约为918.40亿元，同比增加11.7%；经调整净利润约为14.93亿元，同比下降89%，其中核心本地商业板块经营利润同比暴跌75.6%至37亿元，经营利润率从25.1%骤降至5.7%。值得一提的是，美团于8月5日宣布，即日起正式启动中小商户发展扶持计划，向中小商户发放助力金，支持更多餐饮小店实现稳定增长，这意味着美团的运营成本还将提升，短期业绩仍将承压。 实际上，在这场最先由京东掀起的外卖“三国杀”中，京东和美团的大战本来不构成对市场的长期颠覆，但“换帅”后的阿里战略性加入，则令事态演变超出预期。今年4月，阿里整合饿了么资源推出“淘宝闪购”，在强力的补贴下，用户规模和订单量快速飙升。阿里方面透露，淘宝闪购上线仅4个月，月度活跃用户已突破3亿，较4月之前增长200%，日订单峰值达到1.2亿单。 阿里的实力，包括现金流和业务协同，都是京东和美团难以回避的“武器”。阿里财报显示，截至今年二季度末，阿里的现金及其他流动投资为5856.6亿元，这为其提供了充足的弹药。相比之下，美团同期的现金储备约为1711亿元，京东约为2234亿元，与阿里的差距明显。 今年二季度阿里集团收入2476亿元，同比增长2%，但Non-GAAP净利润同比下降18%，外卖大战的巨额投入无疑是首因。阿里大手笔“砸钱”最看重的显然也不只是外卖本身的盈利，而是其带来的生态协同效应。在8月29日的财报分析师电话会上，阿里巴巴中国电商事业群CEO蒋凡强调：“我们不会单独看外卖的盈利情况，考虑到电商的综合收益，我们认为可以在长期保持价格竞争力的前提下，闪购对平台整体产生正向经济收益。” 数据显示，淘宝闪购显著带动了手淘整体用户规模和活跃度，拉动了淘宝8月DAU高达20%的增长，带动了手淘大盘用户活跃天数的提升。流量上涨还带来了广告和CMR业务（客户关系管理）的提升，同时减少了淘宝本身的市场费用投入。 第二季度总营收超3500亿元但净利润接近腰斩的京东，同样试图通过外卖业务带动整体生态。京东财报显示，外卖带来的新用户，在商超品类的交叉销售比例提升了35%，日用百货收入同比增长16%。京东管理层在财报会上也强调：“外卖业务不是孤立板块，而是深度融入京东生态，与零售、物流业务实现协同增效，不追求短期盈利，而是长期市场份额与用户心智的占领”。 而与外卖和即时零售相比，市场更期待看到的，依然是巨头们在AI方面的战略决心和成果——阿里的这份财报以及分析师电话会上高管的回应在这两个方面都给出了超预期的回答。 仅在第二季度，阿里在云计算和AI的资本支出就达到386亿元，同比增长220%，过去四个季度累计投入超1000亿元。这些投入已开始产生实际回报——第二季度阿里云营收近334亿元，同比增长26%，创下三年来最快增速，AI相关产品收入连续八个季度实现三位数增长。同时，通义大模型家族已衍生出超过14万个模型，全球下载量破4亿次，开源生态初步成型。 阿里巴巴集团CEO吴泳铭在电话会上表示，AI技术对所有行业的改变升级以及AI与云计算的深度结合，是未来十年技术领域最大的行业机会；以AI+云为核心的科技平台、购物与生活服务融合的大消费平台，是阿里集团面临的两大历史性战略机遇，“未来三年，阿里巴巴将以创业心态再出发，以驱动业务强劲增长为核心目标，对核心业务持续投入提升竞争优势、获得长期增长充满信心。” 对于三家巨头的表现，资本市场也正在用脚投票。阿里财报公布的当日，其美股股价飙涨13%。以港股来看，今年以来阿里股价累涨近60%，京东和美团股价则分别下跌约10%和32%。这种分化表明投资者似乎更看好阿里的战略布局和长期价值，尽管其短期利润也受到冲击。 不过，随着夏季旺季高峰结束，平台补贴预计也将逐步缓解。瑞银报告指出，当竞争尘埃落定时（可能进入四季度），竞争焦点将转向服务差异化、生态系统协同效应和运营效率。这意味着投资者将看到一场更为复杂的多维竞争：阿里凭借其强大的现金流和生态协同能力持续进攻；美团依靠其成熟的本地生活服务体系努力“守擂”；京东则通过供应链整合寻找差异化优势。 同时值得关注的是，在经过市场监管总局两次约谈后，三大平台近日罕见同步发声，承诺抵制恶性竞争、规范促销行为，同时提出多项限制补贴行为的举措。这场外卖大战的终局，或许不是谁彻底打败谁，而是各自在新的平衡点上找到更适合自己的位置。（财富中文网） 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
2024年12月1日，泰国曼谷市中心的一家PTT充电站，一名比亚迪（BYD）车主正在为其电动车充电。图片来源：Lauren DeCicca/Getty Images • 中国电动车价格已低于燃油车，而在美国，电动车平均仍比燃油车贵约1.4万美元，尽管这一差距自2019年以来有所缩小。与此同时，美国“三巨头”的电动车业务依然未能盈利，规模化扩张也进展缓慢。据专家分析，这与缺乏联邦政府支持性政策密切相关。 中国在电动车竞赛中已经跨越了一个重要里程碑：目前，纯电动车价格低于同级别燃油车。相比之下，美国电动车依然存在高额溢价，汽车数据分析公司JATO Dynamics的最新数据显示，平均溢价高达1.4万美元。 加州大学戴维斯分校交通研究所（UC Davis Institute of Transportation Studies）创始主任丹·斯珀林（Dan Sperling）在接受《财富》采访时表示，1.4万美元的溢价可能被高估了，但他承认确实存在巨大价差。 斯珀林指出，这一巨大价差并不仅仅反映消费者偏好。中国各地政府竞相争夺电动车产业第一，促成了低成本电动车的“白热化竞争”。再加上本土化供应链优势，中国的劳动力和电池成本也更低。 “虽然人们常提到补贴，但目前补贴的影响已微乎其微，”他补充道。 JATO数据显示，中国燃油车的平均售价为2.25万欧元（约合26,205美元），而纯电动车的平均售价则低3%，为2.19万欧元（约合25,509美元）。这与五年前电动车比燃油车贵10%的局面形成巨大反差。 成果随处可见。中国龙头车企比亚迪（BYD）去年销量突破400万辆，是2020年的十倍，如今，从波哥大到布达佩斯的街头，都能看到比亚迪的身影。在欧洲，比亚迪紧凑型车型海豚（Dolphin）的零售价不足2万欧元（约合23,200美元），约是特斯拉Model 3价格的一半。 分析师此前在接受《财富》采访时表示，比亚迪（BYD）通过不断发布新品、严格控制电池供应链以及甘愿牺牲短期利润的战略，正在全面压制传统车企。 斯珀林则警告称，美国过于沉迷于关税博弈，反而忽视了本土电动车产业的发展。他的话印证了那句老话：最好的防守就是进攻。超过100%的高关税虽然挡住了中国车企进入美国市场，但这种保护措施只是在拖延时间，同时也可能让美国车企失去进取心。 “历史早已证明，绝对保护主义往往会削弱而非增强行业竞争力，”他说。 斯珀林指出，由于缺乏直接竞争压力，通用（GM）、福特（Ford）和斯特兰蒂斯（Stellantis）三大汽车巨头对电动车创新的动力不足。 不过，JATO数据显示，过去五年，美国电动车与燃油车的价差已在缩小。2019年，燃油车比电动车便宜44%；到2024年，这一差距缩小至31%。 尽管已有进展，但斯珀林强调，美国仍然缺乏税收抵免、采购配额以及普遍补贴等推动车企规模化生产电动车的结构性政策。 “三巨头”的困境 可以肯定的是，尽管电动车业务亏损不断累积，汽车制造商们仍在大力推进电动车战略。 福特（Ford）本月宣布斥资50亿美元启动新电动车计划，将改造其肯塔基州工厂，以期到2027年投产售价3万美元的电动皮卡，这是其实现电动车规模化生产的雄心勃勃之举。 分析师指出，此举可能成为福特历史性的产业革新，也可能让本已亏损的业务部门再损失数十亿美元。自2023年初以来，福特电动车部门累计亏损已超120亿美元。 通用汽车（GM）今年6月也宣布向其电动车业务在内的本土制造业投资40亿美元。2024年第四季度，通用电动车产品组合实现了“单车利润转正”，即每售出一辆电动车都能覆盖其生产成本（但不包括劳动力或电动车工厂等固定成本）。 通用拥有美国市场第二大电动车产品线，销量仅次于特斯拉。然而，法国巴黎银行Exane（BNP Paribas Exane）高级汽车研究分析师詹姆斯·皮卡列罗（James Picariello）此前向《财富》透露，据其估算，通用去年生产并销售给经销商的18.9万辆电动车共计亏损约25亿美元。 今年早些时候，通用在财报电话会上表示，希望为其电动车业务节约约20亿美元的成本。 斯特兰蒂斯在电动车转型中同样举步维艰。2025年上半年，公司净亏损23亿欧元，营业利润率降至仅0.7%。该车企在美国市场需求疲软，为清理库存被迫大幅下调Jeep Wagoneer S等电动车型的售价。 与此同时，关税壁垒和需求疲软还迫使斯特兰蒂斯延长了意大利特尔莫利工厂的停工时间。尽管如此，公司仍在推进转型：推出的STLA Frame平台采用灵活架构，可适配燃油、混动、电动和氢燃料驱动系统。此外，斯特兰蒂斯还斥资15亿欧元收购中国零跑汽车（Leapmotor）21%的股份，希望通过合作保持竞争力。斯特兰蒂斯期望凭借自身行业优势与知名品牌影响力，结合零跑的创新技术，共同推出更具价格竞争力的电动车。 产业政策失策 业内专家认为，价格差距的部分成因显然应归咎于美国未能通过政策手段有效推动电动车发展。美国总统唐纳德·特朗普（Donald Trump）对电动车的态度反复无常，而其《大而美法案》（Big, Beautiful Bill Act）则彻底取消了对新车、二手车及租赁电动车的税收抵免政策。 斯珀林指出，中国数十年来强制要求外国车企与本土车企合资生产电动车，由此培育出了一支精通电动车技术和软件的人才队伍。他建议美国借鉴类似模式：通过“鼓励合资投资”来加速掌握追赶所需的技术，就像斯特兰蒂斯与中国企业那样的合作模式。 “这样就能培养出大批精通电动车技术的技术人员、工程师和工人，”斯珀林说。“而这正是底特律严重缺失的。” 斯珀林并不认同“底特律注定失败”的说法，但强调其未来完全取决于政策走向。在他看来，美国传统车企目前仍依赖SUV和皮卡“吃老本”，并未在电动车或软件上投入足够资源，而中国竞争对手早已掌握了这些技术。 “如果美国继续阻挡中国车企进入市场，同时打压电动车发展，将需要数十年时间才能赶上，”斯珀林说。“但如果政策发生转变，美国完全可以迎头赶上。”（财富中文网） 译者：刘进龙 审校：汪皓 • China has already made EVs cheaper than gas cars, while U.S. EVs carry a premium of about $14,000, though the gap has shrunk since 2019. Meanwhile, the Big Three remain unprofitable on EVs and slow to scale, a struggle tied to a lack of supportive policies from the federal government, according an expert. China has now crossed a massive benchmark in the electric-car race: battery-powered vehicles are now cheaper than their gas counterparts. In the U.S., by contrast, EVs still face a steep premium; roughly $14,000 on average, according to new data from JATO dynamics, an automotive data analytics firm. Dan Sperling, founding director of the UC Davis Institute of Transportation Studies, told Fortune he thought the $14,000 figure was overestimated – but conceded that there was a strong, real gap. That chasm reflects more than just consumer preferences, Sperling said. In China, there’s also been a “frenzy of competition” to make low-cost EVs, with the heads of different provinces trying to oust each other for the top spot. Labor and battery costs are also lower in China, thanks to its domestic supply chains. “People talk about subsidies, but at this point the subsidy effect is pretty minor,” he added. The average price of a Chinese internal combustion engine is €22,500 (approximately $26,205), whereas a battery electric vehicle costs 3% less, or €21,900 ( $25,509) on average, according to JATO. That’s a big change from just five years ago, when gas EVs cost 10% more. The results are visible everywhere. Leading Chinese automaker BYD sold more than 4 million cars last year—10 times what it sold in 2020—and now dominates roads across the world, from Bogotá to Budapest. In Europe, its compact Dolphin model retails for under €20,000 ($23,200), roughly half the price of Tesla’s Model 3. BYD’s relentless pace of new launches, tight control of its battery supply chain, and willingness to sacrifice short-term profits are overwhelming legacy automakers, analysts previously told Fortune. China’s trade partners also argue that Beijing is fueling overproduction that’s flooding export markets with cut-rate EVs. Meanwhile, Sperling warned that the U.S. is too caught up playing tariff games to develop its own EV industry. His words echoed the old adage that the best defense is offense. Tariffs of more than 100% have kept Chinese cars out of the American market, a protection that may buy time, but also risks making U.S. automakers complacent. “There’s a long history showing that absolute protectionism undermines an industry rather than supports it,” he said. Without the pressure of direct competition, the Big Three of the automotive industry – GM, Ford and Stellantis – have less incentive to innovate with EVs, Sperling said. Still, the U.S. has also improved EV affordability relative to gas cars over the last five years, according to the JATO data. In 2019, gas cars were 44% cheaper than electric cars, and in 2024 the gap narrowed to 31%. But while progress is being made, Sperling said that the U.S. is missing the kind of structural policies – tax credits, purchase mandates, subsidies generally – that spur automakers to build EVs at scale. The struggles of the Big Three To be sure, automakers are attempting large EV pushes, even as EV-related losses pile up. Ford announced a new $5 billion EV initiative this month, where the automaker will reconfigure its Kentucky plant to build a $30,000 electric pickup by 2027, an ambitious attempt to build EVs at scale. Analysts say it could either mark a historic reinvention or sink billions more into an already money-losing division: Ford’s EV arm has racked up more than $12 billion in losses since early 2023. GM also announced in June that it’s investing $4 billion in domestic manufacturing, including its EV wing. In the last quarter of 2024, GM’s electric portfolio became “value profit positive,” meaning that for each electric vehicle sale, GM covers the costs of making each car (but not the fixed costs involved, such as the labor or the EV plants). GM has the second most robust EV portfolio in the American market, sitting behind Tesla in terms of total sales. However, James Picariello, senior automotive research analyst at BNP Paribas Exane, previously told Fortune that he estimated GM lost some $2.5 billion on the 189,000 electric vehicles it built and sold to dealerships last year. Earlier this year, GM said on an earnings call that it hoped to bring about $2 billion in savings improvement for its EVs. Stellantis has also stumbled in its EV transition, posting a €2.3 billion net loss in the first half of 2025 as operating margins shrank to just 0.7%. The automaker has struggled to spark U.S. demand, slashing prices on electric models like the Jeep Wagoneer S to move inventory. At the same time, tariffs and weak demand have pushed Stellantis to extend furloughs at its Termoli site in Italy. Yet the company is still pressing forward: it unveiled the STLA Frame platform, a flexible architecture supporting gas, hybrid, EV, and hydrogen drivetrains. Additionally, Stellantis partnered with China’s Leapmotor in hopes of staying in the game, investing €1.5 billion for a 21% stake in the company. Stellantis hopes that its incumbent advantage and respected brand can combine with the Leapmotor’s innovation to deliver a more affordable EV. Industrial policy failures For industry experts, part of the price gap is clearly attributable to the U.S. failure to promote electric vehicles with policies. President Donald Trump has flip-flopped on his opinion of EVs, but his Big, Beautiful Bill act ended tax credits for new, used and leased EVs. Meanwhile, China’s decades of forced joint ventures – requiring foreign automakers to partner with domestic automakers in EV manufacturing – built a workforce fluent in EV technology and software, Sterling said. For America, he suggested a version of that approach: “encourage joint venture investments” to accelerate the know-how needed to catch up, like the one that Stellantis is doing with China. “You create a whole cadre of technicians and engineers and workers that are adept with the technology,” Sterling said. “Detroit is badly lacking that.” He rejected the idea that Detroit is doomed, but stressed it depends entirely on policy. In his view, legacy U.S. automakers are currently coasting on SUVs and pickups, without making the investments in EVs or software that Chinese rivals have already mastered. “If the U.S. continues to keep out the Chinese and discourages electric vehicles, it will take decades to catch up,” Sterling said. “But if policies change, yes, it can catch up for sure.” 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
特朗普关税战的负面效果正在显现。图片来源：Getty Images 美国人刚刚经历了美国历史上罕见的由国家实施的抢劫，而抢劫的对象是全民。特朗普的关税政策相当于征收了一种全国性销售税，只是换了个名头而已。征收关税能短时间大幅增加财政收入，而债台高筑的美国政府很可能会对此上瘾。特朗普搞的这一套，很可能是因为看欧洲国家眼馋，因为长期以来，支撑欧洲各国政府高额开支的主要支柱就是增值税。虽然特朗普信誓旦旦地说，为关税买单的是外国人，但是最终为关税买单的只会是美国人。 根据各类预测，这项全新“全国性销售税”从美国消费者口袋里抢走的钱，将相当于《大而美法案》延长至2017年的特朗普减税政策所能减免的税额的一半。 现在是时候戳穿关税问题的一些迷雾和不实信息，并详细说明特朗普关税对整个美国经济的真实影响了。 关税：换皮的全国性销售税 除美国以外，几乎所有国家都会对其境内销售的全部或大部分商品征收全国性销售税，这种销费税要么采取统一税率，要么根据商品类型按不同比例征收。这类税主要有两种形式：一是少数国家（如巴基斯坦、缅甸等）会征收一税简单的全国统一销售税，它有些类似美国的州级销售税。二是大多数国家普遍采用的增值税，它在生产的各个环节分阶段征收，德法等老牌经济体和所有欧盟国家都是以消费税为主要税收来源。增值税能带来巨额财政收入，有助于加强欧洲各国政府的支出，但也给这些国家的经济增长造成了巨大拖累。 美国的情况则极为特殊，美国从未征收过增值税，在二战后也从未征收过任何类似全国性销售税的税种。美国的联邦开支占GDP的比重至少比欧洲主要国家低10个百分点以上，不征收增值税可能是一个重要原因。但特朗普关税政策的出台，标志着美国向征收全国性消费税迈出了历史性的一步。而美国的做法也相当简单粗暴，那就是以近年来全球主要经济体罕见的力度，大肆加征关税。 尽管在19世纪80年代至二战刚结束的这段时期，关税也曾一度是美国联邦财政的重要收入来源，但此后它在联邦财政收入中的占比一直相对较低。到2024年，即便拜登政府保留了特朗普第一任期内加征的大部分关税，美国的平均关税率也仅为2.5%，关税收入仅为每年770亿美元，约占联邦税收总额的1.6%。 而特朗普的二次上台，却让关税收入不再是联邦财政收入中无足轻重的“小角色”。需要明确的是，关税与增值税的运作方式存在显著差异，关税由美国海关和边境保护局在入境口岸征收，且完全由美国进口商支付。根据特朗普的计划，即便是同一种进口商品，关税率也会因原产国的不同而存在显著差异。例如，美国对加拿大生产的铝和钢材征收50%的关税，对英国则仅征收25%的关税。通常情况下，进口商会将关税成本转嫁到汽车销售商、连锁超市和大型卖场的商品定价中，最终让美国的普通消费者在消费这些成本时支付更高的费用。从这个角度来看，特朗普关税与德国慕尼黑某家超市的商品价格中所包含的德国增值税，或者明尼阿波利斯市的一家沃尔玛超市的商品价签上所标注的明尼苏达州销售税，并无本质区别。 另外，关税与增值税都是“间接税”，都会被嵌入到零售价格中，而不是在消费时单独向你收的。但是销售税则是在销售环节单独叠加，而且金额会清晰地显示在收据上，因此销售税的税率在上调时往往会引发民众不满。而关税的一大隐患在于，它可能会成为一种类似增值税的“暗箱增税”手段。 总体上看，根据耶鲁大学预算实验室的数据，特朗普政府已将美国的平均实际关税税率提升至18.6%。这一惊人数字已接近1930年《斯穆特-霍利关税法案》所规定的19.8%的水平。《斯穆特-霍利关税法案》在美国历史上臭名昭著，被认为是大萧条时期导致美国经济进一步下滑的重要因素之一。 那么，特朗普的关税政策，究竟会给美国老百姓带来多沉重的成本？根据无党派机构税收基金会（Tax Foundation）的预测，到2025年底，美国的年均关税收入将达2100亿美元。这个数字已经相当庞大了，相当于2024年美国个人所得税总额的9%以上，相当于美国企业所得税总额的40%。而耶鲁大学预算实验室的数据显示，到2034年，特朗普关税总共将从美国老百姓的口袋里掏走2.7万亿美元。 特朗普关税也将成为美国历史上规模最大的增税政策之一。在这种专业问题上，普通老百姓看不懂也是情有可原的。特朗普将《大而美法案》的减税政策延长至2017年，并将其称为“促增长”的核心政策。根据无党派机构国会预算办公室的计算，此举将使美国明年的税收比原本应有的水平减少约4000亿美元。但到 2026年，特朗普关税还会带管额外的2000亿美元税收收入，从而抵销掉其中的一半减免额度。也就是说，特朗普一边看似给美国普通老百姓省了钱，但另一边却又从老百姓的口袋里拿走了一部分。 关税不会导致通胀 当然，批评人士有很多合理理由反对关税，但说关税会加剧通胀的说法却是错误的。 通胀的定义，是一个国家的整体物价水平全面持续上涨。通胀并非是由税收问题引发的——不管是销售税还是关税这种“事实上的销售税”。正如传奇经济学家、诺贝尔奖得主弗里德曼所言，“无论何时何地，通货膨胀本质上都是一种货币现象。”而在所有出现严重通胀（即年通胀率超过4%，且持续至少两年以上）的国家中，在爆发通胀前，无一例外都出现过货币供应量大幅增加的情况。 以美国的上一轮通胀为例。疫情以来，以M2衡量的美国货币供应量出现了自 1913年美联储成立以来在和平时期最快的增长。2021年2月，美国M2的年增长率创下26.7%的纪录。2020年2月至2021年2月期间，M2的年增长率达到了惊人的21.0%。正如白天过后必将迎来黑夜，随着M2的飙升，美国的通胀率也随之飙升。2022年6月，美国年通胀率达到9.1%的峰值。2021年4月至2022年12月期间，美国年通胀率达到7.0%。这场通胀并非短期现象，也并非由供应链问题或其他各种非货币因素导致的，它发生的原因纯粹而简单——货币供应量的激增。 尽管关税会推高进口商品和服务的相对价格，但它并不会改变整体物价指数。特朗普关税只会让美国老百姓在消费汽车、钢铁和铝制品、欧洲葡萄酒以及所有其他需要缴纳高额进口关税的商品上花更多的钱。而除非货币供应量增加，否则美国老百姓花在非进口商品和服务（比如从国内机票、餐饮等）上的钱就会减少。这就是所谓的“此消彼长”。而只有当美联储再次大幅增加货币供应量时，通胀的阴影才会再次浮现。 20世纪70年代日本的一个典型案例就印证了这一点。日本几乎所有的石油都依赖进口。1973年阿拉伯石油禁运后，国际原油价格上涨了近三倍。为了缓解油价飙升带来的冲击，日本央行在1972年初至1973年底期间，将货币供应量增加了近50%。毫不意外，1974年日本的通胀率飙升至23%以上。 而当20世纪70 年代，第二次石油危机于1978年冲击日本时，日本央行从第一次石油危机中吸取了教训，选择不通过增加货币供应量来“缓解冲击”。事实上，日本银行还略微收紧了货币政策。1979年至1980年期间，日本M2年均增长率从1976年至1977年的13.3%降至10.6%。这一次，与第一次石油危机期间的高通胀不同，一直到1980年年底，日本的通胀率始终稳定在4.9%。 关税既无法消除也无法大幅缩减贸易逆差 特朗普经常拿美国的贸易逆差来说事。他声称，美国巨额的进出口逆差给美国经济造成了沉重负担，只有通过加征关税才能大幅缩减。对此，他提出了一套二重解决方案：一方面对进口商品征收高额关税，以减少外国对美出口；另一方面向外国施压，迫使它们开放市场，从而让美国向这些国家出口更多美国制造的商品。特朗普还频繁指责美国的贸易伙伴“作弊”，还说他们无所不用其极地“掠夺”美国的财富。他还说，美国的贸易伙伴正通过种种不正当手段“剥削”美国，而征收高额关税，是消除美国贸易逆差的唯一途径。 但是，特朗普的这些观点都是错误的。道理其实很简单，这是一个数学问题。美国贸易逆差居高不下，根源在于美国人的支出超过了美国国内的产出。一个著名的经济学公式——国民收入恒等式揭示了真相。在国民收入恒等式中，支出可以分为消费、投资和政府开支三大类，商品和服务的总产出价值等于GDP，而总之出之和减去GDP，就等于进出口之间的差额。 这个恒等式可以被数据完美佐证。2024 年，美国的总支出（31.2万亿美元）与总产出（30.33万亿美元）之间的差额为8725亿美元，这与美国当年的整体贸易逆差完全相等。简而言之，贸易逆差源于美国人选择了“消费远超产出”的生活方式，而非外国在剥削你们。自1974年以来，美国每年都存在贸易逆差，而这完全是美国人自作自受的结果。 但事实证明，美国要想找钱填这笔窟窿并不难，外国人非常愿意向美国输送资本，将资金投入以美元计价的资产，比如购买美国股票、美国企业债券和房地产等。这一切都得益于美国拥有“世界主要储备货币”所带来的“超级特权”。 关税将严重拖累经济增长 特朗普有一个观点可能是正确的。如果美国国会保留他2017年推出的减税政策，那么美国经济增长率或将显著提高。但奇怪的是，他并未意识到，关税本质上就是对美国人征收的销售税，而且是一笔高额的销售税。而且关税至少会让《大而美法案》的减税效果抵消一半。与所有税收一样，关税会对经济增长造成拖累。据耶鲁大学预算实验室预测，从长期来看，这些额外关税将使美国GDP年均增长率降低0.4个百分点。而自2000年以来，美国的年均经济增长率仅为2.2%，所以关税的负面影响不可谓不重大。 所以，美国人即将背负的，是一笔很多人都没意识到的“隐形增税”负担。美国本是一个全世界商品最丰富的“大超市”，拥有全球种类最全、价格最优的商品。但特朗普关税的出台，很可能会造成劣币驱逐良币的效果，导致全球各地最物美价廉的商品被逐出美国市场，最终导致美国市场上的商品种类更少，而价格更高。 本文作者史蒂夫·汉克是约翰・霍普金斯大学的应用经济学教授，曾任里根总统的经济顾问委员会委员。他与马特·塞克克合著的新书《让钱发挥作用：如何改写金融体系规则》（Making Money Work: How to Rewrite the Rules of Our Financial System）已于今年5月由威利出版社（Wiley）出版。汉克被誉为 “货币医生”，40多年来一直在为各国总统和总理提供政策建议。（财富中文网） 译者：朴成奎 Buckle up, Americans. You’ve just been ambushed by a federal money grab that this nation hasn’t witnessed in recent history, the equivalent of a national sales tax—branded under another name. This version’s a giant revenue raiser that deficit-ridden Washington is likely to get hooked on. In fact, it’s looking more and more like America’s answer to Europe’s fatal attraction that’s long been the leading enabler for the region’s extremely high levels of government spending: the value-added tax—the VAT. As I’m sure you’ve guessed by now, we’re talking about the Trump tariffs. And you can forget about foreigners picking up the tab for the tariffs. It’s Americans who will. According to a wide range of forecasts, this new nationwide sales levy will yank back from consumers’ wallets about half the projected tax savings from the One Big Beautiful Bill’s extension of the 2017 Trump tax cuts. Unmasking the Trump tariffs as a national sales tax helps clarify the burden looming for Americans’ finances going forward. It’s time to dispel the widespread confusion and misinformation that surrounds the tariffs’ role, and detail their true impact across the economy. Tariffs are a national sales tax by another name Almost all countries except the U.S. impose national sales taxes on all or most goods sold anywhere within their borders, either at a flat rate or at varying percentages depending on the product. These taxes come in two forms: A few nations, including Pakistan and Myanmar, deploy a simple countrywide tax similar to America’s state sales taxes. But the most common type by far is the value-added system imposed in stages at each level of production. It’s a bedrock of the regimes in France, Germany, and virtually every other EU country. The VAT is a formidable revenue generator that has fueled government spending in Europe and imposed a huge drag on its nations’ growth. The U.S. is highly unusual in that it’s never had a VAT or in the postwar era, anything resembling a major national sales tax. This stand-alone status in shunning the former likely explains why America’s federal spending stands at least 10 percentage points lower as a share of GDP than the lions of Europe. But the Trump tariffs mark a historic shift to an effective national sales tax. By employing huge tariffs, Uncle Sam is embracing a tax that’s virtually never been used at remotely this scale by any other major economy in recent history. Although tariffs were an important source of federal funding from the 1880s to just after World War II, they’ve amounted to a relatively small portion of U.S. federal receipts ever since. In 2024, even though the Biden administration retained most of the tariffs from Trump’s first term, the average U.S. tariff ran at just 2.5%, raising only $77 billion, or around 1.6% of all federal tax proceeds. By making tariffs more than a minor footnote in the federal budget, the Trump administration takes this long-fading mode of taxation into uncharted fiscal territory. To be clear, tariffs operate differently from VATs. They’re collected at U.S. ports of entry by the U.S. Customs and Border Protection agency and are paid entirely by American importers. Under the Trump plan, the rates vary greatly even for the same goods, depending on what country they’re arriving from. On aluminum and steel, we’re charging Canada 50% and the U.K. “just” 25%. Typically, importers tack the tariff cost onto the price of their products in auto showrooms, grocery chains, and megastores. Hence, they raise the bill for American consumers at checkout counters. In this sense, tariffs are no different than the German VAT that marks up tabs at the supermarket chain Edeka in Munich or the Minnesota state sales tax that gets added to Walmart price tags in Minneapolis. By the way, tariffs resemble VATs in another sense. Tariffs are “indirect” taxes embedded in retail prices, like VATs, and are not added at checkout. Sales taxes, on the other hand, are tacked on at the point of sale and visible on receipts, and usually spark outrage when they are increased. A danger of tariffs is that they could become a VAT-like way to hike taxes on the sly. All told, according to the Budget Lab at Yale, the Trump administration has raised the average effective tariff rate to 18.6% for Americans. This towering figure is approaching the devastating 19.8% rate imposed by the Smoot-Hawley Tariff Act of 1930, a notorious measure that helped sink the U.S. economy during the Great Depression. Just how big will the tariff tax hit be? According to the nonpartisan Tax Foundation, by the end of 2025, the run rate is predicted to be $210 billion annually. That’s a big number. It equates to over 9% of personal income taxes and 40% of corporate taxes paid in 2024. According to the Budget Lab at Yale, the Trump tariffs will lift a total of $2.7 trillion from Americans’ pockets through 2034. The Trump tariffs will amount to one of the largest tax hits in U.S. history. You can’t blame Americans for being flummoxed. President Trump lauds the One Big Beautiful Bill’s extension of his 2017 tax reductions as central to the POTUS’ pro-growth agenda. According to the nonpartisan Congressional Budget Office, the extension of Trump’s reductions will reduce taxes by around $400 billion relative to what they would have been next year. Trump’s tariffs will grab back half of those savings via the $200 billion–plus tariff smack in 2026. So Trump’s putting more money in families’ pockets with one hand and taking a big chunk of it back with the other. Tariffs don’t cause inflation Critics of tariffs have plenty to legitimately complain about. But they’re wrong in swearing that the duties feed inflation. Inflation is defined as a broad-based, persistent increase in a nation’s general price level. It is not triggered by taxes, whether they’re sales taxes or de facto sales taxes, like tariffs. Instead, as legendary economist and Nobel laureate Milton Friedman famously stated: “Inflation is always and everywhere a monetary phenomenon.” Indeed, we have never observed significant inflation, a period in which the annual inflation rate exceeds 4% and lasts for at least two years, in any country in which the money supply has not significantly increased prior to the outbreak of inflation. Just consider our last bout of inflation in the United States. Following the pandemic, the U.S. money supply, measured by M2, experienced the fastest rate of peacetime growth since the Fed’s founding in 1913. The annual growth rate of M2 peaked in February 2021 at 26.7%, and that annual rate averaged a stunning 21.0% from February 2020 to February 2021. As night follows day, inflation surged. In 2022, it peaked at 9.1% per year in June, and it averaged 7.0% per year between April 2021 and December 2022. The inflation wasn’t temporary and wasn’t caused by supply-chain “glitches,” or a variety of any other nonmonetary factors that were thrown up as explanations for the inflation episode. It was caused, pure and simple, by the surge in the money supply. While tariffs will increase the relative prices of imported goods and services, they will not change the overall price index. The Trump tariffs will simply force Americans to spend more on cars, products made from steel and aluminum, European wines, and all other items heavily taxed at our borders. Unless the money supply is goosed, Americans will have less to purchase products and services that aren’t shipped from abroad—everything from domestic plane tickets to restaurant meals to soft drinks. It will be more or less a wash, with the extra money spent on tariffed goods matching the drop in what’s spent on everything else. The specter of inflation will only reappear if the Fed balloons the money supply. A classic example from Japan in the 1970s illustrates this point. The island nation imports virtually all of its oil. With the Arab oil embargo of 1973, the world price of crude almost quadrupled. In an attempt to soften the blow from this surge in price, the Bank of Japan (BOJ) juiced its money supply by almost 50% from the start of 1972 to the end of 1973. Not surprisingly, inflation spiked to over 23% per year in 1974. When the world’s second oil shock of the 1970s hit Japan in 1978, the BOJ, which had learned its lesson from the first oil shock of the decade, chose to not “soften the blow” by increasing the money supply. Indeed, the BOJ tightened its monetary policy a bit, with average annual growth in M2 falling from 13.3% between 1976 and 1977 to 10.6% between 1979 and 1980. Unlike the inflation episode that accompanied the first oil shock, inflation remained steady at 4.9% through the end of 1980. Tariffs will not eliminate or even significantly shrink the trade deficit President Trump denounces America’s trade deficits. He asserts that the big shortfall between what America exports and imports constitutes a crushing burden on the economy and can only be substantially reduced by the imposition of tariffs. His twofold solution: Slap high tariffs on imports so that foreign nations sell us far less and pressure foreigners into opening their markets so that the U.S. can ship them bigger volumes of stateside-made products. The president regularly blasts America’s trading partners for “cheating” and deploying all manner of punitive practices to “loot” and “fleece” the U.S. He argues that America’s trading partners are treating Uncle Sam unfairly in a ripoff scheme. According to the president, the only way to eliminate America’s trade deficits is with the imposition of steep tariffs. The president’s arguments are incorrect. It is all in the arithmetic. The deficits are bulging because Americans spend more than this nation produces domestically. The truth is revealed by a famous economic identity. Spending falls into three categories: consumption, investment, and government expenditures. Whereas the value of all goods and services produced is equal to the gross domestic product (GDP), the gap between the aggregate spending (C+I+G) and GDP is by definition equal to the difference between exports and imports. And the numbers work perfectly. In 2024, the difference between what the U.S. spent ($31.2 trillion) and what it produced ($30.33 trillion), was $872.5 billion. That precisely equaled the overall trade deficit. Put simply, the deficit arises from a choice we’ve made as Americans to consume much more than we make. It is not the result of nefarious activities by foreigners. In short, since 1974, the U.S. has witnessed a trade deficit each year, all made in the USA. As it turns out, these deficits have been easy for the U.S. to finance. Foreigners are more than willing to send capital to the U.S. so that the funds can be invested in dollar denominated assets by purchasing U.S. stocks, corporate bonds, and apartment complexes, to name a few places where foreign capital flows in and rests in the U.S. All of this is part of the exorbitant privilege associated with possessing the world’s premier currency. Tariffs will prove a big downer for growth President Trump argued correctly that the U.S. would grow much faster if Congress retained his 2017 tax reductions. Strangely, Trump doesn’t realize that tariffs are nothing more than a sales tax on Americans, and a big one at that. Indeed, they promise to negate at least half of the tax reductions contained in the One Big Beautiful Bill. Like all taxes, the tariffs will impose a drag on economic growth. Not surprisingly, the Yale Budget Lab forecasts that the extra duties will slow GDP growth by a substantial 0.4 percentage points per year over the long run. That would amount to a huge hit to the average annual growth rate of 2.2% that the U.S. has experienced since 2000. Americans are about to shoulder a big, unadvertised, stealth tax hike. This nation is the world’s most abundant supermarket, offering the best prices and greatest variety of goods on its shelves. The Trump tariffs are poised to create a far more limited and expensive marketplace by sweeping from those shelves the best bargains arriving from across the globe. Steve H. Hanke is a professor of applied economics at Johns Hopkins University. He served on President Ronald Reagan’s Council of Economic Advisors. His most recent book, coauthored with Matt Sekerke, Making Money Work: How to Rewrite the Rules of Our Financial System, was released by Wiley in May. Known as “the Money Doctor,” he’s advised presidents and prime ministers for over 40 years. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
图为2022年10月7日，美国联邦最高法院大法官的合影。图片来源：Olivier Douliery—AFP via Getty Images • 据Capital Alpha Partners公司合伙人伊恩・卡茨介绍，美联储理事的任期一般是14年，但是他们一般不会干满，而是会在任期结束前主动卸任。不过有的理事也可能会一直等到自己政党这边的总统上台后再卸任。另外美联储在利率问题上全票通过的情况可能将变得越来越少见，意见分歧反而或将成为常态。 随着特朗普上台以来，不断加大对美联储的施压力度，这个一向以保守和共识著称的专业性机构，很可能会变得像美国最高法院一样愈加分裂。 自从特朗普重返白宫以来，特朗普一直要求美联储启动降息，还因为美联储主席杰罗姆·鲍威尔主席拒不降息而频繁对其进行指责。他还暗示自己有可能解雇鲍威尔，不过之后他又收回了这一说法。现在特朗普又威胁称，如果美联储理事莉萨·库克不主动辞职，他就会将其解雇。 对此，丽萨·库克表示，她绝不会在胁迫下辞职。她还表示将对特朗普政府官员对她提出的抵押贷款欺诈指控进行反诉。不管她的官司会打得怎么样，有一个问题都是现实的——她还会选择留任多久。 库克是2022年加入美联储的，当时美联储的一位理事任期未满便宣告辞职，库克在老拜登的提名下临时补缺，后来又获得了连任。因此，她可以在美联储理事会干到2038年，不过理事们通常不会干满整个14年的任期。 Capital Alpha Partners公司的管理合伙人伊恩·卡茨在上周三的一份报告中指出：“现在，美联储已经越来越像一颗政治足球。特朗普已明确表示，他想让忠于自己的人进入美联储理事会。因此，有的理事可能会选择留任至自己政党这边的总统入主白宫——这样一来，美联储的情况可能会越来越像最高法院。” 与此同时，特朗普已提名白宫经济顾问委员会主席斯蒂芬·米兰填补阿德里亚娜・库格勒留下的空缺。库格勒的任期原定明年1月份到期，她也是提前卸任的。 斯蒂芬·米兰支持特朗普的降息政策。更值得注意的是，2024年，他还跟人合著了一篇文章，主张对美联储进行改革，以削弱其独立性。 这一因素很有可能影响库克对自身任期的决定。卡茨指出：“以前理事们在离任时，并不担心总统会安排一个不坚定支持美联储的独立性的人来接替自己。” 同样，鲍威尔本人的离任计划也受到了人们的关注。作为美联储理事会主席，他的任期要到明年5月份才结束。但是作为理事，他的任期则会延续至2028年1月。 美国财政部长斯科特・贝森特表示，鲍威尔应在主席任期结束时主动辞去理事一职，并称这是一贯的传统。但鲍威尔拒绝透露自己的计划。 此事的重要性，影响的不止是美联储的降息幅度问题。摩根大通的分析师甚至警告称，米兰的任命对美联储构成了“生存威胁”，因为这表明特朗普政府有意修订《联邦储备法》，并对美联储的权力进行调整。 决策分歧 目前，特朗普还在物色新一任美联储主席人选，在此背景下，米兰是否会重新被任命为美联储理事，目前尚不确定。但无论如何，美联储很快将有3名特朗普任命的理事。 当然，考虑到联邦公开市场委员会（FOMC）一共有12名委员，这3人光靠自己，还无法左右FOMC会议的利率决策——毕竟参加FOMC会议的还有美联储的各地区联储银行行长。但是如果特朗普能够任命第4名理事，就足以改变由7人组成的理事会中的政治平衡。 正如新闻网站Axios最近所指出的那样，若理事会中特朗普任命的理事占到多数，他们将有权掌控美联储的预算和人事安排，甚至有权决定地区联储银行行长的人选。地区联储银行行长由地区联邦储备银行的董事任命，但需得到美联储理事会的批准。而到明年2月份，所有地区联储银行行长的5年任期都将到期。 随着美联储高层的人事变动，美联储或将迎来一个更加分裂的时代，使它变得越来越像现在的美国最高法院。 以前，美联储的利率决策通常是全票通过的，就连一票反对也极为罕见。相比之下，美国联邦最高法院很少出现全票通过的情况，而基于意识形态分歧的决策分歧也是十分常见的。 7月份的美联储会议或许已预示了未来的形势走向——在这次会议上，两名由特朗普任命的理事投票支持降息，与支持维持利率不变的多数派意见相悖。 尽管鲍威尔为9月降息敞开了大门，但这也不能保证美联储高层能在9月份的会议达成共识。因为FOMC会议的其他委员，比如堪萨斯城联储银行行长杰弗里・施密德，仍然坚定反对降息。 这也就意味着，下次FOMC会议上有可能再次出现反对票。此外，虽然特朗普任命的这几位官员正在努力推动宽松政策，但后续的降息节奏也尚不明朗，这说明美联储内部还有很多争论空间。 与联邦最高法院首席大法官一样，美联储主席也仅有一票表决权，但他仍有超越其他理事的影响力。因此，无论是谁接替鲍威尔，在美联储愈发分裂的环境中，都要依赖自己的说服力来开展工作。（财富中文网） 译者：朴成奎 • Federal Reserve governors typically step down before their 14-year terms expire, but they may stay on until a president from their political party is in the White House, according to Ian Katz at Capital Alpha Partners. Meanwhile, unanimous votes on Fed rates could become less common, with split decisions becoming more of the norm. As President Donald Trump ramps up pressure on the Federal Reserve, the typically staid, consensus-driven institution could take on some qualities of the more bitterly divided Supreme Court. Since returning to the White House, he has demanded that the Fed cut rates and routinely insults Chairman Jerome Powell for not doing so. After teasing that he could fire Powell then backing off, Trump has threatened to fire Fed Governor Lisa Cook if she doesn’t resign. For her part, Cook said she won’t be bullied into stepping down and plans to rebut accusations of mortgage fraud from a Trump administration housing official. That’s raised the question of how long she might choose to serve. Cook joined the Fed in 2022 after being tapped by President Joe Biden to fill an unexpired term that ended in 2024, then getting reappointed. So she can stay on the Fed board until 2038, though governors typically don’t serve out their entire 14-year terms. “However, the Fed has increasingly become a political football,” Ian Katz, managing partner at Capital Alpha Partners, said in a note Wednesday. “Trump has been clear that he wants to put loyalists on the board. As a result, some governors may choose to remain on the board until a president from their same political party is in the White House — making the Fed in that way more like the Supreme Court.” Meanwhile, Trump has named Stephen Miran, chair of the White House’s Council of Economic Advisers, to fill a vacancy on the board left by Adriana Kugler, who stepped down before her term was due to expire in January. He has backed Trump’s call for lower rates. More notably, Miran also cowrote a paper in 2024 calling for an overhaul of the Fed that reduces its independence. That could factor into Cook’s decision on how long she will stay. In his note, Katz observed that “governors in the past have stepped down without concern that the president would nominate a replacement who isn’t a strong believer in Fed independence.” Similarly, Powell’s own plans have come under scrutiny. While his term as board chair expires in May, his term as a governor extends to January 2028. Treasury Secretary Scott Bessent has said Powell should step down as governor when his term as chairman ends, saying that has been the tradition. Powell has declined to say what he will do. The stakes could go well beyond how much the Fed lowers rates. Analysts at JPMorgan have even warned that Miran’s appointment represents an “existential threat” to the Fed as it signals an intention to amend the Federal Reserve Act and alter the central bank’s authority. Split decisions It’s not clear if Miran will be reappointed to the Fed board as the White House looks for someone to replace Powell as chairman. But either way, the Fed will have three Trump-appointed governors. To be sure, that’s not enough to sway rate decisions on the 12-member Federal Open Market Committee, which is also comprised of regional Fed presidents. But if Trump is able to name a fourth governor, that’s enough to tip the balance on the seven-member board. As Axios recently pointed out, a board majority would give Trump appointees power over the Fed’s budgets, staffing, and even selection of regional Fed presidents. Those presidents are appointed by directors of the regional Fed banks, but they are subject to the approval of the board. And in February, the five-year terms for all the bank presidents are scheduled to expire. With composition of the Fed in flux, a more divided era may be looming that also resembles the Supreme Court. Fed rate decisions are usually unanimous with even one dissenting vote being rare. By contrast, the high court rarely has unanimous votes, while split decisions along ideological lines are common. July’s Fed meeting may have been a preview of what’s to come as two Trump-appointed governors voted to lower rates, going against the majority that kept rates steady. And although Powell opened the door to a rate cut at the September meeting, that doesn’t guarantee a consensus either as other FOMC members still sounded hawkish, such as Kansas City Fed President Jeffrey Schmid. That sets up another FOMC meeting with dissenting votes. In addition, the pace of any subsequent cuts isn’t clear, providing more fodder for debate at the central bank as Trump-appointed officials push for dovish policy. Like the chief justice of the Supreme Court, the Fed chair represents just one vote but is also a first among equals who carried outsized influence. So whoever replaces Powell may need to rely on their powers of persuasion on a Fed with more conflicting views. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
Meta计划在里奇兰教区建设的数据中心园区效果图显示，大批设施将为人工智能热潮提供动力。图片来源：Meta 在路易斯安那州东北部那片宁静的昔日农田上，一队挖掘机已将超过2000英亩（约合8093729.6平方米）的红黏土土地推平。此地处于里奇兰教区的乡村地带，过去曾是洪水泛滥的平原，蜿蜒曲折的支流纵横交错，野生芦苇丛肆意生长，黑熊仍在此间出没。当地2万居民中，有四分之一生活在贫困线以下。 如今，全球市值第六大的公司——Meta进驻此地。这家科技巨头渴望将里奇兰教区打造成承载其宏大人工智能愿景的基地，而要达成这一目标，离不开大量新建燃气发电厂提供的能源支撑。该地区土地广袤无垠，且毗邻路易斯安那州庞大的海恩斯维尔页岩气田。 去年12月，Meta启动了其迄今为止规模最大的数据中心建设：这个斥资100亿美元打造的园区，涵盖9栋建筑，将排列着一排排服务器，总占地面积超400万平方英尺（约合37.2万平方米），比迪士尼乐园还要大。 Meta董事长兼首席执行官马克·扎克伯格（Mark Zuckerberg）并未就此止步。7月，他将该项目命名为Hyperion——数据中心“超级集群”，最终能耗可能相当于400万户家庭的总用电量，有望成为全球最大的数据中心项目。扎克伯格表示，Hyperion的占地面积将相当于“曼哈顿的相当大一部分区域”。 该项目规划的算力能耗高达2吉瓦，扎克伯格称最终可能扩展至5吉瓦，用于训练开源大型语言模型。Meta在人工智能竞赛中因此前失败的项目以及耗资数十亿美元却收效甚微的“元宇宙”计划而落后。如今，他将Hyperion项目及大规模建设计划定位为对“超级智能”的追求，同时以2.5亿美元的薪酬待遇挖角人工智能人才，并收购Scale AI 49%的股份。 这无疑是科技巨头在人工智能领域掀起的又一场波澜壮阔的竞争，Meta正与谷歌、微软、亚马逊以及OpenAI等公司展开角逐。 “我们之所以进行这些投资，是因为我们坚信，超级智能将优化我们所做的一切。”扎克伯格在7月30日的Meta财报电话会议上表示。Meta的一位发言人向《财富》杂志透露，目前无法确切阐明该园区将为哪些业务提供动力，因为到2030年园区启用时，人工智能技术将发展到何种程度仍是未知数。 该项目规模之庞大，令这片原本宁静地区的当地人深感震撼。 “我想，和不少人一样，听闻如此偏远的乡村地区被挑中建设这样的项目时，我第一反应是有点难以置信。”附近雷维尔镇第一浸信会教堂的牧师贾斯汀·克拉克（Justin Clark）表示，“随着对项目内容与规模的了解逐渐深入，那种震惊感始终萦绕不散，实在是令人难以置信。” 克拉克期待着迎接新工人来到该地区，但他也承认，实在难以真切想象这一项目的宏大规模。在最近一次商会宴会上，众人得知这是北美最大的建筑工地：“这太不可思议了。”他惊叹道。 总体而言，大型科技公司新建的数据中心对能源和水资源的需求极为巨大。单是维持Hyperion项目服务器冷却及正常运行，所需电力就将达到新奥尔良市用电量的两倍，且未来这一需求还将持续攀升。 随着人工智能热潮加速推进，人们纷纷猜测，公用事业公司将如何满足大型科技公司不断攀升的电力需求。以Meta（《财富》美国500强排名第22位）为例，区域公用事业公司安特吉（Entergy）将新建三座燃气涡轮机，总装机容量达2.3吉瓦——这是该地区数十年来首次扩建此类设施。此举引发双重反对声浪：一方面，用户担心用电成本会上涨；另一方面，气候倡导者担忧这会导致绿色能源目标出现倒退。 在人工智能主导地位的争夺战里，公用事业公司已然成为超大规模科技公司市场的“守门人”。它们必须权衡两方面因素：一方面，是为前景尚不明朗的新兴行业开展大规模资本投资所能获取的收益；另一方面，则是未来数十年可能面临的电费上涨及资产搁浅（即因技术或市场变动，致使投资建成的资产无法收获预期收益）风险。 州监管机构于8月20日批准了安特吉的计划，比预期提前两个月，这一决策可能为未来公用事业公司与大型科技公司在合作新建发电厂方面的交易树立标杆，尤其是在土地成本较低的乡村地区。 路易斯安那州公共服务委员会委员达万特·刘易斯（Davante Lewis）告诉《财富》杂志：“这笔交易可能向其他州发出信号，展示数据中心在管理与运营方面的正确路径。这将成为美国全国范围内的测试案例。我从投资者、信贷机构以及其他数据中心从业者那里都听到过类似的说法——Meta这笔交易的最终成果，很可能会为所有类似项目搭建起框架。” Meta正在推平路易斯安那州里奇兰教区的大片土地，用于建设其数据中心园区。图片来源：Meta 以Meta的Hyperion为范本 Hyperion项目虽获得了当地政界的广泛支持，却也成功促使部分环保主义者和大型石油公司联合起来表示反对，后者担忧炼油厂和石化厂的电力成本会因此上升。 “我们很清楚，当前局势十分复杂。”克拉克表示，同时提到了当地民众立场的矛盾之处，“一些在这片土地上世代生活的居民，因项目开发而觉得自己将流离失所；但与此同时，对于项目是否推进，我们实际上并无话语权。” Louisiana Energy Users Group（包括埃克森美孚（Exxon Mobil）、雪佛龙（Chevron）和壳牌（Shell））指出，该项目将致使安特吉在路易斯安那州的能源需求增加30%，给现有公用事业用户带来前所未有的财务风险。 尽管争议不断，位列《财富》美国500强第355位的安特吉，如今已获得该州公共服务委员会（PSC）的正式批准，得以推进燃气发电厂建设项目。公共服务委员会是由5名民选成员组成的机构，负责监管该州公用事业，而刘易斯是唯一投出反对票的成员。此次听证会也引发了全美范围内悬而未决的问题：究竟多少能源才算足够？各州能否承担拒绝大规模经济开发投资所带来的风险？此外，在中国深度求索（DeepSeek）证明人工智能技术能够以更低成本实现更高效率之后，当前这种对能源的“狂热追逐”是否可能建立在泡沫之上？ 目前美国已拥有约3800个数据中心（其中多数是在早期云计算热潮时期建成的），规模最大的数据中心集群集中在弗吉尼亚州所谓的“数据中心走廊”——那里有500座数据中心，可便捷接入光纤网络以实现高速数据传输。然而，与支撑人工智能运行所需的设施相比，这些数据中心大多规模较小。仅今年一年，超大规模科技公司就宣布投入数千亿美元，以满足生成式人工智能不断攀升的需求。 2025年，亚马逊、谷歌、微软这三家科技巨头各自计划投入750亿至1000亿美元用于数据中心建设——如此庞大的投入规模，在数年前定会让任何一位经济学家都感到难以置信。扎克伯格表示，Meta今年在数据中心方面的预算约为700亿美元（较去年的280亿美元实现了大幅跃升），且随着Meta在“超级智能”领域“下重注”，2026年这一预算预计还会“大幅增长”。 这些项目对新增电力的需求惊人。美国能源部（U.S. Department of Energy）近期发布的一份报告估算，到2028年，数据中心对电网的需求或将增至当前的3倍，耗电量最高或将占到全美总用电量的12%。OpenAI的星门计划于今年1月获得1000亿美元的前期投资，用于在得克萨斯州建设总投资达5000亿美元的数据中心园区，该项目计划新建超100座燃气发电厂，为该园区及其他项目供电——尽管其中多数项目可能永远无法落地。即便如此，行业研究机构Enverus预测，未来五年将有约46吉瓦的燃气发电装机容量投入使用，新建规模将增长20%。 专家一致认为，全美范围内确实亟待提升电力产能。能源经济与金融分析研究所（Institute for Energy Economics and Financial Analysis，简称IEEFA）能源分析师凯西·孔克尔（Cathy Kunkel）表示，关键在于，我们并不清楚确切的增量需求。 美国电力需求在过去15年维持稳定，但去年增长了3%，创下本世纪以来第五高的年度增幅。未来数年，电力需求预计仍将持续攀升。 孔克尔指出，Meta与安特吉为满足这一需求制定的计划“具有开创性意义”。 在Meta项目的推动下，安特吉股价已创下历史新高。与此同时，Meta承担了里奇兰教区项目的大部分前期成本。 根据合同，Meta将在前15年内承担这座总投资达32亿美元的燃气发电厂的电力成本——这一期限比常见的10年合同更长，但未达批评者所要求的25年——同时还将承担部分输电成本。此外，尽管环保组织仍持反对意见，Meta还承诺在路易斯安那州建设总装机容量达1.5吉瓦的太阳能和电池储能设施。 刘易斯表示，这样的合作安排可能会向市场传递出信号——这就是“新黄金标准”。但对反对者而言，这无疑是危险信号。 “问题在于，这将开创先例。”Alliance for Affordable Energy的洛根·伯克（Logan Burke）在8月20日作证时表示，“这样的合作安排使得我们所有人——包括本州全体选民与公用事业用户——都受制于两家企业之间的非公开合同。” 目前，弗吉尼亚州、得克萨斯州、加利福尼亚州已掀起数据中心与燃气发电厂建设热潮，并且这一趋势正逐步向全美各地扩散，包括更多乡村地区。 过度建设风险还是供应短缺担忧？ 该项目的庞大规模及其所需的资源投入已引发路易斯安那州部分人士的警觉，该州电力电网本就十分脆弱。 今年5月，路易斯安那州南部超10万户家庭因电力供需失衡而遭遇停电。 “里奇兰教区的数据中心将成为全球规模最大的数据中心，”塞拉俱乐部（Sierra Club）路易斯安那州分会协调员玛吉·维克奈尔-普雷（Margie Vicknair-Pray）表示，“我们如何保障停电不会愈发频繁地发生？我们至今尚未完全明晰数据中心对土地、资源以及当地居民的影响。” 尽管Meta曾作出建设更多可再生能源项目的非约束性承诺，但路易斯安那州议会近期通过了一项新法案，将天然气纳入“绿色能源”定义范畴——这意味着扎克伯格等人可将安特吉的燃气涡轮机算作“绿色能源设施”。 燃气发电厂还面临其他难题：全球供应链中燃气涡轮机的制造能力出现短缺，未来五年的燃气涡轮机基本已售罄。 由于该州绕过了标准且耗时更长的审批流程，刘易斯对安特吉与Meta是否确实需要额外的涡轮机提出质疑。“坦率来讲，我们为何只将目光聚焦于发电设施的扩建？”他反问，而忽视了电网能效提升与灵活性方面的考量。他警告称，Meta有可能提前退出项目，最终让电力用户承担超支费用。 安特吉公司发言人布兰登·斯卡迪利（Brandon Scardigli）向《财富》杂志透露：“就当下而言，燃气发电是成本最低且合理可行的选项，能够满足Meta这类大型数据中心24小时不间断的电力需求。” 另一个不确定因素在于，市场对计算效率与能源效率提升的预期。孔克尔得出了一个必然结论：这些项目的能耗终将降低，“要么是技术日益高效所致，要么是因效率未达预期而走向破产”。 这或许意味着，公用事业公司与大型科技企业可能会发现，自己投入巨额资金新建的燃气发电设施，最终沦为无人需要的闲置资产。 Meta展示其数据中心内呈蓝色调的冷存储设施。图片来源：Meta 其他考量因素及选址？ 随着大型数据中心在全美乡村地区不断扩张，维克奈尔-普雷对其可能带来的影响提出质疑：空气与噪音污染会给农场主和牧场主带来怎样的影响？尤其是巨大的用水量，是否会威胁到他们的生计？ “水资源将如何分配？”她问道，“倘若农民无法为作物进行灌溉，又该如何？” 无党派智库能源创新（Energy Innovation）建议，超大规模科技公司应优先投资可再生能源与电池储能项目，仅在必要时将部分新建燃气发电作为备用电源来使用。 能源创新的电力政策高级总监迈克·奥博伊尔（Mike O’Boyle）认为，建设过多新燃气涡轮机将带来不必要的风险。“我清楚当前的行业环境，无论是联邦层面还是行业内部，都主张‘大建特建’，力求以最快速度推进项目。”但他强调必须将成本问题纳入考量。“我们正处于资源有限的环境中，供应远低于需求，这导致价格飙升。” 除弗吉尼亚州外，数据中心目前主要集中在得克萨斯州和加利福尼亚州等人口大州。但对开发者而言，数据中心的一大吸引力在于，它能为里奇兰教区这类远离港口与机场的经济欠发达地区带来工业发展机遇。 Enverus能源分析师亚当·罗宾逊（Adam Robinson）研究了未来数据中心建设的潜在走向。他表示，开发者会从多方面进行综合考量：电力与土地的价格及可获得性、电网与光纤网络的连接情况，以及接入电网所需的时长。 罗宾逊预测，PJM Interconnection（宾夕法尼亚州-新泽西州-马里兰州）所在区域（从新泽西州经“铁锈地带”[指美国东北部及中西部传统工业区]延伸至伊利诺伊州）将迎来大量建设项目。该区域凭借具有竞争力的电力市场、良好的连接性及高速数据传输能力，正吸引超大规模科技公司入驻。 他还指出，那些寻求大面积廉价土地的开发者，也将视线投向了美国西部；而提供服务器托管服务的企业及小型开发者，则更关注得克萨斯州与美国南部腹地的廉价土地及税收优惠政策——例如，路易斯安那州豁免了Meta项目的销售税。 克拉克牧师深知，无论是在里奇兰教区还是其他任何地方，科技进步都是不可阻挡的趋势。 “事情已然发生，”他说道，“所以我们期望能充分利用这一机遇。”（财富中文网） 译者：中慧言-王芳 On a quiet patch of former farmland in northeastern Louisiana, a fleet of excavators has leveled more than 2,000 acres of reddish clay earth. This is rural Richland Parish, once a floodplain tangled with meandering bayous and wild canebrake where black bears still wander and a quarter of the 20,000 residents live below the poverty line. Enter Meta—the sixth-largest company in the world by market cap. The tech giant is keen on making Richland home to its wildest AI aspirations—courtesy of a tremendous amount of new gas-fired power. The region has ample land and sits adjacent to Louisiana’s huge Haynesville Shale gas field. In December, construction began on Meta’s biggest-yet data center: a $10 billion complex of nine buildings, housing bank upon bank of servers that will take up over 4 million square feet, an area larger than Disneyland. Meta chairman and CEO Mark Zuckerberg isn’t stopping there. He dubbed the project “Hyperion” in July—a data center “supercluster” that eventually could use the energy equivalent of 4 million homes and become the world’s biggest data center project. Zuckerberg said Hyperion would cover a “significant part of the footprint of Manhattan.” The project entails more than 2 gigawatts of computing capacity—Zuckerberg said it could eventually expand to 5 gigawatts—programmed to train open-source large language models. Meta lagged in the AI race with previous flops and the multibillion-dollar “Metaverse” boondoggle. Now he’s framing Hyperion and his construction spree as the pursuit of “superintelligence,” while poaching AI talent using $250 million pay packages and buying a 49% stake in Scale AI. It’s the latest in a grandiose game of Big Tech one-upmanship in AI, competing with the likes of Google, Microsoft, Amazon, and OpenAI. “We are making all these investments because we have conviction that superintelligence is going to improve every aspect of what we do,” Zuckerberg said in Meta’s July 30 earnings call. A Meta spokesperson told Fortune it’s impossible to say exactly what the complex will power since it’s unclear how AI will have evolved when it opens in 2030. The sheer size has left locals in this quiet region stunned. “I think, like a lot of people, my initial reaction was kind of blown away that a site [so] rural was selected for something like that,” said Justin Clark, pastor of First Baptist Church in nearby Rayville. “As we started learning more about what it was and what the scope entailed, that feeling just continued. An amazement of, ‘Good grief.’” Clark looks forward to welcoming new workers to the area but admits it’s difficult to truly visualize the scope. At a recent chamber of commerce banquet, they were told it’s the largest construction site in North America: “That’s unbelievable,” he marveled. Altogether, Big Tech’s new data centers will be incredibly energy and water hungry. Keeping the Hyperion servers cool and functional will require twice the power of New Orleans—and eventually more. As AI’s boom shifts into ever-higher gears, speculation abounds about how utilities will quench Big Tech’s deepening thirst for electricity. In the case of Meta (22 on the Fortune 500), regional utility Entergy will build three new gas-fired turbines with a combined capacity of 2.3 gigawatts—the first such buildout in decades—sparking pushback from ratepayers worried about consumer costs and from climate advocates who fear a backslide from green energy goals. The scramble for AI dominance has positioned utilities as the gatekeepers of the hyperscaler market, weighing the benefits of massive capital investments for an emergent industry—whose future payoffs remain murky—versus potential rate hikes and the risk of stranded assets for decades to come. State regulators gave Entergy the green light Aug. 20—two months earlier than expected—potentially setting the template for future deals between utilities and Big Tech to build new power plants, increasingly in more rural locales with affordable land. Entergy and regulators called the deal a model for the nation’s data center and power proliferation. “This deal could signal to other states that this is how data centers should be governed and operated,” Louisiana Public Service Commissioner Davante Lewis told Fortune. “This would be a test across the nation. I’ve heard that from investors; I’ve heard that from credit agencies; I’ve heard that from fellow data centers—whatever comes out of the Meta deal may be the framework for them all.” Meta’s Hyperion as the template Hyperion has plenty of local political support, but it also managed to unite some environmentalists and Big Oil in opposition, the latter of which voiced concerns about increased power costs for their refineries and petrochemical plants. “We’re not naive to the fact that it is a complex situation,” said Clark, noting conflicting local loyalties. “Some people who’ve lived in that area for generations feel displaced because of the development. At the same time, we don’t have any real say on whether it’s going to happen.” The Louisiana Energy Users Group—including Exxon Mobil, Chevron, and Shell—said the project increases Entergy’s Louisiana energy demand by 30%, creating unprecedented financial risks to existing utility ratepayers. Regardless, Entergy (No. 355 on the Fortune 500) now has the official go-ahead for its gas plants from the Public Service Commission (PSC), the five-person elected body that regulates utilities in the state. Lewis was the only one to vote in opposition. The hearing raised the same questions looming over the nation: How much energy is enough? Can states risk turning down massive economic development investments? And, after the advent of China’s DeepSeek—proving AI can become cheaper and more efficient—could the stampede for power be built on a bubble? The country already counts about 3,800 data centers—many built during the earlier cloud-computing boom—with the biggest chunk concentrated in Virginia’s so-called Data Center Alley, where 500 facilities find easy access to fiber-optic connectivity for high transfer speeds. But most of those are relatively small compared to what’s needed to power AI. This year alone, hyperscalers announced hundreds of billions of dollars to feed the growing generative AI needs. Amazon, Google, and Microsoft are investing anywhere from $75 billion to $100 billion each into building data centers in 2025—numbers that would have strained the imagination of any economist just a couple of years ago. Meta’s data center budget is about $70 billion—way up from $28 billion last year—and expected to “ramp significantly” more in 2026 as part of Meta’s “massive bet” on superintelligence, Zuckerberg said. These projects depend on an astonishing amount of new power. A recent report from the U.S. Department of Energy estimates data centers’ grid needs could triple by 2028, consuming up to 12% of the nation’s electricity. OpenAI’s Stargate received an upfront investment of $100 billion in January for the $500 billion data center complex proposed in Texas, where more than 100 new gas plants are proposed to power it and other projects—though many will never come to fruition. Still, industry research group Enverus projects the next five years will bring roughly 46 gigawatts of gas-fired electricity online, a 20% jump in new construction. Experts agree some surge in electric capacity nationwide is needed. It’s the exact extent that’s unknown, said Cathy Kunkel, energy analyst for the Institute for Energy Economics and Financial Analysis. Electricity demand in the U.S. held steady for 15 years but, last year, it increased by 3%— marking the fifth-highest rise this century. More jumps are projected for years to come. Meta’s and Entergy’s plans to meet that demand are “precedent setting,” Kunkel said. Buoyed by the Meta project, Entergy’s stock has hit record highs. Meta, meanwhile, has taken on a significant chunk of the upfront costs in Richland. According to the contract, Meta will pay the power costs for the $3.2 billion gas plants for the first 15 years—more than the typical 10-year contract, but not as much as the 25 years critics sought—as well as some transmission costs. Meta also committed to help build 1.5 gigawatts of solar and battery power throughout Louisiana, despite the ongoing opposition from environmental groups. The arrangements could signal to the market this is the “new gold standard,” Lewis said. That’s a red flag for opponents. “The problem here is that this is going to set precedent,” Logan Burke, of the Alliance for Affordable Energy, testified Aug. 20. “This settlement puts all of us, all of your constituents and customers in the state, at the mercy of a non-public contract between two corporations.” Data centers and gas plants are booming in Virginia, Texas, California, and, increasingly, nationwide, including more rural locales. Risks of overbuilding or fears of shortages? The staggering scale of the project and the resource demands it entails have raised alarm bells for some in Louisiana, where the electric grid is already fragile. In May, over 100,000 south Louisiana customers lost power after demand outstripped supply. “The Richland data center is to be the largest in the world,” said Margie Vicknair-Pray, coordinator with the Sierra Club’s Louisiana chapter. “How can we ensure that blackouts won’t become more frequent? What we have yet to fully understand is the impact the data center will have on the land, our resources, and the people.” While Meta has a non-binding promise to build more renewable energy, the Louisiana Legislature passed a new law that adds natural gas to the definition of green energy, allowing Zuckerberg and others to count Entergy’s gas turbines as “green.” Gas-fired plants pose other hurdles. There’s a shortage of turbine manufacturing in the global supply chain. Gas turbines are essentially sold out for the next five years. With the state bypassing the standard, lengthier review process, Lewis questions whether Entergy and Meta need extra turbines. “Why are we only focusing, quite frankly, on generation buildup?” he wondered, rather than grid efficiency and flexibility. He warned of Meta potentially walking away early, leaving ratepayers stuck with excess costs. Entergy spokesman Brandon Scardigli told Fortune that “natural gas-fueled generation is the lowest reasonable cost option available that can support the 24/7 electrical demands of a large data center like Meta.” The other wild card is the expectation for improvements in computing and power efficiency. Kunkel concluded an inevitability. The projects will use less energy, she said, “either because they get more efficient or because they don’t and go bankrupt.” It could mean utilities—and Big Tech—find themselves pouring capital into new gas generation no one needs. What and where else? As huge data centers spread throughout rural locales nationwide, Vicknair-Pray questioned the impact of air and noise pollution on farmers and ranchers, and especially the massive water consumption that could impact their livelihoods. “How will the water be shared?” she asked. “And what happens if the farmers are unable to water their crops?” The nonpartisan think tank Energy Innovation proposes that hyperscalers invest primarily in renewable energy and battery storage developments, with some new gas-fired power used only as needed for backup. Mike O’Boyle, senior director of electricity policy at Energy Innovation, believes building too many new gas turbines poses unnecessary risks. “I know the environment right now, federally and in the industry, is ‘Build, build, build,’ as fast as we can.” But costs must be considered. “We’re in a limited resource environment where supply is much lower than demand, and it’s causing prices to skyrocket.” Beyond Virginia, data centers currently are concentrated in the biggest states, such as Texas and California. But part of what makes data centers attractive to developers is they open industrial development for economically depressed areas that aren’t near ports or airports—such as Richland Parish. Adam Robinson, an energy analyst with Enverus, looked at where the buildout may head next. He said many factors are considered by developers: Power and land prices and availability, grid and fiber-optic connectivity, and the time it takes to connect to the grid. Robinson predicts a lot of development in the PJM Interconnection (Pennsylvania-New Jersey-Maryland) region from New Jersey through the Rust Belt and into Illinois. The region is attracting hyperscalers thanks to competitive power markets, good connectivity, and high data-transfer speeds. Developers looking for large plots of affordable land also are looking West, while co-location and smaller developers are more focused on cheap land and tax incentives in Texas and the Deep South, Robinson said. Louisiana, for instance, exempted the Meta deal from sales taxes. Pastor Clark recognizes that tech progress is inevitable in Richland and everywhere else. “It is happening,” he said, “so we want to make the best of it.” 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
乔治·库尔茨（George Kurtz），CrowdStrike首席执行官兼联合创始人。图片来源：CROWDSTRIKE CrowdStrike首席执行官兼联合创始人乔治·库尔茨表示，网络安全绝不仅仅是软件层面的事情。 “我们在CrowdStrike所做的事业自古就有。”乔治·库尔茨（George Kurtz）对《财富》表示，“这是正义与邪恶的较量，是以技术形态呈现的人性故事。” 随着人工智能的崛起，网络威胁与网络犯罪急剧增加，这场较量的紧迫性与复杂性空前提升。这也让网络安全行业长期存在的并购所面临的风险达到前所未有的高度。毋庸置疑，2025年多宗重大交易都集中在网络安全领域，包括Palo Alto Networks以250亿美元收购CyberArk，以及谷歌（Google）拟以320亿美元收购Wiz。 2019年上市的CrowdStrike同样是并购市场常客。今日，公司宣布以约2.9亿美元收购数据可观测性初创企业Onum。此外，公司今日还公布了2025年第二季度财报，业绩虽超市场预期，但营收前景不及预期，导致盘后股价下跌约4%。 库尔茨就此次Onum收购案及未来并购战略接受了《财富》独家采访。 “我们一般会在恰当时机出手。”库尔茨（George Kurtz）表示，“看看其他收购案例，比如CyberArk，那是一家拥有20年历史的科技公司，整合风险极高。这类企业体量庞大，我早见识过类似操作。我在迈克菲（McAfee）任职期间，公司收购了21家公司，但始终未能实现有效整合……所以归根结底，我们要极度重视客户体验，确保我们以足够严谨的态度完成业务整合。我们在这方面有着良好的整合记录。” Onum是CrowdStrike自去年那次广受关注的IT系统瘫痪事故以来较早达成的交易之一。库尔茨表示，该事件虽未阻碍公司的并购计划，但确实暂缓了并购进程。他透露，事后，CrowdStrike设定了更高门槛，在持续与企业、创业者、风投机构沟通以维持并购渠道畅通的同时，并未避免敲定任何交易。最终，Onum收购在三个月内达成。总部位于马德里的Onum由Dawn Capital与Insight Partners等风投支持，其核心吸引力在于实时数据管道检测能力——能够在数据流入企业系统时分析并识别潜在威胁或异常。 “试想一下，凭借我们掌握的数据量，我们正逐渐成为各类人工智能模型的‘网络安全版Reddit’。”库尔茨（George Kurtz）表示，“数据越大，我们的护城河就越稳固，也越有机会借助AI解决更广泛、更复杂的问题。这正是推动我们构建AI原生安全运营中心（SOC）愿景的核心动力，是顺理成章的延伸。” 在某种程度上，这正是着眼于充满AI智能体的未来愿景。 “我们的目标是守护每个AI智能体。”库尔茨表示，“那么，什么是AI智能体？它本质上就是‘超人’。它可以访问数据，拥有身份标识（即便是非人类身份），能够接入工作流，还能够访问超出既有边界的系统……因此，它完全暴露于我们正在防范的各种风险之中。” 从诸多方面来看，对Onum的收购堪称CrowdStrike的经典操作。自2017年以来，CrowdStrike已完成八笔收购，其中包括2021年以4亿美元收购Humio，以及2024年据称以2亿美元收购Flow Security。 “有些公司的估值明显过高。”库尔茨表示，“我认为有些公司尚未意识到自己正逐渐陷入‘僵尸困境’：其最后一轮融资估值或许对它们而言很理想，但对像我们这样的很多企业而言，其估值过高且缺乏收购可行性……所以，当企业以相对有限的年度经常性收入（ARR）获得数十亿美元的估值时，潜在买家群体就会急剧萎缩。正因为此，我们更愿意在价值洼地出手，既能注入价值，又能为CrowdStrike股东创造回报。” 最终目标依然如初——保障安全，对抗那些如今手中拥有更多武器的恶意攻击者。 “生成式人工智能正在让攻击手段快速普及化。”库尔茨表示，“过去只有极少数人掌握的尖端技术，如今正被广泛普及……最重要的一点是，由于攻击方的行动速度大幅提升，防守方应对威胁的时间窗口被急剧压缩。” 展望未来，库尔茨断言： “我们深知，未来对安全防护的需求只会比今日更为迫切。”（财富中文网） 译者：刘进龙 审校：汪皓 Cybersecurity is more than just software, says George Kurtz, CEO and cofounder of CrowdStrike. “What we do at CrowdStrike is as old as time,” he told Fortune. “It’s good versus evil. It’s a human nature story embodied in technology.” It’s a battle that’s more urgent and complex than ever, as the rise of AI has ballooned the number of cyber threats and cyber criminals. This makes M&A—a longstanding feature of the cybersecurity sector—more high-stakes than ever. To be sure, some of the biggest deals of 2025 have been in cyber, from Palo Alto Networks’ $25 billion acquisition of CyberArk to Google’s proposed $32 billion acquisition of Wiz. CrowdStrike, which went public in 2019, is also a longtime acquirer, and today announced its acquisition of data observability startup Onum for about $290 million. CrowdStrike today also announced its Q2 2025 earnings, beating expectations but offering a softer-than-expected revenue outlook sending its shares down roughly 4% in after hours trading. Kurtz exclusively spoke to Fortune about the Onum deal and CrowdStrike’s M&A strategy going forward. “We like to get things at the right stage,” he said. “When you look at some of these other acquisitions, like CyberArk, you’re talking about a 20-year-old technology company with a lot of integration risk. These are big companies, and I’ve seen the movie before. When I was at McAfee, we acquired 21 companies, and never quite got them integrated… So, when it comes down to it, we’re maniacally focused on the customer experience, on making sure we’re disciplined enough to get this stuff integrated. We have a great track record of doing that.” Onum marks one of CrowdStrike’s early deals since last year’s much-publicized IT outage, which Kurtz says didn’t derail its M&A efforts, but offered a pause. In the aftermath, CrowdStrike set a high bar and refrained from closing any deals, while continuing to talk to companies, entrepreneurs, and VCs, keeping the M&A pipeline active, said Kurtz. The Onum deal ultimately came together in three months. The Madrid-based startup, which counts Dawn Capital and Insight Partners among its VC backers, was especially compelling to CrowdStrike for its real-time pipeline detection—the ability to analyze and detect threats or anomalies in data as it is being ingested into a company’s systems. “If you think about the data we have, we started becoming the Reddit of security data for all these AI models,” said Kurtz. “The more data we get in, the larger the moat we actually have, and the greater the opportunity we have to solve bigger and broader problems from an AI perspective. That’s really driving our vision for AI-native SOC [security operations center]. It’s a natural extension.” In part, this is looking towards a future filled with AI agents. “Our goal is to secure every AI agent,” said Kurtz. “Okay, what’s an AI agent? An AI agent is basically superhuman. It has access to data. It has an identity, though it might be a non-human identity. It has access to a workflow, and it has access to systems that are outside of your own boundaries… So, it has all of the exposure that we’re protecting against. In a lot of ways, Onum is a classic CrowdStrike deal. Since 2017, CrowdStrike has acquired eight companies, including Humio in 2021 for $400 million and Flow Security in 2024 for a reported $200 million. “There are some companies that are obviously richly-valued,” Kurtz said. “I think some of these companies don’t realize that they are starting to move into zombieland: You look at their last round valuation, and it might be great for them, but it’s expensive and it’s necessarily actionable for a lot of companies, even ours… So, you start to hit these big, multi-billion dollar valuations with not a lot of ARR, relatively speaking, and your pool of buyers dramatically shrinks. That’s why we like to catch them in the sweet spot of where we can add value, and that value accrues to CrowdStrike’s shareholders.” The goal, in the end, remains the same—security, and fighting the bad guys (who now have more weapons to play with). “With gen AI, we’re democratizing destruction,” said Kurtz. “We’re taking a very sophisticated topic known by a relatively few number of people … and now you’re making all that expertise available to many more people. … The biggest thing is that you’re really compressing the timeframe that the good guys have to be able to deal with these problems, because the bad actors are moving so much faster now.” What’s one thing Kurtz is sure of, looking to the future? “We know there’s going to be a greater need for security tomorrow than there is today,” he said. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
唐纳德·特朗普。图片来源：Chip Somodevilla/Getty Images 白宫近期试图加强对美联储的控制，这正是特朗普惯用的斗争策略。他向来擅长将矛头精准对准那些通常被视作难以辩驳的个人或机构，不断试探规范、先例、宪法边界及美国价值观的底线。在最新行动中，特朗普试图罢免美联储理事莉萨·库克（Lisa Cook）——此事的起因是上周联邦住房金融局（FHFA）局长威廉·普尔特（William Pulte）收到一则“匿名举报”，指控库克存在抵押贷款欺诈行为。 鉴于初步证据看起来确凿无疑，库克理事并非值得同情的对象。不过，倘若她确实存在不当行为，就其严重程度而言，似乎与特朗普因虚报资产价值以获取优惠贷款及保险费率而被判定银行欺诈罪的行为不相上下。这并非为库克辩护，而是揭示法治屈从于统治者意志所呈现出的双重标准与虚伪性。这位美联储理事尚有十余年任期，尚未被证实存在故意违法行为，其抵押贷款申请中的错误可能仅属文书疏漏。同样重要的是，公平原则与正当法律程序作为核心准则，超越民意或个人道德。 例如，1988年美国最高法院诉拉里·弗林特（Larry Flynt）一案的判决再次明确，即便这位厌恶女性、声名狼藉的色情出版商的作品内容多么令人不适，其言论自由仍应受到保护。该判决明确指出：“言论自由不仅是个人自由的体现……对共同追求真理及维系社会整体活力也至关重要。”同样，在1977年著名的“斯科基事件”判决中，尽管最高法院对伊利诺伊州斯科基市纳粹党人的活动深感厌恶，但仍坚定维护其依据宪法第一修正案享有的权利。 正当程序对保障个人公平至关重要，而对正当程序的尊重则关乎制度自由与法治根基。1913年颁布的《联邦储备法》明确规定，美联储不受美国总统管辖，而是作为一个独立、非政治性、自筹资金的机构存在，完全无需依赖纳税人支持。美联储的决策并非总能尽善尽美——我们也曾在其对经济调控过度时批评其时而反应迟缓。然而，美联储的独立性对美国经济的全球地位及美元作为全球储备货币的地位至关重要。央行独立原则正是美国货币体系赢得信任与尊重、历来不受政治干预而陷入混乱的根源所在。近期土耳其的案例便是反面教材：埃尔多安总统对央行的破坏性干预，导致该国通胀率飙升、里拉汇率暴跌。 就库克理事案而言，目前那些看似有力的指控仍停留在指控层面，尚未形成正式起诉。在没有定罪判决的情况下，鉴于总统本人已被证实存在类似违法行为，当前局面引发了人们对“平等保护权”的质疑。事实上，倘若针对库克的指控确凿有力，那么特朗普的拙劣干预可能打乱美国司法部的节奏——该部门原本可能已准备根据联邦住房金融局移交的材料对库克提起公诉。而今，特朗普的政治干预反而可能成为库克最有力的辩护依据。 一贯特征 这种双重标准已成为特朗普第二任期的一贯特征，其典型表现包括对非法移民的虐待，以及在全美主要城市实现市政执法军事化。在这两种情况下，尽管非法移民中罪犯所占比例极低，且各大城市的犯罪率实则在持续下降，但那些维护公平、坚守正当程序、捍卫宪法精神与美国价值观的人士，仍被描绘成为“入侵的外国罪犯”辩护、为城市犯罪开脱的形象，进而陷入难堪境地。 正直爱国的公众人物面临的挑战在于：如何在捍卫那些构成美国国家特质的基础原则的同时，避免落入特朗普为选定对象精心设置的“流沙陷阱”。传达信息时必须做到有所区分，但也要避免陷入冗长空洞、晦涩抽象的解释，或成为更易遭受诋毁的对象。 这一局面对商界领袖的启示显而易见。首席执行官们必须意识到：若只是置身事外、忧心忡忡地旁观，他们同样可能在无意间沦为“侵蚀美国核心价值观”的帮凶——如今已有同行企业被迫出让巨额利润或股权以在美国开展业务（涉及领域从人工智能、半导体芯片，到农业、保险业），实则遭遇勒索。 以逐步施压的手段，迫使私营企业的治理屈从于单一政治霸凌者的意志，这或许无法赢得所有民粹主义者的同情，但本质上与美联储独立性的逐渐丧失及法治的沦丧并无二致。同样，特朗普对美国城市发起的范围日益扩大的“攻击”，也在侵蚀美国“州权与地方问责制”这一国家根本特质。当前最核心的挑战在于：我们必须认识到，对极权化进程做出渐进式、潜移默化的让步，终将导致国家特质丧失。（财富中文网） 杰弗里·索南菲尔德（Jeffrey Sonnenfeld）是耶鲁大学管理实践莱斯特·克朗教授，耶鲁首席执行官领导力研究所创始人。 斯蒂芬·亨里克斯（Stephen Henriques）是耶鲁首席执行官领导力研究所高级研究员，曾担任麦肯锡咨询公司（McKinsey & Company）顾问及康涅狄格州州长政策分析师。 Fortune.com上发表的评论文章中表达的观点，仅代表作者本人的观点，不代表《财富》杂志的观点和立场。 译者：中慧言-王芳 The White House’s latest effort to tighten control over the Federal Reserve exemplifies President Trump’s typical tactics in battle. He has an uncanny knack for pushing the limits of protocol, precedent, constitutional boundaries, and American values by targeting individuals or organizations that are often seen as difficult to defend. In his latest bid, Trump has tried to remove Federal Reserve Governor Lisa Cook from her position after allegations of mortgage fraud emerged last week, prompted by an “anonymous tip” that landed on the desk of Federal Housing Finance Agency Director (FHFA) William Pulte. Gov. Cook is not a sympathetic target, as early evidence looks damning. Still, any alleged misconduct on her part seems almost as reprehensible as the bank fraud that led to Trump’s conviction for inflating the value of his assets to obtain favorable loan and insurance rates. This is not to defend Cook, but rather to illustrate the parallelism and hypocrisy of the rule of law being subordinated to the law of the ruler. The Fed governor has over a dozen years remaining in her appointed term, she has not yet been proven to have intentionally broken any laws, and she may have made a clerical error in her mortgage applications. Equally important are the issues of fairness and due process as core principles that transcend popularity and personal morality. For example, the 1988 Supreme Court decision in favor of Larry Flynt reaffirmed the freedom of expression for the misogynistic, disgraceful pornographer, no matter how offensive his work was. Notably, that decision affirmed that “the freedom to speak one’s mind is not only an aspect of individual liberty … but also is essential to the common quest for truth and the vitality of society as a whole.” Similarly, in their 1977 decision on what became known as the “Skokie Affair,” the Supreme Court fortified the First Amendment rights of the Nazi party in Skokie, Illinois, however distasteful it found their activities. As important as due process is for individual fairness, respect for due process is for institutional freedom and legality. The Federal Reserve Act of 1913 ensures that the Fed is not subordinate to the President of the United States but rather sits as an independent, non-political, self-funded institution. with zero taxpayer support. The Fed does not always get it right, and we have been critical of their periodic sluggishness, as they oversteer the economy. However, the independence of the Fed is critical to the U.S. economy’s global stature and the U.S. dollar serving as the world’s reserve currency. The principle of central banker independence is why the U.S. monetary system is trusted and respected and historically not disrupted by political interference. As we have seen recently in Turkey, President Erdogan’s disastrous interference has led to soaring inflation and plummeting depreciation of the Turkish lira. In the context of Gov. Cook, the potentially compelling accusations amount at this time to allegations and not indictments. Without a guilty verdict, the situation raises questions about the right to equal protection, considering the president’s parallel proven violations. In fact, if there is a strong case against Cook, then Trump’s clumsy interference could subvert his own Justice Department, which may have been ready to indict Cook based on the FHFA referral. Her strongest defense may be Trump’s political interference. A constant feature Such duplicity has been a constant feature of the second Trump administration, characterized by the abusive treatment of undocumented immigrants and the militarization of municipal law enforcement in our nation’s major cities. In both instances, it places the defenders of fairness, due process, constitutionality, and American values in the problematic position of being portrayed as defending “invading foreign criminals” and excusing urban crime, despite the minuscule share of criminals among undocumented immigrants and the reality of plunging crime rates in metros. The challenge for honorable, patriotic public figures is how to defend American principles that are foundational to our nation’s character without falling into the clever, quicksand-like trap that Trump has set for his chosen targets. Messaging must make distinctions, but also avoid getting lost in parenthetic abstractions or potentially easier-to-discredit targets. The implications for business leaders are clear. CEOs must recognize that they, too, can inadvertently become complicit in the erosion of core American values as they merely watch, worried from the sidelines, as peer companies are subjected to extortion through the surrender of massive profits or equity stakes just to do business in the U.S.—ranging from AI and semiconductor chips to agriculture and insurance. The subjugation of private enterprise governance to a single political bully through gradual stations may not seem to be a sympathetic cause to all populists, but it is equivalent to the gradual degradation of an independent Fed and the rule of law. Similarly, Trump’s expanding war on American cities is eroding our nation’s fundamental character regarding state rights and local accountability. The fundamental challenge is to realize that we lose our character through incremental, creeping concessions to the totalitarian process. Jeffrey Sonnenfeld is Lester Crown Professor of Leadership Practice at the Yale School of Management and founder of the Yale Chief Executive Leadership Institute. Stephen Henriques is a senior research fellow of the Yale Chief Executive Leadership Institute. He was a consultant at McKinsey & Company and a policy analyst for the governor of Connecticut. The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
网飞（Netflix）联合首席执行官泰德·萨兰多斯（Ted Sarandos）出席2025年2月23日在洛杉矶举行的第31届美国演员工会奖颁奖典礼。图片来源：Emma McIntyre—WireImage/Getty Images • 《K-POP：猎魔女团》（KPop Demon Hunters）成为网飞（Netflix）名副其实的现象级爆款，本周以2.36亿次观看量刷新平台历史最高纪录，同时凭借四首歌曲同时跻身《公告牌》（Billboard）榜单前十，创下历史纪录。为打造这部现象级影片，索尼影视动画公司（Sony Pictures Animation，以下简称索尼动画）投入1亿美元，但受疫情时期发行协议限制，据称仅能获得2000万美元的收益。对索尼而言，无疑错失了巨大机遇——如今网飞掌控着这个潜在价值高达数十亿美元的IP。 网飞手握现象级爆款，其火爆程度远超所有人预期。动画电影《K-POP：猎魔女团》讲述了K-pop女团兼猎魔者的故事，该片正式成为网飞有史以来观看次数最多的影片，观看量达2.36亿次，打破了此前由《红色通缉令》（Red Notice）保持的2.309亿次观看量的纪录。这部影片自6月20日首映以来，仅用67天便达成这一里程碑，成为网飞票房榜中攀升速度最快的影片之一。 《K-POP：猎魔女团》不仅刷新了流媒体电影的播放纪录：其原声带中的四首歌曲同时跻身公告牌百强单曲榜前十，这在该榜单67年历史中尚属首次。（具体排名如下：《Golden》位居榜首，《Your Idol》位列第四，《Soda Pop》位居第五，《How It's Done》跻身第十名，您要是好奇的话。）上周末，网飞尝试在院线放映该片的“跟唱版”，尽管观众在家中即可在线观看，但在约1700家影院的放映中，该片仍斩获了约1800万至2000万美元的票房。 影片的巨大成功促使网飞与索尼已就续集事宜展开初步洽谈。对网飞而言，这代表着其多年来一直追求的“现象级动画系列”梦想照进现实；但对影片制作方索尼动画而言，情况则更为复杂——这或许代表着好莱坞近年来最重大的“机遇错失”案例之一。 《K-POP：猎魔女团》：现象级作品的诞生历程 据报道，索尼动画为打造《K-POP：猎魔女团》投入了约1亿美元制作预算，此举是对K-Pop文化与超自然冒险题材全球吸引力的一次重大押注。该片由玛吉·康（Maggie Kang）与克里斯·艾伯翰斯（Chris Appelhans）联合执导，讲述了虚构女团Huntr/X在维持偶像事业的同时对抗恶魔的故事，片中还设置了名为Saja Boys的竞争男团——后续剧情走向不难想象。 截至目前，这场创意豪赌以令人意想不到的方式大获成功。影片原声带不仅完美契合剧情，还掀起真正的音乐现象级热潮：其中《Golden》成为第八首登顶公告牌百强单曲榜的K-Pop歌曲，这是自迪士尼电影《魔法满屋》（Encanto）主题曲《We Don’t Talk About Bruno》以来，首支荣登榜首的动画电影歌曲，也是首支由女艺人演唱的冠军单曲。 截至目前，影片热度仍居高不下：《K-POP：猎魔女团》已连续10周稳居网飞电影榜榜首，仅在最新统计的一周内就新增了2540万次观看量。这种长尾热度对网飞原创作品而言极为罕见，更何况该片还是一部源自不知名IP的冷门动画电影。 索尼的协议及获得的收益 显然，《K-POP：猎魔女团》为网飞带来了巨额收益。按常理，索尼作为这部电影的制作方，本应斩获同样丰厚的回报，对吧？然而事实并非如此。尽管投入约1亿美元打造这部全球现象级作品，索尼动画预计只能从这个潜在价值高达数十亿美元的《K-POP：猎魔女团》系列中获得约2000万美元利润——基本上只是整体收益的一小部分。原因是索尼在2021年与网飞达成一项发行协议，该协议旨在确保疫情不确定时期索尼仍能获得回报。 据《Puck》杂志马修·贝洛尼（Matthew Belloni）披露，索尼同意采用“直接上线平台”模式：网飞支付影片制作成本及每部影片最高2000万美元的额外费用，作为交换，网飞保留该影片的全部版权，并且即便影片大获成功，也无需向索尼支付额外的利润分成。这并非索尼为成品影片寻找发行方，而是网飞实质上承担了制作成本，索尼则负责创意创作。 当时该协议颇具合理性：影院在经历疫情停业冲击后，仍处于恢复阶段，动画电影票房疲软，而索尼自身缺乏主流流媒体平台。该协议确保索尼在无需承担院线失利风险的同时，稳获收益。然而，谁都没有料到——甚至网飞高管也未曾预见——《K-POP：猎魔女团》会如此火爆。 要理解网飞从中所获价值之巨大，不妨看看《红色通缉令》对该平台的意义。这部于2021年上映、由巨石强森（Dwayne “The Rock” Johnson）、瑞安·雷诺兹（Ryan Reynolds）和盖尔·加朵（Gal Gadot）主演的动作片，在网飞播放榜上稳居榜首近四年之久，其2.309亿次的观看量成为网飞成功的标杆。 《K-POP：猎魔女团》不仅超越该纪录，还实现了《红色通缉令》未能达成的目标：衍生IP价值。仅其原声带的成功就开辟出大多数网飞原创作品难以触及的收入渠道。需重申：四首歌曲同时跻身公告牌百强单曲榜前十。“跟唱版”院线试验的成功为网飞提供了另一关键数据点（网飞向来痴迷于数据）。在1700家影院上映的首个周末便斩获1800万至2000万美元票房——约为大片上映影院数量的一半——表明观众对该系列影片的集体观影体验存在真实需求。倘若网飞计划拓展线下体验，这无疑是积极信号。 索尼错失的机遇 倘若索尼保留《K-POP：猎魔女团》版权，那么公司本可坐拥价值高达数十亿美元的IP。然而，真正让情况愈发糟糕、甚至堪称残酷讽刺的是——去年九月，索尼首席财务官十时裕树（Hiroki Totoki）在接受《金融时报》采访时坦言： “无论是游戏、电影还是动漫，我们自主培育的IP并不多。我们缺少处于早期阶段的IP，这正是我们的痛点所在。” 索尼早已承认，除《蜘蛛侠》外，其在打造持久娱乐IP方面困难重重。公司高管承认，亟需更多从早期阶段便着手培育的原创IP——而《K-POP：猎魔女团》恰恰就是这样的作品。如今，索尼只能眼睁睁看着网飞利用该IP开发续集、周边商品等衍生业务。（不过索尼向《财富》杂志表示，公司仍会分享包括专辑销售在内的部分收益，网飞未进一步置评。） 数据更凸显了错失良机的代价。作为参照，据报道网飞曾斥资4.65亿美元购买《宋飞正传》（Seinfeld）的重播权。《K-POP：猎魔女团》作为原创IP，不仅已证明其全球吸引力、影院上映可行性，更催生了真正的音乐爆款。在此背景下，索尼仅能获得的2000万美元收益显得微不足道。 续集初期洽谈与未来展望 网飞与索尼迅速启动续集洽谈就说明了一切。当一个IP能在短短两个月内打破平台纪录、催生冠军单曲并证明其院线需求时，其商业价值不言而喻。网飞意图乘胜追击，而《K-POP：猎魔女团》系列作品蕴藏着巨大潜力。 对索尼而言，续集既是一种肯定，也是一种挫败。工作室证明自身具备打造全球爆款的能力，但财务收益却主要流向网飞。尽管索尼保留制作后续作品的权利，但新协议条款仍需协商——而网飞如今掌握着大部分谈判筹码。 这一事件的启示远超单部影片的范畴。在IP日益成为长期价值驱动力的行业里，“拥有爆款”与“为他人打造爆款”的差距可能以数十亿美元计。未来数年，《K-POP：猎魔女团》可能为网飞带来电影、剧集、消费品及线下体验等多方面收入，而索尼只能继续推进下一个项目，期望好运再次降临。（财富中文网） 为撰写本报道，《财富》杂志使用生成式人工智能协助完成初稿。在发布前，编辑已核实信息准确性。 译者：中慧言-王芳 • KPop Demon Hunters is a bona fide megahit for Netflix, this week becoming its most-watched movie ever with 236 million views, while making Billboard history with four simultaneous top 10 hits. Sony Pictures Animation, which spent $100 million creating the phenomenon, will reportedly net only $20 million owing to a pandemic-era distribution deal. The outcome represents a massive missed opportunity for Sony, as Netflix now controls what could become a multibillion-dollar franchise. Netflix has a monster hit on its hands, and it’s not what anyone expected. KPop Demon Hunters, an animated film about a K-pop girl group who are also demon hunters, has officially become Netflix’s most-watched movie ever with 236 million views, dethroning the previous record-holder Red Notice and its 230.9 million views. The milestone comes just 67 days after the film’s June 20 debut, making it one of the fastest climbs to the top of Netflix’s all-time charts. KPop Demon Hunters isn’t just breaking movie-streaming records: Four songs from its soundtrack are currently sitting in the Billboard Hot 100 top 10 at the same time, something that has actually never happened in the chart’s 67-year history. (“Golden” holds the No. 1 spot, “Your Idol” sits at No. 4, “Soda Pop” is at No. 5, and “How It’s Done” landed at No. 10, since you asked.) And when Netflix decided to test the waters with a sing-along theatrical release last weekend, the film earned an estimated $18 million to $20 million at the box office across roughly 1,700 theaters, despite being available to stream at home. The success has been so overwhelming Netflix and Sony are already in early talks for a sequel. For Netflix, this represents the kind of breakout animated franchise the company has been chasing for years. But for Sony Pictures Animation, which created the film, the story is more complicated—and potentially represents one of the biggest missed opportunities in recent Hollywood history. The making of the KPop Demon Hunters phenomenon Sony Pictures Animation developed KPop Demon Hunters with a reported production budget of around $100 million, positioning it as a significant bet on the global appeal of both K-pop culture and supernatural adventure. Directed by Maggie Kang and Chris Appelhans, the film follows the fictional girl group Huntr/X as they battle demons while maintaining their pop-star careers. There’s a rival boy band called Saja Boys … you can imagine where this is going. The creative gamble, so far, has paid off in surprising ways. The film’s soundtrack didn’t just complement the story—it became a genuine musical phenomenon, with “Golden” becoming the eighth K-pop song to hit No. 1 on the Hot 100, the first time a song from an animated movie reached that spot since “We Don’t Talk About Bruno” from Disney’s Encanto, and the first to feature female artists. So far, the film has sustained its momentum. KPop Demon Hunters has now spent 10 consecutive weeks at No. 1 on Netflix’s movie charts, adding 25.4 million views in just the most recent week tracked. That kind of staying power is rare for any Netflix original, let alone an obscure animated film that isn’t from an established IP. Sony’s deal—and what it walked away with Obviously, KPop Demon Hunters is massive for Netflix. And Sony actually made the movie, so it should be equally massive for them, too, right? Well, not so much. Despite spending roughly $100 million to create what became a global phenomenon, Sony Pictures is expected to net only about $20 million in profit from what is potentially a billion-dollar franchise in KPop Demon Hunters; basically, a fraction of the upside. The reason lies in a 2021 distribution deal Sony struck with Netflix, designed to guarantee returns during the uncertain pandemic era. According to Puck’s Matthew Belloni, Sony agreed to a “direct-to-platform” arrangement where Netflix would pay back the film’s production budget plus an additional fee capped at $20 million per project. In exchange, Netflix retained all rights to the property and owes no additional profit participation, even as the film becomes a massive hit. This wasn’t Sony shopping around a finished film; Netflix essentially funded the production while Sony handled the creative work. At the time, the deal made sense. Theaters were still recovering from pandemic closures, animated films were struggling at the box office, and Sony lacked its own major streaming platform. The arrangement guaranteed Sony would make a profit without risking a theatrical flop. But nobody—not even Netflix executives—predicted KPop Demon Hunters would become as big as it did. To understand the magnitude of what Netflix acquired, consider what Red Notice represented for the platform. That 2021 action film starring Dwayne “The Rock” Johnson, Ryan Reynolds, and Gal Gadot held Netflix’s top spot for nearly four years, with its 230.9 million views becoming the benchmark for Netflix success. KPop Demon Hunters blew past that number, but it also demonstrated something Red Notice couldn’t achieve: franchise potential. The film’s soundtrack success alone opens up revenue streams that most Netflix originals can’t touch. A reminder: four simultaneous Billboard top 10 hits. And the success of the theatrical sing-along experiment provides another data point for Netflix (and Netflix loves its data points). Netting $18 million to $20 million in a single weekend across 1,700 theaters—roughly half the number of theaters a blockbuster release would get—suggests real audience demand for communal experiences around the franchise, which is promising if Netflix is looking at expanding into more physical spaces. The missed opportunity for Sony Had Sony kept the rights to KPop Demon Hunters, the company would be sitting on something potentially worth billions. But what truly acts as salt in the wound, and perhaps some form of cruel irony, is that last September, Sony’s own chief financial officer, Hiroki Totoki, said this in an interview with the Financial Times: “Whether it’s for games, films, or anime, we don’t have that much IP that we fostered from the beginning. We’re lacking the early phase [of IP], and that’s an issue for us.” Sony has been candid about its struggles to develop lasting entertainment franchises beyond Spider-Man. Company executives have acknowledged the studio needs more original intellectual property fostered from the beginning—exactly what KPop Demon Hunters represents. Instead, Sony now watches Netflix leverage the property for sequels, merchandise, and much more. (Sony, for what it’s worth, tells Fortune it will share in some of the profits, including album sales. Netflix offered no further comment.) The numbers make the missed opportunity even starker. For context, Netflix reportedly paid $465 million to acquire the rights to Seinfeld reruns. KPop Demon Hunters is an original property that has already proved its global appeal, demonstrated theatrical viability, and created genuine music hits. The $20 million Sony will earn looks modest against that backdrop. Early sequel talks and what’s next The speed with which Netflix and Sony entered sequel discussions tells its own story. When a property breaks platform records, generates chart-topping music, and proves theatrical demand all within two months, the economics become clear quickly. Netflix wants to strike while the iron’s hot, and there’s a lot of potential for a KPop Demon Hunters universe. For Sony, the sequel represents both a vindication and frustration. The studio proved it could create a global hit, but the financial upside flows primarily to Netflix. While Sony retains the right to produce future installments, the terms of any new deals remain to be negotiated—and Netflix now holds most of the leverage. The broader lesson extends beyond this single film. In an industry where intellectual property increasingly drives long-term value, the difference between owning a hit and creating one for someone else can be measured in billions. KPop Demon Hunters will likely generate revenue for Netflix across multiple films, series, consumer products, and live experiences for years to come. Sony, meanwhile, will move on to the next project, hoping lightning strikes twice. For this story, Fortune used generative AI to help with an initial draft. An editor verified the accuracy of the information before publishing. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
8月28日，牛市氛围渐浓的A股低开低走且一度跌破3800点，但午后上演“V形反转”，最终三大指数集体大涨。无独有偶，在A股上演“大奇迹日”的同时，人民币兑美元在岸与离岸汇率也双双涨破7.13关口，创下年内新高。汇率暖流与股市联动的叙事在沉寂多时后再起涟漪，为似已进入下半场的牛市新增了一个关键的驱动因素。 29日早间，中国人民银行将人民币兑美元中间价定在7.103，为连续第五日上调，同样刷新年度最高水平。人民币兑美元汇率中间价是央行每日公布的官方基准价，市场价则在其±2%的波幅内由供求决定，二者如同“锚”与“浪”的关系，由此来看，央行近日显然在顺势而为。 人民币此番走强并非孤立事件，而是内外因素共同作用的结果。从外部环境看，美联储主席鲍威尔在上周五（8月22日）的杰克逊霍尔全球央行年会上暗示可能调整政策立场，市场对美联储9月降息概率的预期上升，美元指数当日跳水，这给人民币升值积累了较好的外部环境。从内部支撑来看，国内政策主动引导预期，8月25日以来央行中间价加速调升，更强化了市场稳汇率信号。 与此前的波澜不惊相比，人民币汇率开始逐浪而行。民生证券分析认为，中期视角下，人民币兑美元汇率此前存在“滞涨”现象，为本次补涨埋下伏笔。2025年以来，美元指数较1月高点累计贬值约10%，但同期人民币兑美元中间价仅上涨1.2%，这种“美元跌、人民币滞涨”的格局积累了补涨势能。 回顾历史，人民币汇率与股市之间存在显著联动效应。从企业盈利角度分析，由于港元盯住美元，因此当美元兑人民币贬值时，港元兑人民币也会贬值，而大部分港股上市公司是中国内地公司，盈利以人民币计价，这意味着即使公司人民币计价盈利不变，港元计价的报表盈利也会上升，这为港股上行带来了助力。国金证券研报指出，站在资本市场角度，人民币升值有助于直接提振外资信心、修复中国资产估值，这一过程中港股反应往往比A股更快。 同时，基于过往经验，人民币升值与A股市场同样是互为支撑。8月28日的A股V型反弹与人民币兑美元“破位”升值在时间上高度同步，这两者间的同频共振已经初露峥嵘。 股汇联动的背后则是跨境资本流动格局的重构。人民币走强预期促使企业加速结汇，7月份的出口商单月结汇率大幅跳升至54.9%，高于6月的46.1%，创2024年9月以来新高，显示出口商正在加快卖出美元。回看2017-2018年人民币升值周期，当时在人民币单边贬值预期被打破及A股“漂亮50”行情带动下，企业积累的大量的美元头寸开始集中结汇，这反过来又进一步推动了人民币升值和结汇资金流入A股，形成了股汇联动的正向循环。而当前外贸企业手中的美元资金规模更为庞大，其集体转向给市场带来的影响可能也会更强。 同时，外资增配中国权益资产的趋势也愈发明显。今年上半年，外资净增持境内股票和基金101亿美元，5-6月单月增持规模达到188亿美元，创历史同期新高。值得关注的是，对于外资所忧虑的部分领域产能过剩现象，下半年以来全国各行业纷纷发布了“反内卷”倡议。招商证券研报指出，若“反内卷”政策陆续推行、有效落实，中国企业竞争格局将得到显著改善；若人民币重回6时代叠加实际有效汇率升值将放大中国权益资产吸引力，外资大概率流入并强化通胀、内需策略，中国资产进而迎来全面重估，龙头白马标的尤其是消费等内需资产或已迎来配置窗口。 8月29日的美元兑人民币外汇远期汇率交易情况显示，市场预期人民币汇率三年后会升值到6.76左右。如果只看今年，人民币汇率补涨行情也仍有较大空间，尤其是伴随美联储恢复降息以及关税政策对美国经济冲击逐步显现，美元指数将持续承受下行压力。此外，潜在的“中美再会谈”也可能会成为人民币年内进一步升值的催化剂。 然而，考虑到当前出口和贸易环境尚存不确定性以及内需复苏仍需巩固，政策层面大概率仍将谨慎控制人民币升值的节奏，避免过快升值而对实体经济尤其是就业造成冲击。央行在近期发布的2025年第二季度货币政策执行报告中就明确表示，将“坚决对市场顺周期行为进行纠偏，坚决对扰乱市场秩序行为进行处置，坚决防范汇率超调风险”，这意味着如果人民币过快升值，央行可能会出手干预。 尽管人民币升值路径仍面临诸多约束，但对中国资产的新一轮重估或已开始。在这场汇率与股市的“双人舞”中，能够同时把握汇率升值与产业突破两条主线的企业，可能会成为市场重新定价的赢家。国际资本的流向也印证了这一趋势：以今年5月在港股上市后至今已涨逾60%的宁德时代为例，据香港交易所8月26日披露，摩根大通对其H股的多头持仓比例于8月21日从5.76%增至6.06%，摩根士丹利对其持仓比例也从4.96%增至6.05%，表明外资更看好中国那些具有独特产业优势，特别是在高端制造和芯片等已取得突破性进展的领域。 而若跳开企业的层面，从更宏观的角度来看，在地缘政治和企业盈利风险仍存的背景下，人民币汇率与股市的互动之路毫无疑问并不会一帆风顺。这场“双人舞”能否跳得更稳健和优美，最终仍取决于中国经济的内在韧性。（财富中文网） 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
8月21日，名创优品发布中期业绩公告，公司总收入93.93亿元，同比增长21.1%，而期内利润却同比减少23.1%。收入增加而利润反降的一大原因是，董事会主席兼首席执行官叶国富对永辉超市的押注。 去年，名创优品收购永辉超市29.4%的股权。交易达成前，该消息导致公司股价承压。即便如此，叶国富依然表示对这次收购抱有信心。永辉的亏损确实拖累了名创优品的财报表现。 永辉超市何时能从调整中恢复尚不可知，但叶国富已经为名创优品规划了新的增长路径——公司未来将以“国际IP+自有IP”并行。 8月22日，名创优品美股与港股股价均大涨超20%。 业绩会上，叶国富表示，之前名创优品只做国际IP授权，是一条腿走路。半年前，集团已开始进行自有IP布局。他认为，名创优品在与国际IP的合作中，积累了开发、营销和渠道优势，若采取行动将自有IP补齐，集团将“大有前途”。他将“国际IP+自有IP”定位为集团未来的重大战略。 正当红的IP公司泡泡玛特的业绩，支撑了叶国富的这一决策。在他宣布未来方向的前一天，泡泡玛特创始人王宁在业绩会上说，今年营收300亿很轻松。 叶国富表示，看到泡泡玛特业绩高增“非常开心”，因为这体现出了消费者对于潮玩产品的支持，以及资本市场愿意为此而买单，潮玩市场在中国刚刚崛起。 能够与泡泡玛特形成直接对标关系的是名创优品集团旗下的潮玩品牌TOP TOY。 该品牌创立于2020年，定位潮流玩具集合店。截至今年上半年，TOP TOY门店数量为293家，其中包括自2024年下半年启动出海以来，在海外开设的10家门店。在名创优品上半年的业绩中，TOP TOY品牌期内贡献收入7.42亿元，同比增长73%，主要得益于平均门店数量的快速增长。 截至目前，集团已签约首批9位潮玩艺术家，其中名为“右右酱”的IP已于今年6月推出。叶国富表示，“预计‘右右酱’今年会达到4000万销售规模，明年破亿。” 另外，TOP TOY今年以510万元认缴出资，以51%持股比例入主HiTOY海创文化。通过这笔投资，名创优品自有IP矩阵得以扩大，直接获得了HiTOY旗下的三大核心IP——糯米儿Nommi、Honey甜心、霉霉MayMei。叶国富表示，HiTOY所推出的系列产品上半年销售额已过亿。 今年6月，名创优品在港交所公告称，公司正在对以TOP TOY品牌运营的潮流玩具业务潜在分拆上市可能性进行初步评估。该计划仍处早期阶段，是否推进将取决于包括市场条件在内的诸多因素。 若该计划成型，港股将迎来又一个盲盒巨头。而叶国富也将在名创优品、永辉超市之后，收获又一家上市公司。 据财报显示，TOP TOY已收获了来自淡马锡领投的最新融资，投后估值达到约100亿港元。 在集团布局和资本助推下，TOP TOY加快了开店速度，今年上半年净新开98家门店。根据今年3月公司向外传达的规划，预计2025年门店数量将实现50%至60%的同比增长，而业绩增速或将达到70%至80%。 名创优品的野心不止于此——今年TOP TOY宣布，计划未来五年在全球100个核心商圈开设超过1000家店。作为对比，如今市值超4000亿港元的泡泡玛特，目前全球门店数量为五百余家。 泡泡玛特如今的成功与海外市场关系密切，TOP TOY同样也对海外寄予厚望。 TOP TOY创始人兼首席执行官孙元文曾表示，2025年，计划全年新增门店150家，其中100家位于中国。如今国内门店数量的增加已接近预期，而海外市场距目标仍有一段距离。公司预计，TOP TOY未来海外销售预计占比突破50%。 孙元文透露，公司已为海内外扩张准备好百亿储备资金。 伴随着扩张步伐，今年3月，TOP TOY首家全球旗舰店在上海南京东路步行街开业。开业首日客流超3万，单日销售成绩超108万元。孙元文表示，这一顶级商圈位置的竞标对手是安踏、波司登和万代，在资金以外，能拿下这一位置也是品牌影响力的体现。 但即便拥有多年积累的资源作为支撑，对经营自有IP经验有限的名创优品而言，新的道路将面临诸多挑战。成熟IP授权和孵化自有IP存在本质不同。 泡泡玛特旗下多款自有IP的成功，离不开其多年积累的孵化经验和搭建的运营体系，远非将IP直接买下这样简单。王宁经常强调“尊重时间、尊重经营”。根据对IP生命周期的理解，他说，“IP企业最重要的就是持续运营IP，要长期培养与经营IP”。 要做到这点，难度可能不亚于在全球开出1000家门店。（财富中文网）
英伟达加州圣克拉拉总部。图片来源：JUSTIN SULLIVAN—GETTY IMAGES 人工智能芯片制造商英伟达（Nvidia）在周三公布财报时表示，第二季度没有来自中国的H20芯片销售收入，当季营收略高于华尔街预期。 财报业绩虽然证实了对人工智能基础设施的需求依旧强劲，但仍未能令投资者满意。作为全球市值最高的公司，英伟达股价在周三晚间的盘后交易中下跌了4%，至约175美元。 Info-Tech研究集团（Info-Tech Research Group）顾问研究员斯科特·比克利在财报电话会议前对《财富》表示：“（股价波动）大概只是对一个‘一般般’的数据的初步反应。我们居然将一个季度467亿美元的营收视为‘一般般’，这未免太荒唐了。” Visible Alpha收集的数据显示，英伟达第二季度营收较去年同期增长了56%，达到467.4亿美元，高于华尔街预期的465.2亿美元。净利润为264亿美元，较上季度的187.8亿美元增长40.8%。摊薄后每股收益为1.08美元，超过第二季度1.02美元的预期。毛利率升至72.4%，较上季度的61%有了大幅提升。 自今年4月以来，英伟达一直面临针对H20芯片对华出口的贸易限制。美国政府自7月起开始向部分获批的中国买家发放许可证，英伟达表示少数中国客户已获得相关许可。但英伟达表示，第二季度财报中并未包含任何来自中国的H20芯片营收。（英伟达指出，第二季度有部分H20芯片库存销往了中国以外的地区，为营收带来1.8亿美元的增量。） 本月早些时候，特朗普政府宣布了相关计划，允许英伟达及其竞争对手超威半导体（AMD）向获批的中国买家出售特定人工智能芯片，而美国政府将从销售收入中抽取15%的分成。但英伟达表示，目前尚未有任何实质性进展。 英伟达首席财务官科莱特·克雷斯（Colette Kress）在财报电话会议上表示：“迄今为止，美国政府尚未公布任何将此类要求制度化的法规。” 英伟达表示，当前季度的财务预期中未计入H20芯片，但如果“地缘政治”问题得到解决，预计价值20亿至50亿美元的H20芯片将有望出口至中国。公司同时再次呼吁美国政府允许其向中国销售更先进的“Blackwell”系列产品。 首席执行官黄仁勋在财报公告中谈及全球数据中心正在采用的下一代人工智能芯片时表示：“Blackwell Ultra的产能正在全速提升，需求极为强劲。人工智能竞赛已经拉开帷幕，而Blackwell正是这场竞赛的核心平台。” 英伟达数据中心业务营收占据公司主营业务的主要部分，同比增长56%，环比增长5%，达到411亿美元。其汽车与机器人业务板块增速最快，同比增长69%。 思博瑞投资管理公司（Allspring Global Investments）在其部分基金中持有英伟达股份。思博瑞高级投资组合经理兼成长型股票投资团队负责人迈克尔·史密斯对《财富》表示：“市场预期已经极高，但英伟达再次超越了预期。随着Blackwell产能爬坡，利润率不断上升；在出口管制措施实施之后，中国依然存在着巨大的未被开发的机遇；而600亿美元的股票回购，则是在创纪录自由现金流背景下的额外利好消息。”（财富中文网） 译者：刘进龙 审校：汪皓 Nvidia recorded no China sales revenue for H20 chips and reported revenue that narrowly beat Wall Street targets in the second quarter, as the AI chipmaker reported financial results on Wednesday. The results, while confirming that demand for AI infrastructure remains solid, left investors underwhelmed and shares of Nvidia, the world’s most valuable company, declined 4% to around the $175 mark in extended trading Wednesday evening. “[The stock movements are] probably just an initial reaction to a so-so number,” Scott Bickley, an advisory fellow at Info-Tech Research Group, told Fortune before the earnings call. “Which is kind of insane that we’re viewing $46.7 billion in a quarter as ‘so-so,’” he said. Nvidia’s revenue increased 56% from the same period a year ago to $46.74 billion, exceeding Wall Street’s projection of $46.52 billion, per data compiled by Visible Alpha. Profits came in at $26.4 billion, a 40.8% increase from $18.78 billion last quarter. Nvidia posted diluted earnings per share at $1.08, beating projections of $1.02 for the second quarter. Nvidia’s gross margins grew to 72.4%, up significantly from 61% last quarter. Nvidia has been navigating trade restrictions on H20 shipments to China since April. The U.S. government began issuing licenses for approved buyers in China in July, and Nvidia said a few of its China-based customers had received such licenses. But no H20 chip revenue to China was included in its second-quarter revenue, Nvidia said (It noted that some H20 chip inventory was sold outside of China in the second quarter, adding a $180 million benefit to the topline). While the Trump administration announced plans earlier this month to allow Nvidia and AMD, a rival chipmaker, to sell certain AI chips to approved Chinese buyers while giving the U.S. government a 15% cut of the proceeds, Nvidia said nothing concrete has yet come of it. “To date the USG has not publicized a regulation codifying such requirement,” CFO Colette Kress said on the company’s earnings call. Nvidia said it was not including H20 in its financial forecast for the current quarter, though it estimated that $2 billion to $5 billion worth of H20 chips could be shipped to China if “geopolitical” issues were resolved. The company also repeated its call for the U.S. government to allow it to sell its more advanced “Blackwell” generation of products to China. “Production of Blackwell Ultra is ramping at full speed, and demand is extraordinary,” CEO Jensen Huang said of the tech behemoth’s next-generation AI chip, which is used in data centers globally, in the earnings release. “The AI race is on, and Blackwell is the platform at its center.” Nvidia’s datacenter revenue, which accounts for the bulk of its business, grew 56% year-over-year, and 5% sequentially, to $41.1 billion. The company’s automotive and robotics segment grew the most at 69% year over year. “Expectations were sky-high, but Nvidia exceeded them again,” Michael Smith, senior portfolio manager and head of the growth equity team at Allspring Global Investments, told Fortune. Allspring owns Nvidia in some of its funds. “Margins are rising as Blackwell ramps; China remains a massive untapped opportunity post–export controls; and a $60 billion buyback is an extra sweetener amid record free cash flow.” 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
英伟达（Nvidia）首席执行官黄仁勋（Jensen Huang）图片来源：Chesnot—Getty Images 英伟达的财报已不再仅仅关乎公司自身。这家市值高达4万亿美元的芯片巨头的季度财报已成为人工智能热潮的试金石，进而成为整个股市的风向标。作为在标普500指数（按市值加权）中占比达8%的成份股，且在支撑生成式人工智能发展的芯片领域占据无可匹敌的地位，英伟达的业绩在华尔街眼中早已超越“单一公司成绩单”的范畴，更像是宏观经济指标。其财报发布甚至演变成一种文化现象，催生出“观摩派对”。 投资者正屏息等待英伟达周三收盘后发布的最新季度财报。从英伟达期权交易情况来看，市场预期该股将上下波动6%——这意味着其市值可能发生2600亿美元的变动。 自五月份上次季度财报发布以来，英伟达股价已飙升35%。鉴于近期市场对人工智能相关股票存在危险金融泡沫的担忧加剧，本季度最受瞩目的财报盛事的紧张氛围进一步升级。此外，英伟达在华业务的不确定性依然存在。 华尔街分析师预计英伟达第二季度营收将同比增长53%至460亿美元，达到公司业绩指引区间的上限，每股收益为1.01美元。作为英伟达核心业务的数据中心销售额预计将接近400亿美元。鉴于英伟达股价近几个月已累积可观涨幅，若周三公布的财报未达预期，抑或因中国限制政策而发布谨慎业绩指引，其股价恐将暴跌。 英伟达陷中美博弈漩涡 尽管英伟达仍是生成式人工智能浪潮的最大受益者之一，但随着中美争夺科技主导权，该公司关键业务已沦为地缘政治博弈的筹码。今年4月，美国政府开始要求该公司出口H20芯片前必须申请相关许可证。H20芯片是英伟达为遵守2022年末生效、2023年进一步收紧的美国出口管制规定，专门研发的顶级人工智能芯片的简化版本。更为严苛的出口许可证要求迫使该公司在第一季度计提45亿美元的资产减值损失，涉及未售库存和采购承诺。 此后英伟达在华业务愈发复杂。在英伟达首席执行官黄仁勋前往海湖庄园与特朗普会面之后，白宫宣布将批准该公司销售H20芯片。英伟达虽已提交出口许可证申请，但受美国政府强硬立场及中国买家采购时犹豫不决的影响，申请进程遭遇严重延误。本月初，英伟达与超微半导体（AMD）与特朗普政府达成协议，以对华芯片销售收入的15%换取出口许可证。 然而在H20芯片恢复出货之际，中国政府开始劝阻国内企业，推动其放弃采购该芯片，原因是担忧英伟达要求客户提交供美国政府审查的信息可能包含敏感内容。据报道，中国政府还宣称发现证据，表明英伟达芯片可能存在后门，允许美国情报机构提取芯片使用数据。 上周黄仁勋在中国台北宣布，英伟达已开始逐步停产H20芯片，并着手研发性能更强的后续产品。他表示公司正致力于推出“面向人工智能数据中心的新产品”，该产品将按美国要求进行调整，降低部分性能。他表示正在寻求特朗普政府批准该芯片的对华销售许可。 黄仁勋表示：“毋庸置疑，这取决于美国政府。我们正与他们沟通，但目前尚无定论。” 受上述诸多不确定性影响，分析师预测英伟达在本次财报中不会提及中国市场的营收情况。 “我推测他们既不会对中国市场的营收进行统计，也不会作出相关预测，原因是存在太多不确定性。”Cambrian-AI Research创始人兼首席分析师卡尔·弗罗伊德（Karl Freund）指出。 J. Gold Associates创始人兼首席分析师杰克·高德（Jack Gold）向《财富》杂志指出，英伟达如今需同时取悦两大群体：股东和特朗普政府。“他们正陷入进退两难的境地，”他坦言，“当前局势异常诡谲——美国政府实际上已将手伸进了这些企业的钱袋子。” 人工智能泡沫难题 除地缘政治因素外，英伟达还面临另一重挑战：人们愈发担忧人工智能热潮开始步入泡沫阶段。这将直接冲击英伟达的业务核心及其极高估值——该公司当前市盈率超40倍——依赖于对其高性能图形处理单元（GPU）持续增长的需求。英伟达的增长高度依赖少数几家云计算巨头，包括Meta、亚马逊（Amazon）、谷歌（Google）和微软（Microsoft），以及OpenAI等资金雄厚的人工智能初创企业。倘若这些公司放缓支出步伐，英伟达可能突然失去最大买家。 “我确实认为所有人都担心人工智能存在泡沫，”弗罗伊德表示，不过他补充道，这种担忧已持续三年之久。他强调，自己并不认为泡沫会在当下破裂。“我认为未来还有二到五年的增长期。”他说道。 高德对此表示认同，称英伟达“至少在接下来的几个季度内，甚至几年内都能获得可观利润”，但他警告称，在某个特定时刻，倘若市场崩盘，用于购买芯片的资金便会消失殆尽。 “这着实令我忧心不已，”他坦言。“这次，我笃定英伟达的财报数据依然亮眼——（英伟达）以高得离谱的价格出售所能制造的所有产品，这倒也无妨，只要它能一直这般蒙混过关。”但他补充道，从更广泛的市场角度来看，大规模的人工智能数据中心建设“不可能永远持续下去”。 弗罗伊德指出，这正是黄仁勋实际上正着力引导投资者，将关注焦点从以数据中心为中心的视角转向英伟达业务其他板块的原因，包括其汽车和机器人业务：“这便是他当下的策略，即随着人工智能从数据中心走向现实世界，如何让投资者转向以更全面的视角看待人工智能。”（财富中文网） 译者：中慧言-王芳 Nvidia’s earnings aren’t just about Nvidia anymore. The $4 trillion chipmaker’s quarterly financials have become a litmus test for the AI boom—and, by extension, for the whole stock market. Constituting 8% of the market-cap-weighted S&P 500 index and with an unrivaled grip on the chips that power generative AI, Nvidia sees its results treated more like a macroeconomic indicator by Wall Street than as a report card on a single company. The earnings announcement has even become a cultural phenomenon complete with watch parties. Investors are bracing for the company’s latest quarterly results due after Wednesday’s market close, with trading in Nvidia options implying expectations that the stock will move 6%, up or down—equal to a $260 billion change in Nvidia’s market value. In the three months since the company last gave investors a quarterly update, back in May, Nvidia’s stock has surged 35%. But the tension surrounding what is already the most closely watched earnings event of the season has been ratcheted up by recent jitters over what some worry is a dangerous financial bubble in AI-related stocks. And uncertainty about Nvidia’s China business continues to loom large. Wall Street analysts are looking for Nvidia’s Q2 revenue to surge 53% year over year to $46 billion, at the high end of Nvidia’s guidance, with earnings per share of $1.01. Data center sales, the crux of Nvidia’s business, are expected to come in close to $40 billion. But with Nvidia’s shares having gained so much in recent months, a miss on Wednesday, or cautious guidance tied to China restrictions, could send the stock plummeting. Nvidia in the U.S.-China crosshairs Nvidia may remain one of the greatest beneficiaries of the generative AI boom, but a critical part of the company’s business has also become a geopolitical football as the U.S. and China compete for technological dominance. In April, Washington began requiring export licenses for the company’s H20 chips—stripped-down versions of Nvidia’s top-of-the-line AI chips that were specifically designed to comply with the U.S. export controls that took effect in late 2022 and were tightened again in 2023. Those tighter export licenses forced the company to take a $4.5 billion charge in Q1 tied to unsold inventory and purchase commitments. From there, things only got more complicated for Nvidia’s China business. After Nvidia CEO Jensen Huang visited President Trump at Mar-a-Lago, the White House said it would permit the company to sell H20s after all. Nvidia applied for export licenses but faced extensive delays, thanks to the tougher U.S. stance and Chinese buyers hesitating to commit to purchasing. Then, earlier this month, Nvidia and AMD struck a deal with the Trump administration to grant licenses in exchange for a 15% revenue-sharing arrangement on China chip sales. But as shipments of H20 chips resumed, China began discouraging companies from buying them, expressing concerns that the information Nvidia was asking customers to submit for U.S. government review could contain sensitive information. The Chinese government also reportedly claimed it had found evidence that Nvidia’s chips might contain back doors that would allow U.S. spy agencies to extract data on how they were being used. In addition, comments from U.S. Commerce Secretary Howard Lutnick about providing China with Nvidia’s “fourth-best chips” were considered “deeply insulting” by Chinese officials, according to the Financial Times. Finally, last week Huang announced in Taipei that Nvidia has started winding down production of the H20 chip and begun work on a more powerful successor, saying the company was working on offering a “new product for AI data centers,” modified to reduce some of its performance, as required by the United States. He said he was seeking the Trump administration’s approval to sell the chip. “It’s up to, of course, the United States government,” Huang said. “And we’re in dialogue with them, but it’s too soon to know.” As a result of all the uncertainty, analysts predict Nvidia will not allude to China revenue in the earnings report. “I suspect they will not count, nor forecast China revenue; there’s too much uncertainty involved,” said Karl Freund, founder and principal analyst at Cambrian-AI Research. Jack Gold, founder and principal analyst at J. Gold Associates, told Fortune that Nvidia now has two primary groups to please: stockholders and the Trump administration. “They’re caught between a rock and a hard place,” he said. “It’s a really strange situation we’re in now where the government in the U.S. actually has their hands into the pockets, into the wallets of these companies.” AI bubble trouble Beyond geopolitics, Nvidia faces another challenge: growing unease that the AI boom is starting to look like a bubble. This would strike at the heart of Nvidia’s business and its stratospheric valuation—the company trades at more than 40 times its projected earnings—which rely on ever-growing demand for its powerful GPUs. Nvidia’s growth is heavily concentrated in a handful of cloud giants, including Meta, Amazon, Google, and Microsoft, as well as highly funded AI startups like OpenAI. If those companies slow spending, Nvidia could suddenly lose its biggest buyers. “I do believe that everyone’s concerned about an AI bubble,” said Freund, though he added that those concerns have lasted for three years already. He did not, he emphasized, think it would pop now. “I think there are still two to five years of growth left,” he said. Gold agreed, saying there were “at least several quarters, if not a couple of years of good profits” for Nvidia, but he cautioned, at some point, if the market crashed, that money spent on chips would go away. “It concerns me,” he said. “This time, I’m sure the earnings will still be great—[Nvidia is] selling everything they can build at ridiculously inflated prices, which is fine, if you can get away with that.” But from a broader market perspective, he added, the massive AI data center build-outs “can’t go on forever.” That’s why, said Freund, Huang is actually working to get investor attention to shift from the data-center-centric view to other areas of Nvidia’s business, including its automotive and robotics work: “That’s his game right now, how to get investors to shift to a more holistic view of AI as it moves out of the data center and into the real world.” 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
美国联邦储备委员会理事莉萨·库克（Lisa Cook）宣布将就总统唐纳德·特朗普试图免去其职务一事提起诉讼。图片来源：SAUL LOEB—AFP/Getty Images 特朗普解雇美联储理事库克，引发了一场围绕央行独立性的法律对决。但至少在华尔街，一些投资者并不担心制度规范受损，反而对未来融资成本下降的前景感到振奋。 “这是非常利好的消息，”基础设施资本顾问公司（Infrastructure Capital Advisors）首席执行官杰伊·哈特菲尔德（Jay Hatfield）在接受《财富》采访时表示。“简单来说，消除美联储的无能远比捍卫所谓的美联储独立性更重要。美联储一向政治化，只是特朗普把它说到了台面上。” 事实上，市场对这一消息反应平淡，这令一些经济学家感到困惑。《金融时报》Alphaville专栏作家罗宾·维格尔斯沃思（Robin Wigglesworth）直言，市场表现出“令人难以置信的乐观情绪”。 “对制度诚信根深蒂固的期望如今已经荡然无存，”维格尔斯沃思写道。 周二早盘标普500指数和道琼斯指数基本持平，而纳斯达克指数甚至微涨0.3%。特朗普此举后，长期美债收益率上升而短期收益率下滑，导致收益率曲线趋陡，表明投资者押注利率在短期内可能下降，但若政治化的美联储减少对通胀的关注，长期利率将逐步走高。美元指数则下跌0.3%。 在哈特菲尔德看来，这正是问题所在。他认为美联储现行货币政策过于紧缩，当前政策利率“较中性水平高出150个基点”，意味着货币政策对经济增长的抑制过度。 “如果联邦基金利率在3%，而总统要求大幅降息，那才会十分危险，”他说。“但目前远未达到那种境地。” 特朗普已提名斯蒂芬·米兰（Stephen Miran）接替本月早些时候宣布辞职的理事，如果再成功替换库克，加上此前任命的米歇尔·鲍曼（Michelle Bowman）和克里斯托弗·沃勒（Christopher Waller），特朗普将拥有四个理事席位。这将使七人组成的美联储理事会形成共和党占多数的局面。 哈特菲尔德认为，共和党占主导的美联储理事会将推动降息。 “关税就像一次性税收——它在消费者物价指数（CPI）中出现一次便消失，”他说。 哈特菲尔德补充道，沃勒和鲍曼在上月维持利率不变的决议中投反对票时，就明白关税并非持续性通胀推手。相反，由民主党任命的理事却对此认识不足，导致他们“推迟降息并使经济置于风险之中”。 “所以，摆脱鲍威尔领导下的美联储，对股市和债市都是非常利好的消息，”他说。 短期内，哈特菲尔德预计今年至少会有两次降息，与美联储主席杰罗姆·鲍威尔（Jerome Powell）近期释放的鸽派信号相呼应。 “昨夜债市遭到抛售，但如果降息成真，那对债市和股市都是重大利好，”他说。 独立性岌岌可危，市场护栏脆弱难支 其他经济学家则远不如此乐观。布鲁金斯学会（Brookings）学者大卫·韦塞尔（David Wessel）警告称，特朗普“似乎决心掌控美联储——并将动用一切手段争取多数席位”，他称这是削弱民主根基的又一举动。 投行派杰（Piper Sandler）的分析师同样直言，如果投资者认为市场会约束特朗普，那就是自欺欺人。 “凭什么相信所谓的债券义勇军会在危机来临前警告国会？”他们写道，并指出市场既未预见到2022年的通胀冲击，也未预见到全球金融危机前的楼市崩盘。相反，他们认为，股市上涨完全是基于降息预期，“哪怕其中部分动力来自政治压力”。 分析师所强调的更大风险在于制度结构。分析师吉姆·比安科（Jim Bianco）在X平台上解释称，美联储七位理事必须在2026年2月12位地区联储行长五年任期届满时，重新批准或否决所有任命。 随着更多特朗普提名者进入理事会，即便是芝加哥联储主席奥斯坦·古尔斯比（Austan Goolsbee）或纽约联储主席约翰·威廉姆斯（John Williams）这样的核心人物，也可能面临职位不保，从而重塑联邦公开市场委员会（FOMC）的力量格局。 派杰警告称，“长期牛市支柱正逐一被拆除”，自由贸易格局遭逆转，而“稳健货币支柱也正遭到根本性削弱”。他们总结道，市场难以制衡美联储的政治化趋势。 目前市场仍聚焦于短期宽松政策。 瑞银全球财富管理（UBS Global Wealth Management）策略师乌尔丽克·霍夫曼-布查尔迪（Ulrike Hoffmann-Burchardi）在给客户的报告中表示，她的团队仍预期美联储将在接下来的四次会议中累计降息100个基点。 “我们将持续关注美联储面临的政治压力不断加剧的态势，”她写道，“但预计短期内其决策仍将以其使命为指引。” 至于哈特菲尔德，他对所谓的“独立性”并不在意。 “通胀已经得到控制，劳动力市场正在走弱，而我们正走向衰退，”他说。“真正的问题不在于特朗普对阵美联储，而在于美联储数十年来的无能，市场对此早已心知肚明。任何扭转局面的举措，都是利好。”（财富中文网） 译者：刘进龙 审校：汪皓 President Donald Trump’s attempt to fire Federal Reserve governor Lisa Cook has triggered a legal showdown over the central bank’s independence. But on Wall Street at least, some investors aren’t worried about institutional norms; instead, they’re excited about the prospect of cheaper money over time. “This is very positive,” Jay Hatfield, CEO of Infrastructure Capital Advisors, told Fortune. “The simple way to say it is that eliminating Fed incompetence is far more important than defending alleged Fed independence. The Fed has always been political; it’s only Trump who talks about it in public.” Indeed, markets largely shrugged off the announcement, to the bewilderment of some economic experts: Robin Wigglesworth of FT Alphaville argued markets are being “preposterously sanguine.” “Entrenched expectations of institutional integrity are now kaput,” Wigglesworth wrote. The S&P 500 and Dow traded around the flat line Tuesday morning, while the Nasdaq even gained 0.3%. Long-term Treasury yields rose after the Trump move, while short-term yields slipped, steepening the curve, indicating that investors are betting rates may fall in the near term but drift higher if a politicized Fed proves less attentive to inflation. The U.S. dollar index was down 0.3%. For Hatfield, that’s the point. He argued the Fed is already too tight, with policy rates sitting “150 basis points above neutral,” meaning monetary policy is restraining growth more than it should. “If Fed funds were at 3% and the president was pushing for big cuts, that would be dangerous,” he said. “But we’re nowhere near that.” Trump already nominated Stephen Miran to the Fed board after one governor decided to step down earlier this month, so replacing Cook would give him a fourth voice alongside earlier appointees Michelle Bowman and Christopher Waller. That would tilt the seven-member Board of Governors toward a Republican-leaning majority. The prospect of a Republican-leaning Fed board is one reason Hatfield thinks cuts are coming. “A tariff is like a one-time tax—it shows up in CPI once and then disappears,” he said. Waller and Bowman, who both dissented in last month’s decision to hold rates steady, understand that tariffs aren’t a persistent inflation driver, Hatfield added. On the other hand, Democratic-appointed governors misunderstand that, causing them to “delay cuts and put the economy at risk.” “So getting rid of the Powell Fed is very positive for the stock market and the bond market,” he said. In the short term, Hatfield expects at least two cuts this year, echoing Fed Chair Jerome Powell’s recent dovish signals. “Overnight you saw a knee-jerk selloff in bonds, but if we’re going to get cuts, that’s great for bonds and great for stocks,” he said. Independence at risk, with only fragile guardrails from the market Other economic experts are far less optimistic. David Wessel of Brookings warned Trump “seems determined to control the Fed—and will use any lever he has to get a majority,” calling it another step in undermining democratic foundations. Analysts at investment bank Piper Sandler were equally direct, arguing investors are deluding themselves if they think markets will discipline Trump. “What is the basis for believing the so-called bond vigilantes will scold Congress before a crisis is at hand?” they wrote, pointing out markets didn’t foresee the inflation shock of 2022 or the housing bust before the Global Financial Crisis. Instead, they argued, stocks are simply rallying at the prospect of rate cuts, “even if they may come in part due to political pressure.” The bigger risk the analysts are pointing to is structural. As analyst Jim Bianco explained on X, the Fed’s seven governors must reapprove—or veto—all 12 regional Federal Reserve Bank presidents when their five-year terms expire in February 2026. With more Trump appointees on the board, even leaders like Austan Goolsbee in Chicago or John Williams in New York could find their jobs at risk, reshaping the balance of the FOMC. Piper Sandler warned the “pillars of the long bull market are being removed one by one,” with freer trade reversed and “the sound money pillar in the process of being fundamentally compromised.” The market, they concluded, is unlikely to serve as a check on the politicization of the Fed. For now, the markets are focused on near-term easing. UBS Global Wealth Management strategist Ulrike Hoffmann-Burchardi told clients that her team still expects the Fed to deliver 100 basis points of cuts over the next four meetings. “We will continue to monitor rising political pressure on the Fed,” she wrote in a note, “but expect its decision-making to remain guided by its mandate in the near term.” Hatfield, for his part, is unconcerned about what he calls “alleged independence.” “Inflation is already contained, the labor market is weakening, and we’re headed into recession,” he said. “The real story isn’t Trump versus the Fed—it’s that the Fed has been incompetent for decades, and markets know it. Any step toward fixing that is positive.” 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
埃隆•马斯克本周一针对苹果和OpenAI提起反垄断诉讼，指控iPhone制造商苹果与ChatGPT开发商OpenAI合谋阻碍人工智能领域的公平竞争。 这份向得克萨斯州联邦法院提交的61页诉状，兑现了马斯克两周前发出的威胁。当时，他指控苹果在iPhone应用商店的人工智能应用排名中，不公正地偏袒OpenAI及其ChatGPT。 马斯克此前发帖暗示，苹果操纵系统不利于ChatGPT的竞争对手的发展，包括其旗下xAI开发的聊天机器人Grok。如今，在这份由xAI及其另一家公司X Corp联合提交的诉状中，他详细罗列了大量指控，意在通过诉讼获得经济赔偿，并寻求法院禁令制止这些被指控的非法行为。 这场“双管齐下”的法律攻势将近期接连发生的多起事件串联起来，试图将苹果与OpenAI一年前的合作重新解读为一场遏制竞争的密谋——而这一切正发生在一场或将堪比2007年iPhone发布的技术变革之中。 诉状宣称：“这是两个垄断势力的一次联手，目的是确保它们继续主导这个由人类有史以来最强大技术——人工智能——迅速驱动的世界。” 根据诉状的描述，苹果认为人工智能将对其未来成功构成“生存威胁”，因而促使其与OpenAI合谋，试图保护其长期以来最赚钱的核心产品线iPhone。 部分指控还称，苹果试图阻止那些功能齐全的“超级应用”挑战iPhone的地位，这类应用包括马斯克一直以来计划通过X打造的应用。这些指控在一定程度上呼应了美国司法部去年针对苹果提起的反垄断诉讼。 诉状将OpenAI描绘成对人类构成威胁的存在，指控其在2022年底推出ChatGPT后凭借惊人增长势头扩张业务时，一心逐利，置公众安全于不顾。这一指控与马斯克去年提起的另一宗联邦诉讼如出一辙，当时他指控OpenAI背弃了最初“以非营利研究机构身份运作、服务公共利益”的初衷。 OpenAI已提起反诉，指控马斯克“骚扰”，并在周一对这起反垄断诉讼的回应中再次援引了这一指控。OpenAI在声明中表示：“这起最新诉讼与马斯克持续不断的骚扰行为如出一辙。” 苹果方面未立即回应置评请求。 这起诉讼的核心争议在于苹果的一项决定：当设备内置技术无法满足用户需求时，将ChatGPT作为iPhone上的人工智能“问答引擎”。双方于去年宣布的这项合作，原本是苹果以“后来者”身份加入人工智能竞赛的重要举措，计划主要依靠其自研设备端侧技术驱动，但苹果至今仍未能完全兑现当初的承诺。 诉状指出，苹果自身人工智能技术的短板可能正在推动更多iPhone用户使用ChatGPT，而鉴于这是一项独家合作，OpenAI就能因此获得Grok等其他潜在竞争对手无法获得的宝贵数据。 诉状指控，该合作也促使苹果不当提升了ChatGPT在iPhone应用商店人工智能应用中的排名。自苹果宣布与ChatGPT合作以来，DeepSeek和Perplexity等其他人工智能应用曾多次在全球部分地区的App Store AI应用榜单上登顶。 诉状未提及ChatGPT可能对苹果及iPhone未来市场地位构成的潜在威胁。作为扩张计划的一部分，OpenAI已邀请前苹果设计师乔尼·艾夫（Jony Ive）主导开发一款人工智能设备。许多分析人士认为，这款设备未来可能会对iPhone构成挑战。（财富中文网） 译者：刘进龙 审校：汪皓 Elon Musk on Monday targeted Apple and OpenAI in an antitrust lawsuit alleging that the iPhone maker and the ChatGPT maker are teaming up to thwart competition in artificial intelligence. The 61-page complaint filed in Texas federal court follows through on a threat that Musk made two weeks ago when he accused Apple of unfairly favoring OpenAI and ChatGPT in the iPhone’s app store rankings for top AI apps. Musk’s post insinuated that Apple had rigged the system against ChatGPT competitors such as the Grok chatbot made by his own xAI. Now, he is detailing a litany of grievances in the lawsuit — filed by xAI and another of his corporate entities, X Corp. — in an attempt to win monetary damages and a court order prohibiting the alleged illegal tactics. The double-barreled legal attack weaves together several recently unfolding narratives to recast a year-old partnership between Apple and OpenAI as a veiled conspiracy to stifle competition during a technological shift that could prove as revolutionary as the 2007 release of the iPhone. “This is a tale of two monopolists joining forces to ensure their continued dominance in a world rapidly driven by the most powerful technology humanity has ever created: artificial intelligence,” the lawsuit asserts. The complaint portrays Apple as a company that views AI as an “existential threat” to its future success, prompting it to collude with OpenAI in an attempt to protect the iPhone franchise that has long been its biggest moneymaker. Some of the allegations accusing Apple of trying to shield the iPhone from do-everything “super apps,” such as the one Musk has long been trying to create with X, echo an antitrust lawsuit filed against Apple last year by the U.S. Department of Justice. The complaint casts OpenAI as a threat to humanity bent on putting profits before public safety as it tries to build on its phenomenal growth since the late 2022 release of ChatGPT. The depiction mirrors one already being drawn in another federal lawsuit that Musk filed last year, alleging OpenAI had betrayed its founding mission to serve as a nonprofit research lab for the public good. OpenAI has countered with a lawsuit against Musk accusing him of harassment — an allegation that the company cited in its response to Monday’s antitrust lawsuit. “This latest filing is consistent with Mr. Musk’s ongoing pattern of harassment,” OpenAI said in a statement. Apple didn’t immediately respond to a request for comment. The crux of the lawsuit revolves around Apple’s decision to use ChatGPT as an AI-powered “answer engine” on the iPhone when the built-in technology on its device couldn’t satisfy user needs. The partnership announced last year was part of Apple’s late entry into the AI race that was supposed to be powered mostly by its own on-device technology, but the company still hasn’t been able to deliver on all its promises. Apple’s own AI shortcomings may be helping drive more usage of ChatGPT on the iPhone, providing OpenAI with invaluable data that’s unavailable to Grok and other would-be competitors because it’s currently an exclusive partnership. The alliance has provided Apple with an incentive to improperly elevate ChatGPT in the AI rankings of the iPhone’s app store, the lawsuit alleges. Other AI apps from DeekSeek and Perplexity have periodically reached the top spot in the Apple app store’s AI rankings in at least some parts of the world since Apple announced its deal with ChatGPT. The lawsuit doesn’t mention the potential threat that ChatGPT could also pose to Apple and the iPhone’s future popularity. As part of its expansion efforts, OpenAI recruited former Apple designer Jony Ive to oversee a project aimed at building an AI-powered device that many analysts believe could eventually mount a challenge to the iPhone. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
• 当美国劳动者疲于应对工作与生活平衡、不断攀升的育儿成本以及陷入停滞的白领就业市场时，新西兰等国家的劳动者却享有截然不同的待遇：超过30天的带薪休假、长达6个月的全薪产假以及不断上调的最低工资标准。随着越来越多劳动者陷入职业倦怠，他们或许会发现，告别美国企业文化或许能为他们带来更优质的生活。 如果你早已厌倦了在女儿第一次舞蹈表演前还要不停刷新Outlook邮箱，或者被上司拒绝带薪休假申请，那么，移居新西兰或许正是摆脱朝九晚五倦怠的全新答案。 最新发布的《全球生活与工作平衡指数》显示，新西兰凭借最低工资标准的小幅提升，总分较2024年进一步提高，连续第三年蝉联全球生活与工作平衡最佳国家。 全球人力资源平台Remote对全球60个最大经济体进行了研究，评估哪些国家能够让劳动者“工作得好、生活得更好”。该指数综合考量了多个维度，包括带薪休假、病假补贴、产假政策、最低工资、医疗保障、幸福感、工作时长、LGBTQ+（性少数群体）包容性以及整体安全性等，各国以百分制进行评分。 新西兰以86.87分的综合表现领跑全球。当地劳动者享有32天带薪休假、6个月全薪产假，以及位居全球最高水平的最低工资标准（每小时16.42美元）。相比之下，美国联邦最低工资标准自2009年以来从未上调，目前仍仅为平均每小时7.25美元。 研究报告指出：“正如研究数据显示，新西兰以及许多欧洲国家正通过员工优先政策助力劳动者实现生活与工作平衡；但美国等国家依旧推崇长工时和有限带薪休假的职场文化，正面临职业倦怠蔓延的风险。” 紧随新西兰之后的是爱尔兰和比利时。爱尔兰凭借相对较高的最低工资标准和慷慨的产假政策获得81.17分。比利时以75.91分位居第三，其病假补贴和产假薪资制度表现突出。比利时也是欧洲幸福感最高的国家之一，且平均每周工作时间仅（34.1小时）。 相比之下，美国不仅无缘前二十名，更在60个国家中排名倒数第二，原因在于缺乏带薪育儿假以及依赖私人医疗体系。 研究报告指出：“数据显示，随着公共安全水平和LGBTQ+包容性指标下降，美国现已成为全球生活与工作平衡第二差的国家，从2023年的第53位、去年的第55位进一步下滑至本年度的第59位。” 在评定最佳与最差国家时，这项研究摒弃了传统的“工作与生活平衡”的表述，而是颠倒顺序，采用“生活与工作平衡”的表述，旨在强调生活应当优先于工作的基本理念。 全球“生活与工作平衡”最佳的前五国及其得分（百分制）如下： • 新西兰，惠灵顿（86.87分） • 爱尔兰，都柏林（81.17分） • 比利时，布鲁塞尔（75.91分） • 德国，柏林（74.65分） • 挪威，奥斯陆（74.20分） 全球“生活与工作平衡”最差的五国及其得分（百分制）如下： • 尼日利亚，阿布贾（26.67分） • 美国，华盛顿特区（31.17分） • 埃及，开罗（35.77分） • 孟加拉国，达卡（36.91分） • 埃塞俄比亚，亚的斯亚贝巴（37.61分） 美国人渴望更高的工作与生活平衡度 Z世代尤其因重视工作与生活平衡而备受关注。最新一项调查显示，在挑选全职工作时，工作与生活平衡已超越薪资成为他们首要考量的因素。 尽管近年来应届毕业生因追求更多工作之外的自由时间而备受关注，但千禧一代同样不想错过人生的美好时光。根据福特（Ford）的一项调查，超半数千禧一代愿意为优先保障生活质量而降薪20%。77%的受访者表示，他们更重视平衡的个人生活，而不是职场成就或晋升。 美国人深知自己深陷于“内卷文化”的循环中。一项更大范围的海外移民调查显示，70%的美国人认为，美国依然是全球最能赚钱的国家，但同时有68%的人觉得自己“只是生存而非生活”。对于那些考虑移居海外的人而言，美国人最向往的国家是加拿大和英国等英语国家，其次是澳大利亚、法国和意大利；新西兰在其理想目的地中位列第十。（财富中文网） 译者：刘进龙 审校：汪皓 • As American workers struggle to adapt to work-life balance, increased child-care costs, and a frozen white-collar job market, other countries like New Zealand are providing over 30 days of paid leave, 6 months of maternity leave, and increasing minimum wage. As more workers become burnt out, they may find that ditching corporate America could give them a better chance at the quality of life they’re looking for. If you’re tired of refreshing Outlook before your daughter’s first dance recital or your boss denying PTO, moving to New Zealand may be the new answer to combating your 9-to-5 burnout. A new measure of the Global Life-Work Balance Index found New Zealand was crowned the best country for life-work balance for the third year in a row, improving from its score in 2024, thanks in part to a slight jump in minimum wage. Global HR platform Remote studied the 60 countries with the largest economies in the world to measure which ones allowed workers to “live and work well.” The index measured factors ranging from paid leave, sick pay, maternity policies, minimum wage, healthcare, happiness, working hours, LGBTQ+ inclusivity, and overall safety. Each country was ranked out of 100. New Zealand is performing well across each category, with a score of 86.87. Workers there could enjoy 32 days of paid leave, six months of fully paid maternity leave, and one of the world’s highest minimum wages at $16.42 an hour. Compared to the U.S., the federal minimum wage averages $7.25 per hour and hasn’t been changed since 2009. “As our study shows, countries such as New Zealand and many in Europe are helping their employees navigate the balance between life and work with employee-first policies, but nations like the United States risk walking into a burnout epidemic with a culture that continues to prioritize long hours and limited paid leave,” the study said. Following behind New Zealand were Ireland and Belgium. Ireland recorded an 81.17 rating, helped by its relatively high minimum wage and generous maternity leave policy. Belgium scored 75.91, bolstered by its sick pay and maternity-payment rate. Belgium also had one of the highest happiness rates in Europe, along with shorter work weeks (34.1 hours on average). Meanwhile, not only does the U.S. not rank in the top 20, it’s the second to last on the list of 60, due to its lack of paid parental leave and private healthcare system. “With public safety and LGBTQ+ inclusivity decreasing, the United States now has the second-worst life-work balance, according to the data. The U.S. falls to 59th out of 60, having placed 55th last year and 53rd in 2023,” the study said. When ranking the countries with the best and worst ratings, the study rejected the traditional term “work-life balance,” flipping the order to “life-work balance” to emphasize the priority should be living first and foremost. The top five countries with the best ‘life-work’ balance are as follows (scores below are out of 100): • New Zealand, Wellington (86.87) • Ireland, Dublin (81.17) • Belgium Brussels (75.91) • Germany, Berlin (74.65) • Norway, Oslo (74.20) And here are the five countries with the weakest ‘life-work’ balance: • Nigeria, Abuja (26.67) • United States, Washington, DC (31.17) • Egypt, Cairo (35.77) • Bangladesh, Dhaka (36.91) • Ethiopia, Addis Ababa (37.61) Americans yearn for more work-life balance Gen Zers have specifically gained a reputation for valuing work-life balance. A recent survey found work-life balance ranked as the top priority when considering full-time jobs, surpassing salary. Despite recent graduates being in the spotlight for wanting more free time outside of work, millennials want a chance to not miss out on their lives, too. More than half of millennials would be willing to take a 20% pay cut for a lifestyle that prioritizes their quality of life, according to a survey from Ford. 77% of respondents also said they prioritize a balanced personal life over achievements or growth at work. Americans are aware they’re stuck on the hamster wheel of hustle culture in the States. A broader expat survey found 70% of Americans believe the U.S. is the most lucrative country for work, but 68% feel that they are surviving more than thriving. For those who have considered moving abroad, Americans preferred English-speaking countries such as Canada and the UK as the most desirable. Following that were Australia, France, and Italy. New Zealand ranked number 10 on their list. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
上周日，一艘马士基（Maersk）集装箱船停靠在新泽西州纽瓦克港。图片来源：Michael Nagle—Bloomberg/Getty Images • 上周，标准普尔全球（S&P Global）维持美国AA+信用评级及稳定展望，部分原因在于美国“强劲的关税收入”——该收入有助于抵消联邦预算中减税与开支带来的影响。尽管标准普尔全球认为美国财政赤字不会出现显著改善，但也预计其不会急剧恶化。不过，美国的对等关税正面临法律层面的挑战，存在被裁定无效的可能性。 评级机构标准普尔全球针对美国赤字前景给出了“喜忧参半”的评价：好消息是赤字不会进一步急剧恶化，坏消息则是也不会有显著改善。 赤字预测的关键因素是特朗普的关税政策——该政策有助于抵消联邦预算中减税和支出带来的影响。 上周，标准普尔全球维持美国债务AA+评级，其依据在于美国整体经济实力强劲、有效的制衡制度、积极主动的货币政策，以及美元作为全球主要储备货币的地位。 美国当前信用评级较最高级AAA低一个等级，且展望维持稳定，原因是赤字状况不会对整体信用水平造成干扰。 标准普尔全球在一份声明中表示：“我们认为，当前国内外政策的调整不会削弱美国经济的韧性与多样性；反过来，包括强劲关税收入在内的整体财政收入增长，将抵消减税与支出增加引发的财政失衡问题。” 特朗普的《大而美法案》（One Big Beautiful Bill Act）预计会在未来十年间致使赤字增加数万亿美元——该法案新增了多项减税措施，在削减部分政府项目开支的同时，又对另一些项目加大了投入。与此同时，美国国会预算办公室（CBO）预计关税收入将使赤字减少数万亿美元。 标准普尔全球实际观察到美国赤字状况已有所改善：预计2025年至2028年期间，赤字占国内生产总值的比例将降至6%，低于2024年的7.5%，也低于2020年至2023年期间9.8%的平均水平。但这一趋势仍无法阻挡美国总债务突破二战时期创下的历史高点。 此外，标准普尔全球预测美国国内生产总值增速将逐步加快：2025年增速为1.7%，2026年为1.6%，2027年和2028年将回升至2%的平均水平。 标准普尔全球补充道：“《大而美法案》的综合实施与执行情况、关税收入的进一步增长，及其对经济增长与投资产生的影响，将决定美国财政轨迹是改善还是恶化。” 由此可见，关税至关重要。鉴于美国政府不愿通过提高所得税来增加财政收入，分析师指出，每年约3000亿至4000亿美元的关税收入对美国而言数额极为可观，难以割舍，这意味着关税政策可能会长期维持。 然而，美国所谓的“对等关税”正面临法律层面的挑战，其依据的《国际紧急状态经济权力法案》(International Emergency Economic Powers Act，IEEPA)的正当性遭到质疑。 美国联邦上诉法院预计将在9月底前就此事作出裁决，最快或于8月底公布结果。此前，美国司法部官员在一封信中对一旦关税被裁定无效可能引发的灾难性后果发出警告。华尔街部分人士认为，此举暗示美国政府担忧在此次诉讼中败诉。 官员们在信中写道：“在此情形下，民众将被迫流离失所，数以百万计的工作岗位将消失，辛勤工作的美国人将失去积蓄，甚至社会保障与联邦医疗保险（Medicare）都可能受到威胁。简而言之，这将带来毁灭性的经济后果，而非前所未有的成功。” 考虑到关税收入对美国信用评级意义重大，若对等关税最终被裁定无效，将会出现何种局面？美国信用评级是否会被下调？目前标准普尔全球尚未回应置评请求。 与此同时，并非所有机构都像标准普尔全球和美国国会预算办公室那样对关税持乐观态度。惠誉评级（Fitch）上周也维持了美国AA+信用评级，但认为即便有关税收入这一“意外之财”，美国赤字状况仍会恶化。 惠誉评级表示，受经济韧性、股市稳健表现及关税收入推动，美国联邦财政收入将实现增长，因此2025年赤字占国内生产总值的比例有望从2024年的7.7%降至6.9%。然而，随着明年新的减税政策正式生效，整体财政收入将出现下滑，赤字状况实际上会比2024年更为严峻。惠誉评级预测，2026年美国赤字占国内生产总值的比例将飙升至7.8%，2027年进一步升至7.9%。 该评级机构在声明中指出：“尽管惠誉评级预计未来两年关税收入年均将达3000亿美元且呈持续增长态势，但受《大而美法案》中多项政策影响——包括新增小费与加班费免税、提高州和地方税（SALT）扣除上限、65岁以上人群额外扣除——美国政府财政收入仍将出现下降。”（财富中文网） 译者：中慧言-王芳 • S&P Global reaffirmed its AA+ credit rating and stable outlook last week owing in part to “robust tariff income,” which should help offset the impact of tax cuts and spending in the federal budget. While S&P doesn’t see meaningful improvement in the fiscal deficit, it doesn’t expect steep deterioration either. However, reciprocal tariffs face legal challenges and could be struck down. Ratings agency S&P Global had some good news and bad news on the U.S. deficit outlook. The good news is that it won’t get much worse. The bad news is that it won’t get much better, either. A key factor for the deficit forecast is President Donald Trump’s tariffs, which should help offset the impact of tax cuts and spending in the federal budget. S&P reaffirmed its AA+ rating on U.S. debt last week, citing the overall strength of the economy, institutions that provide effective checks and balances, proactive monetary policy, and the dollar’s status as the world’s top reserve currency. The outlook on the credit rating, which is a notch below the top AAA grade, remains stable because the deficit won’t muddy the picture. “This incorporates our view that changes underway in domestic and international policies won’t weigh on the resilience and diversity of the U.S. economy,” S&P said in a statement. “And in turn, broad revenue buoyancy, including robust tariff income, will offset any fiscal slippage from tax cuts and spending increases.” Trump’s One Big Beautiful Bill Act is expected to add trillions of dollars to the deficit over the next decade as new tax cuts were added while spending saw cuts to some programs and hikes to others. At the same time, the Congressional Budget Office sees tariffs shaving trillions of dollars off the deficit. S&P actually sees some improvement in the deficit, which is expected to shrink to 6% of GDP from 2025 to 2028, down from 7.5% in 2024 and an average of 9.8% from 2020 to 2023. But that will not stop the total debt from soaring past record highs last seen during World War II. Meanwhile, S&P sees GDP growth accelerating to an average pace of 2% in 2027 and 2028, from 1.7% in 2025 and 1.6% in 2026. “The combined implementation and execution of the One Big Beautiful Bill Act, higher tariff revenue gains, and their effect on growth and investment will inform whether the fiscal trajectory improves or worsens,” S&P added. So a lot is riding on tariffs. And given Washington’s reluctance to raise revenue via income tax hikes, analysts have pointed out an estimated $300 billion to $400 billion a year in tariff revenue would be too much to turn away, meaning levies are likely here to stay. But so-called reciprocal tariffs are facing legal challenges that dispute their legal justification under the International Emergency Economic Powers Act (IEEPA). A decision from a federal appeals court is expected by the end of September, but could come as soon as late August. And a letter from Justice Department officials with doomsday warnings about what would happen if tariffs are struck down suggested to some on Wall Street that the administration fears a court loss. “In such a scenario, people would be forced from their homes, millions of jobs would be eliminated, hardworking Americans would lose their savings, and even Social Security and Medicare could be threatened,” the officials wrote. “In short, the economic consequences would be ruinous, instead of unprecedented success.” Considering how important tariff revenue is to the U.S. credit rating, what would happen if the reciprocal duties are struck down? Would the U.S. be downgraded? S&P didn’t respond to a request for comment. Meanwhile, not everyone is as sanguine about tariffs as S&P and the CBO are. Fitch ratings also reaffirmed its AA+ U.S. credit rating last week—but sees deficits worsening despite the tariff revenue windfall. The deficit should shrink this year to 6.9% of GDP from 7.7% in 2024, as the resilient economy, solid stock market, and tariff revenues send federal receipts higher. But when new tax cuts take hold next year, the situation will actually become worse than in 2024, as overall revenue drops. Fitch sees deficits spiking to 7.8% of GDP in 2026 and 7.9% in 2027. “Government revenues will fall, driven by additional tax exemptions on tips and overtime, expanded deductions for state and local taxes (SALT), and additional deductions for people over 65 included in the OBBBA, despite the continued increases in tariff revenues, which Fitch expects to average USD300 billion in both years,” the ratings agency said in a statement. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
2024 年，股价与市值持续暴涨的寒武纪，获得了“寒王”这一称号。2024年底其市值突破3000亿元、2025年1月股价摸高777.77元时，市场以为这是“寒王”能触达的极限——直到今年8月21日，其股价罕见涨停且突破1200元关口，次日盘中又涨近14%，总市值近5000亿元首次超过中芯国际而跃居科创板市值第一，市场才再次刷新对“寒王”的认知。 在这一历史性突破过去五天后，寒武纪交出了一份亮眼的半年报，不仅营收规模实现跨越式增长，更首次证明了其盈利能力。这推动其股价盘中跨越1400元关口，最高触及1464元，超过贵州茅台而短暂登顶成为A股市场单价最贵的个股，但收盘回落至1372.10元，贵州茅台则收于1448元，维持其“股王”的地位不变。 作为国产AI芯片第一股，在连续多年因研发投入而巨额亏损后，寒武纪今年上半年营收达28.81亿元，同比增长约4倍，归母净利润录得10.38亿元，自2020年上市以来首次实现半年度盈利。更值得关注的是，其扣非净利润率大幅跃升至约32%，远超2024年全年8%的水平，显示盈利质量出现结构性改善。此外，截至上半年末，其现金及交易性金融资产合计约35亿元，资金面也明显缓解，为后续研发投入和产能扩张提供了支撑。 值得一提的是，寒武纪前身是中科院计算所2008年组建的“探索处理器架构与人工智能的交叉领域”10人学术团队，而作为寒武纪创始人和技术灵魂的中科院博士陈天石，自2016年创立公司之初就几乎倾尽所有投入研发，走上长周期、高风险的硬科技创业之路。2020年寒武纪登陆科创板后曾经历长时间的市场冷遇，但他始终坚守高端AI芯片的自主研发路线。随着国产替代浪潮奔涌，寒武纪股价自2023年以来的最大涨幅超25倍，也凸显出陈天石在国产算力突围进程中的贡献。 在8月26日这份半年报出炉三天前，高盛已将寒武纪的目标价从1223元上调50%至1835元，并称中国信息通信研究院宣布包括中国电信、华为和寒武纪等8家公司通过了DeepSeek适配测试，再次印证了对寒武纪强大研发能力的积极看法。这进一步刺激了市场交易活跃度，而寒武纪的业绩爆发性兑现更是成功将市场AI投资主线从"故事叙事"推向"盈利验证"阶段，彻底激活了国产算力板块的资金配置热情，也为A股这一轮的牛市行情按下加速键。 寒武纪股价今年年内累计涨幅已超120%。显然，推动其股价和业绩爆发的因素是多方面的。美国对华高端AI芯片出口限制的持续收紧，导致英伟达特供中国的H20/B30系列芯片交付受阻，据研究机构Bernstein测算，这留下了高达183亿美元的市场空白。国产芯片厂商若能获取其中30%至40%的份额，就意味着50亿至70亿美元的增量空间。这一替代逻辑正从预期加速走向现实，寒武纪作为国内少数具备云端AI芯片量产能力的厂商，也因此被视为国产替代的核心标的。 此外，8月21日，DeepSeek正式推出V3.1版本，宣布采用自研UE8M0 FP8精度标准，称该精度参数是“针对即将发布的下一代国产芯片设计”。这种低精度格式通过压缩数据位宽，显著提升了计算效率并降低了对显存带宽的依赖，使国产芯片在制程限制下仍能高效运行千亿级模型。平安证券研报称，DeepSeek-V3.1与国产AI芯片的协同进一步加深，将加快推动中国大模型产业链软硬件的协同发展，有利于提高国产AI芯片的市场竞争力。 市场将此视为DeepSeek为国产芯片定制“生态燃料”，直接绑定了寒武纪与国产算力生态的协同关系，也以此作为业绩的重要支撑。从半年报披露的订单和产能布局来看，寒武纪的预付账款达到8.28亿元，存货增至26.9亿元，这些指标指引第三季度单季收入有望突破20亿元，全年收入向80亿元靠拢。有机构推演，若2026年产能按计划落地，公司年收入可能上看350至400亿元区间，并对应约30%的净利润率，意味着利润规模或达到100至120亿元。但这一乐观预期属于市场远期测算，并非公司官方指引，且高度依赖产能假设和订单的实际兑现。 产品层面，寒武纪已实现7纳米MLU-590芯片的量产，其FP16峰值算力达到512TOPS，公司官方表态称其“已追平英伟达A100的训练性能”。而市场高度关注的下一代MLU-690，尽管有传言称其将直接对标H100，但公司已发布公告提示“网传信息不实”，显示出在技术宣传方面的审慎态度。 资金层面，寒武纪新一轮49.8亿元的定增计划已获批准，累计募资金额接近140亿元，将主要用于流片、封测和软件栈的扩充，显示出其为远期产能做准备的努力。 综合来看，市场当前对寒武纪的高估值充分反映了其在国内市场占有率第二的行业地位，而其他主要竞争对手多数仍处于上市辅导阶段，短期内难以对其形成挑战。但在高估值的背后，风险也不容忽视。寒武纪当前的市盈率包含了市场对其2026年甚至更远期利润的百倍假设，任何不及预期的情况都可能引发股价剧烈波动。尤为值得关注的是，其未来业绩高度依赖于供应链的稳定与产能的顺利扩张，中芯国际和华虹半导体等公司为其主要的代工厂商，美国出口管制政策若再度升级，或者流片良率不及预期，都可能影响其技术迭代和产能释放。 无论如何，寒武纪都用一份扎实的半年报首次向市场证明了盈利潜力，叠加外部限制性政策所带来的百亿美元级市场真空，AI芯片国产替代已从概念阶段进入订单与业绩的验证时期，其近期的股价表现正是这种预期和情绪的共同反映。但前方迷雾并未完全散去，对于押注寒武纪及中国AI的投资者而言，这既是一场关于技术突破和国产替代的信仰之跃，也是一次需谨慎权衡风险与回报的理性决策。（财富中文网） 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
2024年3月13日，班达亚齐，穆斯林在斋月期间开斋后购买开斋饭食品饮料。图片来源：GETTY IMAGES 东南亚超过2.8亿人口（约占该地区总人口的40%）信奉伊斯兰教。这催生了对迎合伊斯兰生活方式的商品与服务的需求。这种需求不仅限于清真食品：穆斯林消费者还要求更保守的时装或不使用猪源性成分及酒精的化妆品。 甚至连东南亚金融业也日益清真化。根据伊斯兰私营部门发展公司（Islamic Corporation for the Development of the Private Sector）与伦敦证券交易所集团（London Stock Exchange Group）的研究，2023年，东南亚伊斯兰金融规模达约8,590亿美元，高于2020年的7,540亿美元。 总部位于阿姆斯特丹的云原生软件即服务组合式核心银行平台Mambu，希望开拓这个不断增长的市场。该公司董事总经理兼亚太区销售主管大卫·贝克尔表示："东南亚市场，尤其是马来西亚和印度尼西亚，在伊斯兰银行领域的发展活力非凡。" 该公司已与东南亚客户合作，包括马来西亚最大的伊斯兰教法合规金融产品提供商伊斯兰银行（Bank Islam），以及印尼数字银行Jago银行（Bank Jago）。 图片来源：Courtesy of Mambu 贝克尔指出，伊斯兰金融与传统银行业务增长速度相当，因此Mambu希望提供支持伊斯兰教法合规产品的工具，如利润分享。 与传统银行业不同，伊斯兰金融机构必须避免与涉足有害或被视为“哈拉姆"（禁止）产品的企业往来，如猪肉、酒精或赌博。 伊斯兰银行还不能收取利息，因此必须通过其他机制产生回报，如利润分享或租赁。 贝克尔乐观地认为，东南亚更年轻、更熟悉移动技术的群体将被数字金融解决方案——尤其是体现伊斯兰教义的方案——所吸引。 全球最大穆斯林国家印度尼西亚是伊斯兰金融的明确目标市场。另外一个选择是邻国马来西亚，三分之二人口为穆斯林。新加坡、菲律宾和泰国也存在相当规模的穆斯林群体。 专注于伊斯兰商业研究的新加坡ESSEC商学院（ESSEC Business School）教授塞多米尔·内斯托罗维奇表示，马来西亚是东南亚地区最早采用伊斯兰金融的国家，在该领域的增长方面已“达到顶峰”。相反，印度尼西亚在遵循伊斯兰教义的零售银行和“伊斯兰保险”领域存在更大潜力。 内斯托罗维奇表示：“印尼有广阔的发展空间，因此许多公司希望进入这个市场。” 但他警告称东南亚地区存在特有风险。一方面，与中东更同质化的市场不同，东南亚更具异质性，意味着企业需要针对一系列不同的经济体系、消费者群体和监管制度量身定制产品方案。 Mambu的贝克尔承认东南亚地区存在挑战，包括遵守法规的必要性。但机遇规模大于风险。 他表示：“我们目睹这个市场持续增长，我认为这正是政府和监管机构如此支持该行业的一个因素。”（财富中文网） 译者：刘进龙 审校：汪皓
纽约证券交易所交易大厅的一名交易员，摄于2019年。图片来源：Johannes Eisele—AFP via Getty Images • 受上周五强劲反弹影响（美联储主席杰罗姆·鲍威尔（Jerome Powell）当日释放出下月可能降息的信号），美国股市本周开局料将延续上行走势。然而，市场对人工智能热潮的疑虑正不断加剧，周三英伟达（Nvidia）发布季度财报时，华尔街对人工智能前景的信心或将迎来考验。 上周日晚间，美股期货小幅上扬。华尔街迎来另一关键周——人工智能芯片领军企业英伟达将发布财报，同时市场将迎来最新通胀数据。 上周五，美联储主席杰罗姆·鲍威尔释放出下月可能降息的信号，推动股市大幅反弹，当前市场正延续这一势头。 道琼斯工业平均指数期货攀升24点，涨幅达0.05%。标普500指数期货上涨0.05%，纳斯达克指数期货上涨0.06%。周五，道琼斯指数创下历史新高，而标普500指数和纳斯达克指数也逼近各自纪录高点。 10年期美国国债收益率稳定在4.256%，周五受市场降息预期影响，该收益率大幅下跌。美元兑欧元汇率下跌0.02%，兑日元汇率则维持不变。 黄金价格下跌0.13%，至每盎司3413.80美元。美国原油价格上涨0.2%，至每桶63.79美元；布伦特原油价格上涨0.15%，至每桶67.83美元。 上周五股市反弹是在科技巨头领跌的大规模抛售之后出现的——市场对人工智能热潮及其对企业实际助力的质疑正与日俱增。 此前，麻省理工学院（MIT）的一份最新报告显示，95%的企业人工智能试点项目未能带来显著回报。 加剧这些担忧的还有OpenAI首席执行官萨姆·奥尔特曼（Sam Altman）的言论——他将当前的人工智能热潮与20世纪90年代的互联网泡沫相提并论。 周三收盘后，英伟达将发布季度财报，届时华尔街对“人工智能作为投资主题能否持久”的信心将面临考验。 在该报告公布之前，英伟达与超微半导体（AMD）已达成一项前所未有的协议：两家公司均同意将对华芯片销售收入的15%上缴美国联邦政府。 目前，美国企业对人工智能的需求依然强劲，仅Alphabet、微软（Microsoft）、亚马逊（Amazon）和Meta Platforms等所谓超大规模科技公司今年的资本支出就高达4000亿美元，其中大部分资金都将投向人工智能领域。 周五，美联储青睐的通胀指标如期发布。此前政策制定者一直在密切关注唐纳德·特朗普总统推出的关税政策对通胀的实际影响。 此前发布的消费者价格指数（CPI）与生产者价格指数（PPI）数据喜忧参半。分析师预计，7月个人消费支出价格指数（PCE）环比上涨0.2%、同比上涨2.6%，同比涨幅与6月持平。 但核心个人消费支出价格指数预计环比上涨0.3%、同比上涨2.9%，同比涨幅较6月的2.8%有所加快。 不过，包括鲍威尔在内的部分美联储官员表示，关税对通胀的影响可能是短期的，市场应更多关注劳动力市场——目前劳动力市场已显现出疲软迹象。（财富中文网） 译者：中慧言-王芳
深度求索（DeepSeek）推出的新人工智能模型针对国产芯片进行了优化，且定价低于OpenAI。图片来源：Photo illustration by Cheng Xin—Getty Images 中国人工智能初创公司深度求索（DeepSeek）在今年1月凭借一款名为R1的人工智能模型震惊世界，该模型可与OpenAI及Anthropic的顶级大语言模型（LLM）相抗衡。其研发成本仅为其他同类模型的一小部分，使用的英伟达（Nvidia）芯片数量远少于竞品，且以免费形式发布。如今，在OpenAI最新模型GPT-5发布仅两周后，深度求索再次推出其旗舰V3模型的更新版本——专家称该版本在部分基准测试中的表现可与GPT-5相媲美，且在定价上颇具策略性，低于GPT-5。 深度求索的新模型V3.1是在微信某用户群及Hugging Face平台上悄然发布的。此次发布同时触及当前人工智能领域的多个核心议题：深度求索是中国在不依赖外国技术的前提下，推进先进人工智能系统研发、部署与管控这一整体战略的关键一环。（事实上，深度求索此次推出的新版V3模型专门针对国产芯片进行了优化，以实现卓越性能。） 尽管美国企业对深度求索的模型仍持观望态度，但这些模型已在中国广泛应用，并在全球其他地区逐渐普及，甚至部分美国企业已基于深度求索的R1推理模型开展应用程序开发工作。 中国在人工智能领域的布局远不止深度求索一家：国内还涌现出阿里巴巴的通义千问（Qwen）、月之暗面（Moonshot AI）的Kimi、百度的文心一言（Ernie）等模型。不过，深度求索选择在OpenAI的GPT-5推出后不久发布新版本——后者的推出未能满足行业观察人士的较高预期——凸显出中国科技界力求跟上甚至超越美国顶级实验室的决心。 OpenAI对中国与深度求索感到担忧 深度求索的举措无疑让美国实验室倍感压力。在近期与记者的晚宴上，OpenAI首席执行官萨姆·奥尔特曼（Sam Altman）表示，来自深度求索等中国开源模型的竞争日益激烈，这一现实状况影响了OpenAI两周前发布自有开源权重模型的决策。 “显而易见，倘若我们不采取相应行动，未来全球技术生态或将主要依托中国开源模型构建，”奥尔特曼表示，“这无疑是我们决策时考虑的因素之一，虽非唯一决定要素，但其影响却举足轻重。” 此外，上周美国政府发放许可证，批准英伟达和超微半导体（AMD）向中国出口专用人工智能芯片（包括英伟达的H20芯片），但前提是两家公司同意将相关销售收入的15%上缴美国政府。在美国商务部部长霍华德·卢特尼克（Howard Lutnick）7月15日接受美国消费者新闻与商业频道（CNBC）采访时称“我们不会向中国出售最先进的芯片，也不会出售技术水平次之或处于第三梯队的产品”后，中国政府随即采取反制措施，着手限制英伟达芯片的采购。 通过针对国产芯片进行模型优化，深度求索既展现出应对美国出口管制的韧性，也表明其减少对英伟达依赖的决心。该公司在微信公众号文章中指出，新模型格式已针对“即将发布的下一代国产芯片”进行优化。 在同一场晚宴上，奥尔特曼警告称，美国可能低估了中国在人工智能领域取得的进展，并表示单靠出口管制或许并非可靠的解决方案。 虽未达成质的飞跃，却仍是具有突破性的渐进式进展 从技术层面看，深度求索新模型的亮点在于其构建方式，其中部分技术突破对普通用户而言并不直观。但对开发者而言，这些创新使得V3.1相较于众多封闭且定价高昂的竞品模型更具成本优势与通用性。 例如，V3.1规模庞大，参数数量达6850亿，与众多顶尖“前沿”模型处于同一量级。但其采用的“混合专家”架构意味着，在响应任何查询时，仅需激活模型的一小部分，从而为开发者降低计算成本。此外，与早期深度求索模型——将“可基于预训练数据即时回答的任务”与“需逐步推理的任务”分开处理——不同的是，V3.1在单一系统中同时实现了快速应答功能与推理功能。 GPT-5、Anthropic及谷歌的最新模型也具备类似能力，但目前能做到这一点的开源权重模型仍屈指可数。科技分析师、TechTalks博客创始人本·迪克森（Ben Dickson）向《财富》杂志表示，V3.1的混合架构“是目前为止最大的亮点”。 其他人指出，尽管这款新模型不像今年1月震惊世界的R1模型（由初代V3模型精炼而成的推理模型）那样具有突破性，但全新的V3.1版本仍然令人瞩目。人工智能开发者平台Lightning AI的创始人兼首席执行官威廉·法尔肯（William Falcon）称：“它们能持续实现具有实质意义的改进，这确实令人印象深刻。”不过他也补充道，倘若OpenAI的开源模型“开始出现明显落后”，预计该公司会做出回应，并指出，深度求索的模型对开发者而言在投入生产应用时难度更大，而OpenAI的版本部署起来则相对更为便捷。 尽管技术细节繁杂，但深度求索此次新品发布凸显了一个事实——人工智能正日益被视为中美之间暗流涌动的技术竞赛的一部分。考虑到这一点，倘若中国企业能以其声称的一小部分成本研发出更为卓越的人工智能模型，那么美国竞争对手确实有理由担忧自身能否保持领先地位。 （财富中文网） 译者：中慧言-王芳
6月2日，OpenAI首席执行官萨姆·奥尔特曼出席在旧金山举行的2025年Snowflake峰会。图片来源：Justin Sullivan—Getty Images 就在业内高管和分析师纷纷质疑人工智能是否正在催生又一个泡沫之际，信贷投资者正将数十亿美元资金投入这个新技术领域。 据知情人士本周透露，摩根大通（JPMorgan Chase & Co.）和三菱日联金融集团（Mitsubishi UFJ Financial Group）正牵头承销一笔超过220亿美元的贷款，用于支持Vantage Data Centers建设一个庞大的数据中心园区。据彭博社本月报道，Facebook母公司Meta Platforms Inc.将从太平洋投资管理公司（Pacific Investment Management Co.）和Blue Owl Capital Inc.获得290亿美元资金，用于在路易斯安那州乡村地区建设一个大型数据中心。 未来还将涌现更多此类交易。仅OpenAI一家公司就预计，其未来开发和运营人工智能服务所需的基础设施投入将达数万亿美元。 与此同时，业内关键人物也承认，人工智能投资者很可能将面临阵痛。OpenAI首席执行官萨姆·奥尔特曼本周表示，他看到当前人工智能投资热潮与上世纪90年代末的互联网泡沫存在相似之处。在谈及初创企业估值时，他直言：“总会有人在这里遭受损失。”此外，麻省理工学院（Massachusetts Institute of Technology）的一项研究计划发布报告指出，在企业界，95%的生成式人工智能项目未能带来任何利润。 这些情况足以让信贷市场的观察者们感到不安。 花旗集团（Citigroup）美国投资级信贷策略主管丹尼尔·索里德表示：“信贷投资者自然会回想起2000年代初的情景，当时电信公司可谓过度建设、过度举债，最终我们看到这些资产出现了重大减值。因此，从中期来看，人工智能热潮无疑会引发关于可持续性的疑虑。” 在早期，用于训练和驱动最先进人工智能模型的基础设施建设资金，主要由人工智能公司自身承担，其中包括Alphabet Inc.旗下谷歌（Google）和Meta Platforms Inc.等科技巨头。不过，近期的资金来源正越来越多地转向债券投资者和私人信贷机构。 据彭博情报（Bloomberg Intelligence）近期分析，这类融资风险敞口的形式和规模多样，风险等级也各不相同。许多大型科技公司——即所谓的AI超大规模服务商——一直通过发行优质企业债来为新建基础设施融资。由于这些债务有现有现金流作为担保，因此被认为相对安全。 如今，大部分债务融资正来自私人信贷市场。 瑞银集团（UBS）信贷策略主管马修·米什表示：“过去三个季度，人工智能领域的私人信贷融资规模每季度约为500亿美元，这是保守估计。即便不计入Meta和Vantage的巨额交易，私人信贷市场的资金供给也已是公共市场的两到三倍。” 与此同时，许多新的计算中心正通过商业地产抵押贷款支持证券（CMBS）融资，这类证券并非与企业主体挂钩，而是与园区产生的付款绑定。据摩根大通本月估算，人工智能基础设施支持的CMBS已较2024年全年总额增长30%，达到156亿美元。 索里德与花旗的一位同事在8月8日发布了一份报告，重点分析了公用事业公司面临的特殊风险。这些公司为建设满足高能耗数据中心所需的电力基础设施而大幅举债。索里德及其同事与其他分析师一样，都对当前如此巨额的投入感到担忧，因为人工智能项目尚未证明其具备长期创造营收的能力。 标普全球评级（S&P Global Ratings）私人市场分析全球主管露丝·杨表示：“数据中心项目的融资周期往往长达20至30年，而我们甚至无法确定五年后这项技术会是什么样子。我们会保守评估未来现金流，因为缺乏历史参考。” 瑞银集团指出，压力已经开始显现，表现之一是面向科技领域的私人信贷机构的实物支付（PIK）贷款正在增加。根据瑞银的数据，第二季度，商业发展公司（BDCs）的PIK收入占比升至6%，创下自2020年以来的最高水平。 但这股资金洪流短期内似乎难以停歇。 穆迪全球项目与基础设施融资团队高级副总裁约翰·梅迪纳表示：“直接贷款机构不断在筹集资本，而这些资金必须找到去处。他们把这些资本需求巨大的AI超大规模服务商视作下一个长期基础设施投资标的。”（财富中文网） 译者：刘进龙 审校：汪皓
• 对于那些有幸在当今白领就业市场开启职业生涯的Z世代而言，不要指望能获得加薪。自疫情爆发以来，面对面服务行业需求激增，推动酒店业、医疗保健行业薪资上涨，且涨幅超过通胀水平。与此同时，白领科技类岗位薪资却陷入停滞，人工智能是造成这一现象的原因之一。 Z世代毕业生在抛帽庆祝毕业之后，正面临愈发严峻的现实：他们的技能已被ChatGPT超越，而且加薪不够稳定，无法随意挥霍，只能买杯燕麦拿铁。 而如今又添新打击：他们那些未获得学位、从事调酒师或咖啡师工作的朋友，薪资涨幅竟超过了他们。休闲与酒店业的薪资增速已超越白领岗位，彻底打破了年轻从业者“唯有白领职业能实现收入增长”的固有认知。 Bankrate的最新分析显示，自2021年起，酒店业从业者薪资涨幅已接近30%，较通胀水平高出4%以上；医疗保健行业从业者薪资涨幅也超过了通胀水平，过去四年涨幅约为25%。 然而，专业商务服务、金融及教育领域从业者的薪资增速却未能赶上通货膨胀的速度。以教师为例，其薪资增速比通胀水平低近5%。 不过，即便如此，Z世代也不太可能纷纷涌向当地酒吧或星巴克工作。 初级科技岗位等白领工作的薪资水平仍相对较高——在美国，此类岗位的平均时薪为19.57美元；而酒店业普通咖啡师平均时薪约为16美元。不过，自数年前通胀首次飙升以来，白领群体的薪资增速始终跟不上物价上涨的步伐。零售、贸易、医疗保健、休闲、酒店及食品服务行业，即便从业者时薪较低，但从长远来看，其薪资涨幅却更为突出。 白领就业市场陷入停滞 在白领就业市场，从业者——尤其是像Z世代这样的应届毕业生——正面临另一重严峻现实：金融活动、专业商务服务等白领领域工作人员正面临招聘速度放缓的困境。 尽管初级员工向往科技公司办公室的光鲜亮丽与随时供应的冷萃咖啡，但就在上周，Meta暂停了新人工智能部门的招聘——此前该公司大肆招揽人才、不惜砸下重金聘请大批人工智能研究人员与工程师，甚至开出1亿美元的签约奖金，如今这轮烧钱式招聘已正式落幕。亚马逊（Amazon）首席执行官安迪·贾西（Andy Jassy）也表示，除追求“效率提升”外，人工智能的应用可能意味着白领岗位将面临裁员。 并非只有办公室职员面临挑战：教育行业的薪资与通胀差距最大，建筑行业紧随其后。 即便Z世代足够幸运，获得了梦寐以求的科技行业工作，晋升之路也可能受阻。近期一项调查显示，在“大辞职潮”期间飙升的晋升率已开始回落：2025年5月整体晋升率为10.3%，低于2022年5月14.6%的峰值。（财富中文网） 译者：中慧言-王芳 • For the Gen Zers fortunate enough to start in today’s white-collar job market, don’t anticipate any raises. Since the pandemic, demand for in-person services has pushed up wages in hospitality and health care, outpacing inflation. Meanwhile, white-collar tech jobs are in a freeze, with AI being one of the culprits. Gen Z graduates are facing an increasingly tough reality after tossing their caps into the air: Not only are their skills being outpaced by ChatGPT, but they aren’t getting raises consistent enough to splurge on anything more than an oat-milk latte. But there’s now a new nail in coffin: Their non-degree friends working as bartenders and baristas are seeing bigger pay raises than they are. Wage growth in leisure and hospitality is outpacing white-collar jobs, flipping the script on where young workers can find earning momentum. A new analysis by Bankrate found hospitality workers’ wages have risen by nearly 30% since 2021, outpacing inflation by more than 4%. Health care workers have similarly outpaced inflation and seen their salaries go up by around 25% in the past four years. However, those working in professional and business services, the finance industry, and education have not seen wage gains that keep up with inflation. Teachers, for example, are pacing at nearly 5% below inflation. Yet, Gen Z isn’t likely to flock to work at the local pub or Starbucks. White-collar jobs such as entry-level tech gigs still come with larger paychecks—averaging at $19.57 an hour in the U.S. But in the hospitality industry, an average barista makes about $16 dollars per hour. Still, since inflation first spiked a few years ago, wages have still been falling behind for white-collar workers. Workers in retail, trade, health care, leisure, hospitality, and food services making less per hour, are watching their paychecks grow more over time. The white-collar freeze Across the white-collar job market, workers—especially fresh-faced graduates like Gen Z—are being hit with another tough reality: Workers in white-collar financial activities or professional and business services are encountering a slower pace of hiring. While entry-level workers crave the glam of tech offices and cold brew on tap, just this week, Meta paused hiring for its new artificial intelligence division, ending a spending spree that saw it acquire a wave of costly AI researchers and engineers, and included signing bonuses of $100 million. Amazon CEO Andy Jassy also has said in addition to “efficiency gains,” he expects AI could mean white-collar job cuts. And it’s not just desk workers who are being challenged. Education saw the biggest wage gap relative to inflation, followed by construction. And even if Gen Zers are lucky enough to land that tech job of their dreams, their promotions may not follow. A recent survey found that promotion rates have slowed after surging during the Great Resignation. The overall promotion rate was 10.3% in May 2025, down from a peak of 14.6% in May 2022. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
机器人（和智能体）时代即将来临。图片来源：Getty Images 摩根士丹利（Morgan Stanley）最新深度分析显示，随着AI技术的普及，美国企业界将迎来一场根本性变革，每年可节省近万亿美元成本。该投行估计90%的岗位将在某种程度上受到AI自动化或增强化的影响，成本节省将直接源自人员精简、自然减员以及知识密集型常规任务的自动化处理。 这家华尔街投行估计，大规模部署“智能体AI”软件与具身AI人形机器人技术，可为标普500指数企业创造9,200亿美元年化净效益。分析师指出，节省的绝大部分成本将来自降低薪酬支出和减少重复性或流程密集型岗位的人力需求。 预期节省金额相当于该指数2026年税前收益的28%。分析师认为，这一惊人的效率提升将在各行各业产生连锁反应。但有更多情况值得警惕，因为摩根士丹利主题投资团队警告称，成本节省“可能需要多年才能实现”，且部分企业存在“重大风险”，可能无法达到全面应用水平。他们补充道，9,200亿美元相当于标普500指数企业总薪酬支出的41%，且他们目前仅掌握了约90%标普500指数企业的充足数据。 按照分析师的说法，所谓“经济价值创造”将来自两方面：一是成本削减（通过部署AI减少各类任务的人力投入及执行成本），二是新增收入和利润，因为员工被解放出来之后，可专注于更高附加值的活动，从而创造新的收入并提升利润率。他们发现，不同行业和职业在这两方面的影响比重存在显著差异。根据估值乘数测算，9,200亿美元年经济效益可能会为标普500指数带来约13-16万亿美元的市值增长，相当于当前总市值的近四分之一。 受影响最大的行业 各行各业受影响程度不尽相同。如下表所示，必需消费品分销与零售、房地产管理、交通运输等行业受冲击最大，其AI驱动的生产效率收益可能超过2026年预期收益的100%。医疗设备与服务、汽车及专业服务领域也将面临重大变革与机遇。 相比之下，半导体和硬件等原本人均产值较高的行业，其AI价值创造潜力相对较低。 岗位危机与新兴职业 尽管成本节省将主要来自人力削减，摩根士丹利强调全面自动化与任务级增强存在本质区别。涵盖生成式AI和软件应用的智能体AI更倾向于任务重新分配而非直接取代人类工作岗位，而人形机器人形式的具身AI则在物流实体零售等领域更有可能直接取代人类工作岗位。 报告同时预测，在AI取代人类工作岗位的趋势中，将涌现出从首席AI官到AI治理专家等全新职业类别，这与早期技术革命催生程序员、IT专业人士和数字营销人员需求的历史规律相呼应。 漫长的渐进过程 尽管数字惊人，但分析师警告称，全面应用AI可能需要数年甚至数十年时间。企业将优先依靠自然减员和流程效率优化，而非立即大规模裁员，尤其在面向客户的营收驱动型领域。 但对投资者而言讯号明确：AI已不再是概念主题。其成本节省潜力如此巨大，有望成为2025年后推动美国企业盈利增长的最强引擎之一。（财富中文网） 译者：刘进龙 审校：汪皓 Corporate America is on the brink of a radical transformation as artificial intelligence adoption could unlock nearly $1 trillion a year in savings, according to a sweeping new analysis by Morgan Stanley. The bank calculates 90% of jobs will be touched in some way by AI automation or augmentation, with cost savings flowing directly from reduced headcount, natural attrition, and automation of knowledge-intensive but routine tasks. The Wall Street bank estimates widescale deployment of so-called agentic AI software and embodied AI humanoid robotics could generate $920 billion in net annual benefits for companies in the S&P 500. The lion’s share of those savings, analysts say, will come from lowering payroll expenses and reducing the need for human workers in repetitive or process-heavy roles. The projected savings equate to roughly 28% of the index’s 2026 pretax earnings—a staggering efficiency gain analysts believe will reverberate across industries. There are more caveats, as Morgan Stanley’s Thematic Investing team cautions these cost savings would “likely take many years to achieve,” and they see “significant risk” of some companies not achieving full adoption levels. The $920 billion figure represents 41% of the total S&P 500 compensation expense, they add, and they only have sufficient data to run analyses for approximately 90% of the S&P 500. The “economic value creation,” as they put it, will come in a combination of cost cutting (e.g., lower headcount and lower costs to perform a wide variety of tasks by deploying AI) and new revenue and margin generation, as employees are freed up to spend more time on higher value-added activities that could both increase revenue and enhance margins. They see a wide variety of the balance between these two impacts, based on industry and occupation. The $920 billion in annual economic benefit could translate into a $13-$16 trillion boost in market value for the S&P 500, according to the report, depending on valuation multiples. That figure would amount to nearly a quarter of today’s entire market capitalization. Sectors most exposed Not all industries will feel the effects equally. As you can see from the chart below, Consumer staples distribution and retail, real estate management, and transportation are among the most exposed sectors, with potential AI-driven productivity benefits exceeding 100% of forecast 2026 earnings. Healthcare equipment and services, autos, and professional services also face major disruption and opportunity. By contrast, industries that already run lean on labor relative to earnings—such as semiconductors and hardware—show comparatively lower AI value potential. Jobs at risk, new roles ahead Though the topline savings will come from payroll reductions, Morgan Stanley stressed the distinction between full automation and task-level augmentation. Agentic AI, which spans generative AI and software applications, tends to reassign tasks rather than eliminate jobs outright, while embodied AI in the form of humanoid robots poses more direct substitution risks in industries such as logistics and physical retail. The report also anticipates entirely new categories of jobs—from chief AI officers to AI governance specialists—emerging alongside the displacement trend, echoing earlier waves of technological disruption that boosted demand for programmers, IT professionals, and digital marketers. A long ramp-up Despite the headline number, the analysts caution full adoption is likely to unfold over years, if not decades. Firms will lean first on attrition and process efficiencies rather than immediate mass layoffs, particularly in sectors where customer-facing roles drive revenue. Still, the message for investors is clear: AI is no longer a speculative theme. The cost savings potential is so large it could become one of the most powerful drivers of U.S. corporate earnings growth in the second half of this decade. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
图片来源：Chris Polk—FilmMagic/Getty Images • 已故亿万富翁史蒂夫·乔布斯以苹果（Apple）联合创始人兼CEO的身份闻名于世，他推出了iPhone、iPad和iMac等划时代产品。然而他在苹果公司的经历并非他财富积累的真正原因。实际上，1995年，在iMac上市之前三年，凭借汤姆·汉克斯与蒂姆·艾伦的助力，乔布斯将一次意外的职业瓶颈转化为财富契机，真正实现了亿万身家。 “飞向宇宙，浩瀚无垠！”不仅是《玩具总动员》（Toy Story）中巴斯光年的经典台词，更是乔布斯成为亿万富翁的命运转折点。 1985年，一场权力斗争迫使乔布斯离开苹果公司。次年，乔布斯以1,000万美元收购了卢卡斯影业（Lucasfilm）的图形计算部门。卖家正是刚打造完《星球大战》（Star Wars）帝国的乔治·卢卡斯。这家小公司很快更名为皮克斯（Pixar），即将永远改变好莱坞与乔布斯的财富轨迹。 公司初期经营坎坷，乔布斯因需持续垫付月度资金缺口曾多次考虑是否应该将公司出售。但到1995年，他相信皮克斯已准备就绪。在11月的同一周内，公司不仅推出了首部长片《玩具总动员》，而且启动了首次公开募股。 时任CFO劳伦斯·利维写道，这让他想起奥运会百米赛跑：漫长的训练积累浓缩于瞬间爆发。 他在自己的书《孵化皮克斯：从艺术乌托邦到创意帝国的非凡之旅》中表示：“若全世界爱上《玩具总动员》，皮克斯将开启动画娱乐新纪元。” “如若不然，皮克斯可能只会被视作又一个折戟沉沙的追梦公司。” 让乔布斯成为亿万富翁的IPO 作为持有皮克斯80%股份的控股人，乔布斯在IPO中压下了更高的筹码。若IPO成功，他将收获投资回报；若IPO失败，这不仅可能断送与迪士尼（Disney）未来合作的可能性，而且可能会导致十年的创业心血付诸东流。 所幸结果远超预期：皮克斯的股票发行价预计为12-14美元，首日收盘暴涨175%，达到了每股39美元。这主要归功于由汤姆·汉克斯和蒂姆·艾伦配音的《玩具总动员》票房表现比预期近乎翻倍。乔布斯的持股使他的资产净值随之飙升至逾10亿美元。 1997年重返苹果后，他仍持续参与皮克斯运营。随着《海底总动员》（Finding Nemo）、《超人总动员》（The Incredibles）、《美食总动员》（Ratatouille）等全球票房均破数亿美元的作品接连问世，迪士尼于2006年以74亿美元收购皮克斯的全部股票，乔布斯所持股份价值约46亿美元。 总之，乔布斯凭借直觉投资皮克斯的经历，印证了一条亘古不变的成功箴言：成功的关键在于找到热情所在并全力以赴。 苹果CEO蒂姆·库克在2015年表示：“不管未来你们做什么，这个世界都需要你们的激情、热血、不断向上的进取心。历史很少因个人而改变，但是，不要停止思考、忘记正在发生的事情。” 主业之外的财富传奇 除了乔布斯外，企业领袖通过主业外渠道积累巨额财富者不乏其人。埃隆·马斯克亦有类似经历。 这位世界首富虽因特斯拉（Tesla）与SpaceX的领导者身份而闻名，但他并非在这两家公司积攒下第一笔财富。他以超3亿美元的价格将首家公司Zip2出售给AltaVista，后来他的互联网金融公司X.com与亿万富翁彼得·蒂尔联合创立的Confinity合并成立PayPal，又让他赚得巨额财富。 同样，亿万富翁理查德·布兰森的财富也并非来自他的维珍航空（Virgin Atlantic）和维珍银河（Virgin Galactic）等航空航天业务。这位75岁的英国连续创业者成为亿万富翁，在一定程度上得益于他的唱片连锁店维珍唱片（Virgin Records）。他在1971年创立了该公司，后发展成一家音乐唱片公司，签下了滚石乐队（Rolling Stones）、珍妮特·杰克逊等巨星。1992年，布兰森以10亿美元的价格，将维珍唱片出售给英国索恩-百代集团（Thorn EMI）。（财富中文网） 译者：刘进龙 审校：汪皓 • The late billionaire Steve Jobs is known for being cofounder and CEO of Apple—and introducing the iPhone, iPad, and iMac to the world. However, his time at the computer company wasn’t what helped strike gold for his net worth. Jobs actually made the billions in 1995—three years before the iMac hit shelves—after using an unexpected career roadblock to his advantage, with a little help from Tom Hanks and Tim Allen. “To infinity and beyond!” wasn’t just the catchphrase of Toy Story’s Buzz Lightyear—it was the turning point that turned Steve Jobs into a billionaire. After a power struggle that forced Jobs out of Apple in 1985, Jobs bought Lucasfilm’s computer graphics division the next year for $10 million. The seller was George Lucas, fresh off creating the Star Wars empire. That small acquisition would soon be renamed Pixar—and would change both Hollywood and Jobs’ fortune forever. The company got off to a rocky start, with Jobs questioning whether to sell it multiple times, thanks in part to having to personally cover its monthly cash shortfall. But by 1995, Jobs believed Pixar was ready for primetime. In a week’s span in November, it would release its first major film, Toy Story, as well as launch an IPO. Lawrence Levy, the company’s then-CFO, wrote that it reminded him of the 100-meter sprint in the Olympic Games: a lifetime of training that comes down to a snapshot performance. “If the world fell in love with Toy Story, Pixar would have a chance to usher in a new era of animated entertainment,” he said in his book, To Pixar and Beyond: My Unlikely Journey With Steve Jobs to Make Entertainment History. “If it didn’t, Pixar might be written off as another company that tried but never quite hit the mark.” The IPO that made Jobs a billionaire As the 80% owner of Pixar, the IPO stakes were even higher for Jobs. If everything went well, he was hoping to finally see some return on his Pixar investment. If everything went south, it might have shut the door on any future collaboration with Disney and led to the waste of a decade of his entrepreneurial life. Luckily, all expectations were shattered. Pixar’s initial stock price was predicted to reach between $12 and $14, but at the end of the first day of trading, it was worth 175% more, at $39 a share. This was thanks largely to Toy Story, with Tom Hanks and Tim Allen as lead voices, nearly doubling its box office expectations. Jobs’ stake sent his net worth soaring to over $1 billion. Jobs would later rejoin Apple in 1997, but he remained involved in Pixar as it churned out hit after hit, including Finding Nemo, The Incredibles, and Ratatouille—each bringing in hundreds of millions of dollars worldwide. Disney fully acquired Pixar for about $7.4 billion in stock in 2006. Jobs’ stake was worth about $4.6 billion. Overall, Jobs’ willingness to follow his instincts with Pixar proves the age-old advice that one key to success is finding your passion—and putting all of your energy into it. “No matter what you do next, the world needs your energy, your passion, your impatience with progress,” Apple CEO Tim Cook said in 2015. “History rarely yields to one person, but think and never forget what happens when it does.” Finding fortune beyond their main companies Jobs isn’t alone in being a business leader who gained significant wealth outside of what they’re primarily known for. Elon Musk has a similar story. While the world’s richest person is known today for being the leader of Tesla and SpaceX, that’s not how he first amassed his fortune. Musk sold his first company, Zip2, to AltaVista for more than $300 million. He also made millions through the creation of PayPal, which formed from a merger of Musk’s online financial services company, X.com, with software company Confinity, cofounded by billionaire Peter Thiel. Similarly, billionaire Richard Branson did not make all his money from being focused on his air and space companies, Virgin Atlantic and Virgin Galactic. The 75-year-old British serial entrepreneur actually became a billionaire in part thanks to his chain of record stores called Virgin Records. It launched in 1971 and later expanded into a music label that featured artists like the Rolling Stones and Janet Jackson. Branson later sold Virgin Records in 1992 to British conglomerate Thorn EMI for $1 billion. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
OpenAI首席执行官萨姆·奥尔特曼的财富并非源于他微薄的薪酬。这位科技巨头的19亿美元身家主要来自对优步、爱彼迎、Reddit等企业的早期投资。图片来源：Bloomberg / Contributor / Getty Images • OpenAI即将成为全球最有价值的非上市公司。预计在按计划出售60亿美元股份后，该公司的估值将达到5,000亿美元。但其联合创始人兼首席执行官萨姆·奥尔特曼并不会因这项成就而跻身亿万富豪榜前列。他目前在该AI公司没有持股，年薪仅76,001美元。其19亿美元财富主要来自对Reddit、优步（Uber）、Asana及爱彼迎（Airbnb）等行业巨头的早期投资。 AI竞赛是近十年来最激烈的商战领域，预计到2033年市场规模将达到4.8万亿美元。据首席执行官奥尔特曼透露，OpenAI在这场激烈的竞争中始终保持领先地位，ChatGPT已积累高达8亿活跃用户。 如今，该公司即将成为全球最有价值的非上市公司。公司正在就出售60亿美元股份进行谈判。这笔交易将使公司的估值将从3月份微软（Microsoft）、软银（SoftBank）等投资者注资400亿美元后达到的3,000亿美元，跃升至5,000亿美元。此举将使OpenAI超越埃隆·马斯克旗下的SpaceX。目前，SpaceX以3,500亿美元估值位居非上市公司榜首。 然而即便OpenAI实现如此令人艳羡的估值，奥尔特曼也不会因此迅速提升在亿万富豪榜的排名。这位执掌先锋科技企业的首席执行官当前年薪为76,001美元，较2022年的73,546美元略有增长。据《纽约时报》报道，这位亿万富豪级科技领袖称其薪资“仅够达到医疗保险的最低标准”，更令人惊讶的是，他在OpenAI没有任何股权。 他曾通过红杉资本（Sequoia）投资的与Y Combinator有关的风投基金，持有少量间接股份，但OpenAI发言人向TechCrunch证实，该股份占比不足百分之一，并且已经出售，但金额未公开。 奥尔特曼的零股权现象极为罕见——大多数首席执行官都会持有公司股份，他们的财富与企业业绩直接挂钩。但他却通过其他方式积累财富。 《财富》杂志就此事联系OpenAI寻求置评。 年薪76,001美元的亿万富豪：奥尔特曼的财富密码 虽不持有OpenAI股份，但奥尔特曼绝非依赖微薄薪水度日。这位40岁的科技企业家当前身家19亿美元——虽远逊于竞争对手马斯克的4,100亿美元，但他仍通过跨行业多元化投资积累了可观财富。 奥尔特曼是核聚变初创企业Helion Energy的主要投资人，个人注资3.75亿美元并且目前担任董事长；他向生物技术公司Retro Biosciences投资1.8亿美元；还参与过马斯克的脑机接口公司Neuralink的多轮融资。 除科技领域外，他还是生产力管理平台Asana和Reddit的早期投资人，并在2022年之前一直担任Reddit的董事会成员。Reddit上市后，其持股估值达6亿美元。 2008年，在短期租赁行业初期，OpenAI 首席执行官幸运地在爱彼迎注入10万美元启动资金。此外，他还在优步早期投资10万美元，如今该公司的市值已高达1,940亿美元。 奥尔特曼还投资过创作者变现平台Patreon以及其他AI公司，包括估值350亿美元的行业巨头Humane。 在2011年至2019年执掌Y Combinator的近十年间，奥尔特曼积累了独到的眼光，能够精准判断哪些项目能获得投资、哪些项目会失败以及哪些项目真正具有规模扩张的潜力。这位千禧一代的亿万富豪多元化的投资组合横跨变革性技术、核能、金融科技和社交平台等领域。据《华尔街日报》估算，截至2024年初，奥尔特曼掌控的资产至少价值28亿美元。这位OpenAI的首席执行官透露，截至去年，其风投基金已投资超400家企业。（财富中文网） 译者：刘进龙 审校：汪皓 • OpenAI is on the brink of becoming the world’s most valuable private company, anticipated to reach a $500 billion valuation after a $6 billion planned shares sale. But its cofounder and CEO, Sam Altman, won’t be shooting up the billionaire list for the major accomplishment—he currently holds zero equity in the AI company, earning an annual salary of $76,001. Instead, the bulk of his $1.9 billion fortune comes from his early investments in industry titans, including Reddit, Uber, Asana, and Airbnb. The AI race is one of the hottest business wars this decade, with the market expected to be worth $4.8 trillion by 2033. And OpenAI has been a front-runner in the fierce battle, with ChatGPT amassing a staggering 800 million active users, according to CEO Sam Altman. It’s now on the cusp of becoming the world’s most valuable company, with talks to sell $6 billion in shares that would push its valuation to $500 billion—up from its $300 billion appraisal in March after a $40 billion infusion from backers like Microsoft and SoftBank. That leap would see OpenAI overtake Elon Musk’s SpaceX, which currently tops all other private companies at $350 billion. However, even if OpenAI pulls off the envy-worthy valuation, you probably won’t catch Altman shooting up the billionaires list off the back of it. The CEO currently earns a salary of $76,001, up slightly from $73,546 in 2022, for leading the pioneering tech company. The billionaire tech boss said he makes “whatever the minimum for health insurance is,” according to the New York Times—but more surprisingly, owns zero equity in OpenAI. At one point he owned a “quite insignificant” indirect stake in the company through a Sequoia-backed VC fund associated with Y Combinator. But an OpenAI spokesperson told TechCrunch that the share was reportedly less than a fraction of a percent, and has already been sold for an undisclosed amount. Altman’s lack of equity is quite unusual; most CEOs have some skin in the game, standing to make big wins or crushing losses depending on how they lead their companies. But he has other tricks up his sleeve to bring home the bacon. Fortune reached out to OpenAI for comment. Altman is a billionaire with a $76,001 salary. Here’s how he makes his fortune Just because Altman doesn’t own a part of OpenAI, doesn’t mean he’s living paycheck-to-paycheck on his modest CEO salary. The 40-year-old tech entrepreneur currently boasts a net worth of $1.9 billion—starkly lower than the CEO of the company he’s trying to outpace, as Musk sits on a $410 billion fortune. But he’s still amassed a sizable nest egg thanks to his diverse investments across several industries. Altman is a major backer of Helion Energy, a nuclear fusion startup where he poured in a $375 million personal investment and currently serves as chairman. He’s supported biotech company Retro Biosciences with a $180 million investment, and also participated in funding rounds for Musk’s brain-computer interface maker Neuralink. Aside from these science- and tech-focused firms, Altman was also an early backer of productivity management platform Asana and Reddit, serving as a board member of the latter until 2022. His stake in Reddit was estimated to be worth $600 million after the platform’s IPO. The OpenAI CEO also made a lucky early infusion of $100,000 in Airbnb back in 2008 when the company was in its short-term-rental infancy. Plus, there was his early $100,000 investment in Uber—which today is worth $194 billion. Altman also financially supported creator monetization platform Patreon, even putting his money toward other AI-focused companies including $35 billion titan Humane. After nearly a decade of steering Y Combinator from 2011 to 2019, Altman got firsthand insight into what gets funded, what crashes out, and what can actually scale. The millennial billionaire’s diverse portfolio of investments spans transformational technology, nuclear energy, fintech, and social platforms. All the holdings he controlled up to early 2024 were estimated to be worth at least $2.8 billion, according to reporting from the Wall Street Journal. The OpenAI CEO said his venture funds had invested in more than 400 companies, as of last year. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
Meta首席执行官马克·扎克伯格。图片来源：DAVID PAUL MORRIS—Bloomberg/Getty Images Meta正全力推进所谓的“超级智能”竞赛：随着新成立的Meta超级智能实验室（Meta Superintelligence Labs，简称MSL）逐步成型，公司再度对人工智能部门进行重组。但分析师指出，投资者始终紧盯Meta长期承诺的目标——通过优化产品提升用户参与度，进而推动广告销售额增长。即便超级智能人工智能模型最终落地，也不过是实现这一目标的手段而已。 此次重组由前Scale AI首席执行官汪滔（Alexandr Wang）主导，他于今年6月被马克·扎克伯格（Mark Zuckerberg）任命为首席人工智能官。汪滔目前负责管理由数千名工程师、科学家和产品经理组成的庞大团队，据悉他正计划精简团队，此举预计将导致部分高管离职，至少一个团队面临解散。 Meta聘请汪滔以协助公司招募一支规模虽小但薪酬极高的研究团队——据悉部分研究人员获得的薪酬方案总额超1亿美元（通常分数年发放），这支团队目前已成为Meta人工智能研发工作的中流砥柱。但该团队仅是冰山一角：此次重组将整个人工智能部门整合至超级智能实验室，并新设立四个小组，分别专注于研究、训练、产品以及基础设施，所有调整均以“提速”为目标。这四个小组的负责人均向汪滔汇报工作，其中包括知名投资人、前GitHub首席执行官奈特·弗里德曼（Nat Friedman，将负责产品与应用研究），以及前OpenAI研究员赵晟佳（Shengjia Zhao，将以首席科学家身份领导研究团队）。 在近期一封详尽阐述重组计划的员工邮件中，汪滔承认重组可能会带来混乱，但坚称新架构“将让我们在长期内以更快速度实现超级智能”。（针对《财富》杂志确认邮件内容的请求，Meta未作出回应，该邮件内容此前已由《商业内幕》披露。） 投资者对此似乎反应不一：消息公布当天，Meta股价一度下跌超2%，但收盘前已基本收复失地。 研究公司Futurum Group首席执行官丹尼尔·纽曼（Daniel Newman）表示，股价下跌也反映了整体市场的焦虑情绪——近期人工智能及大型科技股从高位回落，市场热度有所降温。他预计市场将出现“温和修正”，但指出Meta“此前业绩表现极为强劲”，且“最近再次交出了亮眼的季度成绩单”。不过，分析师仍在密切关注两大关键要点：扎克伯格为顶尖人工智能研究人员开出的九位数薪酬，以及公司频繁进行的架构重组，同时留意Meta能否在人工智能竞赛中缩小与竞争对手的差距。“诚然，市场存在担忧情绪，”纽曼说道，并指出OpenAI、xAI、谷歌（Google）等公司推出的多款前沿模型持续迭代升级，而Meta的开源模型Llama似乎陷入停滞。 他表示：“我们认为，在扎克伯格大规模招聘后，Meta团队需要一段适应期，才能逐步提速，研发出更具竞争力的解决方案。” 为Meta的产品机器提供动力 然而，Meta对“速度”的追求，本质上是其“产品引擎”的延伸，而非致力于解决人类面临的最大挑战。尽管Meta曾通过旗下基础人工智能研究实验室[由首席科学家杨立昆（Yann LeCun）共同创立]涉足人工智能登月计划，但OpenAI、Anthropic等竞争对手，以及Thinking Machines Lab、Safe Superintelligence等衍生机构，已将“通用人工智能”（AGI，通常定义为具备人类同等智能水平的人工智能）与“超级智能”（远超人类智能水平的人工智能）确立为核心使命。 相比之下，即便在“超级智能”成为热词之前，Meta的核心使命始终未变：优化旗下产品，提升用户在其利润丰厚的社交媒体平台[包括脸书（Facebook）、Instagram、WhatsApp]上的活跃度。这些平台的广告业务几乎贡献了Meta的全部营收——最新季度营收已达466亿美元。 扎克伯格上月通过一条Instagram短视频及一篇博文强调了这一关键要点。他在发布的内容中提到，人工智能技术正飞速发展，我们开始观察到“人工智能系统自我改进的迹象”，并补充称，超级智能如今“已近在咫尺”。但与竞争对手人工智能公司聚焦“科学突破或经济变革”不同，扎克伯格的愿景始终围绕“个体用户”：打造一款个性化人工智能，助力用户“实现目标、创作出内心渴望在世界中看见的内容、成为更贴心的朋友，成长为理想中的自己”。 这种定位与Meta一贯秉持的发展路径高度吻合——面向消费者的体验设计旨在维持用户粘性（进而推动广告销售）。在扎克伯格看来，超级智能还意味着为“融入人工智能的未来个人设备”提供动力，尤其是能“看见我们所见、听见我们所闻，并全天候与我们互动”的增强现实（AR）眼镜。 纽曼表示，他依然看好Meta的前景，原因在于该公司“对业务中‘研究端’的依赖度较低，而是利用人工智能持续提升日活跃用户数量——营收自然也随之不断增长。” 但弗雷斯特研究公司（Forrester）的迈克·普鲁克斯（Mike Proulx）却持有不同见解。他在接受《财富》杂志采访时指出，“Meta正全力以赴打造‘业界最佳、性能最强大的人工智能模型’，这一点毋庸置疑。当前人工智能竞赛已全面展开，而Meta在与竞争对手的较量中处于落后态势。将重心锁定在超级智能领域，能为Meta在战略与运营层面确立清晰目标，进而凝聚力量。” 扎克伯格在Meta最新财报电话会议上也表达了类似观点，强调人工智能是Meta五大重点领域的核心。但普鲁克斯指出，扎克伯格在电话会议中强调，超级智能融入人们日常生活的“主要载体”是人工智能眼镜，而非Meta旗下的系列应用程序。 总体而言，普鲁克斯表示，他并不担忧Meta人工智能部门看似频繁的变动。“这一领域正以惊人的速度发展。如同任何一场新兴技术竞赛，过程中必然会伴随大量调整。这是行业特性使然。”他说道。 尽管关于超级智能的讨论热度持续攀升，但Meta人工智能部门的重组，表明其核心战略并未发生改变：仍是通过打造个性化产品，促使数十亿用户持续滑动屏幕、广告持续投放——不久的将来，人工智能驱动的眼镜有望成为人人佩戴的日常设备。该公司的发展走向备受关注：“如今的问题在于，团队能否有效达成既定目标。”普鲁克斯说道。（财富中文网） 译者：中慧言-王芳 Meta is doubling down on its so-called race to “superintelligence,” reshuffling its AI organization once more as its new Meta Superintelligence Labs (MSL) group takes shape. But analysts say investors are keeping their eye on the prize Meta has always promised: improved products that increase engagement and, in turn, sell more ads. Superintelligent AI models, if they arrive, are just a means to that end. This time it’s former Scale AI CEO Alexandr Wang—brought on by Mark Zuckerberg in June as chief AI officer—leading the reorganization. Wang, who now oversees a sprawling operation of thousands of engineers, scientists, and product managers, is looking to rein it in, reportedly resulting in some expected executive departures and at least one team shutdown. Wang was hired to help recruit a small, high-priced cadre of researchers—some reportedly offered compensation packages exceeding $100 million, typically spread out over several years—now perched at the pinnacle of Meta’s AI effort. But that group is only the tip of the spear: The new restructuring folds the entire AI organization into MSL, with four new groups focused on research, training, products, and infrastructure, all part of a bid for speed. The quartet of group leaders will all report to Wang, including well-known investor and former GitHub CEO Nat Friedman, who will lead product and applied research, and former OpenAI researcher Shengjia Zhao, who will lead the research team as chief scientist. In a recent email to employees, which detailed the restructuring, Wang acknowledged that reorganizations can be disruptive but insisted the new structure would “allow us to reach superintelligence with more velocity over the long term.” (Meta did not respond to Fortune’s request to confirm the contents of the email, which were published by Business Insider.) Investors, meanwhile, seem to have mixed feelings: Meta’s stock slid more than 2% on the news today, but climbed most of the way back by market close. The share-price slide also reflects broader market jitters, as overheated AI and Big Tech names come off recent highs, said Daniel Newman, CEO of research firm the Futurum Group. He said he expects a “modest correction” but noted that Meta has “had an incredible run” and recently “delivered a great quarter once again.” Still, analysts are eyeing Zuckerberg’s nine-figure paydays for top AI researchers and his repeated reorganizing, and watching for signs that Meta will close the gap in the AI race. “Of course there is some concern,” Newman said, pointing out that numerous frontier models from OpenAI, xAI, and Google continue to improve, while Meta’s open-source Llama models have “seemingly stalled.” “We think the team at Meta, after Zuckerberg’s hiring spree, will need a period of acclimation before it finds the velocity to develop more competitive solutions,” he said. Feeding Meta’s product machine That need for speed, however, is best understood as an extension of Meta’s product machine rather than a bid to solve humanity’s greatest challenges. While Meta has dabbled in moonshot AI through its FAIR research lab (cofounded by chief scientist Yann LeCun), rivals like OpenAI and Anthropic and spinoffs such as Thinking Machines Lab and Safe Superintelligence have made the pursuit of artificial general intelligence (AI generally defined to be as smart as humans) and superintelligence (AI far smarter than humans) their central mission. Meta’s mission, by contrast, has remained the same as it was before “superintelligence” became a buzzword: improving the products that power engagement on its massively profitable social-media platforms, including Facebook, Instagram, and WhatsApp. The advertising on those platforms is the source of nearly all of Meta’s revenue, which reached $46.6 billion in the most recent quarter. Zuckerberg underscored this focus last month with an Instagram Reel and blog post in which he said AI is rapidly advancing and that we’re beginning to see “glimpses of AI systems improving themselves.” Superintelligence is now “in sight,” he added—but while rival AI companies talk about scientific or economic breakthroughs, his vision is aimed squarely at the individual: a personalized AI that helps you “achieve your goals, create what you want to see in the world, be a better friend, and grow to become the person that you aspire to be.” That framing neatly aligns with what Meta has always built—consumer-facing experiences designed to keep people engaged (and sell more ads). To Zuckerberg, superintelligence also means powering the future of AI-infused personal devices, specifically augmented-reality glasses that can “see what we see, hear what we hear, and interact with us throughout the day.” Newman said he continues to like Meta’s prospects because the company “isn’t as dependent on the research end of its business, as it is using AI to continue to create higher daily active user numbers—and of course, the coinciding revenue continues to rise as well.” But Forrester’s Mike Proulx countered that there is no doubt Meta is laser-focused on building “the best and most powerful AI models, period,” he told Fortune. “The race is on, and Meta is lagging against competitors. A concerted focus on superintelligence gives Meta a North Star to rally around both strategically and operationally.” Zuckerberg echoed that sentiment on Meta’s most recent earnings call, stressing that AI is at the center of each of Meta’s five focus areas. But Proulx pointed out that it was AI glasses—not the company’s family of apps—that Zuckerberg highlighted on that call as “the main way” superintelligence will enter people’s daily lives. Overall, Proulx said he is not concerned with the seemingly constant upheaval in Meta’s AI organization. “This space is moving at breakneck speed. As with any emerging tech race, there’s inevitably going to be a lot of pivoting. It comes with the territory,” he said. For all the lofty talk of superintelligence, however, Meta’s AI reshuffling shows its bets are mostly still the same: personalized products that keep billions scrolling, ads flowing—and soon, AI-powered glasses perched on every face. How the company fares will be closely watched: “The question now is whether the team is effectively enabled to deliver, or not,” said Proulx. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
迈克尔·费德尔克在塔吉特（Target）总部，摄于6月。图片来源：Elizabeth Flores—The Minnesota Star Tribune/Getty Images 零售业翘首以盼数月之久的消息终于在周三公布：塔吉特首席执行官布莱恩·康奈尔（Brian Cornell）在执掌公司11年后终于卸任，明年2月，公司运营总监迈克尔·费德尔克（Michael Fiddelke）将接任这一职位。 但该行业翘首以盼的另一则消息——塔吉特恢复销售额增长——仍需等待。塔吉特再次公布了疲软的季度财务业绩，可比销售额下滑1.9%，这家主打平价时尚的零售商重申，预计今年销售额将出现低个位数百分比下滑，这将是其销售额连续第三年下滑。 近年来，塔吉特的发展陷入停滞，而其主要竞争对手沃尔玛（Walmart）却一路高歌猛进——要知道，塔吉特曾在疫情期间实现爆发式增长，成功抢占大量市场份额。分析师指出，塔吉特已失去部分“魔力”：就在不久前，它凭借时尚平价的商品、整洁有趣的门店环境，以及周到的客户服务脱颖而出，从竞争对手（尤其是百货公司）手中夺走大量市场份额。 “曾对消费者需求极为敏锐的塔吉特，如今却难以满足美国消费者的需求。”GlobalData董事总经理尼尔·桑德斯（Neil Saunders）在研究报告中写道。塔吉特面临诸多问题：缺货情况再度出现（2014年康奈尔刚上任时，缺货便是困扰塔吉特的难题，后来得以解决）、结账时等待时间过长，以及桑德斯所说的门店环境愈发杂乱。此外，塔吉特在多元化、公平性与包容性议题上摇摆不定，也加剧了其困境。该零售商先是激怒保守派，面临抵制威胁；随后又因缩减相关举措，惹恼了众多消费者与员工。 据美国消费者新闻与商业频道（CNBC）报道，费德尔克于20年前加入塔吉特，先后担任首席财务官、首席运营官等职。周三与记者通话时，他承认了上述问题，并提出三大核心目标：在消费者心中重塑“塔吉特式魔力”、优化客户体验，以及更好地利用技术提升运营效率、削减成本。此外，他还需凝聚员工士气。《华尔街日报》近期报道称，塔吉特在6月开展的全公司调查显示，40%的员工对公司未来缺乏信心。 尽管49岁的费德尔克强调“深入了解塔吉特”的重要性，但其任命消息公布后，该公司股价在早盘交易中下跌10%。花旗（Citi）分析师保罗·莱朱兹（Paul Lejuez）表示，投资者原本“期待引入外部首席执行官”，以带来全新视角。与此同时，GlobalData的桑德斯也指出，塔吉特需要一位能够打破“群体思维”的首席执行官——他认为，这种思维模式已对公司创新形成阻碍。自四年前触及历史高点以来，塔吉特股价已累计下跌64%（同期沃尔玛股价涨幅超一倍）。 2014年上任的康奈尔将继续担任执行董事长。多年来，康奈尔成功推动塔吉特实现现代化转型：通过整合线上电商与线下门店，将塔吉特打造成电商巨头，升级门店设施，领导团队精准洞察消费者需求，推出Cat & Jack等一系列备受追捧的自有品牌。 在康奈尔的带领下，塔吉特还成功应对了疫情引发的动荡：营收从2019年的780亿美元攀升至2022年1091亿美元的峰值，但此后持续下滑（与此同时，沃尔玛、开市客（Costco）、亚马逊（Amazon）等竞争对手仍保持增长态势）。如今，如何让塔吉特重拾昔日“魔力”，已成为费德尔克亟待解决的难题。 （财富中文网） 译者：中慧言-王芳 The news the retail industry has been anticipating for months was finally announced on Wednesday: Target CEO Brian Cornell is finally stepping down after 11 years at the helm and will be replaced by his operations chief Michael Fiddelke in February. But another piece of news the industry is looking for—Target returning to sales growth—will have to wait. Target reported yet another quarter of weak financial results, with comparable sales down 1.9%, and the cheap-chic retailer reaffirmed its expectation that sales will decline by a low single-digit percentage this year, projecting a third year in a row of declines. In recent years, after a torrid run that included Target making massive market share gains during the pandemic, the retailer has stagnated even as archrival Walmart has soared. Analysts say that Target has lost some of its magic touch. That is, the hip but affordable merchandise; fun, well-kept stores; and attentive customer service that helped it stand out and steal a lot of market share from rivals, notably department stores, in the not too distant past. “Target, which used to be very attuned to consumer demand, has lost its grip on delivering for the American shopper,” Neil Saunders, managing director at GlobalData, wrote in a research note. Those problems have included a resurgence of out-of-stocks (a problem that was plaguing Target when Cornell took the reins in 2014 but had been overcome); long wait times at checkout registers; and what Saunders says are increasingly messy stores. Adding to Target’s woes were its flip-flops on the DEI (diversity, equity, and inclusion) issue. The retailer attracted the ire of conservatives who threatened boycotts, but then angered many shoppers and employees when it curtailed some of its efforts. Fiddelke, who started at Target two decades ago and went on to become finance chief and then COO, acknowledged those problems on a call with reporters on Wednesday, according to CNBC. He has three main goals: reestablish the “Tar-zhay” magic with customers, improve customer experience, and make better use of technology to make operations more efficient and less costly. He’ll also have to rally the troops: The Wall Street Journal recently reported that a companywide survey in June found that 40% of Target workers didn’t have confidence in its future. Though Fiddelke, 49, extolled the value of knowing Target deeply, shares fell 10% in morning trading on the news of his appointment. Citi analyst Paul Lejuez said investors were “hoping for an external CEO” with fresh eyes. Meanwhile GlobalData’s Saunders said Target needs a CEO who can cut through the groupthink he says has impeded innovation. Target’s stock is down 64% since its all-time high four years ago. (During that time, Walmart shares have more than doubled.) Cornell, who took the reins in 2014, will stay on as executive chairman. For a number of years, Cornell was wildly successful in modernizing Target by integrating its e-commerce with its physical stores to turn it into a massive online player, upgrading its stores and overseeing teams that always seemed to know what fun items shoppers might want, with launch after launch of wildly popular store brands like Cat & Jack. Under Cornell, Target also successfully navigated the turmoil of the pandemic, with revenue rising from $78 billion in 2019 and peaking at $109.1 billion in 2022 before slipping ever since. (Meanwhile, Walmart, Costco, and Amazon continue to thrive.) It will be up to Fiddelke to figure out how to put the “zhay” back in Target. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。
图片来源：Getty Images • 主流AI经济举步维艰，但“影子AI经济”却在蓬勃发展。这是麻省理工学院（MIT）一项关于生成式AI在工作场所应用情况的新研究得出的关键结论之一。研究发现，超过90%公司的员工正在使用个人聊天机器人处理日常任务，且通常未经IT部门批准；与此同时，仅有40%的公司实际购买了官方大语言模型订阅服务。 麻省理工学院NANDA项目全新发布的重磅报告《2025年企业AI应用现状》（State of AI in Business 2025）显示，企业AI在企业和员工两个层面的应用可谓“冰火两重天”，一方面，企业官方AI的应用陷入停滞，另一方面，由于员工已大规模应用个人AI工具处理日常工作，强劲的“影子AI经济”正在暗中蓬勃发展。 该研究的核心主题是“生成式AI鸿沟”，MIT发现，尽管企业已在生成式AI项目上投入了300亿至400亿美元，但仅有5%的组织获得了转型回报。绝大多数（95%）企业称，其官方AI投资未对损益表产生任何积极影响。然而，MIT也发现，员工私下对大语言模型工具的使用极为活跃，似已形成广泛的AI“影子经济”。 员工们不再等待企业官方的生成式AI项目克服技术与组织障碍，而是主动使用个人ChatGPT账户、Claude订阅及其他消费级AI工具自动化处理各类日常任务。企业IT部门和高管层对此类活动通常并不知晓。 借助个人AI工具，企业员工已经跨越了生成式AI鸿沟。这种“影子AI”的投资回报率往往高于官方项目，同时可能也是真正能够有效跨越这条鸿沟的路径。 40%对90% 研究过程中，研究人员对300多项公开披露的AI计划进行了分析，对52家组织的代表进行了访谈，并对153名高管进行了问卷调查。 结果显示，尽管仅有40%的公司通过官方渠道购买了大语言模型订阅服务，但超过90%公司的员工会在日常工作中使用个人AI工具。事实上，几乎所有受访者都表示在日常工作流程中以某种形式使用过大语言模型。 许多“影子用户”表示，自己每天工作时都会多次用到大语言模型工具，应用进度远超公司批准的AI计划，后者大多目前仍停留在试点阶段。 NANDA项目分析显示，造成这种差距的关键原因包括： • 灵活方便、即时起效：ChatGPT和Copilot等工具因使用方便、适用场景广泛、效果立竿见影而备受好评，许多定制化企业解决方案恰好缺乏这些优点。 • 契合工作流程：员工可以根据自身需求对消费级工具进行定制，绕过了企业审批流程和系统集成方面的障碍。 • 低门槛：影子AI获取方便，用户可以自由迭代、实验，应用速度进一步加快。 正如该报告所指出的那样：“看出这一模式并加以利用的组织才能代表企业级AI应用的未来。” 企业官方在部署生成式AI时，常因系统集成复杂、操作界面僵化、持续记忆功能缺乏导致推进困难，个人在应用此类工具时就不会面临这些问题，优势十分明显。这种差异也有助于我们理解为何生成式AI在试点与实际生产部署之间存在如此巨大的“鸿沟”。 “基础工作争夺战” 报告指出，影子AI工具的应用造成了一种“反馈循环”，企业员工对符合其需求的个人AI工具用得越多，就越不想用那些死板、僵化的企业AI工具。 “（造成差距的）分水岭并非智能水平”，该报告的作者们写道，并解释称，企业AI的问题在于记忆能力、适应能力和学习能力。 因此，90%的用户表示他们更倾向于由人类来处理“关键工作”，而AI则已在“基础工作争夺战”中胜出，有70%的受访者倾向使用AI来起草邮件，65%倾向于用其进行基础分析。 与此同时，该研究还破除了企业AI领域普遍存在的五大迷思。与热门观点相反，研究发现： • 多数岗位未被AI替代； • 除对岗位影响有限外，生成式AI也并未改变商业模式； • 多数企业已在生成式AI试点项目上投入巨资； • 由监管或模型性能引发的问题相对较少，工具缺乏学习或适应能力才是最大障碍； • 内部AI开发（自建）项目的失败率是外部采购（购买）解决方案的两倍。 话虽如此，我们也发现，过去几年，科技行业裁员已成经济领域的“新常态”，虽然这种现象的出现是否与AI应用相关还有待商榷。此外，关于大学学位薪资溢价缩水的研究表明，劳动力市场正发生根本性转变。 但AI行业的发展可能已遇到瓶颈，OpenAI发布的GPT-5反响平平，一些知名作家由此发出疑问，如果AI的潜力仅止于此，我们应何去何从？ 事实上，美联储曾委托多位经济学家对此问题进行研究，基本结论是，AI至少能大幅提升工作效率。但他们也指出，如能充分发挥AI潜力，该技术或将像100年前的电灯一样驱散阴影，推动实现颠覆性变革。 关于本文，《财富》杂志使用了生成式AI辅助完成初稿。编辑在发布前已核实信息的准确性。（财富中文网） 译者：梁宇 审校：夏林 • The mainstream AI economy is struggling, but the “shadow AI economy” is booming. That’s one of the key takeaways from a sweeping new MIT study on generative AI in the workplace. The study finds that workers at more than 90% of companies are using personal chatbot accounts for daily tasks, often without approval from IT, while only 40% of companies actually have official LLM subscriptions. A sweeping new report from MIT’s Project NANDA, State of AI in Business 2025, has uncovered a dramatic split in the landscape of enterprise artificial intelligence: While official AI adoption in companies stalls, a robust “shadow AI economy” is flourishing under the radar, powered by employees using personal AI tools for day-to-day work. The main thrust of the study is the “GenAI divide”: the finding by MIT that despite $30 billion to $40 billion invested in gen-AI initiatives, only 5% of organizations are seeing transformative returns. The vast majority—95%—report zero impact on profit and loss statements from formal AI investments. Lurking under the surface, though, MIT also finds huge engagement with LLM tools on the part of workers, a shadow economy of seemingly widespread AI adoption. Rather than waiting for official enterprise gen-AI projects to overcome technical and organizational hurdles, employees are routinely leveraging personal ChatGPT accounts, Claude subscriptions, and other consumer-grade AI tools to automate tasks. This activity is often invisible to IT departments and C-suites. Employees are already crossing the GenAI Divide through personal AI tools. This ‘shadow AI’ often delivers better ROI than formal initiatives and reveals what actually works for bridging the divide. The 40% and 90% split The study was based on a review of over 300 publicly disclosed AI initiatives, interviews with representatives from 52 organizations, and survey responses from 153 senior leaders. It reveals that while only 40% of companies have purchased official LLM subscriptions, employees in over 90% of companies regularly use personal AI tools for work. In fact, nearly every respondent reported using LLMs in some form as part of their regular workflow. Many shadow users describe interacting with LLMs multiple times a day, every workday—with adoption often far outpacing their companies’ sanctioned AI initiatives, which remain stuck in pilot stages. Project NANDA’s analysis highlights key reasons for this divide: • Flexibility and immediate utility: Tools like ChatGPT and Copilot are praised for their ease of use, adaptability, and instantly visible value—qualities missing from many custom-built enterprise solutions. • Workflow fit: Employees customize consumer tools to their specific needs, bypassing enterprise approval cycles and integration challenges. • Low barriers: Shadow AI’s accessibility accelerates adoption, as users can iterate and experiment freely. As the report notes, “The organizations that recognize this pattern and build on it represent the future of enterprise AI adoption.” These advantages contrast sharply with official gen-AI deployments, where complex integrations, inflexible interfaces, and lack of persistent memory often stall progress. This helps explain a “chasm” in between pilots and production. The ‘war for simple work’ According to the report, shadow AI usage creates a feedback loop: As employees become more familiar with personal AI tools that suit their needs, they become less tolerant of static enterprise tools. “The dividing line isn’t intelligence,” the authors write, explaining that the problems with enterprise AI have to do with memory, adaptability, and learning capability. As a result, 90% of users said they prefer humans to do “mission-critical work,” while AI has “won the war for simple work,” with 70% preferring AI for drafting emails and 65% for basic analysis. Meanwhile, the study engages in some myth-busting, puncturing five commonly held beliefs about enterprise AI. Contrary to the hype, it finds: • Few jobs have been replaced by AI. • Beyond the limited impact on jobs, generative AI also isn’t transforming the way business is done. • Most companies have already invested heavily in gen-AI pilots. • Problems stem less from regulations or model performance, and more from tools that fail to learn or adapt. • Internal AI development “build” projects fail twice as often as externally sourced “buy” solutions. That being said, the tech sector layoffs of the last several years have become entrenched in the economy, whether they are related to AI adoption or not. And research on the declining wage premium of the college degree suggests that a fundamental shift is occurring in the labor market. But the AI sector may be hitting a plateau, with the underwhelming launch of OpenAI’s ChatGPT-5 leading some prominent writers to wonder: What if this is as good as AI gets? In fact, the Federal Reserve commissioned several staff economists to consider the question, and their base case is that it will significantly boost productivity. But they also said it could end up having an import more like an invention that literally banished shadows when it appeared over 100 years ago: the light bulb. For this story, Fortune used generative AI to help with an initial draft. An editor verified the accuracy of the information before publishing. 财富中文网所刊载内容之知识产权为财富媒体知识产权有限公司及/或相关权利人专属所有或持有。未经许可，禁止进行转载、摘编、复制及建立镜像等任何使用。