内容来源：https://aiweekly.co/issues/ai-slop-a-725b-bet-on-what-no-one-wanted

内容总结：

AI投资热潮遭遇用户信任危机：今年基础设施投入7250亿美元，消费者却开始“用脚投票”

核心矛盾：供给疯狂扩张，需求明显降温

据最新行业报告显示，全球超大规模云服务商今年在AI基础设施上的投入将高达7250亿美元，其中约75%专项用于人工智能。然而，就在资本疯狂涌入的同时，用户端却出现了明确的抵触信号——Gartner调查发现，50%的美国消费者更倾向于选择不使用生成式AI的品牌。维基百科以44票对2票的压倒性结果禁止AI生成内容，Stack Overflow的新问题数量同比暴跌78%，谷歌AI概览功能导致首页点击率下降58%。

需求端多项指标齐跌：消费者不买账、企业零回报

这一结构性矛盾贯穿整个行业：产能增速最快的领域，恰恰是买家逃离最明显的市场。Gartner的2026年消费者调查显示，58%的受访者因品牌使用AI生成内容而降低信任度，仅15%表示更信任。54%的用户已出现“AI疲劳”，AI生成内容的互动率比人工创作低20%-35%。

企业端同样不容乐观。2026年2月NBER研究论文指出，90%的受访企业报告AI部署对生产率“零影响”。MIT的GenAI研究发现，95%的企业项目投资回报率为零。微软自研的Copilot在六个月内市场份额流失39%，用户将“对输出内容不信任”列为首要原因。

平台生态恶化：维基百科禁令、Stack Overflow崩溃

维基百科今年3月正式投票禁止使用大型语言模型生成或改写文章内容，成为全球首个对AI写作采取最强硬制度性立场的大型开放知识平台。Stack Overflow月问题量从2014年高峰期的20万+骤降至2025年12月的3862个，同比跌幅达78%。cURL项目因AI生成的垃圾报告淹没安全团队，不得不终止漏洞悬赏计划。谷歌AI概览功能导致搜索结果首页点击率一年内从34.5%暴跌至58%，全球2500家新闻网站的谷歌推荐流量下降33%。

资本端出现裂痕：软银削减贷款、微软放弃碳中和承诺

尽管资本仍在涌入，但已有松动迹象。软银将向OpenAI提供的保证金贷款目标下调40%至60亿美元，原因是贷款方对估值提出异议。微软据传正考虑放弃2030年碳负排放承诺，因为AI训练周期导致其排放量自2020年以来上升24%。苹果则因延迟Siri AI功能而支付2.5亿美元和解金。

摩根士丹利分析师Todd Castagno预测，2026年资本支出与销售额之比将达34%，2028年增至37%，数据中心债务预计突破1万亿美元——而互联网泡沫高峰期这一比例仅为22%。高盛上周发布报告，承认怀疑论者的估值逻辑，但认为“害怕错过”的心理仍将主导支出。一位华尔街分析师直言：“当卖方银行公开将支出解释为心理而非基本面时，通常意味着周期已近尾声。”

行业警示：这不是“需求终会到来”的故事

数据正同时在两个方向上恶化：产能持续攀升，而消费者信任、平台信任、出版商信任以及CEO信心都在同步下滑。可口可乐和沃尔玛的离任CEO均向CNBC表示，AI浪潮是他们交棒的关键原因——两位全球最稳定的消费品牌掌门人承认自己不是驾驭下一阶段的最佳人选。

关键数据一览：

• 2026年AI基础设施投入：7250亿美元
• 50%美国消费者偏好不使用生成式AI的品牌
• 95%企业AI项目投资回报率为零
• Stack Overflow新问题量：同比跌78%
• 谷歌AI概览致点击率下降58%
• 全球新闻网站谷歌推荐流量降33%
• 微软Copilot半年流失39%市场份额
• 数据中心债务预计突破1万亿美元（互联网泡沫时期为22%）

中文翻译：

超大规模企业今年将在AI基础设施上投入7250亿美元。然而，它们所服务的用户如今正主动拒绝其输出。高盛发现，50%的美国消费者更青睐不使用生成式AI的品牌。维基百科以44比2的投票结果禁止AI生成内容。Stack Overflow的新问题数量同比下降78%。谷歌AI概览将首页点击率拉低了58%。这便是贯穿下文每一则故事的结构性张力：产能增长最快的领域，恰恰是买家最明显退出的市场。

从《AI周刊》获取更多信息
更多信号，更少噪音——选择你的渠道。
你正在阅读的是每周简报。以下为追踪故事的其它途径——每项渠道均免费，随时可退出。

→ 探索16个深度专题
每周特定主题通讯：生成式AI、机器学习、AI在商业中的应用、机器人技术、前沿研究、地缘政治、医疗健康等。浏览全部16个深度专题 →

→ AI重大突发警报
当重大事件发生（如600亿美元收购、监管紧急会议、前沿模型泄露），订阅者将在数小时内收到提醒。通常每天0-2封邮件。获取突发警报 →

→ AI今日新闻（实时）
随着扫描器发现新闻实时更新的仪表板：过去48小时评分故事、每周实体动态、以及覆盖113家AI公司、人物和话题的季度趋势线。打开AI今日新闻 →

速报

1. 浩天AI深度伪造团伙诈骗受害者400万美元。404 Media记录了一起有组织的欺诈行动，该团伙通过WhatsApp、Zoom和Teams进行实时换脸，针对高净值目标实施克隆语音和视频通话。垃圾内容论调已不再是文化评论，它是整个规模化欺诈经济的基质。
2. 出版商起诉Meta，指控其使用盗版书籍训练Llama。五家出版商（爱思唯尔、圣智、阿歇特、麦克米伦、麦格劳希尔）及斯科特·特罗于5月5日提起集体诉讼，声称Meta在扎克伯格个人指示下，通过BT下载了来自LibGen和安娜档案库的267 TB盗版内容。有利于原告的判决将重新定价行业内的每一个训练数据集。
3. Meta因1450亿美元资本支出指引下跌7%，为周期内最严厉的市场反应。彭博社目前预计，到2026年，超大规模企业AI资本支出将超过7000亿美元——Alphabet和微软各1900亿美元，亚马逊2000亿美元。同一周，软银将其OpenAI支持的保证金贷款目标削减40%至60亿美元。资本堆栈的两个不同部分在同一财报周期内出现动摇。
4. 维基百科以40比2的投票结果禁止AI生成内容。英语维基百科编辑于3月20日投票决定，禁止使用大语言模型生成或重写该平台710万篇文章的内容。全球最大的开放知识平台刚刚采取了迄今为止最强硬的制度立场，反对AI撰写的文本。任何文章的基础文本必须由人类撰写。
5. Stack Overflow从每月20万个问题降至近乎零。2025年12月仅发布3862个问题——与2014年每月超过20万的峰值相比，同比下降78%。部分原因是用户迁移至ChatGPT。其余原因是社区信任的侵蚀速度超过了版主团队重建它的速度。
6. 高德纳：50%的美国消费者更青睐不使用生成式AI的品牌。一项2026年调查还发现，58%的人因品牌使用AI生成内容而减少信任；仅15%的人更信任它们。54%的人报告AI疲劳。与人工制作的同类内容相比，AI感知内容获得的参与度惩罚为20-35%。
7. 谷歌AI概览导致顶级搜索结果点击率下降58%。Ahrefs记录了12个月内点击率下降率翻倍（2025年4月为34.5%，2026年2月为58%）。Chartbeat数据显示，谷歌对全球2500个新闻网站的引荐流量下降了33%。网络的根本契约正在破裂。
8. 可口可乐和沃尔玛首席执行官均将AI作为其决定卸任的原因。詹姆斯·昆西（可口可乐，3月31日卸任）和道格·麦克米伦（沃尔玛，2月1日卸任）均告诉CNBC，AI的下一波浪潮是交棒的关键原因。两位全球最稳定的消费品牌CEO承认，他们并非驾驭下一阶段的合适操盘手。这便是标普100指数顶层发出的“我们不知道结局如何”的信号。

每个需求信号都在下行
剔除噪音，结构便清晰可见。超大规模企业承诺2026年资本支出约7250亿美元，其中约75%与AI相关。Anthropic刚承诺五年内向谷歌云投入2000亿美元——自2027年起每年约400亿美元，超过Alphabet披露收入积压的40%。同一时期，软银在其估值遭贷方反对后，将OpenAI支持的保证金贷款目标削减40%至60亿美元。这是来自主要支持者的首次信贷侧退缩。

现在看看需求侧。高德纳2026年消费者调研发现，半数美国成年人会主动选择不使用生成式AI的品牌。一半。2026年2月美国国家经济研究局的一份论文发现，90%的受访企业报告AI部署的生产力影响为零。一项麻省理工学院的生成式AI研究追踪发现，95%的企业项目投资回报率为零。微软自家的Copilot在六个月内失去了39%的市场份额，用户将输出不信任列为首要原因。

平台层面的数据更为尖锐。维基百科于3月禁止AI生成文章。Stack Overflow在十二个月内新问题数量下降78%。cURL在其安全团队被AI生成的垃圾提交淹没后，终止了漏洞赏金计划。谷歌AI概览将排名最高页面的点击率拉低58%，目前58%的搜索以零点击告终。出版商引荐流量平均下降25%，新闻类在全球范围内下降33%。

据报道，微软正考虑放弃其2030年碳负排放承诺，因为AI训练周期使其自2020年以来排放量上升24%。苹果刚就“延迟Siri AI功能”达成2.5亿美元集体诉讼和解，符合条件的iPhone用户可因苹果宣传但从未推出的功能每人索赔最高95美元。

摩根士丹利的托德·卡斯塔尼奥现在建模显示，2026年资本支出与销售额比率为34%，到2028年达37%，数据中心债务预计将突破1万亿美元。互联网泡沫高峰期这一比率为22%。高盛上周发布报告，承认看空者的估值逻辑，同时辩称“错失恐惧症”最终会胜出。当卖方银行公开将支出解释为心理学而非基本面时，这通常意味着周期已近尾声。

这并不是一个“需求终将到来”的故事。数据点在双向同时恶化。产能正在攀升。消费者信任、出版商信任、贡献者信任以及CEO信心都在同一时间线上下降。

关键要点

• 需求在消费者、平台、出版商和企业层面以可衡量的方式收缩。半数美国消费者主动偏好非AI品牌。两个最大的开放知识平台（维基百科、Stack Overflow）内容流失速度超过更新速度。出版商流失三分之一的引荐流量。Copilot市场份额流失。这些并非感觉。
• 资本支出曲线与需求曲线指向相反方向。面对7250亿美元的2026年支出，却有50%消费者偏好非AI品牌，95%企业零投资回报率。要么生产力故事在最后一局到来，要么从一开始供给曲线就是错误的。
• “垃圾内容”现已成为消费者侧的数据点。维基百科禁令、Stack Overflow崩溃、麦当劳撤回AI广告、高德纳数据。将此术语视为消费者AI需求的先行指标，而非模因。
• 能源、法律及采用风险现已与资本支出挂钩。微软的气候转向、Meta图书诉讼、Copilot用户流失、可口可乐/沃尔玛CEO交接：这一切都是同一轮建设的下游结果。这些成本在12个月前均未被计入价格。

值得一读

• 高德纳：53%的消费者不信任AI驱动的搜索。关于消费者信任差距最清晰的一手调查。
• 可口可乐和沃尔玛即将离任的CEO提及AI作为卸任原因（CNBC）。两位最稳定的消费品牌CEO承认他们不是下一阶段的合适操盘手。
• 维基百科禁止AI生成内容（404 Media）。2026年迄今最强烈的机构拒绝信号。
• 高盛：错失恐惧症而非基本面驱动AI支出（《财富》）。卖方银行承认看空者的估值逻辑。
• cURL因AI垃圾关闭漏洞赏金计划（The Register）。垃圾内容经济的开发者侧成本，由全球半数互联网依赖的协议维护者讲述。

上周投票
“当你现在看到一段政治视频片段时，你的第一反应是什么？”（62份回复）

• 34% 在信任前先检查是否真实
• 26% 已完全停止观看政治片段
• 23% 假定其被操纵，除非被证明相反
• 18% 仍然表面信任
  82%的读者不再表面信任AI时代的政治视频片段。同样的默认怀疑正在渗入商业、搜索和信息流。本周的问题延续同一思路。

本周投票
当用户停止接受AI垃圾内容时，泡沫会发生什么？
上周，你们中有62人投票：
当你现在看到一段政治视频片段时，你的第一反应是什么？
当用户停止接受AI垃圾内容时，泡沫会发生什么？

Alexis

英文来源：

Hyperscalers will spend $725 billion on AI infrastructure this year. The users they are spending it on are now actively rejecting the output. Gartner finds 50% of US consumers prefer brands that don't use generative AI. Wikipedia just banned AI-generated content 44-2. Stack Overflow's new-question volume has fallen 78% year over year. Google AI Overviews have collapsed top-page CTR by 58%. This is the structural tension running through every story below: capacity is being added fastest in exactly the parts of the market where buyers are most visibly walking away.
Get more from AI Weekly
More signal, less noise — pick your channels.
You're reading the weekly brief. Below are the other ways to follow the story — every channel free, easy to leave.

→ Explore 16 deep divesWeekly topic-specific newsletters: Generative AI, Machine Learning, AI in Business, Robotics, Frontier Research, Geopolitics, Healthcare, and more.Browse all 16 deep dives →

→ Breaking AI alertsWhen something major breaks (a $60B acquisition, a regulator's emergency meeting, a frontier model leak), alert subscribers know within hours. Typically 0-2 emails per day.Get breaking alerts →

→ AI News Today (live)Live dashboard updated as the scanner finds news: scored stories from the last 48 hours, weekly entity movers, and quarterly trend lines across 113 AI companies, people, and topics.Open AI News Today →
Quick Hits

1. Haotian AI deepfake ring scams victims out of $4M. 404 Media documents an organised fraud operation running cloned voice and video calls against high-net-worth targets via real-time face-swap on WhatsApp, Zoom, and Teams. The slop thesis is no longer cultural commentary. It is the substrate for an entire fraud economy operating at scale.
2. Publishers sue Meta over pirated books used to train Llama. Five publishers (Elsevier, Cengage, Hachette, Macmillan, McGraw Hill) and Scott Turow filed a class action May 5, alleging Meta torrented 267 TB of pirated content from LibGen and Anna's Archive at Zuckerberg's personal direction. A plaintiff-friendly ruling would reprice every training set in the industry.
3. Meta drops 7% on $145B capex guidance, the harshest market reaction of the cycle. Across hyperscalers, Bloomberg now puts 2026 AI capex past $700B — Alphabet and Microsoft at $190B each, Amazon $200B. SoftBank cut its OpenAI margin loan target 40% to $6B the same week. Two different parts of the capital stack flinched in the same earnings cycle.
4. Wikipedia bans AI-generated content, 40 to 2. English Wikipedia editors voted on March 20 to prohibit using LLMs to generate or rewrite article content across the platform's 7.1M articles. The world's largest open knowledge platform just took the strongest institutional stance yet against AI-written content. The base text of any article must be written by a human.
5. Stack Overflow went from 200,000 monthly questions to nearly zero. Only 3,862 questions posted in December 2025 — a 78% year-over-year collapse from a 2014 peak of 200,000+/month. Some of that is migration to ChatGPT. The rest is community trust eroding faster than the moderation team can rebuild it.
6. Gartner: 50% of US consumers prefer brands that don't use generative AI. A 2026 survey also finds 58% trust brands less for using AI-generated content; only 15% trust them more. 54% report AI fatigue. AI-perceived content gets a 20-35% engagement penalty against human-made equivalents.
7. Google AI Overviews drive a 58% click-through collapse for top results. Ahrefs documents the doubling of the CTR decline in twelve months (34.5% in April 2025, 58% in February 2026). Chartbeat data shows Google referral traffic down 33% globally across 2,500 news sites. The web's foundational compact is breaking.
8. Coca-Cola and Walmart CEOs both cite AI in their decisions to step down. James Quincey (Coke, succeeded March 31) and Doug McMillon (Walmart, succeeded Feb 1) both told CNBC that the next wave of AI was a key reason for handing over the role. Two of the most stable consumer-brand CEOs in the world admitting they are not the right operators to navigate the next phase. That is what "we don't know how this ends" sounds like from the top of the S&P 100.
   Every demand signal is pointing down
   Strip the noise and the structure becomes legible. Hyperscalers are committing roughly $725B in 2026 capex, about 75% of it AI-specific. Anthropic just committed $200B to Google Cloud over five years — about $40B/year starting 2027, more than 40% of Alphabet's disclosed revenue backlog. In the same window, SoftBank cut its OpenAI-backed margin loan target 40% to $6B after lenders pushed back on the valuation. The first credit-side flinch from a major backer.
   Now look at the demand side. Gartner's 2026 consumer panel finds half of US adults would actively prefer brands that don't use generative AI. Half. A February 2026 NBER paper finds 90% of surveyed firms report zero productivity impact from AI deployments. An MIT GenAI study tracks 95% of corporate projects at zero ROI. Microsoft's own Copilot has lost 39% of its market share in six months, with users citing distrust of outputs as the leading reason.
   The platform-level data is sharper. Wikipedia banned AI-generated articles in March. Stack Overflow lost 78% of new-question volume in twelve months. cURL ended its bug bounty program after AI-generated slop submissions overwhelmed its security team. Google AI Overviews have cut click-through rates by 58% on top-ranked pages, with 58% of all searches now ending in zero clicks. Publisher referral traffic is down 25% on average, 33% globally on news.
   Microsoft is reportedly considering abandoning its 2030 carbon-negative pledge because AI training cycles are pushing its emissions up 24% since 2020. Apple just settled a $250M class action over delayed Siri AI features, with eligible iPhone owners claiming up to $95 each for features Apple marketed but never shipped.
   Morgan Stanley's Todd Castagno now models capex-to-sales at 34% in 2026 and 37% by 2028, with data-center debt projected to clear $1T. The dot-com peak was 22%. Goldman Sachs published a note last week conceding the skeptics' valuation case while arguing FOMO would win anyway. When a sell-side bank is openly explaining the spend as psychology rather than fundamentals, that is usually late in the cycle.
   This is not a "demand will arrive eventually" story. The data points are getting worse in both directions at once. Capacity is climbing. Consumer trust, publisher trust, contributor trust, and CEO confidence are all moving down on the same timeline.
   Key Takeaways
   • Demand is contracting in measurable ways across consumer, platform, publisher, and enterprise tiers. Half of US consumers actively prefer non-AI brands. The two largest open knowledge platforms (Wikipedia, Stack Overflow) are losing content faster than they can replace it. Publishers are losing a third of referral traffic. Copilot is bleeding share. These are not vibes.
   • The capex curve and the demand curve are pointing in opposite directions. $725B 2026 spending against 50% consumer preference for non-AI brands and 95% enterprise zero-ROI. Either the productivity story arrives in late innings, or the supply curve was wrong from the start.
   • "Slop" is now the consumer-side data point. Wikipedia bans, Stack Overflow collapse, McDonald's pulling AI ads, Gartner numbers. Treat the term as a leading indicator on consumer AI demand, not as a meme.
   • Energy, legal, and adoption risk are now coupled to capex. Microsoft's climate U-turn, the Meta books suit, Copilot's churn, the Coca-Cola/Walmart CEO transitions: all downstream of the same buildout. None of these costs were priced in 12 months ago.
     Worth Reading
   • Gartner: 53% of consumers distrust AI-powered search. The cleanest primary survey on the consumer-trust gap.
   • Coca-Cola and Walmart outgoing CEOs cite AI in stepping down (CNBC). Two of the most stable consumer-brand CEOs admitting they are not the right operators for the next phase.
   • Wikipedia bans AI-generated content (404 Media). The strongest institutional rejection signal of 2026 so far.
   • Goldman Sachs: FOMO is driving AI spending more than fundamentals (Fortune). The sell-side bank conceding the skeptics' valuation case.
   • cURL shutters bug bounty over AI slop (The Register). The developer-side cost of the slop economy, told by the maintainer of the protocol half the internet runs on.
     Last week's poll
     "When you see a political video clip now, what's your first reaction?" (62 responses)
   • 34% check if it's real before trusting
   • 26% have stopped watching political clips entirely
   • 23% assume it's manipulated until proven otherwise
   • 18% still trust it at face value
     82% of readers no longer take AI-era political clips at face value. That same default skepticism is now bleeding into commerce, search, and feeds. This week's question follows the same line.
     This week's poll
     When users stop accepting AI slop, what happens to the bubble?
     Last week, 62 of you voted:
     When you see a political video clip now, what\u0027s your first reaction?
     When users stop accepting AI slop, what happens to the bubble?
     Alexis