[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$frRcXhEi9uTX0Hs31q_f5JFKPxdd3z_lN4bE-Mk5-Zb4":3},{"code":4,"msg":5,"data":6},200,"操作成功",{"id":7,"title":8,"content":9,"digest":10,"source":10,"coverPath":11,"thumbsCoverPath":12,"isTop":13,"isShow":14,"baseClick":13,"clickCount":15,"createTime":16,"typeId":17,"isNewest":18,"newsInfoTypeRespVo":19,"voiceUrl":22,"voiceSize":23,"taskId":24,"releaseTime":25,"titleEn":26,"contentEn":27,"voiceUrlEn":28,"taskIdEn":29,"voiceSizeEn":30},1419,"算力难变现 美国AI陷入困境","\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">日前，美国哈佛大学经济学家研究发现，虽然2025年上半年美国GDP增长1.6%，但几乎完全由数据中心和信息处理技术推动，其他领域增长率仅0.1%。2025年全年美国人工智能（AI）数据中心支出规模预计达5200亿美元，人工智能领域拉动GDP增长的方式主要靠投资，而非相关消费。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">今年以来，作为主要的数据中心投资者，微软、谷歌、亚马逊、Meta这4家企业围绕AI算力疯狂竞争，“砸”下640亿至1000亿美元不等的巨额现金，却缺少可观回报，其性质近乎“烧钱”。近期OpenAI也在自身资金有限的情况下，“画饼”将进行5年万亿美元级别的投资，引发了市场对于美国AI科技产业会不会是泡沫甚至是庞氏骗局的巨大疑虑。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">从AI发展情况来看，好消息是个人与企业使用大模型已相当常见，10月初ChatGPT周活数据已达8亿，消费者AI采用速度超出预期。但一个不好的迹象是，如此规模的用户数据，却仍未产生足够规模的收入，甚至不能覆盖运营成本。一般而言，互联网应用在用户数达到一定规模后，营收前景就会比较清晰，投资机构也会积极提供扩张资金、抢占市场份额，进入皆大欢喜的收获期。但如今，美国各类消费大模型看起来像是个无底洞。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">生成式AI的营收困境，与其自身技术原理有关。互联网产业大多具有很明显的规模效应，基础设施足够后的边际成本基本为零，支持几千万客户的成本与几亿客户差不多，利润率极高。而大模型的每一次回应与推理，都要进行巨量重复运算，用户越多、需要的数据中心与算力就越多，不仅享受不到规模效应，反而对基建和融资的需求越来越高。因此美国AI投资规模扩大不仅不是好现象，反而反映出投资效率可能存在的问题。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">另一个不利于生成式AI商业变现的因素，是较难通过广告变现。互联网广告展示成本不高，利润却极为丰厚，但大模型应用一次只能展现一条回答，难以自然地插入多个广告位，这也是“订阅”为何会成为大模型的主要营收手段。但相比一年千亿美元的投资规模，ChatGPT年化“仅”百亿美元规模的订阅收入显得杯水车薪。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">企业云服务是数据中心另一个被寄予厚望的方向。相比主要起数据存储作用的常见企业“上云”操作，美国企业希望AI数据中心更加智能，能帮助企业提高运营效率、缩减成本、增加营收。然而，麻省理工学院8月的一份报告显示，将AI引入业务的企业中95%并没有赚到钱。甲骨文的GPU云租赁业务毛利率仅14%，远低于公司整体业务70%的水平，在考虑折旧后净毛利率更跌至7%。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">企业“AI引智”进展不力、“幻觉”问题根深蒂固、算力成本居高不下，当人们对AI的期待超出玩具、进入实战阶段，就暴露了目前生成式AI的缺陷。这种缺陷导致了AI技术的商业化进程明显滞后于基础设施建设与金融化进程，给未来营收带来了巨大的不确定性。对此，美国业界和资本形成了“工业泡沫论”和“金融泡沫论”两派，前者认为数据中心建设是实体投资，将最终突破规模化的阈值，成为电力一样的基础能源设施；后者则警告当前投资规模已超出理性范畴，过度乐观的预期可能重演当年的互联网泡沫。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">无论是着眼长远的基建布局，还是科技巨头间的资本游戏，美国AI产业发展模式的不均衡问题是毋庸置疑的。AI推动经济发展绝不能只靠数据中心、算力设施建设的狂飙突进，而需要完成从“硬数据”到“软价值”的转变，在广泛商业场景中证明其经济价值。只有当AI真正渗透到制造业、服务业等传统行业，提升全要素生产率，才能带动各行各业的全面、可持续发展。当资本狂欢落幕，真正带来繁荣的从来不是单一技术的突破，而是技术与社会需求的深度融合和价值创造。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"color: rgb(187, 187, 187);\">【新闻来源】环球时报 作者 陈经 科技与战略风云学会研究员 \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fbaijiahao.baidu.com\u002Fs?id=1845578358302849473&wfr=spider&for=pc\" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(187, 187, 187);\">https:\u002F\u002Fbaijiahao.baidu.com\u002Fs?id=1845578358302849473&wfr=spider&for=pc\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(187, 187, 187);\">（本网转发此文章，旨在为读者提供更多的信息资讯，所涉内容不构成投资、消费建议。文章事实如有疑问，请与有关方核实，文章观点非本网观点，仅供读者参考。）\u003C\u002Fspan>\u003C\u002Fp>","","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fac7996b0b2644aadbe2d7c10c89cfc18\u002FAI领域.jpg","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fthumbs\u002Fac7996b0b2644aadbe2d7c10c89cfc18\u002FAI领域.jpg",0,1,46,"2025-10-13 22:08",2,false,{"id":17,"name":20,"enName":21},"芯位视野","Xinwei Vision","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3A370bfb16-a6d0-446a-b2b0-7df2c3a7df79%3A0.wav?Expires=1760931264&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=hBylZ8I340XtIqzVj3iFj4%2B1fAM%3D",8045978,"370bfb16-a6d0-446a-b2b0-7df2c3a7df79","2025-10-13 22:06","Calculating power is hard to monetize, and the US AI is in a dilemma","\u003Cp class=\"ql-align-justify\">\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">Recently, economists at Harvard University in the United States found that although the US GDP grew by 1.6% in the first half of 2025, it was almost entirely driven by data centers and information processing technology, with other sectors growing only 0.1%. The expected scale of AI data center spending in the US for 2025 is $52 billion, and the way AI drives GDP growth mainly relies on investment rather than related consumption.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Since the beginning of this year, as the main data center investors, Microsoft, Google, Amazon, and Meta have fiercely competed for AI computing power, investing between $64 billion and $100 billion in cash, but lacking significant returns, their nature is close to \"burning money\". Recently, OpenAI also \"promised\" to invest $1 trillion over five years despite limited funds, which has raised great doubts in the market about whether the US AI technology industry is a bubble or even a Ponzi scheme.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">From the perspective of AI development, the good news is that using large models by individuals and enterprises has become very common. By early October, ChatGPT's weekly active users reached 800 million, and consumer AI adoption has exceeded expectations. However, a bad sign is that such a large amount of user data has not yet generated sufficient revenue, and even cannot cover operating costs. Generally speaking, once an internet application reaches a certain number of users, its revenue prospects will be relatively clear, and investment institutions will actively provide expansion funds to capture market share, entering a win-win harvest period. But now, various consumer large models in the US seem like an endless pit.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The revenue difficulties of generative AI are related to its own technical principles. Most internet industries have obvious economies of scale. After the infrastructure is sufficient, the marginal cost is basically zero, and the cost of supporting tens of millions of customers is similar to that of hundreds of millions of customers, with high profit margins. However, each response and reasoning of a large model requires massive repeated computation, and the more users there are, the more data centers and computing power are needed, not only failing to enjoy economies of scale, but also increasing the demand for infrastructure and financing. Therefore, the expansion of US AI investment is not a positive phenomenon, but rather reflects potential issues with investment efficiency.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Another factor that hinders the commercialization of generative AI is the difficulty in monetizing through advertising. Internet advertising has low display costs but high profits, but a large model can only show one answer at a time, making it difficult to naturally insert multiple ad placements. This is why \"subscription\" has become the main revenue method for large models. However, compared to the investment scale of $100 billion per year, ChatGPT's annual subscription revenue of only $10 billion seems insignificant.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Enterprise cloud services are another direction that is highly anticipated. Compared to the common \"cloud migration\" operations that mainly serve data storage, American enterprises hope that AI data centers are more intelligent, helping enterprises improve operational efficiency, reduce costs, and increase revenue. However, a report from MIT in August showed that 95% of companies that introduced AI into their business did not make money. Oracle's GPU cloud rental business has a gross margin of only 14%, far below the company's overall business of 70%, and after considering depreciation, the net gross margin drops to 7%.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The progress of enterprise \"AI intelligence introduction\" is not effective, and the \"hallucination\" problem is deeply rooted. With the rising cost of computing power, when people's expectations of AI exceed toys and enter the practical stage, the shortcomings of current generative AI are exposed. This defect leads to the commercialization process of AI technology lagging significantly behind infrastructure construction and financialization, bringing great uncertainty to future revenue. In response, the US industry and capital have formed two schools of thought: the \"industrial bubble theory\" and the \"financial bubble theory.\" The former believes that data center construction is a physical investment that will eventually break through the threshold of scale and become a basic energy facility like electricity. The latter warns that the current investment scale has already exceeded rational limits, and overly optimistic expectations may repeat the internet bubble of the past.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cbr>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Whether it is about long-term infrastructure layout or capital games among tech giants, the imbalance issue in the US AI industry development model is undeniable. Economic development driven by AI must not rely solely on the rapid advancement of data centers and computing power infrastructure, but needs to complete the transformation from \"hard data\" to \"soft value,\" proving its economic value in a wide range of commercial scenarios. Only when AI is truly integrated into traditional industries such as manufacturing and services, improving total factor productivity, can it promote comprehensive and sustainable development across all industries. When the capital狂欢 ends, what has always brought prosperity is not a single technological breakthrough, but the deep integration and value creation between technology and social needs.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"color: rgb(187, 187, 187);\">【News Source】Global Times Author Chen Jing Researcher of the Institute of Science, Technology and Strategic Studies \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fbaijiahao.baidu.com\u002Fs?id=1845578358302849473&wfr=spider&for=pc\" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(187, 187, 187);\">https:\u002F\u002Fbaijiahao.baidu.com\u002Fs?id=1845578358302849473&wfr=spider&for=pc\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(187, 187, 187);\">（This website re-publishes this article to provide readers with more information and news. The content does not constitute investment or consumption advice. If there are any questions about the facts of the article, please verify with the relevant parties. The views expressed in the article are not the views of this website and are for reference only.）\u003C\u002Fspan>\u003C\u002Fp>","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3Ae8d81045-842f-4558-a2ef-3cbbd34a4081%3A0.wav?Expires=1774838460&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=nGNzCKXeFS2KDY5C0OxLKtUVhEk%3D","e8d81045-842f-4558-a2ef-3cbbd34a4081",10877536]