最新13F文件披露：Leopold的幽灵投资组合与不断变化的AI交易

这周，美国所有知名投资者都公开了他们的投资组合。

This week, every famous investor in America published their investment portfolios for the world to see.

我们说的是超过两千亿美元的投资，涵盖了整个AI产业链，来自沃伦·巴菲特、雷·达里奥，甚至是我们的好朋友利奥波德·阿申布伦纳。

We're talking about over $200,000,000,000 worth of investments across the entire AI stack from the likes of Warren Buffett, Ray Dalio, and even our good friend Leopold Aschenbrenner.

从表面看，这可能显得相当悲观。

And at surface value, it might look like it's quite bearish.

他们卖出了大量主要股票，包括英伟达和一些存储芯片股。

They sold a lot of major stocks, including NVIDIA and some memory stocks as well.

不过仔细一看，在人工智能这个主题下有两个特定的细分领域，他们不仅加大了投入，还把整个蛋糕做大了。

But if you look carefully, there's two specific layers within the AI theme that they've not only doubled down on, but have increased the pie.

人工智能领域的资金规模前所未有，今天这期节目我们就来逐一拆解这些内容，还有更多干货。

There's never been more money in the AI sphere before, and we're gonna unpack all of these and more on today's episode.

我们在7月31日录过一期节目，当时提到了利奥波德的13F持仓报告，还说他的下一份13F文件将会是史上最令人兴奋的一期内容之一。

There was an episode that we recorded on July 31 in reference to Leopold's 13F where we said his next 13F filing is going to be one of the most exciting episodes of all time.

而且说实话，我们现在就在这里

And listen, we are here

这太令人兴奋了。

at this exciting.

确实很让人激动。

It's exciting.

既安全又稳妥。

It's safe and guaranteed.

我得说，这跟我们当初预想的走势不太一样，因为我们现在看到的基本上是一份回忆录式的备忘录。

I will say it's not exactly how we imagined this forecast going because what we're looking at here is basically a memoir memoritum.

有人去世的时候该怎么说？

How do you say it when someone died?

就像利奥波德说的。

Like Leopold said.

暂停期。

Moratorium.

对。

Yeah.

暂停期。

Moratorium.

对。

Yes.

是的。

Yeah.

这就是我们目前看到的情况。

That's what we're looking at.

我们看到的是一份来自一家已经不复存在的公司的文件，因为大家都知道，Leopold 公司悲剧性地被清算了。

We're looking at a sheet of paper from a company that no longer exists because as we know, Leopold was liquidated tragically.

我们先从13F表格开始看吧，这样大家就能对这里面的情况有个大概的了解。

Now '13 f's, we might want to start there just so you kind of have an idea of what's going on here.

简单来说，任何交易超过一亿美元股票的上市公司，都必须按季度披露其持仓情况，而且要在每个季度结束后的四十五天内完成。

Basically, any company that trades equities over a $100,000,000 must disclose their positions quarterly within forty five days of the quarter end.

所以，我们在这期节目里要看的，就是全美乃至全球顶级投资机构在这些快照中的持仓表现，并试图从中得出一些结论。

So what we're looking at here and throughout this episode is what those snapshots look like across all the top investment firms in the country and in the world, and we're gonna try to derive some conclusions from that.

不过开头咱们得先看看利奥波德的持仓，毕竟大家都想知道，到底是哪些股票把他搞垮的。

But to start, we have to go to Leopold's portfolio because everyone's gonna wanna know what were the stocks that he was in that blew him up.

因为上一集里咱们已经讲过了，如果你还没看的话，我强烈建议你去补一下，这位华尔街的宠儿现在已经不再持有公开的头寸了。

Because as we know from the previous episode, which if you haven't watched, I highly suggest, the Wall Street darling no longer owns public positions.

而且我们手头有一些数据，看起来他最大的持仓远远超过其他，分别是闪迪和美光。

And it's because we have some numbers, and it looks like the largest holding by far was, SanDisk and Micron.

他当时可是重度押注内存板块。

He was very much memory pilled.

我觉得最难的地方在于，从方向上看，他其实是对的。

I think the tough thing about this is, like, directionally, I think he was correct.

但他最后彻底爆仓了。

He just got totally liquidated.

不过，SanDisk 的仓位占了整个投资组合的 28.5%。

But SanDisk's position was 28.5% of the book.

Micron 的仓位则占了 28%。

He had Micron at 28% of the book.

然后，我的意思是，加起来的话，美光就占了超过一半的仓位。

And then, I mean, together, that's over half of the book is in Micron.

然后是Bloom Energy，占百分之九点五。

And then there's Bloom Energy at nine and a half percent.

我们知道他在那上面赚了多少。

We know how much he was up on that.

还有台积电、Nebius和Core。

We have TSMC, Nebius, Core.

我们还有几只新云股，其余的加起来大约占投资组合的18%。

We have a couple of the Neoclouds, and then everything else combines for about 18% of the portfolio.

所以他在这类内存股上投入了极大的比重。

So he was heavily, heavily, heavily involved in these memory stocks.

我们知道他用了杠杆，看起来正是这一点导致了他的爆仓。

We know he was using some leverage, and it appears as if that's what blew him up.

对。

Yeah.

所以最让人难过的是，这些仓位基本上都亏光了。

So the saddest part about this is probably all of these positions have been decimated.

我好像看到彭博社一周前报道过一些传闻，说他现在手里大概还有五亿到十亿美元的敞口股票头寸，不过目前大部分资产还是押注在Anthropic上。

I I saw rumors, I think, reported from Bloomberg about a week ago that he now has a casual, you know, 500 to $1,000,000,000 open equity position, but the majority still lies in his Anthropic stake for now.

但我觉得如果我们从Leopold的整体投资逻辑来看，也就是加倍押注AI基础设施，特别是内存和新型云服务商，这个逻辑依然非常成立且有效。

But I think if we look at the Leopold thesis in general, which was doubling down on AI infrastructure, particularly in memory and Neoclouds, that thesis is still very alive and well.

咱们先暂停一下，聊聊他的内存仓位。

And let's like pause on the memory position.

这个月走势挺颠簸的。

It's had a pretty rocky month.

你还记得吗，大概两周前我们报道过SK海力士创下史上最佳季度业绩。

If you remember, I think two weeks ago, we were reporting on SK Hynix reporting their best quarterly earnings ever.

他们一个季度的利润比去年全年都多，但股价在那之前的一周却暴跌了20%左右。

They made more money in a single quarter than they did in an entire year last year, but their stock tanked about 20% the preceding week.

当时的逻辑是，大家觉得内存投资者杠杆加得太高了，而且市场可能已经见顶，因为GPU基础设施、英伟达以及所有购买内存的客户都已经不想再买更多内存了。

And so the thesis behind that was people thought that memory investors were too over leveraged and that we were reaching a top where GPU infrastructure, NVIDIA, and all the people that were buying memory could not want to buy more memory.

SanDisk在投资者日上透露的消息显示，情况不仅不是这样，反而我们今天这期要讲的那些大型投资者都在近期低点加码买入了更多。

Recent news has been revealed from SanDisk over their investing day that not only is that not true, but all these major investors that we're gonna be covering in this episode today have doubled down and bought more, especially at the lows recently.

而且尤其是SanDisk，他们自己开发出了一种新型内存。

And also SanDisk in particular has developed this new type of memory.

它叫做高带宽闪存。

It's called high bandwidth flash.

我想花大概20秒时间解释一下这是什么。

And I just wanna take like twenty seconds to describe what this is.

内存分好几种类型。

There's a couple different types of memory.

有DRAM，就是那种标准内存。

There's DRAM, which is like your standard memory type.

还有HBM，也就是高带宽内存，是专门针对AI GPU设计的。

There's HBM, which is your high bandwidth memory, which is specialized for AI GPU specifically.

另外还有一种叫高带宽闪存的东西，这是SanDisk发明的，坦白说基本上也是他们说了算。

And then there's this thing called high bandwidth flash, which is this new thing, which was created by SanDisk and honestly dominated by them.

它主要用于推理。

And it's used for like inference mainly.

而且最近推理已经成为AI领域的主要趋势。

And inference recently has become like the major trend in AI.

这实际上消耗了谷歌、亚马逊等超大规模企业大量的资本支出。

It's actually consuming so much CapEx spend for the hyperscalers from Google, Amazon, and the likes.

因此，闪迪将从中获益最多。

And so SanDisk in particular stands to benefit the most from this.

他们这款新内存已经接到订单，甚至排到了2027年底。

And they have this new memory type, which is being ordered or backlogged already into the end of twenty twenty seven.

所以这对他们来说是个全新的收入增长点，而且我们以前从未见过需求这么火爆的情况。

So this is a new revenue opportunity for them, and we haven't seen exceeding demand like this ever before.

所以说，内存板块的逻辑依然非常强劲，Leopold的判断是对的。

So the point is memory, the memory argument is still very strong and Leopold was right.

他只是之前杠杆加得太高了而已。

He just, he was too overlevered, man.

这就是最可悲的地方。

That's the tragic part.

就是他的投资逻辑是自洽的，现在也依然成立。

It's like the the thesis is coherent, and it still stands.

而且他的持仓以及他对市场方向的判断都没有问题。

And there was nothing wrong about his positions or his, like, directional ideas of where the market was going.

他只是操作全错了，因为他仓位管理不当，最关键的是用了杠杆。

He just did it all wrong because he was using improper position sizing and most importantly leverage.

我是说，这就是用杠杆会带来的后果。

I mean, this is what happens when you use leverage.

这一点所有人都知道。

Everybody knows this.

所以我觉得这挺有意思，相当于看到了本来可能会发生什么。

So I think it's an interesting peek into what could have been.

就方向来看，那些个股看着都挺强的。

Like, directionally, the stock names look strong.

只是他的仓位配置看起来没那么稳。

It's just the way he positioned didn't look nearly as strong.

这周还有一家公司公布了他们的13F文件，我觉得值得提一下。说到最强劲的投资者，他们绝对是最顶尖的那一档。

There is another company who reported their 13F this week that I think it's worth highlighting who is when you think of the strongest, they're about as strong as it gets.

这些大佬在圈子里待的时间太久了。

These people have been around forever.

说的就是沃伦·巴菲特的伯克希尔·哈撒韦。

This is Warren Buffett's Berkshire Hathaway.

它们总是特别有意思，因为它们的走势跟苹果很像。

They are always really interesting because they move in a way similar to the way Apple does.

它们动得很慢。

They move very slow.

通常都是最后才动的那一个。

They're normally the last mover.

一旦动起来，那体量巨大，而且信心十足。

When they do so, they do so with a tremendous amount of size and conviction.

大家都知道，伯克希尔·哈撒韦现在不再由沃伦·巴菲特掌舵了。

Now as we know, Berkshire Hathaway is no longer run by Warren Buffett.

遗憾的是，查理·芒格也不在了，现在是由格雷格·阿贝尔接手管理。

Unfortunately, no longer run by Charlie Munger either, and Greg Abel is now at the helm of this.

所以格雷格·阿贝尔是这位新掌门人，他加入伯克希尔后，开始以相当高的频率进行资本部署。

So Greg Abel is the new guy in charge, and he has joined Berkshire and has begun deploying capital at pretty high rates.

本季度最大的一笔增持是170亿美元，买的是谷歌（也就是Alphabet），这真的让人挺意外的，因为大家会想，他们为什么不去买那些存储芯片公司，反而去买谷歌呢？

The biggest ad of the quarter, which was 17,000,000,000 added, was for Google or Alphabet, which was really surprising to see because why would they be buying Google instead of buying the memory companies?

我觉得我们有个核心观点。

And I think we we have, like, a thesis.

我先讲讲具体情况，然后再说原因。

I'll cover the rest first before we get into why.

看起来他们大幅增持了达美航空、梅西百货和兰尼尔的仓位，增幅都在44%左右。

But it looks like they've added 44% of Delta, of Macy's, of Lennar.

这说明他们的操作思路很像乔舒亚·库什纳那套风格，意思就是：行，就这么干。

So they're very much moving towards a this is kind of like the Joshua Kushner thing where they're like, okay.

谁已经存在了很久，然后谁会在前沿领域获胜？

Who's been around forever, and then who's going to win on the frontier?

这感觉就像是沃伦·巴菲特和伯克希尔在这里给自己定位的方式。

And that feels like kind of how Warren Buffett and Berkshire are placing themselves here.

但关于谷歌的事情，我想谈谈谷歌的事情，Ejaaz。

But the Google thing I wanna talk about the Google thing, Ejaaz.

他们为什么买这么多谷歌的股票？

Why are they buying so much Google?

所以我认为这是我们俩多年来一直非常喜欢的论点。

So I think it's a thesis that you and I have like been very fond of for over like many years now.

谷歌是世界上唯一一家拥有完整AI堆栈的公司。

Google is the only company on earth that owns the entire stack for AI.

现在，我只想强调这有多重要。

Now, I just wanna emphasize like how important this is.

他们不仅是AI模型的创造者，还是运行和训练这些AI模型的自定义GPU的创造者。

Like, they're not only creators of AI models, but they're also creators of custom GPUs that run and train those AI models.

他们也是分发平台的创造者。

They're also creators of the distribution platforms.

顺便说一句，全球第一的就是谷歌搜索。

The number one in the world, by the way, you've got Google search.

显然，还有Gmail、G Suite、安卓，以及众多谷歌产品，每天覆盖数亿用户。

Obviously, you've got Gmail, you've got G Suite, you've got Android, you've got so many different Google products that distribute to hundreds and hundreds of millions of people every single day.

所以他们拥有终极分发平台，同时也拥有所有基础设施和所需的资金支持。

So they have the ultimate distribution platform, but they also own all the infrastructure, the power that is required to fund all of this.

而且他们的资本支出非常惊人。

And they have amazing amounts of CapEx.

我觉得今年谷歌在支出上确实拔得头筹。

I think Google actually this year takes the crown for spending the most.

他们预计到年底会花费大约2500亿美元。

I think they're on track to spend around $250,000,000,000 by the end of the year.

如果谷歌是唯一一家在AI整个产业链上拥有或主导的公司，那它绝对是一家值得拥有的公司，因为他们可以针对自己的产品优化每一层技术。

If Google is the only company that owns or dominates across the entire stack in AI, that is an amazing company to own because they can fine tune each of the different layers for their particular products.

所以这很重要，比方说你看一下谷歌的搜索数据，发现客户在一年中的这个时候特别喜欢在某一个特定区域买房。

So why that's important is, let's say you look at Google search data and you notice that customers really like buying houses in a specific area at this time of the year.

他们就能让自己的AI模型去生成并推送广告或建议，引导人们以某种特定的方式进行购物。

They can orient their own AI model to produce and serve up adverts or suggestions to people to shop in a particular type of way.

这种统一性或者说极度精准的特性，如果你只是个模型实验室或者只是生产GPU的公司，真的很难做到。

And that kind of uniformity or like hyper specificity is really hard to nail down if you are just a model lab creator or if you are just a GPU producer.

所以说这对谷歌的整体论点来说是非常有力的支撑。

So that's really powerful for the thesis in general for Google.

Josh，我想问你一个问题：你觉得谷歌内部出现动荡、德米斯·哈萨比斯退居二线、杰夫·迪恩离职之前，就已经发生这种情况了吗？

Now, my question to you, Josh, is do you think this happened before the fallout within Google when Demis de Sibis stepped aside, Jeff Dean left?

我感觉是的。

I have a feeling it is.

所以我猜他们从那以后缩减了布局，但我依然认为谷歌是个不错的选择。

So I wonder if they've cut down their position since, but I still think Google's a good bet.

对。

Yeah.

我觉得这件事之前就发生过，而且我也不认为这会真正改变他们的想法，因为我不觉得这会对业务产生实质性影响。

I would bet this happened beforehand, and I would bet that it's also not really going to change their mind because I don't think it materially affects the business.

事实上，这可能还是件好事。

In fact, it may be a positive.

我记得很久之前看到过一篇帖子，大概就是说，谷歌如果不开始清理那些把它带到这个境地的领导层，就不算真正对自己下狠手。

It's like, I remember seeing this post from a long time ago that was basically like, Google isn't taking themselves seriously until they start firing the leadership that put them in this position.

而且，某种意义上，现在发生的正是这个。

And, I mean, in a way, that's kind of what's happening here.

那些让谷歌走到这一步、手握所有技术却没能兑现机会的领导层，正在开始被轮换和淘汰。

The leadership who put them in this position where they had all the technology did not capitalize is starting to rotate and cycle out.

新血液进来，肯定非常有干劲。

There is Fresh Blood, I'm sure is very hungry.

谢尔盖·布林也回来了。

Sergey Brin is back.

他现在几乎是处于战时状态。

He is kind of in wartime mode.

而且，我是说，谷歌真是一家非常了不起的公司。

And, I mean, Google is this remarkably incredible business.

在我们继续看剩下的13F文件时，我们会发现，剧透一下，本季度最大的赢家是亚马逊和谷歌。

And what we're gonna find as we go through the remaining 13F filings here is that I mean, spoiler alert, the two biggest winners this quarter are Amazon and Google.

大家都喜欢超大规模云服务商。

Everyone loves the hyperscalers.

大家都喜欢那些投入巨额资本开支的公司，因为他们很容易预测收入。

Everyone loves the people who are spending huge amounts of CapEx because it's easy for them to project revenues.

对于这些新公司来说，当他们花很多钱时，风险很高。

For these newer companies, when they spend a lot of money, it's high risk.

他们还没有证明自己的长期收入能力。

They haven't proven their revenue over time.

但就谷歌而言，他们对谷歌云服务的利润率预期高得惊人。

But in the case of Google, they're projecting unbelievable margins on the Google Cloud services.

他们越来越有能力自研硬件芯片，比如他们的TPU和Ironwood，而且他们建立了整个生态系统，护城河非常深。

They have the increasing capability of building their own hardware chips like their TPUs and their Ironwoods that they have, and they have this entire ecosystem that they've built with a very strong moat.

而且看起来这种势头短期内也不会消失。

And it seems like that's not going anywhere anytime soon.

所以说谷歌真是家了不起的公司。

So Google is just this unbelievable business.

亚马逊也一样。

Same thing with Amazon.

它们的资产负债表上都坐着巨额资本。

They're companies with huge amounts of capital on their balance sheets.

而且它们也在大力投入这些资金。

They're companies that are spending it.

这些东西可真是被低估了啊，老兄。

These things are pretty underpriced, man.

它们确实被低估了很多。

They're they're pretty underpriced.

相对而言，它们的估值倍数并不算高。

On a relative basis, they're not trading at a very high multiple.

尽管它们是唯一有可靠业绩记录的公司，每个季度都在展示，嘿，我们投入了这么多资本支出，但看看这些资本支出带来的回报。

Even though they're the only companies with proven track records that are showing quarter over quarter, hey, we're spending all this CapEx, but look at the returns that we've had on this CapEx.

关于谷歌，有一个惊人的事实，我不知道具体数字，但我记得读到的时候，心想，天哪。

There's this amazing step about Google, and I have no idea what these specifics are, but I remember reading it and being like, oh my god.

这太对了，谷歌每投入一美元，就能产生大约2.8美元的回报，从公司成立以来一直如此。

This is so right, where every dollar Google invested yielded, say, like, $2.8 in return or something like that, where since since the inception of the company.

所以当你听到像谷歌这样的公司，它是过去三十年里全球最优秀的资本配置者之一，说我们不仅要花很多钱，还要花创纪录的钱，因为我们坚信这些投资长期会带来回报。

So when you hear a company like Google, who is one of the best capital allocators in the world over the course of last three decades saying, not only are we going to spend a lot of money, we're going to spend a record amount of money because we are so high conviction that these investments will pay off in the long run.

我觉得你应该对此感到非常兴奋，因为在资本配置方面，很少有人比谷歌做得更好。

I think you gotta get pretty excited about that because as a capital allocator, there's not many people better than Google.

所以这就是为什么我推测，伯克希尔会在这上面持有相当大的仓位。

So that is why I presume, Berkshire, we see taking a pretty large position in this.

但现在我们得转向一个让我们感觉像家一样的地方。

But now we have to move over to what feels like a home for us.

我觉得这是Brad Gerstner、Gavin Baker和NVIDIA的投资组合。

I think this is the Brad Gerstner, the Gavin Baker, the NVIDIA portfolio.

这些就是我觉得最值得投资的公司，至少对我来说，这种感觉最强烈。

These are what the the companies that I think, like, when we think of the companies we wanna invest in, this feels most true to me at least.

它们的第一大持仓都是英伟达，大概20亿美元，这简直太疯狂了。

The number one holding of all of them, NVIDIA, like $2,000,000,000 is it's just crazy.

布拉德·格斯特纳的持仓大概是19亿美元。

It's like $1,900,000,000 they have.

排在第二位的是Cerebras，持仓16亿美元。

Second for Brad Gerstner is Cerebras at 1,600,000,000.

还有Meta、台积电、CoreWeave和ARM。

Then there's Meta, TSM, CoreWeave, ARM.

而Gavin Baker那边，在SpaceX上有47亿美元的头寸，另外还有美光、Cerebras、Astera Labs、Sienna和Credo。

Gavin Baker, on the other hand, $4,700,000,000 position in SpaceX, then Micron, Cerebras, Astera Labs, Sienna, and Credo.

后面那些公司我压根儿都没听说过，但我们可以这么理解：前两位可能就像Twitter上的那些热门交易员一样。

Those bottom ones, don't even recognize, but that's kind of how we could think of the top two, perhaps, like, Twitter traders.

看看X平台上那些顶级交易员，他们目前的持仓情况大致就是这样。

The the top traders on x, here's kind of the position that they find themselves in.

对。

Yeah.

如果要把这些投资者归为一类，我觉得他们属于那种着眼未来的群体。

Mean, if I were to bucket these investors in a particular group, it's like the forward looking group.

对吧？

Right?

他们算是风险偏好更高的类型，但手里握着的资金盘子也大得多。

They're like the higher risk takers, but they're playing with, like, much bigger capital allocations.

你提到加文·贝克的 SpaceX 46 亿头寸，挺有意思的。

It's funny you mentioned the the SpaceX 4,600,000,000 position from Gavin Baker.

这是他最大的盈利。

This was his biggest win.

我昨天刚听了他在《All In》播客里的节目，他基本上就是说，是的，这确实是他最大的盈利之一，但他现在最兴奋的是电力和光模块这个赛道，这看起来是这些特定投资者特别关注的一个大主题。

I I was listening to a podcast of him on the all in show just yesterday, and he was basically like, yeah, this is like one of my biggest winnings, but like what I'm most excited about right now is the power and optics trade, which seemingly is a big theme amongst these particular investors in particular.

所以简单给你梳理一下现状，当你投资 AI 时，大家通常关注的标的像英伟达之类的。

So just to give you a kind of like lay of the land, when you invest in AI, the typical names that people look at are things like NVIDIA.

他们会关注一些我们已经提到过的超大规模云服务商，比如谷歌和亚马逊。

They look at some of the hyperscalers that we've mentioned already, which is like Google and Amazon.

然后如果你再深入挖掘一层，你会发现哦，这些巨头都需要大量的内存。

And then if you kind of dig a layer deeper, they're oh no, all these guys are going to need a lot of memory.

所以你会投资像闪迪、美光、SK海力士这样的公司。

So you invest in the likes of SanDisk, Micron, SK Hynix.

这些都是我们这档节目之前多次提到过的名字，但有一个话题我们还没怎么深入讨论过，那就是电力。

These are names that we've already mentioned on the show multiple times before, but something that we haven't really opined on is this idea of power.

所以，既然英伟达造了这么多昂贵的GPU，又配了那么多内存，但要想让这些设备跑起来，还是得靠电力。

So if you have all these expensive GPUs that NVIDIA is making, if you have all this memory to make these GPUs, you still need power to turn these things on.

另外，这些GPU还会产生海量数据，需要在它们之间进行传输。

But also there is tons of data that these GPUs are generating that they need to kind of transfer between these GPUs.

加文·贝克就是非常早地意识到了这一点。

And that's something that Gavin Baker specifically has identified really early on.

他通过投资像Coherent Corp这样的公司，赚了不少钱。

And he's winning quite a lot through the likes of companies like Coherent Corp.

那我快速解释一下为什么这很重要。

So let me explain sort of very quickly why this is important.

所以你看，光子学这个领域，简单说就是光。如果你的数据中心里有很多GPU，GPU之间传输数据会变得非常折腾，尤其是当你堆到几十万甚至上百万个GPU的时候——马斯克基本上就是干这件事的人。

So if you look at the sector of photonics, which basically means light, when you have a ton of GPUs in your data centers, it gets really arduous to transfer data between these GPUs, especially if you like rack up like hundreds of thousands and millions of GPUs, which is what Elon Musk is basically doing.

现在，GPU之间传数据靠的是铜线，用的电力。

Right now, the way to transfer data between GPUs is using copper wires and it uses electricity.

它还是挺好的。

It's pretty good.

铜线已经用了好几十年了，但现在已经到了一个节点：GPU太多了，这种方式就变得特别低效，成本上也划不来。

It's been used for like decades now at this point, but we've reached a point where there's so many GPUs that it becomes incredibly inefficient and cost inefficient as well.

结果就是又费电又费钱。

So you waste power and you waste a lot of money.

解决办法基本上就是用光。

The solution to this is basically using light.

你没听错，就是用光，真的光，来传数据。

And you heard me correctly, like literal light to transfer data.

这解决了一个问题，不仅成本大幅降低，还能在很长的距离内实现高速数据传输。

And this solves a problem where it's much, much cheaper and you can transfer data at rapid speeds over very great lengths.

而实现这一点的唯一途径，就是通过像Coherent和GE Vernova这样的公司，它们主要提供材料和那些能安装在GPU上的微型基础设施组件。

And the only way to do this is through companies like Coherent, through companies like GE Vanova, who basically create the materials and the kind of like tiny infrastructure tools that kind of like latch onto your GPUs that'll allow yourself to do this.

多年来，这个交易方向已经逐渐成为市场共识。

And we've seen this trade become quite consensus over the years so far.

你看Coherent Group就是这里的主要制造商之一，Gavin Baker也持有了相当大的一笔仓位。

You look at Coherent Group is kind of like one of the major manufacturers here, Gavin Baker's taken a a pretty big position in.

年初至今，它们的涨幅已经达到了85%。

And year to date, they're up 85%.

如果你看看我们这里展示的这个持仓情况，他重仓了Astera Labs这家公司，主要做GPU之间的互联技术。

If you look at this position actually here on our artifact that we have here, he's taken a major position in Astera Labs, which helps with the interconnects between GPUs.

说白了，就像水管、布线一样，负责在大量GPU之间传输海量数据和信息，资金都流向了这个领域。

So basically like the plumbing, the wiring, the ability to transfer tons of data and information across the GPUs is where all the money has been flowing.

而且这不只是Gavin Baker一个人的操作。

And it's not just a Gavin Baker thing.

这是布拉德·格斯特纳的观点，传统投资者比如瑞·达利欧的看法也差不多。

It is a Brad Gerstner thing, and it is pretty similar across traditional investors as well, including like Ray Dalio.

所以看到这一点挺有意思的，我觉得我也在密切关注这个趋势。

So it's cool to see, and I think it's something that I'm keeping my eye on quite astutely, actually.

你可以大致根据公司规模和风险敞口来判断它们是否已经超出了合理区间。

You could kinda see these companies based on their size and their risk exposure going out the curve.

对，确实如此。

Like, we have Yes.

还有伯克希尔，它可是核心中的核心。

Berkshire who's right at the core.

你买了亚马逊。

You're buying Amazon.

你还买了谷歌。

You're buying Google.

这类公司的增长会更慢、更稳健，稳定性也高得多。

Those types of companies are going to be slower, more steady growth, much more stable.

然后是像Gavin这样的人，还有Brad Gerstner这样的人。

Then there's people like Gavin, people like Brad Gerstner.

他们正在走向这个循环的下一个同心圆。

They're going the next concentric circle out of this loop.

他们转向了基础设施层。

They're going to the infrastructure layer.

生成代币需要什么呢？

What is required in order to generate tokens?

这只是表面现象。

It's optics.

这是为了留住记忆。

It's memory.

就像是不断往技术栈的上层延伸。

It's like going further and further out the stack.

我们看到的正是这种情况，而且效果一直很不错。

That's kind of what we're seeing with them, and it has been playing out really well.

我还想提一下加文投资组合里的另一件事，他持有高达23亿美元的QQQ看跌期权。

Another thing I wanted to mention in Gavin's portfolio is that he has a huge $2,300,000,000 position on QQQ puts.

对于不熟悉的人来说，这基本上意味着他购买了以当前价格拥有QQQ股票的权利，QQQ代表指数，所以这相当于为市场下跌买的保险。

And for those that aren't familiar, that basically means he's buying the right to own shares of the QQQ, which is representative of the index, at the current price, meaning it's insurance in the case that it goes down.

我觉得这很有意思，我们在利奥波德的投资组合里也看到了类似的情况。不是所有人，但确实有几个人在投资组合里加入了这种对冲，这样即使市场下跌，他们也能生存。

So I find this interesting too, and we saw this with Leopold's portfolio as well, where everyone has this or not everyone, but there are a few people that have this hedge baked into their portfolios currently where if the market goes down, they can still survive.

他们实际上是在做一个非常集中的押注，在这个案例中，是押注AI硬件，也就是基础设施层。

They're really just making a very specific hyper concentrated bet, in this case on AI hardware, the infrastructure layer.

所以即使他们判断错了，即使市场真的下跌，但只要AI硬件继续表现良好，他们仍然能赢。

So even if they're wrong, even if the market does go down, but AI hardware still continues to do well, they can win.

我觉得他们设置这种对冲很有意思，挺酷的。

And I find it interesting that they do have this hedge put in place, which is kinda cool.

还有英伟达，我也觉得挺有意思。

There's also NVIDIA, which I found interesting.

显然，他们也提交了13F文件，因为作为一家公司，他们持有不少资产。

And apparently, they they also filed a 13 f because they I mean, as a company, they own quite a bit of things.

第一个就是英特尔。

The first one being Intel.

他们持有价值300亿美元的英特尔股票，这

They own $30,000,000,000 of Intel, which is

是很棒的公司。

Great company.

持仓量很大。

Large position.

对。

Yeah.

我是说，这是个不错的处境。

I mean, good good place to be.

挺有意思的。

It's funny.

你通常会把英特尔和英伟达看作是完全的对手，但实际上英伟达在那家公司里持有相当大的一笔股份。

You think of Intel and NVIDIA as, like, total competitors, but NVIDIA owns, like, quite a big chunk of change in that company.

英伟达的第二大持仓是SpaceX，这个大家都清楚。

NVIDIA also has a second biggest holding being SpaceX, which we all know.

这逻辑上完全讲得通。

I mean, this clearly makes sense.

SpaceX会独家采购英伟达的GPU，用来给所有新建的数据中心供电。

SpaceX is going to be exclusively buying NVIDIA GPUs to power all the new data centers.

另外还有CoreWeave。

Then also there's CoreWeave.

还有Coherent这家公司。

There's Coherent.

还有诺基亚。

There's Nokia.

还有新思科技。

There's Synopsys.

现在有个挺有意思的现象，就是大家开始讨论起类似循环经济这类问题了，就是说，好吧。

There's this interesting thing where there is kinda like the circular economy type questions going on where, like, okay.

英伟达投资了CoreWeave和Nebius，他们能不能用这些资金去买英伟达的GPU，然后这些又算作英伟达的收入呢？

NVIDIA invests in CoreWeave and Nebius, Can they take the capital to buy NVIDIA GPUs, which shows up as NVIDIA revenue?

他们把这些公司放在自己的投资名单上，投资它们，然后从它们那里获得收入。

They have these companies on their cap table that they are investing in and therefore collecting revenue from.

但说到底，这基本上就是那些在AI领域一线打拼的人们的普遍做法。

But, I mean, at the end of the day, this is kind of the general play from the people who are in the trenches in the world of AI.

大量的基础设施，那些更高层次的采购，比如英特尔、SpaceX、还有那些新云服务商。

A lot of infrastructure, a lot of those like higher level purchases into Intel, into SpaceX, into the Neoclouds.

这就是他们考虑的事情。

That's what they're thinking about.

如果你考虑投资这个领域，你始终应该问自己的首要北极星指标是：十年后，对AI产品、AI工具、AI模型的需求会比今天更大吗？

If you are considering investing in this space, the number one North Star you should always ask yourself is, will the demand for AI products, AI tools, AI models be greater ten years from now than it is today?

如果你相信答案是肯定的，那么所有这些基础设施，特别是涉及物理实体的，比如GPU、能源、电力，都会是必需的，并且会增长。

If your belief is yes, then all of these things, infrastructurally, especially involved in physical atoms like GPUs, like energy, like power is all going to be required and going up.

当然，你需要确保你支持的是那些真正能执行这些计划的公司和建设者。

Now, of course, you need to make sure that you are backing the types of companies and builders that are actually gonna execute on this.

说到怎么简单理解该支持什么，我基本上就看英伟达的投资组合，因为没有人比黄仁勋更清楚实际在建设什么、需求在哪里。

And when I think of like the dummies guide to what to kind of like support, I look at NVIDIA's portfolio essentially because there's no one better than Jensen Huang who has a much deeper insight as to what is actually being built and where the demand actually is.

他最近在一次访谈或某个会议上说，人们质疑一些AI实验室或超大规模云服务商是否真的看到了他们提供的AI模型和产品的需求，比如微软和谷歌，在财报发布前大家都怀疑他们。

He said on a recent, I think it was like an interview panel, or maybe it was like at one of these conferences that he spoke at, that people are questioning whether some of these AI labs or some of these hyperscalers are actually seeing demand for the AI models and products that they're offering, like Microsoft and Google, for example, prior to their quarterly earnings, everyone was doubting them.

结果他们发布财报后，不仅收入比上个季度增长了5倍，而且利润率也明显上升了。

Then they released their quarterly earnings and not only has their revenue gone up like 5X over the last quarter, but they've also seen revenue margins specifically go up.

同样的趋势也出现在Neoclouds、超大规模云服务商以及一些CPU公司身上。

And it's the same trend that you see across Neoclouds, hyperscalers, as well as some of these CPU companies.

所以你看他最大的持仓是英特尔，价值300亿美元。

So you see his biggest position is a $30,000,000,000 stake in Intel.

我想提醒大家，我记得去年年中，他向英特尔投资了50亿美元，获得了大约5%的股份，对吧？

Now I wanna remind everyone that I think it was midway last year that he put a $5,000,000,000 investment into Intel to own, I think it was like, was it like a 5% stake in them?

他是跟美国政府一起投资的。

He invested alongside the US government.

当时主要目的还不清楚，但现在很明确了。

And the main reason for that was unclear at the time, but today is very clear.

英特尔做的是最尖端的CPU。

Intel creates bleeding edge CPUs.

要调度这几十万块GPU，你就需要CPU这类东西，确保AI代理能正确调用工具，确保它们运行得足够高效和快捷，这样你才能及时得到你问题的答案。

In order to orchestrate all these hundreds of thousands of GPUs, you need things like CPUs to make sure that your AI agents make the right tool calls and to make sure that they operate efficiently and fast enough for you to get your answer to your prompt.

而英伟达投资的这些公司，做的正是这个。

And this is what all of these companies are that NVIDIA is investing in.

你在很多不同的基金里都能普遍看到这种动向。

And you see this collectively across so many of the different funds.

他们都是在押注AI需求会进一步增长。

They're betting on AI demand just being higher.

如果真的是这样，那你就需要更多的”新云”服务商、更多的超大规模云服务商，还需要多得多的AI模型，并且需要一种统一调度的方式。

And if that's the case, then you're gonna need more Neoclouds, you're going to need more hyperscalers, you are going to need way more AI models, and you're gonna need a way to route across all of them.

你还得有足够的电力来支撑这一切。

You need the power to like power all of them.

所以对我来说，这一切就都说得通了。

So this just makes a lot of sense to me.

对。

Yeah.

不，这挺有意思的。

No, it's pretty interesting.

我想现在我们可以快速过一下，看看还有谁留在这儿。

I guess now we could kind of move down to the rapid fire versions of just like who is left here.

值得一提的是比尔·阿克曼这位投资者，他目前的布局主要集中在支付领域，我觉得这点很有意思。

One of the noteworthy investors, Bill Ackman, we have, he is positioning himself mostly in payments, which I find interesting.

他新买入的股票包括Visa、万事达卡、标普全球和Netflix，这有点奇怪。

His new positions are Visa, Mastercard, S and P Global, and Netflix, which is a little bizarre.

你想想看，就会觉得，为什么是支付板块呢？

And when you think about that one, you're like, why payments?

接着你再想想，Stripe最近收购了谁？

And then you think, well, who did Stripe just acquire?

他们刚收购了OpenRouter。

They just acquired OpenRouter.

所以说，支付基础设施看起来确实需要彻底改革了。如果你押注这个方向，那基本上就是在做空Visa和万事达，而不是买入它们。

It's like it turns out payment infrastructure is due for an entire overhaul, and it seems like if you are betting on that, you are almost shorting Visa and Mastercard, not buying Visa and Mastercard.

所以我很想聊一下原因。

So I'd love to talk about why.

我真想问他，大家都在积极动手搞垮这些公司，他为什么还要跟着投这些公司？

I'd love to ask him, like, why are you going along these companies when everybody's actively working to sabotage and destroy them?

比如信用卡支付平台，可能有些人不太清楚，它们通常每笔交易要抽一个固定美分，再加上2.9%、2.7%左右的手续费。

Like, the credit card providers, for those who don't know, they normally take an x amount of pennies and, like, 2.9%, 2.7% of every transaction.

相当高。

Fairly large.

很多人已经不太愿意付这笔钱了，而且看起来AI代理也需要自己的基础设施层来处理这些支付。

And a lot of people are just kinda sick of paying that, and it seems like AA agents are needing their own infrastructure lay in order to make these payments.

所以我觉得Visa和万事达这事儿确实挺有意思的。

So I find the Visa and Mastercard pretty interesting.

还有几个别的值得关注的。

There's a few other noteworthy ones.

有个叫Sam Klarna的家伙，据说他手里握着35亿美元的资产。

There's, this guy, Sam Klarna, who runs apparently $3.5400000000 dollars worth of money.

他的头号持仓是亚马逊。

His number one is Amazon.

我是说，看起来好多人把亚马逊当成第一重仓股。

I mean, I it seems like there's a lot of people in Amazon being the number one.

还有Tiger Global，我觉得他们也是把亚马逊排在第一位。

Tiger Global, I think they had theirs as number one also.

大卫·泰珀的第一大持仓是亚马逊。

David Tepper, his number one was Amazon.

很多人对亚马逊都非常看好。

A lot of people are really excited about Amazon.

我仔细看了看，感觉大家挺困惑的。

I'm just looking at this, and I'm seeing, like, a lot of confusion.

比如比尔·阿克曼在买入那些金融基础设施公司，但他却把亚马逊的仓位削减了25%。

Like, you've got Bill Ackman buying all the finance rail companies, but he's trimming Amazon down 25%.

然后还有塞斯·科勒曼在买亚马逊。

Then you got Seth Coleman buying Amazon.

接着是特里·史密斯在抛售维萨，抛售，抛售Alphabet。

Then you got Terry Smith dumping Visa, dumping, dumping Alphabet.

所以我心想，好吧。

And so I'm like, okay.

你看，这些基金之间其实有些观点是相通的。

Like some of these funds kind of agree with each other.

有些公司并没有这么做。

Some of them don't.

关于你提到的Appen为什么可能去收购那些金融基础设施，以及Stripe刚刚收购OpenRouter这件事，我意识到整个核心逻辑就是所谓的AI底层架构搭建——我想买算力代币，然后把这些代币转化成能帮我业务多赚钱的智能能力。

I think on your point around why Appen might be buying, like, the financial rails and why Stripe just acquired OpenRouter, I'm realizing that, like, the whole thesis of, like, plumbing AI, I wanna buy tokens, and I want to convert those tokens into intelligence that makes my business earn more money.

Stripe在这个方面位置非常合适。

Stripe is perfectly positioned for that.

Visa和Mastercard也处于完美位置，因为他们已经掌握了连接各大公司财务账目的分发渠道。

Visa and Mastercard perfectly positioned for that because they already have the distribution to all of these different companies' financial books.

对吧？

Right?

他们已经在赚钱了。

They're already getting paid.

所以对他们来说，只要打开一个开关，说，嘿，我们也可以为你提供所有这些不同模型的推理服务，这可能会非常有吸引力。

So for them to flip on a switch and say, hey, we can also provide you and serve you the inference for all these different models might actually be really attractive.

也许这是我的阴谋论猜测，但是

Maybe that's my tin foil hat guess, but

对。

Yeah.

嗯，所有这些事情也需要代理，我们还得再谈谈代理，因为我们的朋友和赞助商LEDGER，他们正在为代理构建基础设施，这是一个三部分的解决方案。

Well, all of these things also require agents, and we're gonna have to talk about agents again because of our friend and sponsor, LEDGER, who is building an infrastructure infrastructure for for agents agents in in which which there's there's a a three three part part solution.

解决方案。

Solution.

如果代理提议，人类批准，那么LEDGER签名者就会强制执行正在进行的交易。

If the agents propose, the humans approve, the ledger signer enforces the transactions that are being made.

他们构建了一个非常简洁的账本代理栈，让你可以用他们的开源工具来与代理协作。

They built this really neat ledger agent stack that allows you to use their open source tools to work with agents.

因为很可能，如果你在用代理构建东西，你会非常担心安全问题。

Because chances are, I mean, if you're building with agents, you're pretty worried about security.

最近像OpenAI和Anthropic这样的公司发生了很多越狱事件。

There's been a lot of breakouts from companies like OpenAI and Anthropic recently.

你想把它锁起来。

You wanna lock that down.

你可以通过Ledger的这个硬件流程来实现。

You do so with this hardware process that Ledger has.

他们与Cloud Code兼容。

They work with Cloud Code.

他们与Codecs兼容。

They work with Codecs.

它与Cursor兼容。

It works with Cursor.

这些都是完全开源的，你今天就可以试试，更多信息可以在下方描述里的链接找到。

It's It's all totally open source for you to try out today, and you can find more in the link in the description below.

感谢Ledger赞助本期视频。

Thank you to Ledger for sponsoring this video.

随着节目进入最后环节，我们得回答一个问题，比如，好吧。

And as we get to the final segment of the show, we have to make the you have to answer the question like, okay.

聪明钱到底在买什么？

What is the smart money actually buying?

综合来看，在看了所有这些投资组合和公司之后，最大的赢家和输家是谁？

In consensus, after going through all these portfolios, after seeing all these companies, what are the largest winners and losers?

我们屏幕上有一个很棒的图表，给那些在Spotify、YouTube或任何地方看视频的人看。

And we have this really nice chart on screen for those who are watching on Spotify, YouTube, wherever you get your videos.

它显示了大家都在加仓什么，我看到Tepper的持仓线一路向下，他似乎什么都加了。

It shows kind of where everyone's been adding, and I see this line down the board of Tepper who seems to have added, like, pretty much everything.

Tepper在全面加仓。

Tepper's going across the board.

他是新来的，对。

He's the new Yeah.

抱歉。

Apology.

看起来Tepper真的在大举投资亚马逊、Alphabet、台积电、电力能源、新云和SpaceX。

It seems like Tepper's, like, really making a strong play into Amazon, Alphabet, TSMC, Power Energy, Neoclouds, SpaceX.

但我们看到的是，在所有基金中，Alphabet获得的投资最多。

But what we do see is the most amount of dots out of all these funds across Alphabet.

大家都非常看好谷歌。

Everyone's really bullish on Google.

人们想拥有这家公司，它今年要花几千亿的资本支出，云基础设施利润率很高，而且新的领导团队可能比旧团队更有干劲。

People want to own the company that's spending how many hundreds of billions of CapEx this year with how large profit margins in their cloud infrastructure with, like, a new leadership team that's coming in probably a little more hungry than the old.

这几乎像是谷歌的新开始，希望很多人能从中获利。

It seems like this new beginning for Google almost, and the hope is that a lot of people can capitalize on that.

接下来最受欢迎的，我想这里有三家。

The next most popular, I guess there's three of them here.

就是亚马逊、台积电和SpaceX。

It's Amazon, TSMC, and SpaceX.

这些看起来都是很不错的机会。

Those all seem like pretty amazing opportunities.

亚马逊，我们知道，基本上是七大科技巨头中在AI领域布局最大的公司之一。

Amazon, we know, is pretty much one of the largest mag seven companies for AI exposure.

你有AWS，还有公司对Anthropic的隐秘投资，再加上Tranium芯片，这三个都是能对亚马逊盈利产生重大影响的巨大业务板块。

You get AWS, you get the sneaky Anthropic stake for the company, and then you get Tranium chips, which are all three huge verticals that can play a large role in the earnings of Amazon.

我觉得大家对这类大型科技公司挺兴奋的。

I think people are pretty excited about these large cap companies.

SpaceX，不用说，我们做了25期节目讲为什么它会成为世界上最有价值的公司之一。

SpaceX, again, goes without saying, we have 25 episodes on why they're gonna be one of the most valuable companies in the world.

大家对太空中的AI数据中心感到兴奋。

People are excited about AI data centers in space.

他们对Grok进入前沿领域感到兴奋。

They're excited about Grok getting on the frontier.