无人叫停,大型科技公司 AI 基建就不会停

正在准备对话…
除非有人出手叫停,大型科技公司的 AI 大建设就不会停止
Amazon、Microsoft、Alphabet 和 Meta 都要为各自的数据中心计划负责。
Amazon、Microsoft、Alphabet 和 Meta 都要为各自的数据中心计划负责。
图片来源:Cath Virginia / The Verge,Getty Images
2026 年秋季财报观察
是一位独立记者,专注于商业调查、分析和新闻。她此前在 Business Insider 报道 AI 数据中心热潮。
在 AI 竞赛中四家最大的竞争对手于上月末召开季度财报电话会议之后,有两点已经很清楚:大型科技公司并未停止在 AI 数据中心建设上投入创纪录的资金,而 AI 数据中心建设也并未更接近兑现其所承诺的回报。
第三件事也开始变得清晰——大型科技公司不会放慢脚步,至少不会主动放慢。
亚马逊、微软、Alphabet 和 Meta 告诉投资者,仅今年一年,它们就计划在资本支出上投入至少 7200 亿美元。其中大部分将投向 AI 基础设施。这还是在未来数年已安排的至少 1 万亿美元支出承诺之外。
大型科技公司的支出计划上周正式惊动了华尔街,引发科技股大幅抛售;随着投资者将注意力转向周二的中期选举,抛售开始逐渐减弱。在全国选民选出支持暂停数据中心建设及其他限制措施的候选人之后,这些支出计划可能会发生变化。
鉴于科技股周二开始重新上涨,华尔街或许会把数据中心监管前景带来的新不确定性视为一件好事。毕竟,大型科技公司还有另一个问题必须解决。尽管投入了巨额资金,它们至今还没有真正从 AI 上赚到钱,却仍在继续加大押注。
迄今为止,尽管大型科技公司不断加大支出为 AI 建设提供资金,投资者一直没有真正动摇。但三年多过去,华尔街开始对 AI 所承诺的回报感到不安——以及他们究竟何时才能看到投资回报。
并不是说营收没有增长——增长其实相当强劲。亚马逊公布了五年来最快的季度增长,Meta 的广告收入正在激增,Google Cloud 的收入在一年内翻了一番多。但听起来令人惊讶的是,这可能仍不足以证明大型科技公司在 AI 上投入的巨额资金是合理的。
“如果这类加速和机会是在 25 年里逐步发生、你可以把资本支出分摊到 25 年,会不会更方便?是的,也许会方便一些。但我们所处的世界并非如此,机会也不是这样。”
上周,Alphabet 给出了这方面最令人不安的例证:Google 母公司 Alphabet 的高级副总裁兼首席商务官 Philipp Schindler 在公司财报电话会上承认,Google Cloud 的收入增长是以其核心搜索业务为代价换来的。
Schindler 表示:“值得注意的是,AI Mode 查询量的强劲增长带来了影响,因为我们在 AI Mode 上看到的变现率较低。”
第三季度,Google 搜索广告收入增长了 12%,较该部门去年同期 17% 的增速明显下滑。
这对 Google 来说是一个奇怪的悖论。公司核心业务 Google 搜索的强劲表现,此前正是让投资者对公司在 AI 上激进的支出计划抱有信心。但如今,Google 新推出的 AI 搜索正在蚕食搜索业务,而公司尚未真正想清楚如何让 AI Mode 变现。
在谈到 Google 为 AI Mode 变现所做的努力时,Schindler 表示:“需要说明的是,我们对目前的进展感觉非常好。我们的广告实验显示出令人鼓舞的结果。”但他没有给出任何具体例子。
上周,华尔街对 Google 尤为不满:周四,Alphabet 股价经历了公司历史上最惨的单日下跌,一天之内重挫 13%。
本周该股已开始回升,但这恐怕无济于事——到目前为止,大型科技公司唯一能确定的事,就是预计还会看到更多支出。
Alphabet 计划今年资本支出在 1950 亿至 2050 亿美元之间,据 CFO Anat Ashkenazi 称,该公司还计划在 2027 年“大幅”增加支出。
Amazon CEO Andy Jassy 上周在公司的财报电话会议上表示,Amazon 计划在 2027 年的支出“明显高于”今年计划的 2200 亿美元资本支出。他将这笔支出——公司历史上规模最大的一次——称为把 Amazon 发展成万亿美元公司的“一生一次的机会”。
“如果这种加速和机会是在 25 年里发生,你可以把资本支出分摊到 25 年,会不会更方便?是的,可能会更方便,”Jassy 说。“但这不是我们所处的世界,也不是这个机会。”
Meta 预计今年支出在 1350 亿至 1450 亿美元之间,它没有披露明年打算支出多少,但主要金融机构预计其支出将比今年增加。
在推介公司推出 Muse 时,Meta CEO Mark Zuckerberg 指出,这个智能体式 AI 项目只有在 Meta 能把算力成本降下来之后才会实现盈利。
Zuckerberg 指出,Meta 预计明年算力效率将提升“十倍”,他预计这一努力将需要“重大投资”。
烧钱的并不只有大型科技公司。Anthropic 本周在 S-1 招股说明书中更新了第三季度营收,显示算力成本和净亏损大幅上升。这家 AI 初创公司该季度营收增长至 170 亿美元,同比增长超过 1000%,但其算力成本达到 175 亿美元——尚不足以覆盖这一亮眼的增长。
Anthropic 还承诺了近 9000 亿美元的未来算力与基础设施支出,并计划启动自建数据中心这一耗资巨大的工程。
这 9000 亿美元中的一部分将流向 Google,后者正向 Anthropic 出售其 TPU 芯片。在 Google 的电话会议上,两位不同的分析师就 TPU 业务向高管施压。毕竟,Google 与其他超大规模厂商一样,都表示自身产能受限。如果情况属实,Google 为何要耗费自身资源,把自己的产能卖给其他实验室?
当被问及出售 TPU 所得是否有助于 Google 削减自身部分支出时,Ashkenazi 没有给出明确的肯定或否定回答。不过,答案很可能是否定的。Google 是先垫付成本来制造这些卖给 Anthropic 等 AI 实验室的硬件。这些合同由财务义务支撑,将在硬件可以投入使用到期,但它们并不代表 Google 当前口袋里的现金。
值得注意的是,上周 Microsoft 是唯一躲过华尔街怒火的大型科技公司。该公司本季度资本支出为 452 亿美元,低于此前预期的 500 亿美元。然而,支出减少并非因为 Microsoft 方面缩减开支,而是由于“第三方产能延迟”。
直到上周,Microsoft 一直是唯一没有通过举债来为其 AI 建设提供资金的超大规模厂商。“鉴于我们的评级水平,我们有幸能够进入债券市场,而我要说,我们尚未动用这一渠道,”CFO Amy Hood 在周三晚间的公司财报电话会议上表示。
到第二天早上,情况就变了:Microsoft 规模 250 亿美元、分为 10 部分的债券发行消息传出。
截至本文撰写时,大型科技公司的股价已经回升。也许华尔街正押注中期选举的“蓝色浪潮”会迫使大型科技公司在未来几年放缓部分支出。放缓也可能给大型科技公司多一点时间,去琢磨如何让 AI 实现盈利。
话又说回来,尽管上周中期选举迫在眉睫,但大型科技公司的财报电话会议上都没有提到它。
关注本文涉及的话题和作者,即可在个性化首页信息流中看到更多类似内容,并接收邮件更新。
这是原生广告的标题
这是原生广告的标题
登录以查看你的通知,或创建账户参与讨论。
译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。
查看英文原文(Big Tech’s big AI buildout will not end until someone stops it)
Big Tech’s big AI buildout will not end until someone stops it | The Verge
Getting the conversation ready...
Big Tech’s big AI buildout will not end until someone stops it
Amazon, Microsoft, Alphabet, and Meta are on the hook with their data center plans.
Amazon, Microsoft, Alphabet, and Meta are on the hook with their data center plans.
Nov 5, 2026, 10:45 AM UTC

Image: Cath Virginia / The Verge, Getty Images
Fall 2026 earnings watch
is an independent reporter focused on business investigations, analysis and news. She previously covered the AI data center boom for Business Insider.
After four of the biggest competitors in the AI race held quarterly earnings calls late last month, two things are clear: Big Tech is not done spending record amounts on the AI data center buildout, and the AI data center buildout is not any closer to yielding its promised returns.
A third thing is also starting to become clear — Big Tech will not be slowing down, at least not on its own.
Amazon, Microsoft, Alphabet, and Meta told investors they’re planning to spend at least $720 billion on CapEx this year alone. Most of that will go toward AI infrastructure. That’s on top of at least $1 trillion in spending commitments that have been arranged for the coming years.
Big Tech’s spending plans officially spooked Wall Street and triggered a sharp selloff in tech stocks last week that began to taper off as investors shifted focus to Tuesday’s midterm elections. Those spending plans could change after voters across the country elected candidates who favor moratoriums and other restrictions on data centers.
Wall Street might see the new uncertainty around the future of data center regulation as a good thing, given that tech stocks began to rise again on Tuesday. After all, there’s another issue Big Tech companies have to address. Despite enormous amounts of spending, they haven’t really made money from AI yet, and they’re continuing to double down on it anyway.
Until now, investors haven’t really blinked as Big Tech has ramped up its spending to fund the AI buildout. But more than three years in, Wall Street is starting to get antsy about the promised payoff of AI — and exactly when they might expect to see a return on investment.
It’s not that top-line revenue isn’t growing — it very much is. Amazon reported its fastest growth in five years for the quarter, Meta’s ad revenue is surging, and Google Cloud’s revenue more than doubled in a year. But as astonishing as it sounds, it might not be enough to justify the massive amounts Big Tech is spending on AI.
“Would it be more convenient if this type of acceleration and opportunity happened over 25 years and you could spread the CapEx over 25 years? Yeah, it might be more convenient. But that’s not the world we live in, and that’s not the opportunity.”
Alphabet offered the most uncomfortable example of this last week when Philipp Schindler, Google parent Alphabet’s senior vice president and chief business officer, acknowledged on the company’s earnings call that Google Cloud’s revenue growth has come at the expense of its core Search business.
“It’s worth noting that the strong growth in AI Mode queries had an impact because of the lower monetization we were seeing in AI Mode,” Schindler said.
Revenue from Google’s Search ads grew 12% in the third quarter, a marked drop from the 17% growth rate the division posted for the same quarter last year.
It’s a weird paradox for Google. The strength of the company’s core business, Google Search, is what previously gave investors the confidence in the company’s aggressive spending agenda for AI. But now, Google’s new AI-enabled search engine is eating into theSearch business before the company has really figured out how to monetize AI Mode.
“To be clear, we feel very good about our trajectory here. Our ad experiments have shown promising results,” Schindler said of Google’s efforts to monetize AI Mode. He did not offer any specific examples.
Wall Street was especially unhappy with Google last week: On Thursday, Alphabet stock saw its worst one-day drop in history, sinking 13% in a single day.
The stock has started to pick back up this week, but it probably isn’t helping that the only thing Big Tech can say with certainty so far is to expect to see more spending.
Alphabet, which plans to spend between $195 billion and $205 billion in capital expenditures this year, is also planning to increase spending “significantly” in 2027, according to CFO Anat Ashkenazi.
Amazon intends to spend “meaningfully more” in 2027 than the $220 billion in capital expenditures it is planning this year, CEO Andy Jassy said on the company’s earnings call last week. He framed the spending — the largest ever in the company’s history — as a “once in a lifetime opportunity” to grow Amazon into a $1 trillion company.
“Would it be more convenient if this type of acceleration and opportunity happened over 25 years and you could spread the CapEx over 25 years? Yeah, it might be more convenient,” Jassy said. “But that’s not the world we live in, and that’s not the opportunity.”
Meta, which is expected to spend between $135 billion and $145 billion this year, didn’t disclose how much it intends to spend next year, but major financial institutions are expecting it to increase its spending from this year.
In touting the company’s rollout of Muse, Meta CEO Mark Zuckerberg noted that the agentic AI program will only become profitable if Meta can get its compute costs down.
Zuckerberg noted that Meta expects compute efficiency to increase “tenfold” next year, an endeavor he expects will take “significant investment.”
It’s not just Big Tech that is burning cash. Anthropic updated its S-1 prospectus with third-quarter revenue this week, revealing a sharp uptick in computing costs and net losses. The AI startup’s revenue grew to $17 billion in the quarter, a more than 1,000% increase from last year, but its compute costs came in at $17.5 billion — not quite enough to cover the impressive gains.
Anthropic has also committed nearly $900 billion in future compute and infrastructure costs, and it plans to begin the costly endeavor of building its own data centers.
Some of that $900 billion will go to Google, which is selling its TPU chips to Anthropic. Two different analysts on Google’s call pressed executives on the TPU business. After all, Google, like the other hyperscalers, has said it is capacity-constrained. If that is true, why would Google spend its resources selling its own capacity to other labs?
When asked if the money made from selling TPUs would help Google tamper some of its own spending, Ashkenazi didn’t give a clear yes or no answer. However, the answer seems likely to be no. Google is fronting the costs of building the hardware it sells to AI labs like Anthropic. Those contracts are supported by financial obligations that will come due when the hardware is ready to be used, but they don’t represent cash in Google’s pocket right now.
It may be worth noting that Microsoft was the only Big Tech company to escape Wall Street’s ire last week. The company posted $45.2 billion in capital expenditures for the quarter, down from an expected $50 billion. However, the lesser spend wasn’t due to any austerity on Microsoft’s part, but rather “third party capacity delays.”
Until last week, Microsoft had been the only hyperscaler that hadn’t taken on debt to fund its AI buildout. “We are fortunate to have access in the debt markets, given where our ratings are, and we have not yet, I would say, tapped that,” CFO Amy Hood said on the company’s earnings call Wednesday night.
That had changed by the next morning, when news of Microsoft’s $25 billion, 10-part bond sale broke.
By the time of this writing, Big Tech stocks were back up. Maybe Wall Street is banking on the “blue wave” of the midterms to force Big Tech to slow down some of its spending in the coming years. A slowdown might also give Big Tech a little more time to figure out how to make AI profitable.
Then again, despite the midterm elections looming last week, they did not come up on any of Big Tech’s earnings calls.
Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.
This is the title for the native ad
This is the title for the native ad
Sign in to see your notifications or create an account to join the conversation.
原文为英文,正文由自动抓取并保留原文。
原文发布于 The Verge AI(Big Tech’s big AI buildout will not end until someone stops it),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。