本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
我直到去年还在做公关公司，算是以合同工的形式在做。
Like, I ran a PR firm until, like, last year, and I was, like, doing that as kind of contract labor thing.
然后我从2020年开始写东西，就一直写个不停。
And then I started writing this in 2020, and I just kept going and going and going and going.
一开始我主要写管理理论方面的内容，不太喜欢远程办公，觉得那些文章对它的看法太负面了。
And then, initially, I was writing about, like, management theory stuff, not much I liked remote work and how evil the articles were about it.
后来我又写了加密货币，还写了埃隆·马斯克。
Then I wrote about crypto, I wrote about Elon.
到了这两件事的尾声，我对它们俩都感到彻底沮丧。
And by the end of those two things, I was so thoroughly depressed with both.
我当时就想，我到底要
Was just like, what am I gonna
写点什么呢？
write about?
是说选题方面，还是写作本身呢？
Well, with the topics or with writing?
是话题的问题。
With the topics.
对。
Yeah.
那些话题太让人沮丧了。
The topics were so depress.
比如，每周都要关注埃隆·马斯克在做什么，真是让人无语。
Like, every week following what Elon Musk does and just ugh.
然后元宇宙出现了。
And then the metaverse happened.
写那个还挺有意思的，但那是在那之前。
That was kinda fun to write about, but that was before that.
但后来OpenAI，抱歉。
But then OpenAI got like, sorry.
山姆·奥特曼被OpenAI解雇了。
Sam Altman got fired from OpenAI.
我在推特上看到一些记者，他们简直像天塌了一样。
And I saw on Twitter that there were journalists who were like, oh my god.
他们希望他能被请回来。
I hope they bring him back.
他们谈论他的方式，就像他是个摇滚明星一样。
They were, like, talking about him like a rock star.
就好像，哦，然后每个人都在那里，有人说，哦，我哭了，因为看到大家这么支持山姆·奥特曼，真是太感人了。
It was like, oh, then everyone's there was someone like, oh, I'm crying because how beautiful it is that Sam that everyone's showing support for Sam Altman.
乔什都哭了。
Josh cried.
记者们。
Journalists.
对。
Yeah.
记者们。
Journalists.
这太疯狂了。
It's crazy.
所以我看到这个，我甚至都没去评判他们。
So I saw this, and I just I didn't even judge them for it.
我当时就想，行吧。
I was like, fine.
无所谓。
Whatever.
你。
You.
但我觉得这很奇怪。
But I was like, this is weird.
对。
Yeah.
你们为什么表现得像……这到底是什么意思？
Why are you acting in like, was like, what does this mean?
你为什么这样表现？
Why are you acting this way?
所以我一直很专注地看，然后我立刻发现大家都在说，AI是最大、最厉害、最好的东西。
And so I kept looking really intently, and I just immediately saw this thing where everyone was saying, AI is the biggest, most hugest, best thing ever.
这是不可避免的。
It's inevitable.
等等等等。
Blah blah blah.
但我找不到任何收入？
But I couldn't find any revenues?
就像，我找不到，每次我都想，好吧。
Like, I couldn't find and every time I was like, okay.
太好了。
Great.
但它能赚多少钱呢？
But how much does it make?
然后每个人都说，我、我实在没法说。
And everyone was like, I I couldn't possibly say.
微软赚了多少？
How much does Microsoft?
他们在资本支出上花了数十亿美元。
They're spending billions of dollars on CapEx.
这能赚多少？
How much does this make?
没法告诉你。
Can't put couldn't tell you.
没人愿意告诉你。
No one wants to tell you.
对。
Right.
但每个财报季都有这些文章说微软的人工智能押注得到了回报。
But you kept having these articles every earning season saying Microsoft's AI bet pays off.
谷歌的AI押注也见效了。
Google's AI bet pays off.
尽管他们从未提过AI带来的收入。
Even though they never mentioned their AI revenue.
我就从这儿开始自学经济学了。
And I just kind of taught myself economics from there.
说实话，很多东西都是自学的，这也让那些讨厌我的人很头疼，毕竟这些都是我自己摸索出来的。
Like, a lot of this like, everything's self taught, which to the chagrin of my many haters is just, like, stuff I learned.
是的。
Yeah.
不过通过读书之类的途径，你能学到很多东西。
But you can learn a lot by reading books and such.
好的。
Okay.
那你还好吧？
So you alright.
所以你用的是，我猜，是Substack吧？
So you're I guess, it's it's a Substack?
或者是Ghost。
Or Ghost.
好的。
Okay.
所以你现在做的事情就是到处被人阅读。
So what you're doing is being read everywhere.
对。
Yeah.
人们在分享它，看好AI的人都在读。
People are sharing it, and bulls on AI are reading it.
因为我觉得，我对市场的看法是，大多数专业投资者都足够聪明，会听取双方的意见。
Because I think so my view of the market is most professional investors are intelligent enough to listen to both sides.
是的。
Yeah.
他们不需要同意读到的每一条内容，才会想继续读下去。
And they don't have to agree with everything they read in order to wanna keep reading it.
对。
Right.
我觉得散户投资者的情况可能不是这样，但我们不用管那个。
I don't think that's the same as true for retail investors, but we're not gonna concern ourselves with that.
我认为专业人士确实喜欢听两边的观点，尤其是当他们投入了大量资金的时候。
I think professionals do like to hear both sides, especially if they have a lot of money on the line.
这确实是我和一些银行、机构打交道时的亲身经历。
That's actually been my experience talking to, like, some of the banks and the institutions.
他们，即使是那些非常看好AI的，也愿意听
Like, they they wanna even the ones who are very pro at
听这些。
with that.
是的。
Yeah.
好的。
Alright.
我们也是这样。
We're the same way.
所以我们一直在做这个节目。
So we've we've been doing this show.
我们从2021年夏天开始，每周四录制。
We tape every Thursday going back to the summer of twenty twenty one.
五年来，我们的大多数嘉宾都是看涨的。
And for five years, the majority of our guests are bullish.
到目前为止，大部分时候都还不错。
And so far, so good for the most part.
对。
Yeah.
好的。
Okay.
我们也请过一些看空的人，但他们不是针对某个特定领域的看空。
We have had people who are bearish, but they are nonspecifically bearish.
比如我们请过杰里米·格兰瑟姆，他因为不认同“永久看空”这个标签而出名，但他几乎总是看空的。
So we've had people like Jeremy Grantham, who's become famous for he he takes issue with the permabear label, but he's almost always bearish.
我们请过这样的人上节目，我一直觉得这很有意思，但我们还真没请过专门针对AI的看空者。
So we've had people like that on the And I always find that fascinating, but we haven't really had specifically an AI bear.
对。
Right.
就是说，那些来上节目的人，大多数是投资组合经理或者分析这个领域的分析师。
Somebody that like specific you know, most of the people that come on portfolio managers or they're analysts covering the space.
嗯。
Mhmm.
而且你知道，他们都有动力去……对。
And you know all the incentives to Yeah.
看到更光明的一面。
See the brighter side.
所以这将是对我们在节目中做过的所有事情的一个有趣考验，对吧。
So this will be an interesting test of all of the things that we've done on the show Yeah.
听听另一方的观点。
To hear the other side.
所以我们真的很期待。
So we're really excited about it.
我很兴奋。
I'm pumped.
好的。
Okay.
太棒了。
Awesome.
多头们觉得你是白痴，还是他们通常喜欢你对他们的说法？
Do bulls think you're an idiot, or do they generally, like, like what you're saying to them?
所以这已经到了一个奇怪的地步。
So it's got to a weird point.
所以如果我们把各种类型的牛市都混在一起的话，我猜很多都是散户。
So if we're mixing in every kind of bull here, so I imagine a lot of them are retail bags.
他们之所以这样，是因为我以前是做公关的。
They are driven like, because I used to be in PR.
他们就觉得，他不过是个做公关的。
They're like, he's just a PR boy.
他就是个公关人员。
He's just a PR guy.
他不该谈论这些。
He shouldn't talk about it.
别跟他说话了。
Stop talking to him.
他可能会说些外行话，好像金融专家从来没犯过错似的。
He could say something as if financial experts have never got anything wrong before.
有意思的是，有些人其实对半导体信贷挺看好的，比如在推特上关注我的那个Bubble Boy，他在半导体领域真的很厉害，非常聪明。
There are some of like, what's weird is, like, there are some of them who are quite bullish on semi credits, like, who follow me on Twitter, like Bubble Boy, who's really great and very, very smart in semis.
我不确定他是否完全同意我的观点，但他非常友善和友好，而且像他这样的人有很多。
I'm not sure he agrees with me on everything, but very kind and friendly, and, like, there are lots of them who are.
有些人是卑鄙可怕的，但这种情况比比皆是。
Some of them are vile and horrible, but so say many such cases.
对。
Yeah.
所以关于看多者，那些说‘我爱这项技术’的人，
And so the thing is with the Bulls is the ones who are like, I love this tech.
我对未来有一些疯狂的梦想，但确实存在泡沫。
I have some crazy dream what it'll be, but there's a bubble.
我尊重那些人。
I respect those.
如果你能热爱AI，我仍然认为他们没考虑到巨大的环境和社会问题，但至少你活在现实中。
If you can, like, love a I still think there are massive environmental and social problems that they're not thinking of, but at the very least, you're living in reality.
我认为那些认为Anthropic和OpenAI会增长到2840亿美元收入的人。
I think the ones who are, like, Anthropic and OpenAI are gonna grow to $284,000,000,000 in revenue.
这是OpenAI对2030年的预测。
That's what OpenAI is projecting in 2030.
那些看到这个预测的人会说，太好了。
The people who read that and they're like, yes.
我觉得没问题。
Sounds good to me.
这些人没有活在现实中，我也不需要它成为
Those people are not living in reality, and I don't need it to be
那个数字？
the number?
或者，如果你相信他们的预测，OpenAI将在三年半内变得比Meta还大，而且他们还要在运营支出是Meta两倍多的情况下做到这一点。
Or OpenAI is gonna become the gonna become bigger than Meta in three and a half years if you believe their projections, and they're going to do that while spending more than twice of Meta's OpEx.
我觉得，微软的运营支出大概是1500亿美元，而OpenAI在2030年将在计算上花费2000多亿甚至更多。
I think well, mate, I think Microsoft's OpEx is, like, 150,000,000,000, and OpenAI is gonna spend 200 something billion or more on compute in 2030.
要达到3800亿美元
To get to $380,000,000,000
的收入。
in revenue.
284。
284.
2800亿美元。
280,000,000,000.
那你是彻底看空，还是觉得那些家伙在抽大麻，还是我说的是一回事？
So are you outright bearish, or you think the balls are smoking crack, or am I saying the same thing?
可能是一回事。
Possibly the same thing.
我觉得在2022年之后，科技行业变得非常惨淡，不知道你还记不记得。
I think the after 2022 when everything got really grim in tech I don't know if you remember.
比如，英伟达在2023财年业绩平平。
Like, Nvidia had a flat year of fiscal twenty three.
当然记得。
Of course.
那段时间真的很难熬，大家都在拼命吹牛。
Really rough time, and everyone was kind of blowing hard.
微软、谷歌、亚马逊、Meta 都提高了价格，还改了广告拍卖的规则。
Microsoft, Google, Amazon, Meta, cranked up prices, changed ad auction stuff.
亚马逊其实做出了一个意外健康的广告业务。
Amazon actually made a surprisingly healthy ad business.
他们得想办法增加收入，而且这时候 ChatGPT 也出现了。
Like, they had to find ways to grow revenue and also ChatGPT happened.
于是他们就说，哦，那我们买一堆 GPU 吧。
So they went, oh, we'll buy a bunch of GPUs.
市场马上就反应说，所有这些收入增长都来自 AI。
And the market immediately was like, all of this revenue growth is coming from AI.
因为这些公司从不披露他们的 AI 收入，记者和分析师也没那个本事去追问，所以大家就说，哦，这是 AI。
And because the companies never disclosed their AI revenues and journalists don't have the teeth to bother them or analysts, especially, people were like, oh, it's AI.
每次他们花这笔钱，都说是 AI。
Every time they spend this money, it's AI.
这确实是个好消息。
And this was really good.
这让他们的股价疯狂上涨。
It helped their stocks pop crazy style.
比如，对我工作的一个合理批评是，如果你根据我的工作来做交易——我从来没说过要这样做，我不提供财务建议。
Like, the fair one fair criticism of my work is if you traded off of my work, which I've never said to do, I do not give financial advice.
对。
Yeah.
你大概
You probably
嗯，但如果你最终在这件事上是对的，他们会拿枪指着你的头，让你开个对冲基金。
Well, but if you end up being right about this, they're gonna put a gun to your head and say start a hedge fund.
我的意思是，天哪。
I mean oh, Christ.
我甚至不知道怎么做。
I don't even know how to do that.
但我只是告诉你这个。
But I'm just telling you that.
那听起来挺有意思的。
That sounds fun.
但重点是，我真的很喜欢写作。
But the point is, I'm I like writing a lot.
我真的很享受深入其中的过程。
I really enjoy actually getting into it.
这确实在智力上非常吸引人。
It's genuinely intellectually fascinating.
但我要说的是，直到2025年左右，这基本上就是买GPU，然后希望OpenAI和Anthropic能发展起来，也希望他们能盈利。
But the point I'm making is up until about 2025, it was mostly just buying GPUs and hoping OpenAI and Anthropic grew into it and also hoping they'd become profitable.
但后来发生的是，超大规模云服务商现在在财务上已经依赖OpenAI和Anthropic的增长了。
But then what happened is the hyperscalers have now become financially dependent on the growth for OpenAI and Anthropic.
分析师预期，比如瑞银、巴克莱和富国银行，预计谷歌、亚马逊和微软的云收入中有4400亿美元将来自OpenAI和Anthropic，而这两家都是需要不断融资的未盈利初创公司。
Analyst expectations, from, like, UBS, Barclays, and, Wells Fargo have $440,000,000,000 of cloud revenue across Google, Amazon, and Microsoft coming just from OpenAI and Anthropic to unprofitable startups who need to constantly raise money.
而且，大公司还需要建立数据感知能力才能赚钱。
And, also, the big companies need to build the data sense to make the money.
所以，所有这些事情都需要发生，而这些事情都非常不可能，甚至是不可能发生的。
So there's all of these very improbable, if not impossible, things that need to occur.
而且，OpenAI 和 Anthropic 还需要有现在十倍的需求量。
And, also, OpenAI and Anthropic need to have 10 times the demand they have right now.
不管你对这些公司怎么看，它们都还不够大。
It's not even however you may feel about these companies, they are not big enough.
它们现在产生的现金流，连它们承诺的1.1万亿美元以上的债务都远远覆盖不了。
They are not generating enough cash right now to even get close to covering their 1.1 or more trillion dollars in commitments.
而且，它们还占据了90%的AI基础设施。
And on top of that, they're taking up 90% of AI infrastructure.
这造成了一种虚假的需求信号，因为现在GPU很难买到。
This is creating an illusory demand signal because right now it's difficult to get GPUs.
人们会说，哦，哦，这是因为AI的需求太大了。
People are saying, oh, oh, it's because there's so much demand for AI.
这上面有个大大的巴里·邦兹式的星号。
There's a big Barry Bonds asterisk at the top of that.
基本上，AI的需求主要来自两家公司。
There's so much demand from AI from two companies pretty much.
Meta也算一个，但他们其实没在做什么。
Meta as well, but they're not doing anything.
他们只是在自己那摊烂泥里打滚。
They're just they're just rolling in their filth.
所以现在的情况是，大家都在说，哇，AI需求巨大。
But so what's happening is everyone's going, oh, there's tons of demand for AI.
我要建一堆数据中心。
I'm gonna build a bunch of data centers.
尽管在2026财年，微软向彭博社透露他们在AI上赚了约343.3亿美元。
Even though when you so in fiscal year twenty six, Microsoft, they made going to Bloomberg about $34,330,000,000 on AI.
其中241亿美元来自OpenAI，这意味着微软的OpenAI软件和GPU租赁服务其实只是几十亿美元的小生意。
$24,100,000,000 of that is OpenAI, which means the OpenAI, Microsoft software and GPU rental services are single digit billion dollar businesses.
而且他们在这个上面投入了超过2600亿美元的资本支出。
And they have 260 plus billion dollars in CapEx in this.
这简直是一场灾难，而且不是那么容易解决的——我觉得根本解决不了。
It's a disaster, and it's not something that gets easily I don't I don't think it gets fixed at all.
而且，AI GPU还挺专业的。
And on top of it, AI GPUs are pretty specialist.
它们主要用于AI、数据分析和三维建模，这些业务规模都很小。
They're useful for, like, AI and data analytics and three d modeling, which is a very small business.
这之后不会再有互联网泡沫那种事了。
There's no .com bubble thing after this.
电费会一样贵，甚至更贵。
The electricity will be just as expensive, if not more.
哦，等等。
Oh, wait.
你真是滔滔不绝啊。
You're on a roll.
想给大家一个机会平掉他们的多头仓位，我们先把节目正式介绍一下，然后马上回来继续深入讨论。
Wanna give people the chance to close out their long positions, and we're gonna introduce the show properly, and then we're gonna dive right back in.
谁，谁来给我倒计时？
Who's who's doing my countdown?
妮可小姐？
Miss Nicole?
好的。
Alright.
本期节目由富达投资赞助播出。
This message is brought to you by Fidelity Investments.
当时机至关重要时，你需要强大的工具和研究，能够随时满足你的需求。
When timing is everything, you need powerful tools and research that can meet you in the moment.
借助全新的富达交易者Plus平台，你的图表和偏好设置会保持一致，并在所有设备上同步，这样无论你在哪里交易，都能迅速行动。
With the all new Fidelity Trader Plus platform, your charts and preferences show up consistently, synced up across all your devices so you can act fast whenever and wherever you're trading.
你可以在家里的台式电脑上保存订单，然后
You can save an order on your desktop at home, get
在上班时收到手机提醒，然后直接在Fidelity应用里完成交易，不用重新开始。
a mobile alert when you're at work, and complete the trade in the Fidelity app without starting over.
而且通过可下载的Fidelity Trader Plus桌面平台，你可以用多显示器视图、增强工具和自定义选项获得更多控制，还能与Fidelity交易专家进行集成屏幕共享。
And with the downloadable Fidelity Trader Plus desktop platform, you have more control with multi monitor views, enhanced tools, and customization options, and integrated screen sharing with Fidelity Trading Specialists.
试试Fidelity目前最强大的交易平台，访问fidelity.com/traderplus。
Try Fidelity's most powerful trading platform yet at fidelity.com/traderplus.
Fidelity Investments和The Compound没有关联。
Fidelity Investments and The Compound are not affiliated.
讨论的观点、意见、产品、服务和策略均未得到Fidelity Investments的认可或推广。
Views, opinions, products, services, and strategies discussed are not endorsed or promoted by Fidelity Investments.
Fidelity Brokerage
Fidelity Brokerage
Services LLC。
Services LLC.
NYSE SIPC成员。
Member, NYSE SIPC.
好的。
Alright.
各位听众，欢迎收听全球最棒的投资播客。
Ladies and gentlemen, welcome to the world's greatest investment podcast.
我说这话可不是在开玩笑。
I say that without a trace of irony.
我是认真的。
I really mean it.
今天，你们将享受到一份特别的惊喜。
Today, you're in for a very special treat.
我们即将进行一场以前从未在这里展开过的对话。
We are having a conversation that we have not really had here before.
关于人工智能，最极端的看空观点是什么？
What is the ultra bear case on AI?
为什么有人在鼓掌？
Why are people clapping?
这太疯狂了。
It's crazy.
这可是关系到你的退休金。
It's your retirement on the line.
怎么回事？
What's going on?
各位，冷静点。
Comp guys, calm down.
好的。
Alright.
今天我们的嘉宾是Ed Zitron。
Joining us today, Ed Zitron.
Ed是EZPR的创始人兼CEO，这家研究和媒体分析公司是他2013年创立的。
Ed is the founder and CEO of EZPR, a research and media analysis firm he launched in 2013.
他还主持了获得Webby奖的Better Offline播客，这个节目是和iHeartRadio以及Cool Zone Media合作的。
He also hosts the Webby award winning better offline podcast with iHeartRadio and Cool Zone Media.
埃德写了一份名为《你的埃德在哪？》的通讯。
Ed writes a newsletter called Where's Your Ed At?
他在2020年创办了这份通讯，其中他2023年的《腐烂经济》文章——我待会要讲这个——主张科技公司为了增长而牺牲了创新和用户，这篇文章已经火了。
That he launched in 2020, where his 2023 rot economy essay, I'm gonna get into that, Arguing tech firms have chased growth at the expense of innovation and users has gone viral.
如今，他是最响亮、被引用最多的AI怀疑者之一，专注于OpenAI、Anthropic以及超大规模资本支出的单位经济学。
Today, he's one of the loudest and most cited AI skeptics focused on the unit economics of OpenAI, Anthropic, and the hyperscaler CapEx build out.
埃德，感谢你来到我们节目。
Ed, thank you for being here.
我们很感激。
We appreciate it.
谢谢你们邀请我。
Thanks for you for having me.
好的。
Alright.
好的。
Alright.
所以我们得从英伟达说起。
So we have to start with Nvidia.
对。
Yeah.
我觉得他们正在证明你说的那些都不对。
I feel like they're they're proving everything that you've said to be not correct.
我知道。
I know.
但是，不。
But just No.
不。
No.
不。
No.
不。
No.
给我，给我，给我出去。
Give me give me give me out.
对。
Yeah.
他们刚刚发布了财报。
They just they just came out.
这是他们昨晚做的事情。
Here here's what they did last night.
迈克尔会详细分析这些数字，然后我们会给你机会发表评论。
And Michael will jump into the numbers, and then we're gonna give you a chance to comment on it.
但他们刚刚首次表示，我们将提供一整年的业绩指引。
But they just said for the first time ever, we're actually gonna give a full year's worth of guidance.
他们之前是按季度来给的。
They were going quarter by quarter.
他们实际上给出了到2028财年的指引，也就是2028财年，不是日历上的2028年，但是
They actually gave guidance through 2028, fiscal twenty twenty eight, which is not calendar twenty eight, but
现在是2027年2月，对吧？
It's February February twenty twenty seven often, what's
所以他们说的是，营收大概要增长70%。
So they're saying, like, we're gonna grow revenue by about 70%.
GPU和Vera CPU的销售都已经锁定了，等等等等。
We've got the sales locked in for GPUs and, Vera CPUs and etcetera, etcetera.
我们知道他们大约90%的业务是卖给数据中心的。
We know that like 90% of their business is selling to data centers.
其中很大一部分是卖给同样的四五家客户，就是那些超大规模云服务商。
A huge chunk of that is selling to the same four or five customers, the hyperscalers.
他们正在谈论扩大客户范围。
They are talking about broadening that out.
他们现在更多谈论自动化和机器人，但数据中心业务才是核心。
They're talking more about now automation, robotics, but the data center business is the business.
好的。
Okay.
那他自己的预测会不会完全不准，还是说他们能完成目标？但问题的一部分就在于买家是谁，以及他们的动机是什么。
So is he gonna be completely wrong in his own forecast, or will they make their number, but that's part of the problem because of who the buyers are and what their motivations are?
比如说，让我们给大家一个更细致的解读，说说你到底在讲什么，因为英伟达确实是整个AI故事的核心。
Like, what let's give people the nuanced take on exactly what you're saying because Nvidia really is at the heart of the entirety of the AI story.
我也想先说明一下。
I just wanna be clear as well.
英伟达确实赚了很多钱。
Nvidia made a bunch of money.
这一点我不质疑。
Like, that's I'm not questioning that.
那也太离谱了。
That'd be crazy.
当然。
Of course.
不过，他们最近一个季度收入的16%来自一个客户。
However, 16% of their late latest quarter revenue was one customer.
2027财年上半年，44%的收入来自三个客户。
44% of their first half of fiscal twenty twenty seven was three customers.
五个客户占了他们账款的70%。
Five customers make up 70% of their accounts
我能插一句吗？
Can I jump in right there?
这些客户都是优质客户。
They're good customers.
哇。
Woah.
哇。
Woah.
我能插一句吗？
Can I jump in right there?
因为他们没有透露客户名字。
Because they don't name them.
哎呀，我们知道他们是谁。
Well well, we know who they are.
我们知道吗？
Do we?
嗯，我们可能不知道谁是16个，谁是12个，但我们知道，是的。
Well, we might not know who's 16 and who's 12, but we know the Yeah.
但如果其中一个是甲骨文。
But if one of them is Oracle.
但听我说完。
But hear me out.
那是他们的直接客户。
That's their immediate customer.
对。
Yeah.
那些客户有数百万的客户。
Those customers have millions of customers.
换句话说，这就是为什么超大规模云服务商不会为了自己而购买英伟达芯片，除了Meta的情况。
And that's what this in other words, you don't have a hyperscaler spending on Nvidia chips to serve itself, except in case of Meta.
对。
Yeah.
没错。
Right.
他们服务的是财富500强企业、政府、主权基金，对。
They're serving Fortune 500 businesses, governments, sovereigns Yeah.
等等。
But etcetera.
那些才是他们的客户。
That those are their customers.
但我刚才提到的关于微软AI收入的观点，就是在这里体现的。
But I just the point I made about Microsoft's AI revenue is where this comes in.
所以2026财年，刚刚结束，AI收入达到了343.3亿美元。
So fiscal twenty six, which just ended, $34,330,000,000 of AI revenue.
其中241亿美元来自OpenAI。
24,100,000,000 of that is OpenAI.
这意味着，除了他们那个大客户之外，向其他任何人销售GPU的业务规模只有几十亿美元。
That means that selling GPUs to anyone else other than their large fail son is a single digit billion business.
微软是软件销售领域的顶级掠食者。
It's Microsoft is the apex predator of software sales.
他们有数十万经销商。
They have hundreds of thousands of resellers.
他们有数万名销售人员，却连Copilot和其他所有东西的销售额都凑不到几十亿美元。
They have tens of thousands of salespeople, and they can't scrape together more than single digit billion dollars of sales for Copilot and all the stuff.
而且，GitHub Copilot是他们为数不多的成功AI产品之一，现在也改为按token计费了，以前人们每月花40美元就能用价值5000美元的token。
And this is on top of the fact that GitHub Copilot, one of their only successful AI products, has now gone token based billing, which means people were before able to spend $40 a month, spend $5,000 of tokens.
现在他们按token计费。
Now they're token based billing.
现在按使用量计费。
Now it's usage.
没了。
Gone.
按用量计费。
Usage.
没了。
Gone.
那生意已经死了。
That business is dead.
对。
Right.
而且除此之外，哇。
And on top of all of that, wow.
2600亿美元的资本支出，加上所有那些财富500强的需求，不管具体是多少，也就只有——多少？
$260,000,000,000 in CapEx and all of that Fortune 500 demand, true on like, whatever it may be, is single is what?
100亿美元。
$10,000,000,000
就是说，给我们几年的时间。
say, give us a couple of years.
为什么我得道歉？
Why do I have to give sorry.
但重点是，我们已经给了他几年时间，而这就是他能给我们的最好的结果。
I'm but the point is is, like, we've given him a few years, and this is the best he's got for us.
我们有个谎言，OpenAI现在是个负担。
We have a lie OpenAI is a liability now.
就是说，他们足够重要，以至于微软实际上不得不——我想，他们拥有大约27%的股份。
Like, that's they are material enough that Microsoft actually has to I think, what's, like, 27% they own of them.
这实际上是一个很低的负担。他们——
It's actually a deeply low bearing They
必须把他们的财务合并到他们自己的损益表上。
have to consolidate their financials onto their their own income statement.
另一件事是，微软只有在OpenAI能继续筹集资金的情况下才能得到回报。
The other thing is is Microsoft's only getting paid if OpenAI can keep raising money.
而OpenAI，能给它投资的人越来越少了。
And OpenAI, it has a shrinking amount of people to raise it from.
他们上一轮融资是1220亿美元，其中只有120亿来自风险投资，而且就是那些投资数据中心的资产管理公司。
Their last round, 122,000,000,000, Only $12,000,000,000 of that came from venture capital and the very same asset managers who were funding data centers.
其中1100亿来自英伟达，30。
110,000,000,000 was Nvidia, 30.
软银投了30，这让他们陷入了财务困境，现在还没缓过来。
SoftBank, 30, which put them into financial dire straits, and they're still there.
还有亚马逊投了500亿。
And, Amazon, 50,000,000,000.
这种情况不可能反复发生。
That's not gonna happen multiple times.
或者如果英伟达再投一次，他们说过不会了。
Or if it if Nvidia invests again, they said they wouldn't.
他们确实说过这话。
They literally said the words.
如果他们再投资，那接下来的日子就不好过了。
If they invest again, it's bad times ahead.
所以微软的增长依赖于OpenAI和现在的Anthropic的支出。
So Microsoft's growth is dependent on OpenAI and now Anthropic spending.
除此之外，他们没有明显的需求。
Outside of that, they don't have significant demand.
这甚至不是个人观点。
This isn't even an opinion.
这是彭博社的一篇文章明确指出的。
This is a Bloomberg article that spelled this out.
如果你去
And if you go and
看彭博社的观点？
look Bloomberg opinion?
不。
It no.
我记得好像是关于布罗迪·福特的。
It was it was about Brody Ford, I think.
但这也是基于实际的财报文件。
But it was and it was also based on the actual actual earnings documents.
微软自己的AI原生产品，并没有来自自家客户的需求。
Was Microsoft doesn't have the demand for its own AI native products from its own customers.
要达到他们需要的规模，我不是说完全没有需求。
At the scale that they would need to justify like, I'm not saying there's zero demand.
他们确实找到了一些，但这还是在骚扰客户和大量销售之后。
They have found but this is also after them harassing their customers and massive sale.
我得说清楚。
I must be clear.
微软的销售团队，比他们更烦人的只有甲骨文了，而甲骨文是另一个大问题。
Microsoft sales teams are there's only one level higher of more annoying, and that's Oracle, and Oracle is a whole other problem.
但关键是，如果他们的收入有30%来自OpenAI，那我可能会说，好吧。
But the point is is if this was a case where 30% of their revenue was OpenAI, I would be like, alright.
展开剩余字幕（还有 480 条）
也许我的生意有点要垮了。
Maybe my business is kinda falling apart.
他们70%的AI收入来自OpenAI，这简直是致命的。
70% of their AI revenue being OpenAI is existentially bad.
他们承诺了2500亿美元的Azure支出。
They have $250,000,000,000 of Azure spend committed to.
这太多了，微软上个财年我记得是3310亿美元。
That's so much that's Microsoft was $331,000,000,000, I think, last fiscal year.
但现在他们已经解除了与OpenAI的独家协议。
But now they've divorced themselves from the exclusivity agreement with OpenAI.
这只会让情况更糟，因为现在OpenAI需要在八年内向亚马逊支付1380亿美元，向Cerebras支付200亿美元，向CoreWeave支付224亿美元，向Oracle支付超过3000亿美元。
Which only makes it worse because now OpenAI needs to spend 138,000,000,000 over eight years with Amazon, 20,000,000,000 with Cerebras, 22,400,000,000 with CoreWeave, 300,000,000,000 plus with Oracle.
他们花的这些钱，真是最棒的事情。
They're spending this is the greatest thing.
UBS的Steven Ju有一份分析师报告。
Steven Ju from, UBS had an analyst note.
他们每年在谷歌那边要花大概125亿美元。
They're spending, like, 12,500,000,000 a year at Google.
这个我也是最近才知道的。
I didn't know this until very recently.
奇怪的是这事儿没怎么被报道，但问题是，OpenAI的Sam Altman真正擅长的就是签合同。
Weird that it's not being reported very well, but the thing is is OpenAI, Sam Altman has one real talent, and that's signing his name.
他特别爱签合同。
My man loves signing contracts.
而且现在大家面临的问题是，你只能考虑未来一两个季度的事。
And the thing is when the problem that everyone has right now is you can only rethink a quarter, two quarters in the future.
你没法想太远，因为就算说2028年会怎样。
You can't because even like, oh, it's gonna happen in 2028.
根据共识预测，英伟达在2028财年的市值可能会达到6740亿美元左右。
Nvidia will get to set to 600 and something 674,000,000,000, I guess it would be, based on consensus, for fiscal twenty eight.
所以英伟达能达到那个目标，因为现在它赚得很多。
So Nvidia will get there because right now Nvidia makes lots of money.
现在这不是问题。
It's not a problem now.
要达到那个目标，英伟达基本上得拿下三到五个客户。
To get there, Nvidia is going to have to basically take three to five customers.
他们得找到更多的债务融资，而且要多得多，因为债务成本在上升，而且英伟达刚刚把价格提高了17%。
They're going to have to find way more debt, like way more because the price of debt is increasing, but also Nvidia just bumped up prices 17%.
内存成本正在飙升。
The cost of memory is skyrocketing.
除此之外，还有人才短缺、电工钢短缺的问题。
And on top of that, you have a shortage of talent, shortage of electrical grade steel.
现在有各种不利因素都冲着英伟达来。
There are all sorts of things stacked against Nvidia here.
他们也许能办到，但我不知道他们怎么筹到那笔钱，需要的资金简直是天文数字。
They could pull it off, but I don't know how they're going the the actual money that needs to get raised is astronomical.
我看到了丹尼尔的图表五。
I saw Daniel, chart five.
埃德，如果两年前有人跟你说会发生这种事，你也会像现在一样说这根本不可能。
Ed, if if somebody said to you two years ago that this is going to happen, you would be saying the same things that you are right now about how implausible it would be.
所以，听众朋友们，我们现在看到的是，英伟达过去十二个月的净利润刚刚超过了苹果。
So what we're looking at for people that are listening is the trailing twelve month net income for Nvidia just screamed past Apple.
嗯哼。
Uh-huh.
但其中有多少是股权收益呢？
But how much of that is equity gains, though?
那是什么意思？
What does that mean?
就是说，不是他们的运营收入。
As in their Oh, not operating earnings.
Anthropic、OpenAI。
Anthropic, OpenAI.
他们投资了CoreWeave、Nebius、Iron。
They've invested in CoreWeave, nebious, iron.
他们有一堆股权投资，帮他们把数字抬上去了。
They have a bunch of equity investments that help bumping them up.
说清楚点，我在2024年的时候确实判断错了。
To be clear, I was wrong back in 2024.
为什么？
And why?
因为我当时太天真了。
Because I was fucking naive.
我当时觉得市场是个理性的地方。
I was like, the market's a sensible place.
微软、谷歌、亚马逊，不会无缘无故砸几千亿美元。
Microsoft and Google, Amazon, wouldn't spend hundreds of billions of dollars for no goddamn reason.
等等。
Wait.
所以你错是因为市场太蠢？
So you were wrong because the market's stupid?
我错是因为我太蠢了。
I was wrong because I was stupid.
我错是因为我对这个世界的运作方式太天真了，而且我不得不接受一个残酷的
I was wrong because I was naive about how the world works, and I've had to take a hard
你没想到他们会把事情做到那种程度，
You didn't think they would take it as far as
对，他们确实做到了。
they've Exactly.
而且我没想到债务体系能撑得住。
Taken And I didn't think that the debt system would support it.
所以，是的，我……
So, yeah, I
那你觉得现在市场哪里看错了？
so what do you think the market is getting wrong right now?
因为我觉得，我觉得大家都知道OpenAI可能是个问题。
Because I think I think everyone knows that OpenAI is a potential problem.
对。
Yeah.
看看甲骨文的股票。
Look at Oracle stock.
它跌了70%，还在底部趴着。
It's down 70%, still on the mat.
即使软件股在反弹，
Even as as software stocks are bouncing
对。
Yeah.
甲骨文也没跟上反弹。
Oracle has caught no bounce.
那你觉得市场现在误解了什么，而你看到了什么？
So what do you think the market is misunderstanding that you're that you're seeing?
我之前说的数据中心需求集中化的问题。
The thing I said about the centralization of data center demand.
我觉得OpenAI大概占了所有AI数据中心需求的60%。
OpenAI, I think, is, like, 60% of all AI data center demand.
没有OpenAI的话，就没有另一个像OpenAI那样规模的支出了。
Without OpenAI, there is not another OpenAI sized spender.
而且，之所以会有像Anthropic或OpenAI那样在AI计算上大手笔投入的公司，完全是因为有风险投资资金的支持。
And the only reason there is an Anthropic or OpenAI sized spender on AI compute is because of the availability of venture capital dollars.
所以没有这些资金，你做不到，因为需求根本不存在。
So without that, you can't because, like, the demand isn't there.
市场没看到这一点，因为这种情况还没发生。
The market isn't seeing that because it hasn't happened yet.
这一两个季度，他们可能觉得，钱还在进来，所以没意识到问题。
The the quarter, two quarters think they're doing where they're like, well, the money's still coming in.
我之前也说过，Anthropic和OpenAI的支出规模必须增加大约10倍。
Because what I've kinda said this already, but the amount that Anthropic and OpenAI has to spend has to increase by, like, 10 x.
嗯，他们的收入在增长，而且那些是实实在在的收入。
Well, their revenues are increasing, and those are real revenue dollars.
而且他们两家大概在未来六个月里都会上市。
And they're both going public in the next, call it, six months.
等等。
Wait.
你刚才做了个表情。
You just made a face.
Anthropic 的营收是 600 亿美元，这是
Anthropic is doing $60,000,000,000 Is that
不对吗？
not right?
不。
No.
不是。
Not.
是年化收入。
Run rate.
继续。
Go.
继续。
Go.
继续。
Go.
运行率。
Run rate.
好的。
Okay.
如果真是这样，那这就是最大的骗局了。
It that's this is the biggest scam of them all.
什么？
A what?
年度经常性收入运行率，而不是那个季度的数据。
ARR run rate as opposed to less Did that quarter.
这就是骗局的第一部分。
That's the first part of the scam.
所以ARR以前指的是年度经常性收入，意思是如果我有十个一亿的合同，那我一年就有十亿美元的收入。
So ARR used to mean annual recurring revenue, referring to if I have ten, one hundred million contact contracts, I've I've got, like, a billion dollars a year.
轻而易举。
Easy peasy.
这就是年度经常性收入。
That's annual recurring revenue.
哇。
Wow.
太厉害了。
How amazing.
这就是SaaS。
That's SaaS.
SaaS。
SaaS.
我可能把数学算错了。
I probably up the math there.
我意识到了。
I realized.
不过无所谓。
But whatever.
继续吧。
Moving on.
对。
Yeah.
而且他们怎么确认收入的，他们可能
And how they're booking the revenue, they probably
不喜欢。
don't like.
嗯，现在他们管这叫运行率，运行率可以有。
Well, now they're calling it run rate, and run rate can have.
Anthropic从来没有定义过这个。
Anthropic has never defined this.
就连彭博社在650亿美元那篇报道里也没定义。
Even Bloomberg in the 65,000,000,000 story didn't define it.
它可能指四周乘以12，或者四周乘以13。
It can mean four weeks times 12 or four weeks times 13.
问题是，Anthropic和OpenAI都在把token消费年化，这可不是软件订阅那种经常性收入。
The problem is is Anthropic and OpenAI are both annualizing token spend, which is not a re it's not a software subscription.
它不是经常性的。
It's not recurring.
客户这个月可能花1000，下个月不花，再下个月花50，尤其是如果他们转向开源模型的话。
A customer could spend 1,000 this month and then not spend and then spend 50 next month, especially if they're moving to open source models.
抱歉。
Sorry.
这百分之百是真的。
That's a 100% true.
嗯。
Mhmm.
但是，任何一家上市公司，只要对华尔街提到自己的支出，嗯。
But every publicly traded corporation that says anything to Wall Street about its spend Mhmm.
你听到的永远只有一句话：我们的算力预算已经用完了，需要追加投入。
The only thing you ever hear is we already ran out of our compute spend and need to spend more.
对。
Yeah.
那是因为Anthropic和OpenAI占用了大部分算力基础设施，所以不够用。对。
That's because Anthropic and OpenAI are taking up most of the compute infrastructure, so there's not enough Right.
现在的需求确实超过了供给。
There is more demand than there is right now.
问题是超出多少？
The question is how much more?
因为我不认为……我是说，我大概在一周前的通讯里做过一个估算。
Because I don't think that I mean, I kind of estimate in an estimate in a newsletter a week ago.
我觉得算力需求大概有220亿美元。
I think there's about $22,000,000,000 of compute demand.
SiteOne Climate在二月份的时候说，他们规划中的容量是190吉瓦。
SiteOne Climate said back in February that it was a 190 gigawatts in planning.
这意味着他们每年需要的算力需求大概在150亿到3000亿美元之间。
That's about 1.5 to $3,000,000,000,000 a year in compute demand they'd need.
这里存在一个非常明显的经济错配。
There is a complete economic mismatch, and it's stark.
你不喜欢用年度经常性收入（ARR）的年化速率，因为它只是简单的推算吗？
You don't but you don't like ARR run rate because it's just a extrapolation?
它只是某个时间段的快照，而且
Of a it's a snapshot of a period in time And
然后他们把它扩展到一年或者
then they define it out to, like, a year or
之类的。
so.
所以我们也不清楚这个周期到底是什么，甚至都不明白它具体指什么。
So it's we also don't know what the period we don't even know what it means.
他们从来没给过明确定义。
They never define it.
Anthropic在自己的公告里也用过好几次这个词。
Anthropic has used it several times in their own announcements.
但他们始终没给出定义。
They never define it.
它可能指的是，嘿。
It could refer to, hey.
我们刚发布了这个模型，大家都在试用。
We just released the model, and everyone's trying it.
嘿。
Hey.
我们刚上线了一个新功能，所有人都跑去买新订阅了。
We just did a new feature, and everyone's buying a new subscription.
哦，就在这一天，我们流失了一大批订阅，所以我们就把这个周期挪过来，好让我们看起来规模更大。
Oh, on this day, we lost a bunch of subscriptions, so we'll pull the period over here so we look bigger.
这根本不是一个衡量企业的可信指标，而且它居然被接受，简直是对投资者智商的侮辱。
It is not a trustworthy measure of a business, and the fact it gets accepted is an insult to investors' intelligence.
那这里谁是傻瓜呢？
So who's the idiot here?
是那些为所有这些资本支出负责的CEO和董事会吗？
Is it the CEOs and the boards for all of this CapEx?
是那些股权投资者吗？
Is it the equity investors?
是那些借钱来建这些数据中心的人吗？
Is it the people loaning the money to build these data centers?
就是说，现在谁的目光这么短浅？
Like, who is not seeing past their nose right now?
放贷的人，所有对AI数据中心的投资都让我从骨子里感到恐惧。
Lenders, all of the investment in AI data centers terrifies me to my core.
这太糟糕了。
It's so bad.
这些东西就像微型城市一样。
These things are like miniature cities.
它们把英国布里斯托尔市的规模压缩到大约12.5亿平方英尺，就像拼车一样。
They condense city of Bristol in England, using this in the carpool, about 1,250,000,000 square feet.
需要大约800兆瓦的电力。
Takes about 800 megawatts of power.
Stargate Abilene，OpenAI的数据中心，功率1.2吉瓦，占地99.8万平方英尺。
Stargate Abilene, OpenAI data center, 1.2 gigawatts of power in 998,000 square feet.
所以它们把一座城市的电力压缩到千分之一的空间里，而且这个项目还严重落后于进度。
So they're condensing a city's power into a thousands of the space, and that thing is way behind schedule.
每一个这样的项目，即使是小型的，都像……
Each one of these, even the smaller ones, are like
那是德克萨斯州的Oracle那个吗？
Is that the Oracle one in Texas?
其中一个就是德克萨斯州的那些数据中心之一。
One one of the it was one of the ones in Texas.
所有这些数据中心都非常雄心勃勃，而且规模越大，就越离谱。
And all of these data centers are very ambitious, and the bigger they are, the more ridiculous they are.
千兆瓦级的数据中心是个新概念。
Gigawatt data centers are a new idea.
另外，要说明一下，千兆瓦级数据中心其实是由多个较小的建筑组成的园区。
Also, a gigawatt data center is a campus of smaller buildings, just to be clear.
但它们是新的。
But they're new.
它们完全是新的，而且他们想在那种用订书钉和胶带拼凑起来的电网上建这么多数据中心，这真是……而且他们总是说，哦，这能帮我们升级电网。
They're brand new, and they're trying to build so many of them in a power grid kinda put together with, like, staples and tape and such, which is I mean and the thing they always say is, oh, it's helping us upgrade the power grid.
不。
No.
根本不是。
It's not.
这是策略性地把电力放在一个你也许能租给别人的地方。
It's strategically placing power in a place that you can maybe rent it to someone.
问题是，要让一切都
And the problem is is that for everything to be
但这是恶意的，还是他们只是不像你那样了解需求？
But is it nefarious, or do they just not understand the demand as well as you do?
听我说，事情是这样的。
Little little here's the thing.
因为他们看起来像是在
Because it seems like they're
承担风险的一方。
the ones taking the risk.
这是恶意加无知。
It's nefariously ignorant.
我觉得他们的无知危害太大了。
I think that their ignorance is so harmful.
星门阿比林，那个信息里有关于这个的报道。
Stargate Abilene, the information had a report about this.
蓝猫头鹰同意在十分钟内投资。
The Blue Owl agreed to invest in ten minutes.
尽职调查就是这样做的，兄弟们，就用了十分钟。
That's how that's how the due diligence gone, boys, with just ten minutes.
得了吧。
Come on.
这就像……而且那是一个耗资数百亿美元的项目。
It's it's just like and that is for a project that will cost tens of billions of dollars.
为什么就是没人愿意花时间呢？
And it's just why is no one putting the time in?
等等。
Wait.
等一下。
Hold on.
那不可能。
There's no way that's true.
我，我，我很乐意说十分钟？
It's I will I will glad ten minutes?
十五分钟？
Fifteen minutes?
我打赌是十分钟。
I bet it was ten
也许是十五分钟。
maybe it was fifteen.
不。
No.
不。
No.
我现在就能把文章给你，如果你想要的话。
I can get I can get you the article right now if you'd like.
我相信那篇文章是真的，但我不相信那件事是真的。
I believe that the article is real, but I just don't believe that that is actually true.
你读过关于Blue Owl的东西吗？
Have you read anything about Blue Owl?
我两秒钟都不信，兄弟。
I don't believe it in two seconds, man.
比如，我很抱歉。
Like, I'm sorry.
我看了Blue Owl做的事情，是的。
I'd like having seen what Blue Owl does, I yeah.
我会相信的。
I I would believe that.
而且，这也是个狂热。
And, also, this is a mania.
而且数字在涨。
And number go up.
嘿。
Hey.
最大的理由，如果你不仔细想一秒，听起来是合理的。
The biggest the justification makes sense if you don't think for a second.
世界上最大的公司不会无缘无故花一万亿美元。
The biggest companies in the world wouldn't spend a trillion dollars for no reason.
我得建一个这样的东西。
I gotta build one of these things.
我会赚很多钱。
I will make so much money.
这就是赌徒的逻辑。
It's it's gambler logic.
我们来看看这些数字吧。
Let's go through let's go through some of these numbers.
丹尼尔，你想看第一张图表吗？
Daniel, you wanna do chart one?
回到英伟达的话题，因为这就好比，我称黄仁勋为AI美联储主席。
So back on Nvidia, just because this is like well, I call Jensen Wang the in the, the chairman of the AI Federal Reserve.
基本上，他就是那个出来让大家安心的人。
Basically basically, he's the person that comes out and makes everybody feel better.
对。
Yeah.
几乎每个季度，我们都会看到超大规模企业的财报，然后有三周的延迟。
Pretty much every quarter, we get the hyperscaler earnings, and then there's a three week lag.
然后英伟达在财报季结束时出来，确认了我们之前说的一切。对。
And then Nvidia comes out at the end of the earnings season and ratifies everything that we Yeah.
好。
Okay.
所以单看表面，收入是960亿美元。
So just like on its face, dollars 96,000,000,000 in revenue.
我们知道数据中心的成本大约有50%花在芯片上。
We know that the cost of a data center is approximately 50% on chips.
对。
Yeah.
现在更是这样，因为内存的原因，但没错。
More now, thanks to memory, but yeah.
好。
Okay.
那就是75%毛利率下的16%营收增长。
That's 16% revenue growth at 75% gross margins.
对。
Yeah.
现在的英伟达，我是说，这可是真金白银。
Nvidia right now, I mean, this is real money.
对。
Yeah.
所以你的问题不在于支出是否真实。
So your issue is not whether or not the spending is real.
你真正要质疑的是，买英伟达产品的钱到底是从哪儿来的。
Your issue is the source of where the capital is coming from to buy from Nvidia is really what's in question.
对。
Yes.
而且说清楚一点，这些收入和各种资金的来源，确实是真实的。
And just to be clear, the source of the revenue and the money into all the this stuff, real.
但真正流入AI软件的钱——也就是这些硬件想要支撑的东西——如果去掉那些补贴的部分，其实非常少。
The the money going into AI software, which this is meant to prop up, very small when you remove the subsidized stuff.
每一家AI初创公司都在靠补贴。
Every AI startup is subsidized.
每一家都在亏钱。
Every single one loses money.
比如，直到不久前，OpenAI和Anthropic还让你每月花200美元就能烧掉几千美元的token。
Like, up until fairly recently, OpenAI and Anthropic let you burn thousands of dollars of tokens for $200 a month.
我觉得现在还可以，但没那么多额度了。
I think you still can, but not as much.
这些公司，当客户面对真实成本时，就会大幅缩水。
These companies, when their customers are exposed to the real cost, they shrink full away.
而且很简单，如果他们觉得能按真实成本收费，他们就会这么收。
And just very simply, if they thought they could charge the real cost of their services, they charge the real cost of their services.
他们不会搞这种奇怪的把戏。
They wouldn't do this weird dance.
所以问题不在于找到……嗯，其实，现在这确实开始成为问题了。
So the problem isn't finding well, I mean, actually, it is becoming one now.
这很多是……我觉得那个大客户，那个百分之十六的，我想是SpaceX。
This a lot of this was I think that that the, the one big customer, the sixteen percent one, I think it was SpaceX.
记住，SpaceX，一堆债务。
So remember, SpaceX, bunch of debt.
所有这些，都是一堆债务。
All of this, bunch of debt.
除了微软，现在其他公司都得背上债务。
Other than Microsoft at this point, they're all having to take on debt.
这就是问题所在。
And this is the thing.
这一切其实并不是真正的需求，因为他们没法用现金流来支付这些。
Everything it it's not actually about real demand because they can't pay this out of cash flow.
他们显然赚的钱不够，没法用现金流来支付，然后他们就会说，哦，这是新的工业革命之类的废话。
They're clearly not making enough money to pay this out of cash flow, and they would say, oh, it's the new industrial revolution or some such wank.
好吧，显然进展得不太顺利。
Well, it's clearly not going very well.
哦，我们以后会解决的。
Oh, we'll we'll work out in the future.
什么时候？
When?
真正的需求在哪里？
Where's the real demand?
到底在哪里？
Where is it?
而且因为所有东西都越来越贵，不管是债务本身还是债务买来的东西，他们需要的钱会越来越多，而这时候整个系统已经有点撑不住了，但系统唯一想投的，偏偏是最蠢的东西。
And so they're and because everything's getting more expensive, both the debt and the things that the debt's buying, they're gonna need more and more and more money at a time when the system's kind of straining itself, but the only thing the system wants to invest in is the dumbest thing possible.
其实，抱歉。
It it's actually sorry.
把担心的事说清楚。
Spell out the worry against.
Anthropic 是什么时候开始有收入的？
When did Anthropic first turn their revenue?
我记得是2023年3月。
I think it was March 2023.
什么？
What?
他们什么时候，什么？
When was when did they what?
Anthropic 是什么时候赚到第一块钱的？
When did they first book their dollar first dollar of revenue for Anthropic?
大概是2023年的某个时候。
It was sometime in 2023.
对。
Yeah.
也许吧。
Maybe.
我记得他们2021年就成立了，但我不记得他们具体是什么时候……
I they they found it in 2021, but I can't remember when they
但你说这话的时候，感觉你好像已经厌倦了反复说它什么时候会来。
actually But you you say that, like, you're sick and tired of of saying where, when it's gonna come.
你在这件事上似乎没有开放的心态，没有考虑到现在确实还早。
You don't you don't have any open minds on this to the fact that it genuinely is early.
我们现在看到的需求，跟未来将要出现的需求相比，占了多少？
How much of the demand that we're going to see are we seeing now?
因为我觉得这只是极小极小的一部分。
Because I think it's a tiny, tiny fraction.
那么，人们会突然停止在这上面花钱吗？
Well, will people all of a sudden stop spending on this?
客户。
Customers.
有个叫尼克·苏雷什的人，他有个很好的博客叫“AI正在摧毁全球决策”，他是一名技术顾问，也是一位天才软件工程师，跟很多CEO聊过。
So there's a guy called Nick Suresh who has a really good blog called AI is eviscerating global decision making, and he's talked to he's a tech consultant, gifted software engineer, talks to a lot of CEOs.
很多这种情况，你有没有看过那个
A lot of this is you ever see the
《斯大林之死》？
death of Stalin?
好的。
Okay.
所以很多做AI的人，是不是那个
So a lot of people doing make Is that
在Netflix上？
on Netflix?
不是。
No.
嗯，也许吧，其实。
Well, maybe, actually.
但是，不管怎样，开头有一段他们围着斯大林的尸体，说，他看起来真好。
But, nevertheless, there's a bit at the beginning where they're sitting around Stalin's corpse and going, he's so he looks great.
他看起来很健康。
He looks very healthy.
哦，是的。
Oh, yes.
很多
Lot of
这种情况就是人们说，哦，AI让我生产力提高了100倍之类的。
this is people saying, oh, AI is making me a 100 x more pro productive or whatever.
这不是真的。
It's not true.
大多数AI集成都会失败。
Most AI integrations fail.
大多数
Most
好的。
Alright.
不过我想给你一个反例，因为我每个财报季都会听大概20个电话会议。
So I wanna give you a counter to that, though, because I listen to, I'm gonna say, 20 calls every earning season.
对。
Yeah.
通常是我持有的股票，但有时也会听那些波动很大的股票。
And usually the stocks I own, but sometimes just stocks that make a big move.
我就是想了解原因。
And I just wanna learn like why.
明白。
Mhmm.
为什么突然间大家要么非常看好，要么非常看空？
Why all of a sudden are people really bullish or really bearish?
所以其中一只股票，我今天刚好聊到它，印象还很深，就是爱彼迎。
So one of the stocks, and I happen to have been talking about this today, it's fresh in my mind, is Airbnb.
嗯哼。
Uh-huh.
然后切斯基出来说，AI是我们遇到过的最好的东西。
And Chesky comes out and says AI is the best thing that ever happened to us.
接着他举了具体的例子，其中一个就是上个季度，他们45%的客服电话都是由AI处理的，完全没有人工参与。
And then proceeds to cite specific examples, one of which is that last quarter, 45% of all their customer support calls were closed by AI with no human involvement whatsoever.
好的。
Okay.
这提升了现金流。
It boosted cash flows.
还降低了开支。
It reduced expenses.
他们有没有说具体是多少？
Did they say by how much?
他们给出了具体数字。
They gave the numbers.
对。
Yeah.
那挺好的。
That's good.
所以呢，
So
他们还转向了开源。
They also moved to open source.
切斯基特别明确地表示，不付钱给OpenAI。
They very Chesky very specifically Not paying not paying the OpenAI.
算是早期的那些公司之一。
One of the early ones.
所以对我来说，这是我最看空的一点，但我们一会儿会谈到这个。
So for me, that's the thing that I'm most bearish about, but and we'll get into that in a second.
我不同意你的地方在于——
I'm not so here's what I'll here's where I don't agree with you.
对。
Yeah.
我真的不认为作为AI用户，这件事会突然停下来。
I really don't think this is gonna stop on a dime as an AI user.
它会慢下来。
It's gonna slow down.
公司里，每周我们都会发现新的东西可以接入AI，而且比上周做得更好。
Company, At every week, we're finding new things that we can plug into AI and do better than we did the week before.
我很好奇。
I'm curious.
我们必须管理90多名员工，4000个客户家庭。
We have to have to manage 90 some odd employees, 4,000 client households.
在Salesforce、我们的投资组合会计软件和我们正在进行的交易之间，需要持续不断地协调运营活动。
There's a constant orchestration of operations activity that has to take place between Salesforce and our portfolio accounting software and the trading that we're doing.
从我们的角度来看，客户根本不在乎我们是否使用AI。
And the general idea from our standpoint is the customers don't give a shit about our AI use.
这是运营层面的问题。
This is operations.
我们怎样才能把公司运营得更好？
How do we run the firm better?
对。
Right.
我们怎样才能做得更快、更高效？
How do we do things faster, more efficiently?
我认为这正是现在美国每家公司都在努力做的事情。
And that's what I think every company in America right now is trying to do.
我们越来越多地听到他们在与分析师的问答中一个个地说，为什么我们会有超预期的表现？
And we are increasingly hearing from them one by one on Q and As with analysts saying, why did we have an upside surprise?
好，很高兴你问这个问题。
Well, glad you asked.
我们大概在23年或24年就开始投资AI了。
We invested in AI probably in '23 or '24.
对。
Mhmm.
现在突然之间，我们能告诉你我们具体在用AI做什么，而且这些确实在提升我们的盈利。
And now all of a sudden, we're able to tell you specific things that we're doing with AI that are improving our earnings.
我觉得这部分不会停。
I don't think that part's gonna stop.
所以我觉得这种情况会减少，因为我觉得很多
So I think it will stop happening as much because I think a
都是虚构的。
lot of it is make believe.
AT&T是最糟糕的例子。
AT and T being the worst one.
我希望那些谎言，
I hope that lying,
说到AT&T，这里有个巧妙的地方。
With AT and T, if here's the clever thing.
他们在《华尔街日报》上发了篇文章，说使用开源能节省80%到90%的成本。
So they had this article in the journal saying 80%, 90%, savings on using open source.
然后你往下读这篇文章。
Then you go down the article.
文章说在某些功能上能节省80%到90%。
It says 80 to 90% savings in some fee some functions.
然后下一段又说AT&T有成千上万的东西接入了AI。
And then the next paragraph is AT and T uses has thousands of things it has plugged into AI.
所以就像，哦，所以有些东西可能帮你省了点钱。
So it's like, oh, so some things and some things may have saved you some money.
这就是问题所在。
This is the problem.
我之所以提起尼克·苏雷什的文章，很大程度上是因为他提出了一个观点：很多人这么做是因为害怕如果不这样做会显得不好看，或者管理层已经焦头烂额，不得不把AI当作他们面临的救命稻草。
A lot of this the reason I brought up Nick Suresh's article is because he makes a point where a lot of people are doing this because they're afraid of how it will look if they don't, or indeed the management is so cooked that they must see AI because it's the thing they're facing.
确实。
With that.
我完全同意，这种情况确实很多。
I definitely agree that there's a lot of that going.
而且关键在于。
And here's the thing.
我们可以讨论一个点：即使情况如此，这些收入是否足以维持OpenAI和Anthropic的运营？这两家公司目前都在亏损。还是说，这会不会转向托管GPU系统加上开源模型的方向？
On there is a point we can come to, which is even with that being the case, is that going to be the revenue sufficient to keep OpenAI and Anthropic alive who are both unprofitable, or is that going to move on to, I don't know, a hosted GPU system with an open source model?
有很多种可能，比如Anthropic和OpenAI必须持续增长，而且必须增长得更快得多。
There are so many way like, Anthropic and OpenAI must keep growing, and they must keep growing so much faster.
这离他们需要的水平还差得远。
This is not even close to where they need to be.
他们需要在未来三年半内负担得起4400亿美元的开支。
They need to be able to afford $440,000,000,000 in the next three and a half years.
所以你的很多看空观点都建立在这样一个想法上，就是标普十强依赖那两个最知名的基础模型来交付他们的业绩。
So a lot of your bear case rests on that idea that the two most well known foundation models are being relied upon by the S and P ten to deliver their okay.
对。
Yeah.
这个我不否认。
I I don't dispute that.
那Gemini呢？
But what about Gemini then?
你怎么看Alphabet的Gemini情况？他们不需要OpenAI或Anthropic来花钱跟他们合作。
What do you do with the Alphabet Gemini situation where they do not they do not require OpenAI or or Anthropic to spend with them?
我觉得你问这个问题，是基于我看到的关于Gemini的所有报道。
I think asking that question based on all the press I've seen around Gemini.
Gemini也就是另一家LLM公司。
Gemini is just another LLM company.
问题是，尽管人们喜欢假装，哦，OpenAI和Anthropic很特别，也许他们喜欢Sol什么的。
The problem is is that as much as people love to pretend, oh, OpenAI Anthropic special, maybe they like Sol, whatever.
大多数情况下，人们其实不太在乎用哪个模型。
For the most part, people don't really give a shit which model they use.
每个模型之间的区别，感觉就是，我为什么要用这个？
And each model, the differentiation between them is kinda, why do I need this?
然后又有新的更好的出来了。
And then there's a new better one.
你试试这个。
You try this.
每次换模型，都要改一堆配置和提示词。
And every time you move it, there's a bunch of harness changes and prompt changes.
这些东西的功能太不稳定、太善变了，我觉得经常用真的让人很累。
The actual functionality of this stuff is so malleable and volatile that I think it's exhausting to engage with on the regular.
实际上，跟那些专业的AI人士聊，我接触的那几个，他们也有这种感觉。
And actually talking to pro AI people, the few that I talk to, they experience this.
对这种东西已经感到疲惫了。
There's an exhaustion around it.
你得不停地跟上，他们就像任何糟糕的关系一样，把它合理化。
You constantly have to keep up, and they kind of rationalize it like any bad relationship
就是浪费。
just waste.
就像你说的，我需要OpenAI发布最新模型。
Like you're saying I need to OpenAI releases the newest model.
必须把这个错误隔离起来。
Must insulate this error.
对。
Yeah.
对。
Yeah.
而且我必须，特别是如果他们经营一家面向客户提供模型的公司，他们必须立刻做到，这会推高成本。
And I must and especially if they run a company that has models facing the customer, gotta they have that immediately, which bumps up the cost.
因为问题是，token的成本可能差不多，但消耗的token数量在不同模型之间会随意变化。
Because the thing is the cost of the tokens may stay around the same, but the amount of tokens being burned changes arbitrarily between models.
所以对于模型来说，价格一样，但消耗的令牌更多了。
So for models, the same price, but it's burning more tokens.
就像一辆车，每加仑英里数一样，但你多开了200英里。
It's like having a car that miles per gallon is the same, but you're driving 200 extra miles.
所以我的看法是，我能看到未来会有大语言模型，但我不确定它们能有多赚钱，甚至不确定即使有定制芯片，这些模型能否实现盈利。
So my thing is I could see a world where there are LLMs in the future, but I don't know how profitable is to well, I don't know if it's possible to make these things profitable even with custom silicon.
我不确定这些公司是否还有未来，也不确定超大规模云服务商是否还会以现在的价格提供这些服务。
I don't know if there's a future for these companies at all or indeed whether hyperscalers will still offer these services at the current prices.
他们可能会把它们作为某些企业的高价服务来提供。
They may offer them as a really expensive thing for certain businesses.
他们可能会那样做。
They may get that.
那是有可能的。
That might happen.
我觉得这很现实。
I think that's realistic.
比如Gemini，它可能会变成那样。
Gemini, for example, could become that.
这可能会让你更看跌。
This might make you more bearish.
好的。
Okay.
这些公司太大了，不能倒。
There is too much these things are too big to fail.
太多事情都押注在成功上，不只是这两家公司，而是整个概念不能崩盘。
There is too much riding on the success of maybe not just these two companies, but this whole idea not imploding.
而且看跌的投资者已经一次又一次地学到了教训。
And bearish investors have learned over and over and over again.
这个系统就是被设计成会成功、会往上涨的。
The system is rigged to work out and to go up.
这可能会让你更看跌，但我很想听听你对这个的看法。
Now that might make you more bearish, but I'm curious to hear what your take would be on that.
所以问题在于，人们都在谈论救助，但我实际上认为私人信贷才是那里的警示信号，沃尔特先生，该进手术室了。
So the problem with this is people talk about a bailout, and I actually think private credit is the warning sign there, mister Walter to the operating room.
就是，这简直，唉。
Like, it it just, like, ugh.
私人信贷领域发生了很多糟糕的事情。
Lot of bad things happening with private credit.
对于人工智能和数据中心，你有多个问题。
With AI and data centers, you've got multiple problems.
你有OpenAI和Anthropic。
You have OpenAI and Anthropic.
你要救助它们吗？
You bail these out?
好吧。
Okay.
但这并不能解决他们三年半后无法支付4400亿美元的事实。
But that doesn't fix the fact that they won't be able to afford to pay $440,000,000,000 in three and a half years.
这并不能改变一个事实，那就是很明显，微软、谷歌、Meta 和亚马逊的整体增长正在放缓。
It doesn't fix the fact that it's very clear that Microsoft, Google, Meta, and Amazon, their overall growth is slowing.
否则，他们绝不会这么做。
Otherwise, they would have never done this.
他们正在减速，而我之前提到的微软 2026 财年收入的 7% 是来自 OpenAI 的。
They are slowing down, and my 7% of Microsoft's fiscal year twenty six revenue was from OpenAI.
这个我得反驳一下。
I have to challenge that.
他们的云收入正在加速增长，这得益于 Anthropic 和 OpenAI 的支出。
Their cloud their cloud revenues are accelerating this Based on Anthropic and OpenAI spend.
但你说他们在放缓，实际上并没有。
But you're saying they're slowing, but they materially are not.
当你把 OpenAI 和 Anthropic 的支出剔除后，但为什么非要剔除呢？
When you remove the OpenAI and Anthropic spend But why do you have to remove?
为什么要剔除它们呢？
Why would you remove them?
我的意思是，是你在付钱给他们。
I'm saying you paying them.
抱歉。
Sorry.
你说得对。
You're right.
好吧。
Okay.
但以后，你不能指望这个。
In the future, you can't count on it.
我本来应该那样说的。
I should have put it like that.
以后不能指望这个。
You can't count on it in the future.
我之前说的那个4.40的数字，其实会大得多，因为，是的，他们现在确实在增长。
When I said that $4.40 number, that is exponentially larger because, yeah, they are growing right now.
书都摊开了。
Books are open.
AI和Anthropic。
AI and Anthropic.
他们的支票正在清算。
Their checks are clearing.
这个
This
话说回来，所有那些超大规模云厂商的云计算业务增长数据都远超预期。
Well, all of the hyperscalers, all of them reported way better than expected growth numbers for cloud.
他们刚公布了第二季度的财报数据。
In the quarter, they just reported second quarter numbers.
除了Meta之外。
Other than meta.
所以Meta的情况完全不一样。
So meta is a whole other basket.
对，但我不想问那个。
Yeah, but I don't want ask that.
不过这确实是个关键点。
Is actually a point though.
好的。
Okay.
没错。
Right.
不过确实，AWS、Azure和谷歌云的增长都在加速。
But it is true that AWS and Azure and Google Cloud are seeing accelerating growth.
这一点你也没法否认吧。
You're not questioning that.
你的意思是这里有个大前提。
You're saying there's a big asterisk.
主要是有两家公司在大幅加码投入。
It's two companies that are ramping up all that spend.
那第二季度还发生了什么？
So what also happened in q two?
就是那巨大的token消耗，还有在云算力上的巨额投入。
The massive token maxing, the massive spend on cloud compute.
这跟它们的时间点完全吻合。
It times directly with them.
Anthropic和OpenAI的数据一上去，它们的数字也跟着涨，而且那些钱是实打实的。
Their numbers went up when Anthropic and OpenAI, and that money is real.
其中一部分钱直接来自那些超大规模云服务商自己。
Some of that money came directly from the hyperscalers themselves.
我提Meta不是为了抬杠，而是你会发现Meta并没有像其他公司那样爆发式增长。
The reason I brought up Meta was not not to be argumentative is you'll notice that Meta didn't grow like gangbusters.
甲骨文倒是涨了。
Oracle did.
OpenAI和微软也涨了。
OpenAI, Microsoft did.
Meta跟这些完全没关系。
Meta's none of that business.
但这就是我要说的点。
But that's the point I'm making.
你会发现，AI并没有让Meta大幅增长。
You'll notice that AI is not making Meta grow massively.
只有那些从Anthropic和OpenAI拿到钱的公司才在增长。
It's just the ones who get the money from Anthropic and OpenAI.
而这就是问题所在。
And this is the problem.
问题是，目前来看，你往前看一个季度、两个季度，所有东西都……
The problem is is that for now, everything you'd look a quarter, two quarters in the future.
哇。
Wow.
钱不是钱，好。
Money not money good.
钱在这里。
Money here.
嗯，Meta的看多者，我不是其中之一，但他们会说，他们不是在租算力。
Well, meta Meadows the Meta Bulls, and I'm not one of them, but the Meta Bulls would argue, they're not renting compute.
他们认为他们想优先用它来改进Reels。
They think they wanna prioritize it to improve reels.
嗯。
Uh-huh.
他们随时可以改变主意，说，哦，对了，现在我们是个数据中心，我们要租算力。
They could at any time do an about face and say, oh, yeah, by the way, now we're a data center, and we're gonna rent compute.
对。
Yeah.
如果他们这么做，我想你会同意我的，至少那天股价会上涨。
If they do that, I think you would agree with me, that day at least the stock goes up.
对。
Yeah.
好的。
Okay.
他们还没这么做。
They haven't done it yet.
不过有传闻说他们会这么做。
They rumored, though.
他们还没这么做，但他们的核心业务，也就是广告投放，现在状态比以往任何时候都要好。
They haven't done it yet, but their core business, which is serving ads, is in better shape than it's ever been.
那不是说他们第一次出现用户下滑吗？
That's not they had their first decline in users.
不是。
No.
不是。
No.
不是。
No.
不是。
No.
不是。
No.
核心广告业务，就是他们广告业务的盈利能力。
The core ad business, like the profitability of their ad business.
对。
Right.
但我的意思是，如果……
But So I'm really saying if
用户和支出。
the users the spending.
不过用户数量在下降，那也会跟着下降。
The users are declining, though, that will decline too.
再说，我明白你的意思。
Again, it's I understand what you're saying.
我是说，目前来看，情况还不错。
I'm saying for now, it looks good.
这是我一直以来的老问题。
And this is the eternal problem I've had.
就是，总有一天会出问题，而我不再给时间表的原因就是因为我之前判断错了。
It's like, at some point, this goes wrong, and the reason I'm no longer giving timelines because I was wrong.
我不怪你。
I don't blame you.
总有一天，所有事情都会出问题。
At some point, everything goes wrong.
但这次出问题的方式会非常糟糕，因为不只是回到需要救助的那种情况。
But this is the way this goes wrong is astronomically bad because it's not just getting back to the bailout thing.
不只是说，哦，OpenAI和Anthropic得活下去。
It's not just, oh, OpenAI and Anthropic have to survive.
他们得变得像谷歌或微软那么大，而且每年光这两三家公司就要花得比赚得多。
They have to become 10 they basically have to become the size of Google or Microsoft and spend more than they make every year just on those two two or three companies.
所以这个没法救。
So you can't bail that out.
而且，就算我们假设他们缩减规模，只建四分之一的数据中心，那也还是有607万亿美元的数据中心需求需要满足，但实际上满足不了，因为大多数AI初创公司都不赚钱，我是说，所有AI初创公司都不赚钱。
On top of that, if we build let's say, they cut back the the process and they only build a quarter of the data centers, we're still talking $607,100,000,000,000 dollars of data center demand that needs to exist that won't because most AI startups are on well, I mean, all AI startups are unprofitable.
那钱从哪儿来呢？
So where's the money gonna come from?
你要救数据中心吗？
Do you bail out the data centers?
你要救所有数据中心吗？
Do you bail out all of them?
救完之后它们怎么办？
What happens to them afterwards?
还有，怎么解决我说的云增长的问题？亚马逊、谷歌和微软的大部分增长都来自云服务，虽然亚马逊的广告业务意外地好，但云增长主要靠Anthropic和OpenAI。
What how do you bail out the fact that the point I'm making around cloud growth, which is most of the growth for Amazon, Google, and Microsoft, though Amazon has a surprisingly good ad business, comes from cloud, and most of that growth comes from Anthropic and OpenAI.
没有它们，增长故事就结束了，而且我其实觉得最终会……
Without that, their growth story ends, and they I actually think will eventually
你觉得他们能上市吗？
Do you think do you think they'll be able to come public?
我觉得Anthropic可以。
I think Anthropic will.
我觉得OpenAI会面临Anthropic上市带来的问题，而且我认为他们的财务状况会很糟糕。
I think OpenAI is gonna have the problem of Anthropic going public and having I think their economics are gonna be bad.
我看过OpenAI的实际数据。
I've seen OpenAI's I've seen their actual numbers.
真的很糟糕。
They're bad bad.
而且我——
And I
糟糕的意思是，是的，收入增长很快，但支出增长得更快。
and Bad meaning, yes, there's a lot of revenue growth, but the expenses are ramping up even faster.
对。
Yep.
收入130.7亿美元，亏损209亿美元。
$13,070,000,000 of revenue, loss of $20,900,000,000.
如果它表现还行的话。
If it does okay.
SpaceX也在亏很多钱。
SpaceX losing losing a lot of money too.
对。
Yeah.
如果它们真的能上市，并且在华尔街受到还算不错的欢迎，不管是这两家公司中的哪一家，而且它们能筹集到可能创纪录的IPO股票销售，它们真的能筹到那么多钱，那会不会减轻你对未来数据中心需求来源的一些担忧？
If it does manage to go public and get a reasonably warm reception on Wall Street, either or both of those companies, and they do raise what will probably be record setting IPO share sales, they do raise that kind of money, Does that alleviate some of your concern on the sources of all the data center demand going forward?
我的意思是，短期内，因为1000亿美元……
I mean, in the short term, because a $100,000,000,000
所以这确实会推迟问题。
So it does push it out, though.
会推迟。
It pushes out.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。