The markets are starting to act seriously weird. Blame AI市场开始出现非常奇怪的现象。这都怪人工智能。
AI’s shaky balance sheets raise questions about circular financing that investors have long questioned but the industry and many traders have swept under the rug.

Spencer Platt/Getty Images
On the surface, the markets seem stable. Stocks are near record highs, and bonds are right about where they’ve been for the past couple of months.
But check under the hood, and you’ll see signs that investors are growing increasingly nervous — mostly because of AI.
This time, the anxiety isn’t because AI is going to kill us all (that was SO last month). The latest concern: AI may be a financial quagmire.
And if the air is starting to come out of the AI bubble, it could threaten to knock over the foundation supporting America’s economic boom.
The stock market is so big and diverse that it’s sometimes useful for investors to measure it in creative ways, comparing two measurements to figure out if there’s a trend that the headline numbers can’t show you.
For example, you can measure the benchmark S&P 500 stock index in a number of ways. The most common way is to weight each stock by its market value — giving gigantic tech companies worth trillions of dollars considerable influence over the ups and downs of the stock market.
But if you look at the S&P 500 another way, weighting each stock equally, you can sometimes get a clearer picture of how the market is performing across industries. Equally weighting stocks ensures companies in smaller industries aren’t drowned out by massive ones, like Big Tech.
When comparing those two measurements on October 22, the result was seriously bizarre — so strange it had only happened twice before in the past three decades.
The main index, weighted by market value, fell 1.2%. But the equal-weighted index gained 0.5%. Since 1995, that massive divergence in the two measurements had happened only twice before: April 14, 1999, and March 30, 2000, noted James Reilly, senior markets economist at Capital Economics.
What does it mean? A clue can be found in another weird thing that happened that same day.
Also on Thursday, the 10-year Treasury yield slammed on the brakes and went into reverse, falling below 5% and tumbling in the largest single-day decline since March 10, 2023 — the day that Silicon Valley Bank imploded.
So investors were pouring money out of tech and into Treasuries and non-tech stocks at a historic pace.
It’s a sign that AI is making some folks on Wall Street kind of nervous.
Bonds had been tumbling (and yields, as a result, had been surging to multi-decade highs) for three core reasons:
America’s ballooning debt
The Iran war sent inflation on a serious rebound, cutting off the world’s oil and fuel supplies and sending prices through the roof. Bond investors, when confronted with rebounding inflation, demanded higher yields to compensate for the chunk of their investment that inflation has been eating into.
Atta Kenare/AFP/Getty Images
Meanwhile, America’s debt has skyrocketed to $40 trillion, and its budget deficit surged to $2 trillion over the past fiscal year. Politicians across the political spectrum continue to advocate for populist policies, including lower taxes and higher spending, giving investors no hope for a wave of fiscal responsibility soon to take over Washington. That means Treasury will need to keep borrowing — a lot, flooding the market with bonds.
And enormous AI investment has underpinned America’s sustained economic growth in recent years. But the infrastructure needed to power AI is expensive — around $1 trillion a year and growing. Tech companies are borrowing huge amounts of money from investors, in the form of bonds, crowding out demand for other debt, including Treasuries.
None of this has fundamentally changed.
In fact, the Iran war, inflation and the debt problems are potentially about to get even worse. Oil prices are at four-year highs, and fuel prices are exploding higher in some parts of the United States. And President Donald Trump has signaled the costly Iran war is about to kick into high gear again after the midterms.
Even with that backdrop, government bonds, which investors have turned their backs on for months, suddenly grew popular again.
Investors recently started asking some serious questions about the AI part of this three-legged stool that had been supporting bond yields. That wobbly AI leg made folks worried.
And when traders feel nervous, they come running home for safety. And the safest home on the planet offers the security of the US dollar and Treasuries.
If we are in the midst of a paradigm shift, the turning point may have begun with CoreWeave .
The neocloud company, which rents out computing power to AI companies, is fueled by debt. Its borrowing costs ballooned to an astounding 15% earlier this month as investors grew concerned about its financial prospects. That dwarfs its rate of return on its investments.
In other words: CoreWeave is losing a ton of money and may not be able to pay off its debt.
CoreWeave may be an outlier — or it may not be. A Wall Street Journal report on Anthropic’s draft IPO filing showed a nearly $81 billion loss in the first half of 2026 — on pace to dwarf the record for the largest-ever annual loss reported by a public company (if Anthropic goes public this year). Meta reported that it spent $2 billion more cash than it produced in the third quarter, sending its stock tumbling 7% Thursday.
AI’s shaky balance sheets raised questions about that whole circular financing thing that investors have long questioned but the industry and many traders have swept under the rug. If every tech company is invested in every tech company, is this all just a highly leveraged house of cards? And what happens when one of those cards falls?
So some traders are pausing to re-think how much value Wall Street has been placing on AI — and whether these debt levels are sustainable.
“What we’re seeing is elevated volatility around something that looks and feels and smells like a bubble,” said Joe Brusuelas, chief economist for RSM US.
Stephanie Keith/Getty Images
What’s not clear is how long this AI fizzle-out will last. The Nasdaq is in the midst of a two-week-long, 6% slide, but we’ve been down this road before, with investors growing nervous around past earnings seasons that tech companies are spending too much on AI — only to pour money back into AI stocks when some companies report signs of sustained demand.
And even if the AI bubble popped, it wouldn’t necessarily mean the economy would go down with it, Reilly noted. Each financial bubble has its own distinct character.
In recent years, economists have described the economy as in a “rolling recession,” as various sectors boomed and others busted, balancing one another out and preventing the entire economy from entering a downturn.
This time may be the same. Or different. No one knows.
So the market went a little haywire. Bond yields tumbled. And the stock market did something it has done only three times since 1995.
It may be the start of something –– or it may be just another bump in the road.
Spencer Platt/Getty Images
表面上看,市场似乎很稳定。股票价格接近历史高位,债券价格也与过去几个月基本持平。
但深入了解后,你会发现投资者越来越紧张的迹象——这主要是因为人工智能。
这次的焦虑并非源于人工智能会毁灭人类(这种说法早就过时了)。最新的担忧是:人工智能可能会引发金融危机。
如果人工智能泡沫开始破裂,可能会威胁到支撑美国经济繁荣的基础。
股市规模庞大且种类繁多,因此投资者有时需要采用创造性的方法来衡量它,比较两种衡量方法,以找出主要数据无法显示的趋势。
例如,衡量标普500指数的方法有很多种。最常见的方法是按市值对每只股票进行加权——这使得市值数万亿美元的科技巨头对股市的涨跌拥有相当大的影响力。
但如果你换个角度来看待标普500指数,给每只股票赋予相同的权重,有时就能更清晰地了解市场在各个行业中的表现。等权重分配可以确保规模较小的行业公司不会被大型科技公司等巨头的光芒所掩盖。
10 月 22 日对这两次测量结果进行比较后,发现结果非常奇怪——这种奇怪的现象在过去三十年中只发生过两次。
按市值加权的主要指数下跌了1.2%,但等权重指数上涨了0.5%。凯投宏观(Capital Economics)高级市场经济学家詹姆斯·赖利(James Reilly)指出,自1995年以来,这两种衡量指标出现如此巨大分歧的情况此前只发生过两次:分别是1999年4月14日和2000年3月30日。
这是什么意思?或许可以从当天发生的另一件怪事中找到线索。
周四,10 年期美国国债收益率急刹车并开始逆转,跌破 5%,创下自 2023 年 3 月 10 日(硅谷银行倒闭之日)以来最大的单日跌幅。
因此,投资者以前所未有的速度将资金从科技股抛售到国债和非科技股。
这表明人工智能让华尔街的一些人感到有些不安。
债券价格暴跌(收益率也因此飙升至数十年来的最高水平)主要有三个原因:
美国不断膨胀的债务
伊朗战争导致通货膨胀急剧反弹,切断了全球石油和燃料供应,使物价飙升。债券投资者面对反弹的通胀,要求更高的收益率,以弥补通胀侵蚀的投资损失。
Atta Kenare/AFP/Getty Images
与此同时,美国的债务飙升至40万亿美元,上个财政年度的预算赤字也高达2万亿美元。各政治派别的政客继续鼓吹民粹主义政策,包括减税和增加支出,这让投资者对华盛顿即将迎来财政责任浪潮不抱任何希望。这意味着财政部需要继续借贷——而且借贷规模将大幅增加,从而向市场注入大量债券。
近年来,巨额人工智能投资支撑了美国经济的持续增长。但支撑人工智能发展的基础设施造价不菲——每年约1万亿美元,而且还在不断增长。科技公司正以债券的形式从投资者那里借入巨额资金,挤占了对其他债务(包括国债)的需求。
这一切本质上都没有改变。
事实上,伊朗战争、通货膨胀和债务问题可能还会进一步恶化。油价已达到四年来的最高点,美国部分地区的燃油价格也在飙升。唐纳德·特朗普总统已暗示,中期选举后,代价高昂的伊朗战争将再次升级。
即便在这样的背景下,投资者几个月来一直冷落的政府债券突然又开始受到欢迎。
投资者最近开始对支撑债券收益率的这三条腿中的人工智能部分提出一些严肃的问题。人工智能这“腿”的不稳引发了人们的担忧。
当交易员感到紧张时,他们会寻求避险。而地球上最安全的避风港就是美元和美国国债。
如果我们正处于范式转变之中,那么转折点可能就始于 CoreWeave。
这家向人工智能公司出租计算能力的云服务公司主要依靠债务驱动。本月初,由于投资者对其财务前景感到担忧,其借贷成本飙升至惊人的15%。这远远超过了其投资回报率。
换句话说:CoreWeave 正在亏损巨额资金,可能无法偿还债务。
CoreWeave或许是个例外,也或许并非如此。《华尔街日报》报道Anthropic的IPO申请草案显示,该公司2026年上半年亏损近810亿美元,如果Anthropic今年上市,这一数字将远超上市公司有史以来最大的年度亏损纪录。Meta公司公布第三季度现金支出比收入多出20亿美元,导致其股价周四暴跌7%。
人工智能公司摇摇欲坠的资产负债表引发了人们对循环融资模式的质疑,投资者长期以来一直对此抱有疑虑,但行业和许多交易员却对此视而不见。如果每家科技公司都投资于其他科技公司,这难道不是一个高杠杆的纸牌屋吗?当其中一张牌倒下时,又会发生什么?
因此,一些交易员开始停下来重新思考华尔街对人工智能的重视程度,以及这些债务水平是否可持续。
RSM US首席经济学家乔·布鲁苏拉斯表示:“我们看到的是围绕着一个看起来、感觉起来、闻起来都像泡沫的东西而出现的剧烈波动。”
Stephanie Keith/Getty Images
目前尚不清楚人工智能热潮会持续多久。纳斯达克指数正经历着为期两周、跌幅达6%的下跌,但我们以前也经历过类似的情况:在过去的财报季,投资者们对科技公司在人工智能领域的投入过高感到担忧,但当一些公司公布人工智能需求持续增长的迹象时,他们又会重新将资金投入人工智能股票。
赖利指出,即便人工智能泡沫破裂,也未必意味着经济会随之崩溃。每个金融泡沫都有其独特的特征。
近年来,经济学家将当前经济描述为“滚动衰退”,各个行业蓬勃发展,而其他行业则衰落,彼此相互平衡,防止整个经济陷入衰退。
这一次或许会和上次一样,或许会不一样。谁也说不准。
于是市场出现了一些混乱。债券收益率暴跌。而股市也出现了自1995年以来仅出现过三次的情况。
这或许是某个新事物的开始,也或许只是前进道路上的又一个坎坷。