利奥波德回归：13F文件揭示新投资

华尔街最著名的AI多头刚刚宣布，整个AI市场已经见顶。

The most famous AI bull on Wall Street just called the top on the entire AI market.

24岁的前OpenAI研究员利奥波德·阿申布伦纳，曾被开除后创办了基金，两年内把两亿五千万美元翻到了一百四十亿美元，现在他回来了，而他的最新投资组合完全出人意料。

Leopold Aschenbrenner, the 24 year old ex OpenAI researcher who got fired, started a fund, and turned $250,000,000 into $14,000,000,000 in less than two years is back, and his latest investment portfolio is not what you'd expect.

他对整个股市彻底转为看空。

He's gone completely bearish the entire stock market.

他做空了AI领域所有巨头，总规模高达80亿美元。

He has taken out an $8,000,000,000 short across the biggest names in AI.

涉及的包括英伟达、AMD、博通，以及整个半导体供应链。

We're talking about NVIDIA, AMD, Broadcom, and the entire semiconductor supply chain.

但事情还没到绝望的地步。

But all is not lost.

他还透露了下一个最大的AI投资方向在哪里。

He also revealed where the next biggest AI investment is going to be.

那就是电力和内存。

It's in power and memory.

他进一步加大了对数据中心的投资，同时还押注了三家全新的公司。

He's doubled down on his investments in data centers, as well as three brand new companies.

我们待会儿会详细聊这些，但先说说最大的变化。

We're gonna get into all of this, but first, let's talk about the biggest changes.

全球最大的公司、人工智能革命的象征、在这轮行情中让无数投资者赚得盆满钵满的英伟达，现在成了他的目标。

Ejaaz, the largest company in the world, NVIDIA, the poster child for the AI revolution, the stock that has made so many investors so wealthy over this run is now in the crosshairs.

这是莱奥波德最大的空头头寸，但你从他的持仓文件里不容易看出来——因为当你看他最看空的资产时，排第一的是VanEck半导体ETF，紧随其后的就是英伟达。

This is the largest short position that Leopold has, and it's not obviously apparent when you're looking at the filings because when you look at the portfolio of what he's most short, we see Van Eck semiconductor ETF is number one, and just beneath that is NVIDIA.

目前，他对英伟达的空头敞口达到了15亿美元，而且是通过看跌期权实现的。

Now currently, he has $1,500,000,000 of short exposure to NVIDIA, and this comes through the form of a put.

如果你不太了解期权，简单说，看跌期权就是给了莱奥波德一个权利，但不是义务，让他能在约定的价格卖出标的资产。

And those from not familiar with it, a put basically just gives Leopold the option but not the obligation to sell the underlying asset at a predetermined price.

也就是说，他买的是一个权利：如果英伟达股价下跌，他就能以更高的价格卖出。

So he basically buys the right to sell NVIDIA stock at a higher price should it go lower.

在他上面还有一个20亿美元的头寸，可能很多人没听说过，代码是SMH，也就是VanEck半导体ETF。

Now there is a $2,000,000,000 position sitting just above this, which is a stock that most people may not have heard before, the ticker is SMH, and it goes by the name of Van X Semiconductor ETF.

在这个ETF里，

Within this ETF,

我查了它的持仓，最大的持仓其实是英伟达，占了20%，也就是说，把前两大空头头寸加起来，他对英伟达的空头敞口达到了19亿美元。

I was looking through the holdings, and the largest holding is actually NVIDIA at 20%, which means if you combine the top two shorts, you get a $1,900,000,000 short position on NVIDIA.

这可能会让很多人感到意外，因为大家都觉得英伟达只会一路上涨，但利奥波德的想法恰恰相反。

And this is probably disappointing to a lot because everyone seems to believe that NVIDIA is on a one way trajectory north, but Leopold seems to think otherwise.

除了这个头寸之外，他还做空了博通、甲骨文、AMD、美光、ASML、英特尔和康宁。

In addition to this position, we have Broadcom, Oracle, AMD, Micron, ASML, Intel, Corning.

这些现在都是他的空头头寸。

These are all short positions now.

你知道吗，英特尔是他曾经最赚钱的股票之一。

And what you'll know is that I mean, Intel, one of the ones that he made his bread and butter on.

英特尔给他带来的收益，比投资组合里任何一只股票都要多。

Intel made him more money than any stock in, I think, the portfolio's history.

可现在他反而做空了英特尔，这些全部都是新建立的仓位。

He is now short on Intel, and these are all new positions.

他做空了博通。

He's short on Broadcom.

如果你不太了解，博通主要是负责构建OpenAI的Stargate项目的公司。

For those not familiar, Broadcom is the company that is responsible mostly for building out OpenAI's project Stargate.

这意味着他实际上是在做空OpenAI和Stargate项目。

That means he's essentially pulling out a short position on OpenAI and project Stargate.

如果你一直看好AI市场，这些名字可就让人担心了。

So there's some concerning names here if you've been bullish on the AI market for a while.

就连康宁这样的光学玻璃公司也不例外。

Even something like Corning, the optical glass company.

这公司在半导体热潮后一直是个高贝塔的标的，而他现在建了很大的空头仓位。

This has been a a big kind of beta play after the semiconductor trade, and he's pulled up a big short position on this.

所以市场上正在出现大量做空头寸。

So there's a lot of shorts that are coming on the market.

就像你说的，空头敞口达到了80亿美元。

There's, like you said, $8,000,000,000 of short exposure.

这比十八个月前基金的价值高出四十倍。

That's 40 times more than the fund was worth just eighteen months ago.

所以他这次押注的规模真的很大。

So this is a huge position he's taking.

是的。

Yeah.

这举动非常激进，尤其是当你意识到他的整个基金策略，都建立在他写的一篇六十四页的文章《情境意识》之上。

It's it's extremely aggressive and it becomes more apparent when you realize that his entire fund thesis was based on an 64 page essay that he wrote called situational awareness.

如果你还记得的话，乔什，核心观点是大力押注半导体，特别是认为未来十年的计算能力将提升好几个数量级。

And the core thesis, if you remember, Josh, is a big bet on semiconductors, specifically that compute flops will increase on multiple orders of magnitude over the next decade.

而他这次明确做的这笔交易——八十亿美元的空头头寸，实际上就是在赌这个观点不成立。

This explicit swing trade that he's made, this nine this $8,000,000,000 short position is effectively a bet against that now.

所以这只能说明两种可能。

So it either indicates one of two things.

要么是他觉得这个交易太拥挤了，预期短期内会有波动和价格下行压力；要么是他自己觉得当初的逻辑出了问题，只是还没公开说明具体哪里出了问题。

One, he thinks that the market is too crowded for this particular trade, and so he's expecting there to be short term volatility and downwards pressure on price, or he just believes something's broken in his thesis and he hasn't, like, spoken about what that might be.

不过事情还没到无可挽回的地步。

Now not all is lost.

如果我们看这张图的右侧，其实还有一份看涨清单。

If you look on the right side of this chart that we're showing, there is a bull book as well.

他仍然持有大量特定类型公司的股票，同时也买入了不少看涨期权。

So he does hold still massive positions in stock equity positions for specific types of companies, as well as taking on call options as well.

那我们来看看他看好的方向。

So let's look at what he's positive on.

CoreWeave 他一直保留着仓位，这公司是他最早期就开始布局的云基础设施投资之一，从基金成立之初就一直在投。

So CoreWeave, he's maintained his position and CoreWeave has been one of his biggest data center or Neo Cloud investments for the longest time, since the start of the fund actually.

他还通过多种方式加杠杆投资 CoreWeave，比如自己私人投资或收购了 Core Scientific，这家公司能帮 CoreWeave 做好他们的业务。

And he's taken levered bets on CoreWeave in many different ways through his own private investment or acquisition of Core Scientific, which helps CoreWeave do its thing.

对不太了解的人解释一下，CoreWeave 是一家新兴云服务商，专门搭建 GPU 集群，然后租给最大的那些人工智能实验室。

So for those of you who aren't familiar, CoreWeave is basically a neo cloud that creates and sets up GPUs and provides them to the biggest AI labs.

他们已经和 Meta、Anthropic 这类公司签了数十亿美元的长期合同。

They've signed multi billion dollar deals with the likes of Meta, Anthropic, and the likes of that.

再往下看，Bloom Energy，这是上个季度他最受关注的投资标的。

And then if you look just below that, Bloom Energy, this was his biggest- Our fan Our fan favorite of last quarter.

Bloom Energy 生产的是可移动的燃气涡轮机，你可以直接运到任何数据中心旁边，现场发电。

Bloom Energy creates these portable gas turbines, which you can kind of like fly into wherever your data center is and generate energy.

现在 AI 数据中心最大的瓶颈之一就是，你有这么多高性能显卡，却没法开起来，因为美国现有的电网根本撑不住。

Of course, one of the biggest constraints for AI data centers right now is that you have all these fancy GPUs, but you can't power them up because the energy grid that is currently here in The US does not work.

效率太低了。

It's not really effective.

所以你需要额外的电力供应方案。

So you need to kind of have supplemental resources.

这就是 Bloom Energy 的作用。

That is Bloom Energy.

他还没清仓，但已经减仓了整整十亿美元。

He hasn't exited this position, but he did trim off a cool $1,000,000,000.

说实话，我完全理解他的做法。

And to be honest with you, I don't blame him.

他的持仓从大约八亿美元，涨到了可能二十五亿美元，就在过去三个月里。

His position went from, I believe, $800,000,000 to about $2,500,000,000 potentially over the last three months.

所以他选择套现一部分很合理，但他还是保留了超过十亿美元的Bloom Energy股票。

So it makes sense that he's taking some money off, but he's still maintaining about just over $1,000,000,000 worth of Bloom Energy stock specifically.

再往下看，他还加仓了多家公司。

And then if you look below this, he's increased a bunch of different companies.

CleanSpark，他增持了。

CleanSpark, he's increased.

Riot Platforms、Applied Digital，还有Iron，也都加了。

Riot platforms, Applied Digital, and Iron.

如果你觉得这些名字有点耳熟，那是因为它们和CoreWeave一样，都押注在新兴的云算力市场。

Now if these names seem a little familiar, that's because they play in the same Neo Cloud market as CoreWeave itself.

他正在全力加码数据中心和新兴云算力领域。

So he's really doubling down on data centers and Neo Cloud specifically.

他注意到Anthropic和OpenAI这些公司不断推出新模型，算力需求还在持续攀升。

He's observed that the likes of Anthropic and OpenAI are releasing newer models and that compute scaling laws are just continually increasing.

所以他主要押注的是，GPU依然必不可少，但眼下大家面临的是交付问题，而这些公司解决的是交付环节，而不是真正制造GPU的厂商，这大概就是他做空英伟达、博通这类公司的原因。

So his big bet is that GPUs is still needed, but right now it's a delivery function that they're facing and these companies solve that versus the actual GPU manufacturer, which is why I guess he short NVIDIA, Broadcom, and all the likes.

这确实挺有意思的。

So interesting to see.

对。

Yeah.

这正在形成一种新的交易逻辑：我们正从半导体本身转向基础设施领域，比如电力、数据中心和内存。他延续了上个季度的思路，但这次还同时做空了那些他认为前景不妙的公司。

This is a new narrative trade that we have forming, where we're kind of moving away from the semiconductors into more of that infrastructure, into the power, into the data centers, into the memory, and he's very much doubling down what we saw last quarter, but doing so in a way where he's also getting short exposure to the companies that he thinks might not do so well.

需要注意的是，这份13F文件只是个快照。

Now it's important to note that this 13 f filing is a snapshot.

它反映的是上个季度交易结束后的一个时间点。

It is a single moment in time that is taken based on the previous quarter's trades.

这份13F文件里列出的持仓，都是从去年1月1日到3月31日之间的操作。

So the trades that are in this 13 f filing, the holdings in there, are from January 1 to March 31.

这基本上就是他在季度末的持仓状态。

This is basically where he ended the quarter.

而且利奥波德几乎每次判断都对了。

And Leopold has been right pretty much every single time.

我们看到他的投资组合从2.2亿美元增长到了现在的137亿美元名义价值，不过他可能也有判断失误的地方。

And we've seen his portfolio grow from 220,000,000 to that, what, $13,700,000,000 in in notional value currently, but there are some things that he might be getting wrong.

我回过头看他的做空名单，AMD是他做空的主要标的之一。

When I go back to that shortlist and I think about the companies that he shorted, AMD is one of those large shorts.

AMD在过去一个月里涨了74%，而他却做空了它。

AMD is up 74% in the last month and he shorted it.

所以他挑了个可能最贵的时机来押注这次风格轮动，不过话说回来，他确实认为轮动正在发生。

So he picked perhaps one of the most expensive moments to bet against this rotation, but nonetheless, he thinks that the rotation is happening.

那这是时机问题吗？

So is it a timing thing?

还是说这是个宏观逻辑的问题？

Is it a general thesis thing?

另一个让人意外的标的就是阿斯麦（ASML）。

Another one that was surprising is ASML.

在我看来，ASML 仍然是唯一能做光刻的公司。

I mean, as far as I'm concerned, ASML is the still the only company that can do lithography.

这些芯片的市场完全被它垄断，可他偏偏做空了它。

It's a 100% monopoly on these chips, and he's shorting that.

很明显，他的核心观点是更看好内存、电力和基础设施，而不是半导体，这点值得提一下。

So clearly, the thesis really is strongly presented towards memory and power and infrastructure over the semiconductors, and I think it's just noteworthy to mention.

现在我们有这两张图表，是按数据整理的，显示了他的股票持仓和主要期权持仓。

Now we have these two charts, the book by the numbers that shows the stock only positions as well as the top options positions.

我们不妨简单过一遍，看看他目前具体持有哪些标的，分别在哪儿。

Maybe we wanna walk through that briefly just to do a little trace over of what he currently holds and where.

我觉得对乔什来说，更有价值的可能是先理清他为什么这么布局，对吧？

Well, I think what actually might be more useful, Josh, is if we walk through what his thesis might be for all of these positions, right?

这些仓位确实很惊人，他到底凭什么这么操作？

Because these are pretty impressive, Why is he doing this?

他现在有个巨大的空头头寸，高达九十亿美元。

We've got like a massive short position, dollars 9,000,000,000.

这可不是个小数目。

That is not a small like number.

然后他还持有一个很小的、奇怪看涨的仓位，但我们不太确定，因为那些都是些新潮的云服务和电力公司，我们根本没怎么听说过。

And then we have this like small, weirdly bullish position, but we're unsure because like they're just a bunch of neo clouds and power companies, which we haven't really heard of.

这些公司到底行不行？

Like, are these good?

这些公司是不是不靠谱？

Are these bad?

我的看法是这样的。

So here's my take.

他其实是在为自己的观点做双向押注。

He's doing a bidirectional trade here for his thesis.

他做空的是半导体。

And that is he is short silicone.

他认为这个方向已经太拥挤了。

So he thinks it's an overcrowded trade.

他认为像英伟达、博通这样的GPU设计公司，还有台积电这类芯片制造商，现在都太拥挤了。

He thinks GPU designers such as NVIDIA, Broadcom, as well as maybe manufacturers of the chips themselves, TSMC, they're all like overcrowded trades right here.

我不觉得他看空这些东西。

I don't think he thinks he's bearish those things.

我只是觉得他认为这些东西目前被高估了。

I just think that he thinks that they're overvalued right now.

而相反，他大量押注电力领域。

And then conversely, he's very long power.

他对数据中心和GPU的理解，没人比得上。

He understands data centers and GPUs unlike nobody else in this entire market.

他研究这些话题已经不知道多久了。

He has researched these things for goodness knows how long.

所以，在所有人里，他最清楚下一个瓶颈会在哪儿。

And so he, of all people, will know where that next constraint is.

很明显，他认为瓶颈在电力和能源上。

And obviously, he thinks it's power, energy.

他认为目前的能源供应，以及把电力输送到GPU的方式，根本不够用。

He doesn't think there's enough energy or ways to get this power and energy to the GPUs itself.

所以，他可能并不是明确看空半导体板块，而是觉得钱花在追逐下一个瓶颈上更划算。

So maybe it's not that he's explicitly bearish on the semis specifically, but he thinks that his money is better spent chasing that next constraint.

顺便说一句，他提到的瓶颈不只是电力，还包括内存。

And he's expressed that as not just power, by the way, it's memory as well.

他还加仓了闪迪，要知道，过去一年它的股价涨了将近四万倍。

He doubled down on SanDisk as well, which by the way, is up like 40000% over the last year.

按理说，这么热的标的早就人满为患了，但他显然看到了我们没看到的东西。

So you would think that if any trade was overcrowded, it would be Sandisk, but obviously he sees something that we haven't.

闪迪以生产一种叫NAND闪存的内存技术闻名，这种技术能为AI模型提供临时存储空间。

Sandisk is famous for creating a specific type of memory known as NAND flash, which allows you to kind of store temporary memory for AI models.

比如你跟AI对话时，它需要记住你的信息，才能在后续交流中回忆起来，这靠的就是这种存储。

Like when you talk to an AI model, it needs to store temporary memory about you so that it can remember stuff and recall stuff when you're conversing with it.

而这正是闪迪所提供的核心能力。

That is explicitly something that SanDisk provides.

所以我觉得他并不是对GPU看空。

So I think that he isn't necessarily bearish, the GPUs.

他只是觉得现在这个交易太拥挤了，钱花在电力和内存上更值得。

He's just short term thinking that it's an overcrowded trade and his money's better spent on things like power and memory.

我现在在想，他对整个市场到底是看多还是看空？

And now I'm wondering, is he bullish or bearish on the market as a whole?

因为这可是这支基金历史上第一次，空头头寸的名义价值和风险敞口超过了多头。

Because, I mean, this is the first time in the fund's history where the short side of the book is actually larger than the long side in terms of notional value and exposure that they have.

对于一个一直只做多、从不做空的人来说，这变化实在太明显了。

And this is a a really stark difference for someone who's been really up only and long only.

我刚开始分析的时候就在想，这难道只是个对冲策略？

And when I was first evaluating this, was wondering, well, is this just a hedge?

是为了保护他的投资，因为涨得实在太多了？

Is this protecting his investments because they've gone up so much?

也许他只是想落袋为安，提前防范下行风险。

Maybe he's just locking it in and he's kind of protecting his downside.

但如果只是对冲，通常只会持有较小的仓位来抵消多头部分，上个季度他就有一些对冲头寸，但那时候主要还是对冲。

But if it was a hedge, you'd really only expect kind of smaller positions sized to offset the long book, and we saw this last quarter where he was he had some hedge positions, but it was mostly a hedge.

那并不是一种方向性押注。

It wasn't a directional bet.

但现在看，他的看跌期权规模已经超过多头了，这明显是在押注市场会下跌。

But now that the puts are larger than the longs, it's kind of a directional bet on the market going down.

所以这看起来有点奇怪，像是他预期AI市场可能会走低。

So it seems like this weird pseudo thing where he expects the AI market to perhaps go down.

但即便如此，像内存、基础设施和能源这些领域还是会继续上涨，这才是他真正押注的方向。

But even as a result of that, some of these things, the memory, the infrastructure, the energy will continue to go up, and that's the bet.

所以我很好奇，这会对整个市场产生什么影响？因为他现在处于一种矛盾的位置：市场整体下跌，但某些持仓却不会跌。

So I wonder how that impacts the broader market as a whole because it puts him in this, like, weird juxtaposed position where the market goes down, but yet some positions will not be going down.

这样理解说得通吗？

Does that make sense?

我们其实有一些证据支持这个观点，我觉得你正好点出了关键——那就是不确定性。

We actually have some proving points around that because I think you're touching on something, which is basically uncertainty.

他也不确定在某些情况下自己是对的，还是市场会上涨或下跌。

He doesn't know whether in some cases he's right or whether the market will go up and down.

你还能看到，他把不少看跌期权和看涨期权配对了。

And you see that with he's matched a bunch of his put positions with call positions as well.

换句话说。

AKA.

是的，这张图非常好。

Yeah, this is a great chart.

对。

Yeah.

他实际上在做双向交易。

He he's trading a bidirectional book.

也就是说，按照对冲基金的典型做法，如果你不确定市场会涨还是会跌，就会对冲头寸，靠这些头寸之间的权利金差价赚钱。

So what that means is in typical kind of like hedge fund mannerisms, if you don't know where the market is gonna go, whether it's gonna go up or down, you hedge positions and you can cream off profit from the premiums that you make between these positions.

比如说，你买了一笔价值一千万美元的某公司看跌期权，另一边再买等值的看涨期权，然后纯粹靠这两者之间的价差获利。

So let's say you take out a, I'm just making this up, a $10,000,000 put position on one company, you can take out the equivalent on the other side and then like just earn from the margins between those different positions.

不管价格上涨还是下跌，你都能拿到权利金，对吧？

If the price goes up or the price goes down, you still get paid premiums, right?

这专门被称为领口策略。

It's known as like a collar trade specifically.

他已经在四家公司这么操作了，其中最大的一笔是美光。

So he's done that on four companies, the biggest one being Micron.

如果他看好闪存，我不明白他怎么会在基本面层面看空美光——毕竟美光可是美国市场上最大的标的。

And if he's bullish Sandisk memory, I don't understand how you would also be bearish from a thesis level on Micron, which is like the biggest US play.

我觉得利奥波德是个很纯粹的人。

You know, Leopold is a purist, I believe.

所以他非常看好美国股票。

And so he's very bullish American stocks.

他赚得最多的一笔就是靠英特尔的交易。

That's why he made his biggest amount of money from the Intel trade.

他现在也在对布卢姆能源、英伟达这些公司做同样的操作。

He's doing it with Bloom Energy, NVIDIA, and the like.

所以，他反而做空美光，这看起来有点奇怪，但我的理解是他这么做是为了中性布局。

So it would seem weird that he's like taking out a short position on Micron, but the way I understand it is he's doing this as a flat trade.

他并不确定股价会往哪个方向走。

He doesn't know which way it's gonna go.

他确实觉得市场现在太拥挤了。

He does believe the market is overcrowded.

他相信长期来看，这股票表现会很好，但他不想去玩那种动不动就上十亿波动的投机交易。

He does believe long term it's gonna do very well, but he can't play around with, you know, a crazy billion dollar swing trade.

所以他选择对冲市场风险。

So he's decided to hedge the market.

我觉得这其实挺聪明的。

And I think that's that's pretty smart, actually.

对。

Yeah.

而且我觉得，可以把这个新观点归纳为四个关键主张。

And I guess you could kind of reduce this down into four key claims to this new thesis.

这是新的所谓Leopold观点。

This is the new, like, Leopold thesis.

第一个是说，瓶颈已经从芯片转向了电力，而这一点我们确实已经看到了。

The first one being that the bottleneck has moved from chips to electrons, and we we have that.

我们知道，现在芯片其实并不缺。

We kind of know that a lot of chips are available.

真正的问题在于，该怎么把它们用起来，最近SpaceX和Anthropic的合作就说明了这一点——Anthropic对算力的需求太迫切了，甚至愿意跟竞争对手联手来搞到。

The problem is figuring out where to plug them into, and we see that with the most recent deal that was announced between SpaceX and Anthropic, where Anthropic is so desperate for compute, they're willing to partner up with their rivals in order to get it.

这根本不是芯片不够的问题。

That is not a matter of not having enough chips.

而是缺乏能大规模部署这些芯片的基础设施。

It's a matter of not having the correct infrastructure to deploy these chips at scale.

第二个观点是，芯片公司的估值还停留在一个早已过时的世界里。

The second claim is that chip valuations are priced for a world that doesn't really exist anymore.

我们之前提到的SMHETF，今年到目前为止已经涨了66%。

That ETF, the SMH ETF that we mentioned earlier, it's up 66% year to date.

与此同时，英特尔的股价已经涨了200%。

Meanwhile, Intel is up 200%.

所以，当市场还在按一种假设来定价，认为所有半导体公司都能从AI需求中同等受益时，利奥波德却指出了这种看法的问题所在。

So while the market is pricing in a world where every name and semis benefits equally from this AI demand, Leopold is taking account of that.

他认为事情根本不是这么运作的。

He's saying that's not how this works.

市场里既有赢家也有输家，而早期的赢家会一直赢下去，他打算顺着这个逻辑继续深挖下去。

There are winners and there are losers, and the early winners are the ones that are gonna keep on winning, and he's going to continue to pursue that as far as he could take it.

我也刚注意到一件特别离谱的事，Josh。

I just noticed something crazy as well, Josh.

如果我们回顾一下他之前做的那些多头仓位。

If we if we go back to these, like, long positions that he's taken.

对。

So he's Uh-huh.

他继续持有 CoreWeave 的仓位，而且还在几家新兴云服务商上加倍下注。

Maintained his CoreWeave position, and he's doubled down on a bunch of Neo Clouds.

对吧？

Right?

这些新兴云服务商，正好契合他在这支新投资组合中所押注的核心逻辑。

These Neo clouds, these companies stand to benefit from the exact thesis that he's trading with this new portfolio.

假设半导体股票下跌，对吧？

So let's say that semi stocks go down, right?

这些新兴云公司的股价也会跟着跌，对吧？

Their stocks would also go down, right?

因为它们虽然拥有GPU，但CoreWeave和其他这些公司拥有的是英伟达目前没有的东西——电力资源的获取能力。

Because they own the GPUs, but CoreWeave and these other companies own something else that NVIDIA currently doesn't have, which is the power access.

别忘了，他投资这些新兴云公司，不只是因为它们能运行GPU。

Remember, he invested in a bunch of these neo clouds, not just because they can run GPUs.

任何数据中心都能做到这一点。

That's something that any data setter could do.

只要你有资金，就能办到。

You just need capital to do it.

但更重要的是，它们拥有许可证，并能接入现有的电网基础设施来为这些GPU供电。

But more importantly, they have the licenses and access to existing energy grid infrastructure that can serve these GPUs.

所以他通过一家公司就巧妙地同时押注了两个方向：既看好电力供应，又看空半导体板块。

So he's playing both sides of the trade pretty intelligently through just a single company that can express both his interest in the power trade, but also his bearishness on the semis trade as well.

这样一来他就能双赢，这很合理，因为这正是他重点加仓的那类公司。

Like he can have a win win and that makes sense because that's the companies that, or type of companies that he's doubled down on.

真正有优势的，正是那些能拿到电力资源的数据中心新贵。

It's exactly these data center neocards that have access to power.

这招确实挺聪明的。

It's a pretty smart trade.

他还进一步加仓了一个有趣的细节——真正能拿到电力和电网资源的地方，其实是比特币矿场。

And he's also doubled down on this fun little Easter egg on where you can actually get this power and get this grid capability, and those are Bitcoin mining companies.

对。

Yes.

上个季度我们简单提过，但他这次又大举押注这些公司了。

We talked about this briefly last quarter, but he's going big on them again.

规模巨大。

Huge.

今年，美国的比特币矿企将新增约30吉瓦的并网电力容量。

And this year, US Bitcoin miners are going to approximately put 30 gigawatts of interconnected power capacity online.

这个数字有多大呢？对比一下，这差不多是微软、谷歌、亚马逊和Meta四家公司宣布的总和。

That's roughly I mean, for comparison, that's roughly the total amount of Microsoft, Google, Amazon, and Meta combined in what they announced.

所以，他们正在上线大量数据中心，而这些正是所有人都需要的。

So this is a tremendous amount of data centers that they're putting online that everyone's going to need.

由于大家的关注点正从比特币转向人工智能，这些矿企已经掌握了关键的基础设施。

And because people are kind of pivoting from Bitcoin to AI, they already have a lot of the critical infrastructure.

他们有电。

They have the power.

他们有数据中心的规模。

They have the data center size.

设施也都建好了。

They have it built out.

他们只需要换上专门为AI设计的新芯片，就能立刻转型。

All they need to do is swap in new chips that are built for AI, and they're on their way.

这是一个非常独特又有趣的现象，我几乎没看到其他人讨论过，除了利奥波德。

And that is a really unique, interesting case that I don't think I've seen a lot of people explore other than Leopold.

其实就是把比特币和加密货币的转型思路拿过来用。

It's just taking the the Bitcoin pivot, the crypto pivot.

很多比特币矿工，他们本质上就是追着钱走的。

A lot of Bitcoin miners, they're there to follow the money.

当钱都流向AI领域，而他们又能一年内上线30吉瓦的电力，这规模可不小。

And when the money's in AI and they could put 30 gigawatts online in a single year, that's a huge amount.

对。

Yeah.

当我从更高的角度来看这件事的时候，

When I zoom out from this, right?

我们到目前为止聊的所有内容，其实都清楚地表明，他正在全力押注实体基础设施。

Everything we've discussed so far, there's like a clear view that he's taking here, which is he's doubling down on physical infrastructure.

他认为这部分不可能被商品化，但他提出的一个重要观点是，半导体的设计层面，也就是芯片设计这一块，已经太过拥挤了。

He doesn't believe that can get commoditized, but what he's saying, and this is a big statement, is he thinks the design layer of semiconductors, the chip side of things, is overcrowded.

不过我要提醒大家一下，英伟达其实并不自己生产芯片。

Now, may I remind everyone, NVIDIA doesn't actually make the chips.

他们是一家设计公司。

They're a design company.

他们负责设计，然后把图纸交给台湾的台积电，由台积电实际制造和生产芯片。

They create the design and they send the blueprints to this company in Taiwan called TSMC, and they actually manufacture and build the chips for them.

对吧？

Right?

博通也是这么做的。

Broadcom does the same thing.

英特尔生产CPU和GPU。

Intel creates CPUs and GPUs.

AMD也是。

AMD as well.

他最近的文件里做空了这两家公司。

These are two companies that he's short on this recent filing.

但再说一遍，他们只是负责设计这些芯片。

But again, they create the designs for these things.

并不亲自生产实物。

They don't actually build the thing.

英特尔和AMD确实打算将来自己做生产，但目前还没建成必要的工厂和基础设施。

Now Intel and AMD's intention is to eventually do this, but they haven't got the necessary factories or infrastructure to be able to do this.

他们的计划是在未来五年内实现这一点。

That's their plan in the next like five years.

所以他明确押注：芯片设计领域已经过于拥挤，真正的钱都在硬件基础设施这一层。

So he's making an explicit bet, which is like the design space for chips is overcrowded, but the hardware infrastructure layer is where all the money's going.

而他们最需要的底层支撑，莫过于电力。

And one thing that they need as a substrate more than anything is power.

因此他才下了这个赌注。

And so he's making that bet.

是的

Yeah.

好的

Okay.

这里有一部分讲到了交易可能出现问题的地方，也就是这个模式可能出故障的地方。

So we have this one section that talks about where the trade breaks, where this thing can start to break down.

我之前提到过，他做空了AMD。

Now I mentioned earlier, AMD, he was short on.

现在股价涨了75%。

It's up 75%.

这可真够刺人的。

That's got a sting.

这个说法对吗？

Is this correct?

这里还列出了几个可能出现问题的环节。

And we have a few things listed here on where it breaks.

第一个是英伟达，他在这两只持仓中最大的空头头寸高达19亿美元。

The first being around NVIDIA, the largest short position through these two holdings that he has of $1,900,000,000.

对。

Mhmm.

而这个策略可能出问题的地方在于，英伟达的护城河其实比他想象的要深得多。

And it could break in the sense that NVIDIA's moat is actually stickier than he thinks.

目前他赌的是，英伟达的芯片会逐渐商品化，像谷歌和亚马逊这样的公司，通过他们的TPU或Tranium芯片，会慢慢蚕食英伟达的垄断地位。

So currently, he's betting on the fact that NVIDIA is going to become kind of commoditized as it relates to chips, likely that other companies like Google and Amazon through their TPUs or their Tranium chips are going to slowly start to chip away at the NVIDIA monopoly.

但实际情况可能没这么简单。

And the reality is is that that may not be entirely true.

你看看那些采购订单，再看看他们GPU高达80%的利润率，大部分订单还是流向了英伟达，而这很大程度上是因为一个叫CUDA的平台锁定效应。

When you see a lot of the purchase orders coming in, when you look at the the margins that they have around 80% on these GPUs, a lot of the volume is still coming into NVIDIA, and a lot of that is due to this thing called CUDA, which is the platform lock in.

这是运行在硬件之上的软件生态，非常专有。

It's the software stack that runs on top of this hardware, and it's very custom.

能在这个平台上开发的人，其实非常有限。

It's very kind of niche in the people that can build for it.

还有一种情况是，这会成为一个非常坚固的护城河，让投资英伟达的人根本不想离开。

And there's a world in which that becomes a pretty strong molt a moat in which people who are investing in NVIDIA, they don't want to leave.

那些建设数据中心的人，因为之前已经建过了，所以继续用会更方便，而为这些新芯片打造定制化基础设施可没那么简单。

People who are building up these data centers, it's just easy because they built it before, and building custom infra for all these new chips is going to be complicated.

这到底对不对呢？

Is that true or not?

我们还不知道。

We don't know.

Anthropic 就是走的这条路，认为这并不成立。

Anthropic is kind of taking the route of it not being true.

他们和亚马逊合作，使用他们的训练芯片。

They've partnered with Amazon for for training chips.

他们也和谷歌合作，使用 TPU，同时还在用英伟达的芯片。

They've partnered with Google for TPUs, and they're using NVIDIA.

但你看像 XAI 和 Colossus 这样的公司，他们的整个数据中心全是英伟达的 GPU，全是干活的主力，而且正全力采用最新的 Blackwell 芯片，拼命扩充规模，完全依赖 CUDA 生态。

But then you see a company like XAI and Colossus, their entire data center is purely NVIDIA GPUs and just workhorses, and they're taking the new Blackwell chips, they're building them up as fast as they can, and they're very much leaning into that CUDA mode.

所以这事儿咱们还得等着看。

So it's something that we're gonna have to see.

这可能是其中一个可能的发展方向，但实施起来可能会有点难。

This is one of the thesis that that might play out, but it could be a little difficult.

而且，再说一遍，英伟达是目前全球最有价值的公司。

And, I mean, again, NVIDIA is the most valuable company in the world.

现在要让这么一家大公司开始走下坡路，可没那么容易。

This is a big company to start to fall apart now.

是的。

Yeah.

我们这儿有些英伟达显卡都用了六七年了，但合同还没到期，客户就已经提前一年预订了。

I mean, we have NVIDIA GPUs that are six to eight years old that are being rented out a year in advance of their contract expiring.

对吧？

Right?

而且还有人愿意出钱租

And people are paying

比几年前值钱多了

much value than they were years

比最初的时候还值钱。

than they originally were.

几年前的老型号H100，如今反而比两年前还值钱。

Old h one hundreds from years ago are actually worth more today than they were two years ago.

这简直难以置信。

That's unbelievable.

我的意思是，利奥波德的交易风格让我想起几个月前我们聊过的另一个交易员，迈克尔·布里，他就在英伟达股价刚起飞的时候狠狠踩了雷。

I mean, Leopold's trading style kind of reminds me of another trader that we spoke about a few months ago that got burned pretty badly, Michael Burry, who went incredibly perish on NVIDIA right at the point that the stock absolutely set.

所以我希望利奥波德别重蹈覆辙。

So I hope the same thing doesn't happen to Leopold.

不过咱们也得看看他思路里其他可能的漏洞或风险——利奥波德是做对冲基金的，对吧？

But to kind of, like, also pick apart at some of the other kind of like gaps in his potential thinking or risks here is Leopold runs a hedge fund, right?

他的‘情境感知基金’不是那种通常只做多头的风投基金。

Situational awareness fund isn't a VC fund, which is typically like long only.

很少见到对冲基金像他这样大胆地满仓做多，对吧？

It's actually kind of rare to see a hedge fund go super long as aggressively as he did, right?

所以关键在于，我们看的13F文件——他每季度必须提交的交易明细和投资组合——未必反映他当前最新的操作。

So the point being is what do you see or what we're speaking about in the 13F filings, which is something that he has to submit, his trade breakdown, his investment portfolio every three months may not be the latest and greatest trades that he's currently made.

事实上，就在我们聊这个的时候，他三月底提交报告后，可能已经把所有这些仓位都换了。

In fact, today, as we're speaking about this, after he's filed the report at the end of March, he could have changed all these trades.

他完全可能做了完全不同的安排。

He could have done completely something different.

对吧？

Right?

我还想知道，他是什么时候建的这些看跌期权仓位？

Another thing I think about is when did he take these put positions?

他具体是什么时候建的这些仓位？

When did he take these specific positions?

这些仓位说不定是年初就建好了的。

They could have been at the start of the year.

而且，这个基金本来可能亏得很惨。

And like, you know, the fund could have suffered pretty badly.

现在最明显的证据是，他的基金价值从三个月前的55亿美元涨到了140亿美元，说明他确实赚了钱。

Now the obvious evidence to prove that this isn't the case is the fact that the value of his fund went from $5,500,000,000 three months ago to $14,000,000,000 So the point is he's made money.

他已经从利润中套现了。

He's taken off money from the top.

需要指出的是，这些看跌期权和看涨期权本质上都是杠杆赌注。

And it's important to point out that these put positions, these call positions, these are kind of like levered bets.

所以我们说总共有80亿美元的看跌仓位时，他实际投入的本金可能也就十亿美元左右。

So when we talk about a $8,000,000,000 put position in total, he's typically probably only put up like a billion dollars worth of actual capital, right?

不过，他也为此支付了大量费用和期权费。

Now, he's also paying a lot of fees and premiums on that.

这本质上是一笔短期交易。

So it's like a short term trade.

再说一遍，他可能已经平掉其中一些仓位了。

Again, must say like he might've exited some of these trades already.

所以如果你正在看这段内容，或者在听，心里想着天哪，我得彻底改掉我的股票组合。

So if you're reading this, if you listen to this and you're thinking, oh my God, I need to change my entire stock portfolio.

记住，你未必会像他那样交易。

Remember, you may not necessarily be trading like him.

你不是在做短期或者高频交易。

You're not doing short term or high frequency trades.

你可能是打算长期持有，这完全是另一种思路。

You might be in it for the long term and that's a very different type of kind of like approach.

也许乔什，现在是个好时机，聊聊普通投资者能做些什么，真正的逻辑是什么，未来该把钱投到哪里。

And maybe Josh, this is a good time to talk about what the retail audience can do about this and like what the actual thesis might be and where you might wanna invest your money going forward.

是的。

Yeah.

我们其实有来自PolyMarket的一些数据可以佐证，情况可能没我们想象的那么糟，毕竟普通投资者和他做的完全不一样。

So we actually have some data to back this up through PolyMarket, which shows us that things might not be as bad as we're perceiving them because, again, retail is different than what he's doing.

我是说，利奥波德是个交易员。

I mean, Leopold is a trader.

如果你是个人投资者，情况就不太一样了。

If you're a retail investor, things are a little bit different.

如果你觉得AI泡沫即将破裂，根据PolyMarket的数据，这种想法其实是错的。

If you think that the AI bubble is going to pop, and that's what this implies, according to PolyMarket, that's sadly mistaken.

到今年12月31日之前，AI泡沫破裂的概率只有24%。

There's only a 24% chance of the AI bubble bursting by December 31.

概率非常低。

Very low probability.

我还想提一下另一个市场预测，它显示的是今年5月底之前市值最高的公司。

There's also a second market that I wanted to highlight, which shows the largest company by the end of May this month in a few more weeks.

目前，英伟达是全球市值最高的公司，紧随其后的是谷歌。

Now currently, NVIDIA is the largest company in the world, and they're pretty closely followed by Google.

但根据PolyMarket的数据，到这个月结束时，英伟达仍保持榜首的概率高达93%，这说明市场波动其实没他那些财报和13F文件暗示的那么大。

The reality is, though, according to PolyMarket, there's still a 93% chance that it stays this way, that NVIDIA is going to continue to take the crown throughout the course of this month, and I think that's a testament to a a little bit less volatility than he may be implying with these earnings and with this 13 of filings.

所以，要清楚一点，这些信息反映的都是上个季度的情况。

So, again, it's important to note that this is last quarter's news.

局势已经发生了相当大的转变。

The tides have turned pretty considerably.

我们不知道他过去几个月交易时发生了什么，但我觉得这恰恰说明，情况并没有表面上看起来那么糟。

We don't know what's happened over the last couple of months that he's been trading, but I think it's a good testament to the fact that things aren't quite as bad as it may seem on the surface.

他只是采用了新的策略，稍微改变了这个投资组合的走向。

He's just applying a new strategy that kind of alters the trajectory of this portfolio.

感谢 PolyMarket 展示这些图表。

And thank you to PolyMarket for showcasing these charts.

所以，我们还是回到刚才的问题吧：作为个人投资者，你该怎么应对这种情况？

So yeah, guess we should get back to the question of how do you as a retail investor adjust to this?

你的策略是什么？

What is your strategy?

你是怎么应对的？

How do you navigate this?

你看好后市吗？

Are you bullish?

你是看空吗？

Are you bearish?

Ejaaz，你对这个新情况有什么直觉？你个人打算怎么调整仓位？或者你建议别人怎么应对？

Ejaaz, do you have any ideas on like kind of your your gut take on how you personally plan to position yourself or how people should consider positioning themselves around this new information?

我给你两个答案。

So I'll give you two answers.

如果你刚进入这个市场，只看Leopold的13F报告就据此交易，那你就该谨慎一点。

If I was someone who is kind of like new into this market and is just reading Leopold's 13F filing and and are like basing their trading decisions off of that, you would be tentative.

这可不是冲进去all in一只股票的时候。

You wouldn't be this isn't a time to to go crazy and go all in on a single stock.

反正我从来不会建议这么做。

I would never advise that anyway.

但关键是，我觉得他这么保守是有原因的——过去两年，整体市场已经涨了百分之几百，这在正常股市里简直是天文数字。

But the point is I think he's being conservative for a reason, which is the market on average has probably run up a couple 100% over the last two years, And that in a regular stock market is absolutely huge.

如果你看看标普500的主要涨幅，其实主要来自前五到七家巨头，它们都在AI上投入了大量资金，非常激进。

If you look at like the major increases in the S and P 500, it has primarily been through five top companies in the max seven, which have all invested extremely heavily and aggressively in AI.

这笔资金会流向我们之前提到的许多公司。

And that money flows downstream into a lot of these companies that we've spoken about already.

所以他可能只是想说，这个交易已经太拥挤了，大家得小心谨慎。

So he may just be suggesting that it is an overcrowded trade, so just be cautious and careful.

不过说真的，我对乔什一直持乐观态度。

That being said, I always have a bullish cap on Josh.

我现在最关注的是电力和能源领域。

And where my mind goes to right now is in the power and energy side of things.

关于布卢姆能源和数据中心这块，我和利奥波德的看法一致。

Now I'm aligned with Leopold on the Bloom energies and the data centers side of things.

事实上，我觉得这招太聪明了——就算你投资那些顶级云服务商，他们正和Anthropic、Meta签下数十亿美元的合约，即便半导体行业下滑，你照样能受益，因为他们掌控着电力资源。

In fact, I think it's genius that even if you invest in some of these top neo cloud providers who are signing, by the way, multi billion dollar deals with Anthropic and Meta, you still get to benefit if the semiconductor space goes down because they own the power capacity.

这是我从这次分析中学到的新观点，让我特别看好。

That's something new that I've learned from this that I'm feeling extremely bullish on, right?

所以这可能是我愿意投钱的方向。

So that's something that I might pop my money in, right?

同样地，我也在关注他做空的一些公司，比如康宁，这家公司也是个关键瓶颈。

Equally so, I'm looking at some of his short positions on the likes of companies such as Corning, which is also a bottleneck.

对吧？

Right?

这跟光通信有关，英伟达刚和他们签了巨额合作协议，可他却在做空他们。

It's on the optic side of things, and NVIDIA just signed a massive multibillion dollar partnership with them, and he's short on them.

所以他是在挑着选，看准哪个瓶颈下手。

So he's kind of picking and choosing which bottleneck that he wants.

目前我更看好能源方向，但他对这些光纤网络和其他做空标的的判断，我还不敢说完全对。

I'm kind of more bullish on power at this point, but I don't know if he's completely nailed it when it comes to some of these optical fiber networks and some of the other short positions that he have.

我也说不准。

I don't know.

你呢？

What about you?

是的。

Yeah.