AI泡沫并不在你想象的地方

我是伊娜·加藤，每周在《欢乐时光》节目里，我都会邀请新老朋友来纽约的厨房桌边，一起喝杯酒，聊聊天。

I'm Ina Garten, and each week on Happy Hour, I invite new and old friends for a drink and a fun conversation at my kitchen table in New York City.

本周的节目，我请来了主厨保罗·卡迈克尔。

On this week's episode, I'm joined by Chef Paul Carmichael.

我们聊了很多，包括他工作过的那些了不起的餐厅，餐厅卫生间里最适合放的音乐，还有当他得知自己的餐厅卡巴瓦被《纽约时报》评为第一名时的反应。

We talk all about the incredible restaurants he's worked in, the best music to play in the restaurant bathrooms, and how he reacted when he found out his restaurant, Kabawa, had been ranked number one by The New York Times.

《欢乐时光》的新剧集每周三上线。

New episodes of Happy Hour are available every Wednesday.

你可以在YouTube上看，或者在你常听播客的地方听。

You can watch on YouTube or listen wherever you get your podcasts.

干杯。

Cheers.

我是梅根·拉皮诺。

Megan Rapinoe here.

本周的《你为什么这样？》节目里……

This week on Why Are You Like This?

我这次采访的是搞笑的乔什·约翰逊。

I am talking with the hilarious Josh Johnson.

我们深入聊了他职业生涯中的各种起起伏伏，这些经历让他最终成为《每日秀》的轮值主持人。

We dig into all the twists and turns of his career that landed him on "The Daily Show" as a rotating host.

他告诉我为什么每周都会发布一个新的喜剧专场，以及他更像妈妈、爸爸还是宝宝。

He tells me why he releases a new comedy set each week, and if he's more mommy, daddy, or baby.

快去听听最新一集的《你为什么这样？》

Check out the latest episode of Why Are You Like This?

你可以在任何播客平台和YouTube上收听。

wherever you get your podcasts and on YouTube.

大家好。

Hey, everybody.

我是苏·伯德，本周又带来了全新一集的《伯德视角》。

Sue Bird here, back with another brand-new episode of Bird's Eye View this week.

和我一起上节目的是杜克大学女子篮球队和美国队的教练卡拉·劳森。

Joining me on the pod is Kara Lawson, head coach of Duke University women's basketball and Team USA.

我和卡拉一起分析了2026年WMA决赛的进展。

Kara and I break down how the 2026 WMA finals are unfolding.

我们回顾了美国队夺得世界杯金牌的历程，并讨论了女子篮球运动的发展程度。

We revisit Team USA's path to World Cup gold and discuss how far the women's game has come.

另外，卡拉还透露了她执教过的最喜欢的球员。

Plus, Kara reveals her favorite player that she ever coached.

请在YouTube和所有播客平台上收听《Bird's Eye View》。

Check out Bird's Eye View on YouTube and wherever you get your podcasts.

欢迎收听《简单来说》。

Welcome to Simply Put.

我是埃德·埃尔森。

I'm Ed Elson.

本周，我们来聊聊两个泡沫的故事。

This week, a tale of two bubbles.

这个月初，人工智能领域简直乱套了。

Earlier this month, all hell broke loose in the world of AI.

事情始于一些令人不安的头条新闻。

It started with some unsettling headlines.

杰富瑞全球股票策略主管预测，引用原话，“AI领域将出现大规模资本毁灭。”

Jefferies global head of equity strategy predicted, quote, "massive capital destruction in AI."

后来，最大的芯片制造商之一博通的CEO警告说，AI投资者“根本不知道自己在投资什么。”

Later, the CEO of one of the largest chipmakers, Broadcom, warned that AI investors have, quote, "no clue what they're investing in."

这两句话之所以吓人，是因为它们很可能说的是实话。

Both statements were scary because they were probably true.

然而，压垮骆驼的最后一根稻草是CoreWeave，全球最热门的新数据中心提供商之一，也就是所谓的“新云”公司。

The straw that broke the camel's back, though, was CoreWeave, one of the world's hottest new data center providers or neo clouds.

简而言之，这家AI公司的债券陷入了危机状态。

In short, the AI company's bonds went into crisis mode.

其无担保债券的收益率飙升至15%，信用违约互换的利差也急剧扩大，暗示该公司在未来五年内违约的概率高达62%。

The yield on its unsecured notes exploded to 15%, and the spread on its credit default swaps also skyrocketed, implying a 62% probability that the company would default in the next five years.

信贷市场在响亮而明确地发出警告：这家公司可能即将崩溃。

The credit markets were screaming loud and clear, this company might be about to implode.

投资者为什么这么担心？

Why were investors so worried?

简单来说，这家公司借的钱比它可能承受得起的要多。

Simply put, the company borrowed more than it can probably afford.

正如D.A. Davidson的Gil Luria上周在《市场》节目中所解释的，CoreWeave不仅承担了500亿美元的债务，而且是以9%的利率借的，却只产生了1%的回报。

As D.A. Davidson's Gil Luria explained on Markets last week, CoreWeave had not only taken on $50 billion in debt, they'd done so at a 9% interest rate while generating a 1% return.

这就像用保证金贷款去买国债，Gil告诉我们。

That's like taking out a margin loan to buy Treasuries, Gil told us.

从财务角度看，这根本说不通。

From a financial perspective, it doesn't make sense.

事情看起来一切正常，直到突然出了问题。

Things seemed fine until they didn't.

由于内存价格上涨和政治反弹，数据中心建起来比最初预期的更贵、更难。

Thanks to memory prices and political backlash, data centers turn out to be more expensive and difficult to build than originally anticipated.

与此同时，流入AI生态系统的资金数量也在放缓。

Meanwhile, the number of dollars flowing into the AI ecosystem is also slowing.

中国是主要原因。

China is primarily responsible.

他们极其便宜的开放权重模型把AI代币价格砍了一半，迫使像OpenAI这样的前沿实验室降价高达80%。

Their extraordinarily cheap open weight models have cut AI token prices in half, forcing frontier labs like OpenAI to reduce their prices by up to 80%.

总之，建造AI的成本在上升，而销售AI的价格在下降。

In sum, the cost to build AI is going up, and the price to sell it is going down.

这可不太好。

Not good.

如果你关注Ed Zitron，我们Markets节目的常客之一，你早就等着这一天了。

If you follow Ed Zitron, one of our frequent Markets guests, you have been waiting for this.

他长期以来一直在警告像CoreWeave这样的新型云服务商可能会崩盘。

He has been warning about the implosion of neo clouds like CoreWeave for a long time.

虽然那个预测不一定成真，但警告信号确实开始显现了。

And while that prediction hasn't necessarily come true, the warning signs are definitely starting to show.

你可能会因此认为AI泡沫即将破裂，股市也快要崩盘了。

You might assume then that the AI bubble is about to burst, and therefore, that the stock market is about to crash.

我不会怪你有这种想法，因为泡沫通常都是这样发展的。

And I wouldn't blame you for that assumption, as that is generally what happens with bubbles.

但这一切都取决于你认为AI泡沫到底在哪里。

But all of this depends on where you think the AI bubble actually is.

如果你和我一样，你其实不太担心股市。

And if you're like me, you're not so worried about the stock market.

不是因为不存在AI泡沫，泡沫确实存在，而是因为泡沫不在股市里，它藏在别的地方。

Not because there isn't an AI bubble, there is, but because the bubble's not in the stock market, it is hiding somewhere else.

首先，让我解释一下为什么股市不会马上崩盘。

First, let me explain why the stock market isn't about to crash.

用一句话来说，AI崩溃的风险已经被市场消化了。

In one sentence, the risks of an AI implosion are already priced in.

虽然今年AI股票确实涨了很多，但那是因为它们的利润确实暴涨了。

While it's true that AI stocks have risen significantly this year, that is only because their profits have actually exploded.

与此同时，当你看看它们的倍数，也就是投资者愿意为这些盈利支付的溢价，你会发现估值实际上已经下降了。

Meanwhile, when you look at their multiples, i.e. the premium investors are willing to pay for those earnings, you realize that valuations have actually come down.

拿英伟达来说吧，它的市盈率自去年以来已经下降了48%。

Take Nvidia, for example, whose price-to-earnings multiple has fallen 48% since last year.

这说明投资者确实在担心些什么。

That means that investors are worried about something.

他们对英伟达未来每一美元的现金流都打了更大的折扣，因为他们觉得这些现金流有些地方不太可持续。

They're placing heavier discounts on every dollar of Nvidia's future cash flows because they believe that there is something unsustainable about them.

而且他们是对的。

And they're right.

数据中心占了英伟达业务的90%以上，考虑到AI公司的客户集中度以及他们融资成本的上升，认为有些客户可能会消失也不是没有道理。

Data centers make up more than 90% of Nvidia's business, and considering the customer concentration of AI companies as well as their rising financing costs, it's not unreasonable to think that some of their customers might drop off the map.

那是个很大的风险，但投资者也在给这个风险定价。

That is a significant risk, but it is a risk that investors are also pricing.

其他主要的芯片制造商也是如此。

The same is true for the other major chip makers.

在过去一年里，博通的市盈率下降了54%。

Over the past year, Broadcom's earnings multiple has fallen 54%.

美光的股价也下跌了56%。

Microns has fallen 56%.

SK海力士的股价下跌了49%。

SK Hynix's has fallen 49%.

他们的股价确实上涨了，但估值并没有。

Their stocks have gone up, yes, but their valuations have not.

与此同时，所有的新兴云计算公司都遭受了重创。

Meanwhile, all of the neo clouds have been taken to the cleaners.

CoreWeave在过去一年里下跌了52%。

CoreWeave is down 52% in the past year.

Iron的股价下跌了53%。

Iron's down 53%.

Cipher的股价下跌了55%。

Cipher is down 55%.

这些是典型的AI热门公司，一直是所有AI怀疑论者关注的焦点，但投资者普遍认为它们的未来充满不确定性。那大型科技公司呢？

These are the quintessential hot AI names that have been on every AI skeptic's watchlist, but investors are mostly in agreement that their futures are highly uncertain What about big tech?

就像我夏天写的，大科技公司今年大部分时间都处于AI的低迷期。

As I wrote over the summer, big tech has spent much of the year in the AI doldrums.

它们的估值倍数大幅收缩，因为担心管理层不完全理解AI，虽然我理解这些担忧，但我也觉得它们被夸大了。

Their multiples contracted significantly out of fears that management didn't fully understand AI, and while I understood those concerns, I also found them to be overdone.

因此，我在看起来是底部的时候，大量投资了最受惩罚的两家公司，Meta和微软。

As a result, I loaded up on the two most punished names, Meta and Microsoft, at what now appears to be the bottom.

从那以后，我的投资分别回报了25%和40%，虽然这对我来说很棒，但也引发了担忧，即热情再次变得过高。

Since then, my investments have returned 25% and 40% respectively, and while that is awesome for me, it also raises concerns that exuberance has once again gotten too high.

我的看法是，我们还好。

My take, we're still okay.

Meta、Alphabet、亚马逊和微软的平均盈利倍数，在剔除它们的私人投资收益后，略高于30，比S&P 500的平均值高出不到10%。

The average earnings multiple of Meta, Alphabet, Amazon, and Microsoft when you strip out their private investment gains is a little over 30, which is less than 10% higher than the S&P 500's average.

在我看来，这些不再是买入，但也不是卖出。

These are no longer buys in my view, but they're not sells either.

它们反映了一种谨慎的乐观，与互联网泡沫时期的狂热相比，简直微不足道，那时思科的市盈率超过100倍。

They reflect a cautious optimism that pales in comparison to the dot-com boom hysteria that resulted in Cisco trading at more than 100 times earnings.

那泡沫到底在哪里呢？

So where is the bubble then?

答案就在你眼前。

The answer is right in front of you.

AI领域最大的两个名字，同时也是估值最离谱的。

The two biggest names in AI are also the most overvalued.

OpenAI和Anthropic计划上市，估值分别是1.4万亿和2万亿美元。

OpenAI and Anthropic plan to go public at $1.4 trillion and $2 trillion respectively.

这样一来，它们将成为全球第7和第11大最有价值的公司，几乎相当于过去45年所有科技IPO的总和。

That will make them the 7th and 11th most valuable companies in the world, almost equal in value to every tech IPO of the past 45 years combined.

如果这些AI实验室每年能带来几千亿美元的收入，我还能接受，但事实并非如此。

I would be fine with this if the AI labs were bringing in hundreds of billions of dollars a year, but they're not.

恰恰相反，它们每年亏损几十亿美元，而且它们的商业模式能否真正行得通，还不清楚。

Quite the reverse, they're losing billions of dollars a year, and it's not clear if their business models are ever going to actually work.

但Anthropic不是刚说他们盈利了吗？

But didn't Anthropic just say they're profitable?

不，其实不是。

No, not really.

上周投资者日公布的财务数据显示，Anthropic 只是在所谓的“调整后”基础上盈利。

The financials offered at last week's Investor Day revealed that Anthropic is profitable only on a, quote, "adjusted basis."

“调整后”到底是什么意思？

What does adjusted actually mean?

没人知道，但它很可能剔除了收入分享协议以及他们最大的支出——训练成本，也就是他们用来构建产品所需的资金。

No one knows, but it very likely strips out revenue-sharing agreements as well as their largest expense by far, which is their training costs, i.e. the money they need to pay to build their product.

忽略这些成本简直愚蠢至极，几乎可以说是犯罪。

Ignoring those costs is so stupid, it almost seems felonious.

这就好比我在鞋里垫了增高垫，然后说“我身高六英尺六，但这是调整后的数据”。

It's equivalent to me inserting lifts in my shoes and saying, "I'm six foot six, but on an adjusted basis."

尽管有这些会计上的把戏，投资者们还是趋之若鹜。

Despite these accounting gymnastics, investors have gobbled it up.

按照2万亿美元的估值，Anthropic 的股价是其年化营收的近30倍。

At a $2 trillion valuation, Anthropic trades at nearly 30 times its revenue run rate.

我们没法用盈利来给它定价，因为它根本没有盈利。

We can't price it against earnings because it doesn't have any.

OpenAI 也是一样，估值1.4万亿美元。

OpenAI is in the same boat at $1.4 trillion.

这种价格只有奢侈包包的设计师才敢定。

These are the kinds of prices only luxury handbag designers can get away with.

和加密货币一样，只有存在愿意买单的更大傻瓜时，它才行得通。

As with crypto, it only works when there is a greater fool who is willing to pay it.

那为什么投资者对某些AI公司那么谨慎，对另一些却盲目乐观呢？

So how could it be that investors are so cautious about some AI companies, but so delusionally optimistic about others?

答案在于私人市场和公开市场之间的差异。

The answer lies in the divide between the private markets and the public markets.

私人市场的投资者完全是另一类人。

Private market investors are simply different people.

作为私人初创公司，OpenAI 和 Anthropic 的估值是由硅谷的风险投资家决定的。

As private startups, OpenAI and Anthropic have their prices set by venture capitalists in Silicon Valley.

那些人关心的是创始人的愿景和科技人才。

Those people care about things like founder vision and tech talent.

与此同时，像英伟达这样的上市公司的股价是由华尔街的投资者决定的。

Meanwhile, the prices of public stocks such as Nvidia are set by investors on Wall Street.

他们关心的是利润率和盈利。

They care about things like margins and profits.

这些性格差异导致了截然不同的价格。

These personality differences have led to vastly different prices.

换句话说，我们面临的不是人工智能泡沫，而是人工智能初创公司泡沫。

In other words, we don't have an AI bubble, we have an AI startup bubble.

如果Anthropic在公开市场上交易，它的估值会变成什么样？

What would happen to Anthropic's valuation if it traded in the public markets?

答案是，它的估值会减半。

Answer, its valuation would be cut in half.

当我们把英伟达当前的市值与销售额的倍数应用到Anthropic上时，这家AI实验室的估值会降到大约1万亿美元。

When we apply Nvidia's current price to sales multiple to Anthropic, the AI lab's valuation drops to roughly $1 trillion.

对OpenAI也做同样的操作，它的估值也会被砍半。

Do the same thing to OpenAI, and its valuation gets halved as well.

当我们审视公共市场与私人市场的分歧时，我们会发现Anthropic和OpenAI一直在价格发现的游乐场里玩耍。

When we examine the public versus private markets divide, we learn that Anthropic and OpenAI have been fighting in a playpen of price discovery.

许多其他AI初创公司也是如此。

The same is true of many other AI startups, too.

风险投资投资者最近认定，Thinking Machines的价值是Snap的四倍，尽管其收入只有Snap的六十分之一。

VC investors recently determined that Thinking Machines was four times more valuable than Snap despite generating a 60th of its revenue.

他们给另一家AI公司定价为320亿美元，尽管该公司还没有产品。

They priced another AI company at $32 billion despite not having a product.

私人市场的投资者一直在用啤酒漏斗狂饮AI的酷爱饮料，甚至不知道里面是什么。

Private markets investors have been guzzling AI Kool-Aid through a beer bong without even knowing what's in it.

好消息是，直到这些初创公司真正上市，他们才需要面对现实。

The good news is they won't have to reckon with the reality until these startups actually have to go public.

坏消息是，这即将发生。

The bad news is that that is about to happen now.

Anthropic 一个月后就要上市了。

Anthropic goes public in a month.

公司已经提交了招股说明书，据说一切准备就绪。

The company has filed its prospectus and everything is reportedly ready to go.

唯一的问题是管理层是否有胆量真正执行。

The only question is whether management has the stones to actually follow through.

一旦进入公开市场这个竞技场，Anthropic 的投资者就会明白价值的真正含义。

Once they enter the coliseum that is the public markets, Anthropic's investors will find out what value truly means.

我向你保证，Dario Amodei 已经考虑过这一点。

Dario Amodei has considered this, I assure you.

他知道初创公司的估值都是胡说八道。

He knows that startup valuations are bullshit.

他知道风险投资家生活在另一个现实中。

He knows that VCs live in a different reality.

他把公司建立在一个充满乐观主义者的星球上，这些人接受一切，不质疑任何事。

He has built his company on a planet of Pollyannas who accept everything and question nothing.

但他即将离开那个世界，进入一个更残酷的世界。

But he is about to exit that world and enter a much harsher one.

他知道这一点，我敢保证，他对此很紧张。

He knows this, and I guarantee you, he is nervous about it.

这就是为什么如果下周我们看到这样的头条新闻，我也不会感到惊讶："Anthropic 推迟 IPO，理由是市场不确定性。"

That is why I wouldn't be surprised if sometime next week we read the following headline, "Anthropic delays IPO, cites market uncertainty."

因为要么是这样，要么就得降价，我怀疑阿莫迪先生对后者不感兴趣。

Because it's either that or the price must come down, and I doubt that Mr. Amodei is interested in the latter.

本期《简单说》由克莱尔·米勒和艾莉森·韦斯制作，娜塔莉亚·卡布雷拉编辑，本杰明·斯宾塞负责工程。

This edition of Simply Put was produced by Claire Miller and Allison Weiss, edited by Natalia Cabrera, and engineered by Benjamin Spencer.

我们的研究助理是丹·沙隆。

Our research associate is Dan Chalon.

我们的设计编辑是奥利维亚·雷尼·霍尔。

Our design editor is Olivia Rennie Hall.

特别感谢薇薇安·姜和罗希特·博拉。

Special thanks to Vivian Jiang and Rohit Bhola.

感谢收听Ed Elson的Simply Put节目。

Thank you for listening to Simply Put by Ed Elson.

你可以在Substack上找到这篇文章以及更多内容。

You can find this piece and many more on Substack.

只需访问edwardelson.substack.com。

Just go to edwardelson.substack.com.