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。
你知道吗，如果我说对了，对冲策略是什么？
You know what the hedge is if I'm right?
就是去买比特币。
It's to go buy Bitcoin.
所以如果你相信你会被AI颠覆，那你就会开始问，如果我会被颠覆，那谁不会被颠覆呢？
So if you believe that you're gonna get disrupted by AI, then you start asking if I'm getting disrupted, who isn't gonna get disrupted?
真的很难想出哪家公司能幸免，所以我把注意力集中在代理计算这一块。
It's really hard to come up with with any company, which is why I focus my attention on the agent to compute side.
我非常有信心，在未来三年里，作为半导体、光学公司、传输公司的组合……
I have total confidence over the next three years that as a basket of semiconductors of the optical companies of the transport...
那些制造数据中心的工业公司，在未来三年里，随着人形机器人的出现和其他一切的发展……
You know, the industrial companies that make the data centers, that over the next three years with humanoids coming and everything else that...
大家好，最近怎么样？
What's going on, guys?
今天，我们和Jordi Visser进行了一场非常精彩的对话。
Today, we've got an excellent conversation with Jordi Visser.
在这次对话里，他会谈到对冲基金的消亡，以及为什么代币化可能会让这种冲击直接降临到它们头上。
In this, he's gonna talk about the death of hedge funds and why tokenization may be the thing that brings that right to their doorstep.
我们会聊到长期资本管理、比特币、代币化、CPI 的现状以及为什么我们不再关心它、AI 如何变得不可避免、金融市场正在发生什么，以及这对你的投资组合意味着什么。
We talk about long term capital management, Bitcoin, tokenization, what is going on with CPI, why we don't care about it anymore, how AI is now inevitable, what's going on in financial markets, what it means for your portfolio.
然后，当然，我们得在某些事情上意见不合。
And then, of course, we gotta disagree on something.
所以我们会展开一场非常激烈的辩论，对两位绅士来说算是很激烈的。
So we're gonna get into a very heated debate, heated for two gentlemen.
在那场对话中，我想你会学到很多关于 AI 如何运作、代币价格，以及我们如何看待世界随时间演变，无论是大公司还是小公司，以及你如何围绕这些主题进行投资。
And in that conversation, I think you're gonna learn a lot about how AI is working, token prices, and how we see the world evolving over time, both for big companies, small companies, and how you can invest along all these themes.
这些以及更多内容，尽在今天与 Jordi Visser 的对话中。
That and much more in today's conversation with Jordi Visser.
好的，Jordi。
Alright, Jordi.
这周你对长期资本管理有一些非常独特的看法，以及我们可能正在经历一个非常相似的时期。
You've got some very unique thoughts this week on long term capital management, and how we may be living through a very similar time right now.
你能给我们讲讲，你是怎么看待当前这个时局的吗？
Can you give us context on how you're thinking about this moment in time?
对于我的订阅观众来说，你们应该看到我周四发了一个简短的视频。我之所以发那个视频，是因为很多人联系我，问我现在的情况，担心这是否值得忧虑。虽然我这一周挺忙的，但大家的反馈让我觉得有必要回应一下。
And for those of you who are my subscribers, you saw me do a very brief video on Thursday, and the reason I put it out is because so many people had reached out, and I had a busy week, but because people had reached out and said, what's going on, is this something I should be worried about?
关于这次股市抛售，以及大家围绕人工智能产生的种种担忧。
The sell off and all of the fears around the AI.
半导体板块已经连续下跌七天了。
Semiconductors are down seven days in a row.
Core Weave、CDS，还有这些乱七八糟的东西。
Core Weave, CDS, all this stuff.
当然，每当市场终于开始爆发式上涨时，那些看空者的回声就会越来越大。
And of course, the echo chamber of bears gets louder and louder whenever there's a, oh, it's popping now finally.
打个简单的体育类比吧，这就像你参加一场比赛，决定押注一支被让28分的橄榄球队。
To give you like a brief sports analogy, this is like you go into a game and you've decided to bet on a football team that, you know, is is given 28 points.
也就是说，这支球队是四分达阵的弱队，而……
So there are four touchdown underdog, and the...
你得了第一个达阵，就开始捶胸顿足。
You get the first touchdown, and you start banging your chest.
在我看来，那些看空AI泡沫破裂的人，在标普指数离历史高点只差几个百分点的时候，就表现得像这样。
That is what bearish people on the AI bubble popping when the S and P is a couple percent off the all time highs feels like to me.
我从来没见过谁能预测泡沫的终结，当然我也不认为那些每天等着事情发生的人能做到。
Like, I've never met anyone who predicted the end of a bubble, and I certainly didn't don't think anyone who's waiting every day for anything to happen.
所以这周确实引发了很多恐慌。
So this week brought out a lot of fear.
我想分享一些关于LTCM如何与此相关的想法，因为我觉得让大家理解杠杆真的很重要。
And I I wanna share some thoughts on how LTCM fits into this, because I think it's really important for people to understand leverage.
我坚信这一点，这也是为什么我不在对冲基金行业。
So I'm a believer, and this is the reason why I'm not in the hedge fund industry.
我对和人工智能竞争毫无兴趣，而所有留在行业里的人都在和它竞争。
I have no interest in being in a competition against artificial intelligence, which everyone who's staying in the industry is competing.
他们是在和AI代理竞争。
They're competing against AI agents.
现在任何散户都能立刻设立一个多空对冲基金，回测一下就能上手。
The fact that any retail person can set up a long short hedge fund, instantaneously now back test things and do it.
所以我早就觉得这段时间不会轻松，但我百分之百确定，就像互联网泡沫之前一样。
So I didn't believe this was gonna be an easy period, but there is no doubt in my mind with 100% certainty just like before the dot com bubble.
互联网泡沫之前我们就知道，互联网会改变我们的生活。
We knew before the dot com bubble that the Internet would change our lives.
但我们不知道的是，这需要多长时间。
What we didn't know was how long it would take.
所以当互联网泡沫破裂时——那是在LTCM之后不久，我待会儿会讲到——人们必须明白，互联网上的每一个想法最终都被证明是对的。
So when the dot com bubble happened, which was right after LTCM, which I'll get into, people have to understand every single thought on the Internet ended up being right.
问题在于，为了建设必要的基础设施而借来的资本，花了很长时间才到位。
The problem was the capital that was borrowed to build the pipes necessary for it took a long time to get there.
现在让我们回到LTCM。
Now let's go back to LTCM.
我当时在巴西。
I was in Brazil.
我在巴西亲眼看到了LTCM交易的另一面。
I got to see LTCM on the other side of their trades.
给你讲讲他们做的交易类型吧，1997年全球交易量最大的股票之一Telebras正在进行私有化。
And to give you the kinds of trades they did, Telebras, which was one of the largest trading stocks in the world in 1997, was going through a privatization.
所以这家公司基本上要么被私有化，要么不……
So the company basically was gonna be or not...
他们会被私有化，最终的结果是普通股和优先股的价格会趋于一致。
They were gonna be privatized, what was gonna end up happening was ordinary and preferred shares would converge to the same price.
所以LTCM做的其中一笔交易就是做空那些比普通股溢价交易的优先股，随着私有化进程推进，它们会趋同。
So one of the trades that LTCM did was to basically short the preferred stocks trading at a premium to the ordinary shares, and they'd converge as the privatization got...
日期变得更加确定。
The date became more certain.
但在那个日期到来之前，价差反而扩大了。
Well, before that date happened, the spread widened out.
那为什么价差会扩大呢？
Now why did it widen out?
为什么会扩大呢？
Why would it widen out?
嗯，有很多原因会导致价差扩大，这也是现在对冲基金正在经历的事情。
Well, there's a variety of reasons that go into why a spread would widen out, and that's what hedge funds are going through right now.
我之所以说确定，是因为人工智能及其计算需求是不可避免的。
And the reason I say with certainty, AI and the compute needs for it are an inevitability.
你可以争论它是否需要二十年，我们会经历某种形式的互联网泡沫。
You can debate whether it's gonna take twenty years and we'll go through some form of a .com bubble.
你可以进行那样的争论。
You can you can have that debate.
我……
I...
大家都知道我的观点，只要需求大于供给，我认为就不可能出现类似于互联网泡沫的崩盘，因为提供资金的人和电信公司完全不同。
Everyone knows my views, which are as long as demand is greater than supply, there's no chance in my mind of having an unwind that is very similar to the .com bubble because the people that are financing it are very different than the telecom companies.
我周四在视频里开了个玩笑，把它比作房地产市场。
I made a joke on my video on Thursday when I compared it to the housing market.
我就说，好吧。
I'm like, okay.
我们来看看Core Weave。
Let's go through Core Weave.
Core Weave的资金需求可不是靠脱衣舞娘来满足的。
Core Weave doesn't have strippers funding their needs for capital.
实际上，资金来自微软、OpenAI，还有……
It's actually coming from Microsoft, from OpenAI, from...
再看看Meta，看看那些大公司名单。
Go through Meta, go through the list of really big companies.
我拿脱衣舞娘开玩笑，是因为房地产泡沫时期就是那样。
And I made the joke about the stripper funding because that's what was going on with the housing bubble.
比如，他们开始拖欠还款。
Like, they started defaulting on their payments.
微软可不会拖欠。
Microsoft's not defaulting.
Meta 不会违约。
Meta is not defaulting.
OpenAI 也不会违约。
OpenAI is not defaulting.
所以对于 CoreWeave 来说，问题在于 Gartner 集团表示，他们认为到 2030 年，当公司公布财报时，他们的订单已经排到 2030 年，而且看得清清楚楚。
So for CoreWeave to be an issue, the Gartner group said that they think by 2030, when companies report earnings, they have they have visibility with orders sold out to 2030.
这种情况已经反复出现。
This has been coming on repeatedly.
所以，在市场的历史上，当然也是我人生中第一次，大家都在担心五年后我们能不能达到收入目标。
So for the first time in the history of markets, and certainly the first time in the history of my life, everyone is worried about whether we're gonna hit revenues in five years.
这并不会阻碍 AI 的必然发展，而且我觉得我们几乎不可能达不到，因为我听 Jensen Huang 这样的人说过。
That is not gonna drive the inevitability of AI, and I find it almost impossible that we won't hit there, because I listen to people like Jensen Huang.
我听那些在科技领域待了很久的人说，他们意识到每天都有订单源源不断地涌来，都是关于算力、算力、算力的。
I listen to people that have been in the technology field for a long time that realize that every day the orders keep coming in for compute, for compute, for compute.
所以每个人都可以理解其中的细微差别。
So everyone can take the nuance.
我看空Anthropic。
I'm bearish Anthropic.
我看空OpenAI。
I'm bearish OpenAI.
我看空这个。
I'm bearish this.
大家要明白的问题是，我看到系统中唯一被击垮的杠杆就是对冲基金。
The problem people have to understand is the only leverage that I see in the system that is being knocked out is hedge funds.
我们上周五录制的时候，本世纪最优秀的对冲基金之一——不是那种靠运气或者刚入行的新手，而是Lone Pine Capital——上半年涨了40%以上。
We recorded last week at the end of that day, on Friday of last week, one of the best hedge funds of this century, this century, not situational awareness, not some guy who just started managing money, Lone Pine Capital was up 40 plus percent in the first half of the year.
但到了九月底，我想说，在一个标普只涨了2%的季度里，VIX指数从未出现过任何有意义的飙升。
And as of the end of September, and I wanna say, in a quarter that the S and P was up 2%, never did the VIX go higher in any meaningful way.
整个期间，月度级别的大波动一次都没有，但他们却把全部收益都吐了回去。
Never in an entire time was there a big move on a monthly basis, and yet somehow they gave up the entire performance.
所以大家应该明白这意味着什么。
And so people understand how that means.
如果你涨了40%以上，我不知道他们的费率是多少，但假设你收20%的费用，管理费先不算，那意味着你实际上涨了超过50%，因为你把收益全吐回去了，现在净回报是零。
If you're up 40 plus percent, and I don't know what their fees are, but let's assume you're charging 20%, and let's leave the management fee out, that means by definition you were up more than 50% because you gave it all back, and now your net return is is zero.
这只能说明有大事正在发生。
That just says that something big is happening.
这不仅仅是市场感知的问题。
It wasn't just situational awareness.
这是有史以来最优秀的一些对冲基金，在我看来，现在正在发生并且会持续到不可避免的结局的整个过程，就像我当年看到的LTCM一样。
It was some of the best hedge funds that have ever existed, and in my opinion, what is going on and will persist for the entire five years to the inevitable is what I watched with LTCM.
你是在赌你的投资者能陪你多久。
You are making a bet as to how long your investors will stay with you.
你是在赌主经纪商能给你多长时间的杠杆，无论你用了多少杠杆。
You are making a bet on how long the prime brokers will give you leverage for whatever leverage you're using.
我们看到杠杆正在从系统中撤出，对冲基金行业，在我看来，正在经历一个过程，这个过程会淘汰系统中大量的杠杆，而且这个过程会持续下去，因为AI赢家的集中度太小，而输家的规模太大，这个过程会让人们不得不投资于确定性。
We are watching leverage come out of the system, and the hedge fund industry, in my opinion, is going through a process which is gonna weed out a lot of leverage in the system, and this process is gonna continue and continue because the concentration of the winners in AI is so small, and the losers are so big that this process will leave people needing to invest in certainty.
而有了确定性，你就是在赌时间和你在那条路上导航的能力。
And with certainty, you're making a bet on time and your ability to navigate that path.
所以我觉得这周是个大事件，而且当你听到那些看空的人说泡沫要破裂的时候，讽刺的是，泡沫其实在那些长期用杠杆赚钱的聪明人身上。
So I think this was a big week, and when you listen to the bears going to bubbles unwinding, the irony is the bubbles are in really smart people that have used leverage to make money for a long time.
你看投资策略，有巴菲特时代，买好东西，永远持有。
When you look at investment strategies, there is the Buffett, you know, era, buy great things, hold them forever.
还有黄金爱好者、比特币玩家，买好资产，永远持有。
There are gold bugs, Bitcoiners, buy great asset, hold it forever.
对冲基金，他们其实……
Hedge funds, they really...
除了那些收租的，还有像城堡和做市商那样的，大多数对冲基金都是靠结构性定价错误或者结构性套利。
Outside of the rent tax and maybe the citadels and market makers, most of the hedge funds have found, you know, kind of structural mispricing or structural arbitrages.
他们用了很多杠杆，然后赚了不少钱。
They've used a lot of leverage, and they've been able to generate significant returns.
我听到你的意思是，这部分投资现在每天都在快速变化，一周一周，一月一月地变。
What I think I hear you saying is that section of investing is now rapidly evolving on a day to day, week to week, month to month basis.
所以你不能坐在这儿说，嘿。
And so you can't sit here and say, hey.
你知道吗？
You know what?
我打算每天收盘时买入，开盘时卖出，然后坚持几个月甚至几年，而且我会用大量杠杆来做。
I'm going to buy the close and sell the open every day, and I'm gonna do that for months or years, and I'm gonna do it with lots of leverage.
这样我每天就能赚个100个基点，最后累积成很可观的回报。
And so I'm gonna clip my 100 basis points, you know, each day, and I'm gonna end up with this great return.
而且因为市场变化太快，这些人现在真的必须跟着市场不断调整，这样一来，他们犯错的可能性是不是就越来越大了？
And because everything is evolving so much that that's really where now these people have to evolve with the market, and you're just increasing the odds that they make mistakes?
是不是应该这么理解？
Is is that the way to think about it?
我想再往前追溯一点，这么说吧。
So I'm gonna I'm gonna go a little bit further backwards and say this.
如果你问我什么伤害了私人信贷，仔细想想，其实是人工智能。
If I asked you what has hurt private credit, if you really think about it, it was AI.
人工智能？解释一下。
AI Explain.
所以，就像我们了解到的，大多数私募信贷投资都是21、22年的，而且都是SaaS公司。
So the majority, as we learned, of the private credit investments, they're from 2122, and they're SaaS companies.
AI摧毁的正是SaaS公司。
What did AI destroy was SaaS companies.
它并没有让它们全部归零，但如果你投资了这些公司，而它们又在和AI竞争，那就会不断侵蚀你。
And it didn't move them all to zero, but what it did was if you make investments in these companies and they're in competition with AI, it just eats at you.
我的观点一直是，人类不再是智能食物链的顶端了。
My thesis has always been that human beings are no longer the top of the intelligence food chain.
如果说以前所有的事情都像长期资本管理公司那样，天才也会失败，那么现在我们到了这样一个阶段：AI代理能做的，对冲基金都能做。
If everything that was made was about LTCM when genius failed, we are now at the point where an AI agent can do everything that a hedge fund can do.
所有事情。
Everything.
没有一件事例外。
There is not a single thing.
而且，你更愿意把钱借给谁？是那种能即时看到所有事情、能读懂公司所有信息、能通过13F文件判断自己是否处于拥挤交易位置的风险管理，还是别的什么？
And who would you rather lend to risk management that is instantaneous in seeing everything happening that can read every single thing from a company can understand, am I in a crowded position or not by going through the 13 f filings.
嘿。
Hey.
我听说……
I heard...
有一家新的对冲基金，叫Vara……
I mean, there's a new hedge fund, Vara...
也不算新，就是Vara AI，据说有26……
Not new, but Vara AI, reportedly with 26...
他们七月份跌了百分之四十多，所以这是今天的新消息。
They were down 40 something percent in July, so this is a new story today.
我对他们一无所知。
I don't know anything about them.
我不知道他们现在的实际情况如何，但他们正在经历类似的情况。
I don't know what the reality is in terms of where they are, but they're going through a similar situation.
而且我敢肯定，就像LTCM一样，如果你回头看，LTCM平掉的所有头寸都让接手的人赚了钱。
And I'm sure just like with LTCM, if you go back, all of the positions that were unwound from LTCM made the people that took them on money.
所以肯·格里芬得到了整个LEO的认可。
So Ken Griffin got the whole LEO, like, acknowledgment.
别以为人们知道LTCM倒闭时，美联储并没有救他们。
Don't think people realized when LTCM went out, the Fed didn't bail them out.
他们召集了14家银行、对冲基金之类的机构，共同出资……
They gathered 14 banks, hedge funds, whatever, that contributed...
我记得大约是不到40亿美元，用来接管他们的资产。
I think it was slightly less than $4,000,000,000 to take their stuff out.
当时我从巴西回到纽约后，参与了LTCM的清算工作。
Now I was part of the LTCM closed down when I went back to New York after Brazil.
所以我站在LTCM事件的一方。
So I was on one side of LTCM.
我也看到了另一面。
I saw the other.
我不想针对对冲基金行业说三道四。
I I don't wanna pick on the hedge fund industry.
我想说的是，流动性差的资产受到了影响。
What I wanna pick is that illiquid assets have been impacted.
风险投资受到了影响。
VC's been impacted.
还有……
Peace...
私人信贷受到了影响。
Private credit's been impacted.
私募股权受到了影响。
Private equity's been impacted.
债券受到了影响。
Bonds have been impacted.
房地产也受到了影响，比如远程办公。
Real estate's been impacted, meaning remote work.
人工智能会带来数字代理。
AI is gonna bring digital agents.
这些座位我们该怎么处理呢？
What are we gonna do with all these seats?
我只是觉得，当你看到相对于利率水平所获得的回报时，你……
I just think that when you look at the returns you're getting relative to where interest rates are, you...
你的目标收益率越来越高，而AI同时也在推高利率。
Your bogey is getting higher and higher, and AI is pushing rates higher at the same time.
所以AI正在淘汰所有低效的企业，但它也会淘汰那些依赖市场波动相关性以及头寸拥挤程度的杠杆对冲基金。
So AI is crowding out all inefficient businesses, but it's also gonna crowd out hedge funds who use leverage that depends on the volatility correlation of the market and the crowded nature of the positions.
我只是觉得，这个行业在杠杆基础上相对于现有的流动性来说太大了。
And I just think the industry is too big on a levered basis relative to the liquidity that exists.
这将会与代币化融合，因为最后一块拼图——我这周去了摩根士丹利，最后我们谈到了代币化。
This will merge with tokenization because the final piece of this, and I was at Morgan Stanley this week, and and at the very end, we talked about tokenization.
如果你是对冲基金，正在听这个节目，而你还没有想过代币化将如何改变你的生活，先别管你的业务，想想你的生活。
If you're a hedge fund listening to this, and you haven't thought about the way tokenization will change your life, forget your business, your life.
全天候、全年无休地进行风险管理，对我来说这简直难以想象，但这就是我们前进的方向。
Risk managing twenty four seven, three sixty five, it's it's an unthinkable process to me, but that's where we're headed to.
资本会立刻发生变化，这对市场波动性影响巨大，也对你如何对冲你的投资组合以及主要经纪商提供杠杆的能力产生重大影响。
The capital that will be immediately changing, it has huge implications for the volatility in the market, has huge implications for the way you hedge your book and the and the the the ability for prime brokers to provide leverage.
这可能不是今年的业务，但就像人工智能一样，这些变化发展得太快了，随着时间的推移，你的目标就变成了，嘿。
Now that may not be this year's business, but just like with AI, these things are moving so fast that over time, your bogey becomes, hey.
资本成本在哪里？
Where's the cost of capital?
这就是为什么当你和我谈论我个人时，我的资本成本与比特币相关。
That's why when you and I talk about for me personally, my cost of capital is related to Bitcoin.
这与这无关，因为这就是我看到的结局。
It's not related to this because that's the way I see this ending.
所以对于人工智能，如果你真的回顾对冲基金行业进入这个低谷的时候，那是从六月开始的。
And so for AI, if you really go through when the hedge fund industry entered this abyss, it started in June.
从六月到现在，就说整整四个月前吧，看看比特币的表现。
And since June, let's say exactly four months ago, Go look at what Bitcoin has done.
它上涨了将近30%。
It's up close to 30%.
看看以太坊的表现。
Go look what Ethereum has done.
涨幅超过了30%。
It's up more than 30%.
标普指数，持平。
The S and P, flat.
纳斯达克，持平。
Nasdaq, flat.
我刚跟你说过，很多对冲基金已经跌了超过40%，但市场却没什么动静。
Hedge funds, I mean, I just told you a lot of these funds are down 40 plus percent, and nothing has happened.
这应该让人感到害怕，因为唯一的结局就是这种情况不会持续。
This should be scaring people because the only way this ends is if this isn't the case.
所以每次再出现像这周这样的小插曲，Core Weave那件事简直离谱。
So every time we get another one of these little episodes like this week, the Core Weave thing is ridiculous.
他们有来自大公司的现金流。
They have cash flow coming from big companies.
这和电信行业的情况不一样。
This is not the same situation as telecom.
它们是好的股权投资吗？
Are they good equity investments?
我不知道。
I don't know.
它们是好的债券投资吗？
Are they good bond investments?
我不知道。
I don't know.
但我知道一点，根据高德纳的说法，要到2028年、2029年、2030年我们才能确定它们是否会陷入困境。
But I know this, we won't know for sure whether they're gonna be in trouble according to Gartner until 2028, twenty twenty nine, twenty thirty.
那不会从那时起推动市场。
That's not gonna drive the market from there.
代币化，你认为它带来了流动性和透明度，而且你是在说代币化可能会终结对冲基金吗？
Tokenization, you think brings both liquidity and transparency, and you're arguing that tokenization could be the death of the hedge funds?
我觉得随着时间推移，这会给对冲基金带来很大的压力。
I think as time goes on, it will put so much pressure on the way hedge funds...
再说，那些大型对冲基金能撑得最久，没有别的原因，就是它们已经……
And again, there are the mega hedge funds, which will sustain the longest over time, and for no other reason, they've just met...
大多数都已经改成五年锁定期了。
Most of them have gone to like five year lockups.
所以它们把资金锁定了。
So they've locked capital in.
我觉得这种代币化和把锁定资金释放出来的概念……
I think this tokenization and this concept of of bringing in locked up capital...
那我们回到私人信贷。
So let's go back to private credit.
如果你进入私人信贷，看看BDC的交易情况，也就是实时的估值，假设现在BDC的交易价格比那些进入私人信贷的散户投资者……低了20%。
If you go into private credit and you look where the BDCs are trading, so real time marks on this, and let's assume right now the BDCs are trading at 20% below where retail investors who got into private credit...
所以基本上，在接下来的四年里，假设就这样，他们最终会在投资上亏损20%。
So they're basically over the course of, let's say, the next four years, let's just use all this, they end up losing 20% on their investments.
对吧？
Okay?
这不会要了他们的命。
It's not gonna kill them.
但问题是，如果他们当时有那笔资金，能在低于现在20%的位置卖掉，然后把钱重新投回市场，那影响就翻倍了。
It hurts in the fact that if they would have had that capital and been able to sell out at 20% below here, and then put the money back in the market, well, that's double the impact.
问题在于资金被困住了。
It's the trapped nature.
当代币化来临时，人们会想，好吧，我该怎么退出？
When tokenization comes and people are like, okay, where can I get out of this?
嗯，总有人愿意以某个价格买入私人信贷，所以你有被困住的资本，而它已经不想待在那里了。
Well, someone is willing to buy into private credit at some level, So you have trapped capital that doesn't wanna be there anymore.
捐赠基金也一样，很多其他机构也面临同样的情况。
This is the same thing for endowments, same thing for a lot of...
随着代币化的到来，它将释放出为资产定价的能力。
As tokenization comes, it will free up the ability for there to be a price on things.
一旦有了清算价格，资本就会流向它该去的地方。
And once there's a clearing price on things, then capital will go to where it should be.
它还会带来的是……
What it will also do is...
我更愿意拥有什么？
What would I rather own?
是花75美分买一美元的Core Weave债务，还是我手里那些21、22年份的私人信贷垃圾？
Core Weave debt at 75¢ on the dollar, or the private credit garbage that I own from the '21, '22 vintages that I'm sitting in?
接下来几年，我更愿意持有Core Weave，而不是这些。
I'd rather own Core Weave right now for the next whatever years than this.
所以资本会流入，Core Weave的价格也会回升。
So that capital would come in, and Core Weave would go back up.
还记得迈克尔·塞勒的产品Stretch跌到75的时候吗？
Remember when Stretch, Michael Saylor's product, traded down to 75?
我听到过同样的……
I heard the same...
就是那些在Core Weave上发XPost的同一批人。
The exact same people sending out XPost on Core Weave.
那些Stretch的东西最后又涨回到面值了。
The stretch things went all the way back up to par.
所以那些只看短期、说这就是末日的人，这些说法……
So these arguments of people looking at short term things and saying this is the end of it, this is the the...
这些人主要是对冲基金的人，或者那些靠杠杆赚了大钱的人。
These are mainly hedge fund people or people that made most of their money on leverage.
对冲基金赚钱大概有两种常见的方式。
There are two ways, like, general ways that hedge funds end up making money.
德鲁肯米勒从来不谈杠杆，我也不确定他到底用过多少杠杆。
So Druckenmiller never talks about, like, leverage, and I'm not sure what he's ever used in leverage.
他谈的是集中持仓，他承担的风险就是集中度风险。
He talks about putting a position on being concentrated, and that's the risk he's taking, his concentration risk.
而像LTCM那样用杠杆的多元化对冲基金，杠杆率高达30比1，他们赌的是相关性和波动性会保持某种状态，这样就不会因为杠杆太高而收到追加保证金的通知。
On a diversified hedge fund that's using leverage like an LTCM that's levered 30 to one, they're betting on correlations and volatility staying in some kind of way that they don't have margin calls come in because of the amount of leverage.
嗯，大多数多策略基金都是这么设置的。
Well, that's the way most of the multistrats are set up.
所以如果我们遇到这种情况，生物科技在九月份就崩了。
And so if we end up in a scenario, biotech blew up in September.
所以每个在生物科技领域做多空投资组合的人，我说每个人，我不喜欢绝对的说法，但据我所知，很多基金，尤其是那些独立交易团队，都遇到了麻烦。
So everyone who runs a long short portfolio in biotech, and I say everyone, I don't like absolute statements, a lot of funds from what I've heard, pods in particular had trouble.
可转债。
Converts.
可转债是Core Weave的一部分。
Converts are part of Core Weave.
它们也是MicroStrategy的一部分。
They're also part of micro strategy.
当你遇到这些小情况，人们用某种杠杆，比如做多这个可转债，做空那个可转债，他们依赖这些不会像LTCM那样扩大价差。
When you get these little things happening where people are running some kind of leverage where it's like I'm long this convert, I'm short this convert, they're depending on these things not to widen out like LTCM.
嗯，这又是一次价差扩大。
Well, this has been another widening out.
所以我觉得对于那些正在交易的人来说，我想给我的订阅者一些建议，比如如果你需要在未来一个月内靠某个仓位赚钱，那你就麻烦了。
So I think for people that are trading and like, I'm trying to give my subscribers things like, if you need to make money on a position over the next month, you're in you're in trouble.
要好好利用你现有的时间。
Like, take advantage of the time that you have.
所以对我来说，最令人惊讶的积极方面是，尽管股市疲软，新云和AI交易也都很弱，但比特币却一直保持得非常强劲。
So maybe the most shocking thing to me in a positive way, all of the weakness in the stock market, all the weakness in the neo clouds and, you know, the AI trade, Bitcoin has held in really strong.
它一直在8.2万到8.5万美元之间交易，感觉它几乎是在逆势而行，其他所有东西都在下跌，而比特币却稳住了。
It's been trading pretty much between 82 and $85,000, and it almost feels like it's defying the market momentum where everything else was trading down and Bitcoin holds in.
你怎么看这个现象？
What do you read into that?
我的看法是，随着持有其他资产变得越来越难，不管是电脑决定的还是什么，人类和电脑都在寻找好的夏普比率。
So what I read in is as it gets harder and harder to own other things, whether it's the computer decided, whatever it is, human beings and computers look for good Sharpe ratios.
比特币在过去四个月的夏普比率非常突出，在全世界都处于混乱状态的时候，它显得格外亮眼。
Bitcoin Sharpe ratio right now over the last four months stands out, and it stands out as the rest of the world is kinda going through this chaos.
而且我想说，这不仅仅是AI交易的问题。
And and I do wanna say, this is not just the AI trade.
就像我说的，纳斯达克四个月来没什么变化。
Like I said, the Nasdaq's unchanged over four months.
标普四个月来也没什么变化。
The S and P is unchanged over four months.
债券市场在下跌。
The bond market is down.
现在想找到过去一段时间表现好的投资，真的越来越难了。
It's getting very hard to find investments just using a trailing period that do well.
我觉得这是因为这个领域的规模太小了。
I believe it's because the size of the space is too small.
代币化已经是现实了。
Tokenization's a a reality.
就像我说的，我创建了自己的加密生态系统。
Like I said, I created my crypto ecosystem.
我们聊聊比特币吧。
You know, we talk about Bitcoin.
我也可以说，我每天都关注NEAR，把它当作一种AI代理型的项目，我觉得它应该会表现不错。
I also can say that I watch NEAR every single day as kind of one of these AI agentic where I think it should do well.
以太坊也是一样。
Ethereum, the same thing.
不只是比特币一直坚挺。
It's not just Bitcoin that's that's hung in there.
这是加密货币中最优质的部分，我仍然认为这会是明年的一个现实趋势。
It is the most quality parts of crypto, and I still believe that this is a reality of something for next year.
如果你问我AI交易什么时候……
If you ask me when the AI trade...
因为如果AI交易是不可避免的，那么它上涨的时机就是当人们刚刚撤出风险的时候。
Because if the AI trade is an inevitability, the time for it to rally is when people have just taken off the risk.
我们第一次学到这一点是什么时候？
And when did we learn this first?
通过情境意识。
With situational awareness.
刚才有人给我发了个消息，说摩根士丹利主经纪业务名单上的AI多空交易现在处于过去一年的第七百分位。
Right now, someone sent me something saying the AI long short trade on the Morgan Stanley prime brokerage list is now in the seventh percentile of the last year.
第七百分位。
Seven.
所以风险已经完全释放了。
So the risk has been completely taken off.
现在它处于第七百分位。
Now it's in the seventh percentile.
它可能仍然很高，但那是相对于过去一年来说的。
It could still be high, but that is relative to the last year.
对我来说，这意味着人们已经撤出了。
For me, that means that people have bailed out.
有两个……
There are two...
我在关注三个触发点，来把握下一波行情的时机。
Three trigger points that I'm looking in terms of timing for the next wave.
而且今年，如果你看标普500，年初的头两个半月非常令人沮丧，然后突然大跌了9%。
And this year, if you look at the S and P 500, it was very frustrating start to the year, first two and a half months, and then all of a sudden we had a big fall, 9%.
然后出乎意料的是，在油价要涨到200的预期下，标普开始反弹，纯粹是因为空头回补之类的。
And then out of nowhere with the expectations that oil was going to 200, the S and P started to rally purely from short covering and things like that.
所以今年标普真正上涨的窗口其实只有大约三个月，而且其中一半时间还笼罩在油价恐慌中。
So there was really only a window of about three months this year where the S and P rallied, and during half of that time was the fear over oil.
那是什么推动标普上涨的呢？
Now what drove the S and P higher?
全都是AI交易。
It was all the AI trade.
所以只有三个月的窗口期，也就是从三月到六月。
So it was only a three month window, and aside from that, that takes you from March to June.
今年整年标普都没有再涨过。
There's been no rally in the S and P the entire year.
所以你能创造超额收益的窗口越来越小，而这就是英伟达过去三年的走势。
So your windows of making alpha have gotten smaller and smaller, and this is the way NVIDIA's traded for the last three years.
就像……
There's like...
他们是在超额收益的基础上赚钱的。
They they make money on an alpha basis.
大概只有20%的时间是这样。
It's like 20% of the time.
我觉得整个市场都会变成这样。
I think the entire market's gonna be like that.
所以对于关注比特币的人来说，从相对角度来看，它开始按照我预期的方向走了。
So for people that are watching Bitcoin, on a relative basis, it is starting to do what I thought it should do.
到了年底，我仍然会说，对我来说最大的积极因素是，每天都有很多非常聪明的人开始打电话来问，你能跟我聊聊代币化吗？
And when we get into the end of the year, I still will say the biggest positive for me is that the amount of really smart people are starting to call up every day going, will you talk to me about tokenization?
你会……
Will you...
就像，有……
Like, there...
如果我写足够多有深度的文章，讲它怎么影响M2、怎么影响对冲基金，就会有更多人想听，因为他们好奇万一他说对了呢？
If I write enough thoughtful pieces on how it affects m two and how it affects hedge funds, more people wanna hear because they're interested in what if he's right?
你知道如果我说对了，对冲的办法是什么吗？
You know what the hedge is if I'm right?
就是去买比特币。
It's to go buy Bitcoin.
所以如果你觉得自己会被AI颠覆，那你就会开始想，如果我会被颠覆，那还有谁不会被颠覆？
So if you believe that you're gonna get disrupted by AI, then you start asking if I'm getting disrupted, who isn't gonna disrupted?
很难想出哪家公司能幸免，所以我才把注意力放在智能体计算这块。
It's really hard to come up with any company, which is why I focus my attention on the agentic compute side.
我非常有信心，未来三年，半导体、光通信、传输这些公司作为一个整体……
I have total confidence over the next three years that as a basket of semiconductors, of the optical companies, of the transport...
还有那些造数据中心的企业，未来三年，随着人形机器人和其他东西的出现，这些公司的盈利一定会增长。
You know, the industrial companies that make the data centers, that over the next three years with humanoids coming and everything else, that earnings will come to those companies.
而且我相信，相比房地产开发商和汽车公司，它们会表现更好，也就是说这些公司的股价会涨。
And I have faith that relative to homebuilders and autos, they're gonna outperform, which means those companies will go up.
我只是觉得，如果你指望它每个月都会发生，那你会很难受，因为我觉得对冲基金也在玩同样的游戏，而且他们都挤在完全相同的交易里。
I just think if you're betting for it to happen at any month, you're gonna have a really hard time because hedge funds, I think, are playing the same game, and they're crowded into the exact same trades.
今天的节目由Figure赞助播出。
Today's episode is brought to you by Figure.
如果你喜欢比特币，你已经知道卖出是你最不想做的事情。
If you're into Bitcoin, you already know selling is the last thing you want to do.
但与此同时，生活总会发生点事，兄弟。
But at the same time, life happens, man.
机会来了，账单来了，市场也在动。
Opportunities arise, bills come up, and markets move.
这时候Figure就派上用场了。
That's where Figure comes in.
不用卖掉你的比特币，Figure让你用它来借款，这样你就能拿到现金，同时不失去你的仓位，还可能避免触发税务事件。
Instead of selling your Bitcoin, Figure lets you borrow against it so you can get cash without losing your position and may also avoid triggering a tax event.
现在，他们的贷款利率大约是8.91%。
Right now, their loans are around 8.91% interest.
那是9.9%的年利率，贷款价值比是50%。
That's 9.9% APR at a 50% loan to value.
这样你既能保留上涨的潜力，又能获得流动性。
So you keep your upside but still get liquidity.
我知道大家第一个问题就是，对，但这安全吗？
And I know the first question people ask is, yeah, but is it safe?
问得好。
It's a good question.
Figure用的是MPC托管，基本上就是说你的比特币会放在自己独立的钱包里。
Figure uses MPC custody, which basically means your Bitcoin stays in its own segregated wallet.
它不会被借出去、被集中管理，或者放在某个交易所里。
It's not being lent out, pooled, or sitting on some exchange.
他们还有可选的清算保护，在市场波动大的时候能帮上忙。
They also have optional liquidation protection, which helps if the market gets volatile.
如果你反过来，手里有现金想赚点收益，他们有个叫Democratized Prime的产品。
And if you're on the other side, like you've got cash and you want a yield, they've got something called Democratized Prime.
它的年化收益率最高可达9%，按小时支付，而且有真实资产支撑，不只是代币通胀。
It's earning up to 9% APY, paid hourly, and is backed by real world assets, not just token inflation.
所以无论你是想长期持有比特币，还是真的想用资金去赚钱，这都是一个相当有意思的方案。
So whether you try to hold your Bitcoin long term or actually put capital to work, it's a pretty interesting setup.
如果你想了解的话，我在描述里放了一个链接，你可以点进去，或者直接访问figuremarkets.co/pump。
If you wanna check it out, I've got a link in the description that you can go to, or you can go to figuremarkets.co/pump.
当你运动的时候，或者玩棋盘游戏的时候，总是会有……
When you play sports, when you play a board game, there's always...
在某些时候，你想进攻，其他时候你想防守。
At certain times, you wanna play offense, other times you wanna play defense.
而且真正理解什么时候在这两者之间切换，是非常重要的。
And actually understanding when to flip back and forth between those is pretty important.
我每周和你聊天最喜欢的一点是，你来自一个和我完全不同的世界，但我们最终都达到了同样的认识。
What I love about having conversation with you every week is you come from a very different world than I come from, but we both have arrived at the same thing.
多年来，我一直觉得这些复杂的金融趋势或发展很难向人解释清楚。
And for years, I've always thought about these complex financial trends or developments are very hard to communicate.
我一直在说的那句话就是，AI是进攻，比特币是防守。
The phrase that I keep using is AI is offense, Bitcoin is defense.
而且感觉如果你回到那个棋盘游戏的比喻，你想用AI去颠覆市场、颠覆竞争对手、颠覆某个东西。
And it feels like if you go back to that board game analogy, you want to use AI to go disrupt the market, disrupt a competitor, disrupt something.
所以用这项技术去进攻，机会非常多。
And so that offensive move of using this technology, immense amount of opportunity to do that.
但同时，如果你百分之百进攻，你就会暴露自己。
At the same time, if you play 100% offense, you expose yourself.
就像拳击一样。
Almost like boxing.
如果你不护住下巴，就可能被击倒。
If you don't cover your chin, you can get knocked out.
所以比特币就成了防守。
And so Bitcoin then becomes defense.
而且感觉如果你能用AI进攻，去获取收入、获取利润，然后把这些带回来。
And it feels like if you can use AI offensively to go and get revenue, get profit, you bring that back.
现在你得想办法保护它不被破坏，这就是你把它存在比特币里的原因。
Now you have to somehow protect it from being disrupted, and that's where you stored in Bitcoin.
这是不是个挺好的总结……
Is that kind of a a good summer...
关于你的世界观，它是怎么整合到一起的？
Summary of your worldview as to how this comes together?
那我用一个真实的例子来说吧，而且我确实想提一下
So let me use a real life example, and I do wanna give
silvia.com。
silvia.com.
这里插个广告。
Plug here.
在播客刚开始的时候。
Early early in the vodcast.
彼得，彼得去哪儿了……
Peter, where did Peter...
彼得·蒂尔刚搬到哪里去了？
Where did Peter Thiel just move to?
嗯，我知道真相，但你先说。
Well, I know the truth, but go ahead.
可以按公开的说法讲。
Can go with the public
他告诉
What did he tell
大家什么？
people?
他……
He was...
他告诉大家什么？
What did he tell people?
新闻是怎么报道的？
What's the news reporting?
说他搬到阿根廷了？
That he moved to Argentina?
他在阿根廷。
He's in Argentina.
好的。
Okay.
你们去谷歌搜一下他的地址。
You guys Google his address.
好的。
Okay.
我们就按新闻报道说的来，还有背后的原因。
Let's just use the reported news story and the reasons why.
好的。
Okay.
人工智能要毁掉世界了。
AI is gonna take down the world.
展开剩余字幕（还有 480 条）
人类会被毁灭的。
Humanity is gonna be destroyed.
好的。
Okay.
他搬到那里是为了经济机会吗？
Did he move there for economic opportunities?
假设你搬到了那里。
Assuming you moved there.
你会为了经济机会搬到阿根廷吗？
Do you move to Argentina for economic opportunities?
马来做得很好，但我觉得不是这样。
Malay is doing a great job, but I don't think so.
你搬到那里是为了安全吗？
Do you move there for safety?
你搬到那里是为了……
Do you move there for...
我在巴西住过。
I lived in Brazil.
二战结束后不久，人们搬到巴西的原因之一，就是害怕核战争和各种事情发生，想找个最安全的地方，躲起来，或者找个地方熬过去。
One of the reasons that people moved to Brazil shortly after World War two is because of the fear of nuclear war and everything happening, and where they would be the safest, and where to hide, and where to go through it.
所以我一直相信，也一直这么说，这就是为什么贬值派那边的人对我有意见，而可编程派那边的人也不满意。
So I do believe, and I've always believed this, this is where the debasement crowd gets mad at me on one side, the programmable side gets mad at me on the other.
我相信现在世界正在加速走向通缩。
I believe in a deflationary world accelerating right now.
所以与其说是颠覆，不如说我认为通缩会撕裂一切，真的是所有东西。
So instead of disruption, I believe that deflation is just going to tear apart everything, literally everything.
杠杆，整个体系。
Leverage, the whole thing.
你不需要杠杆。
You don't need the leverage.
你不需要那些能……的回报。
You don't need the returns that are able...
在一个AI即将带来通缩的世界里，你从这些东西上能获得的回报。
That you're able to get from these things in a world where AI is coming to just be deflationary.
所以对我来说，比特币更多是关于智能体做决策时，嘿，
So Bitcoin to me is more about where agents make decisions going, hey.
我今天不想冒险。
I don't wanna take risks today.
我把钱转到哪里才能保值，同时还能有不错的回报，而且安全，还有……
Where do I move my money for a store of value that is creating good returns and it's safe and...
或者说比其他东西更安全。
Or safer than everything else type thing.
所以我确实相信这种情况正在发生。
So I do believe that's happening.
我想确保回答你之前说的一个事情，这正好契合这个话题。
I wanna make sure I answer one thing that you said before that fits into this.
那么，比特币的下一次大涨会是什么时候？
So when will Bitcoin have its next rally?
我觉得这取决于AI交易什么时候开始下一轮反弹。
Well, I do believe that comes to when the AI trade starts to have, let's say, its next bounce.
要么是在超大规模企业发布财报的时候，要么是在那之后，因为这会和两件事同时发生。
Either right during or after the hyperscalers report, because it'll coincide with two things.
那是在月底。
It's happening at the end of the month.
所以人们开始意识到，情境感知是什么时候发生的？
So people start realizing when did situational awareness happen?
AI上一次触底是什么时候？
When did the last bottom happen in AI?
是在月底。
It was at the end of the month.
当对冲基金陷入困境时，月底非常重要，因为它们在月初之前都是短筹码操作。
The end of months are really important when hedge funds are in distress because they play short stacked until there's a new month.
如果人们没意识到这一点，作为一个在对冲基金行业待了很久的人，12月和1月对人们的压力是完全不同的游戏。
And if people don't realize this, as someone who's in the hedge fund industry for a long time, the stress on people in December versus January Different ball game.
完全不同的局面……
Different ball...
就是……
It's...
哦，天哪。
Oh, oh my gosh.
当你在十二月，手里有一大堆筹码，你会想，这个月最后一个月我不想冒任何风险。
When you're in December and you've got a big stack of chips, and you're like, I don't wanna take any risks this last month.
所以这个月底很重要。
So the end of this month is a big one.
还有中期选举，时间上也差不多重合。
You have the midterms, so right around there also kinda coincides.
所以如果让我猜的话，我会选两个时间点，下一次AI……
So if I had to guess, I'm gonna take two times where the the next AI
十一月的第二周？
Second week November?
不是。
No.
不是。
No.
不是。
No.
我会说就在……
I would say right...
但通常是在事件发生之前。
But it usually happens before the event.
所以这会在超大规模企业发布财报时发生。
So this would happen when the hyperscalers report.
而且我确实认为超大规模企业有机会，因为股价表现这么好，这个ROIC对我来说太愚蠢了。
And I do think there's an opportunity for the hyperscalers to, because the stocks are trading so well, this ROIC thing to me is so stupid.
过去两周关于Anthropic和OpenAI的事情真是太愚蠢了。
The Anthropic and OpenAI thing last two weeks was just so stupid.
大家快去听黄仁勋和布拉德·格斯特纳在BG2播客上的对话。
Everyone, please go listen to Jensen Huang with Brad Gerstner on the BG two podcast.
这周末的视频里，我至少会用五六页幻灯片来聊这个。
I will be covering it for at least five or six slides in this weekend's video.
去听听真正懂行的人怎么说吧，他把收入来源这件事讲得多简单明了。
Just go listen to a guy who actually knows what's going on and how simple he makes the whole thing about where the revenues are going to come from.
不是只有OpenAI和Anthropic。
They're not all about OpenAI and Anthropic.
第二次会是在十二月。
The second time would be December.
而且会发生在十二月，因为那时候很多人觉得这一年已经结束了，然后事情就拖到明年去了。
And it would happen in the month of December because that is the time when a lot of people just say, I'm done with the year, and it gets away from them going into next year.
就是那种情况，如果我觉得AI交易一年里只有三个月能行——我觉得最多也就这样——那可能连续三个月，也可能像标普今年那样，或者一个月暴涨，然后没动静，再过一个月又涨。
Those are the kind of channels where if I only think the AI trade is gonna work like three months of the year, which is what I think maximum, it could happen three months in a row, or it could...
就像标普今年那样，或者一个月大涨，然后没动静，再过一个月又涨。
Like the S and P did this year, or it could happen one month, it goes up dramatically, then nowhere, then another month.
这就是这个游戏的本质，所以我认为今年剩下的时间里，就是这两个时间点。
That has been the nature of the game, and so I do believe that those are the two times for the rest of the year.
所以，去吧。
So go.
对于比特币，我预计它会迎来下一次上涨，冲向高点，但我还是要再说一遍，我的看法是，明年，因为从现在到年底在代币化方面发生的所有事情，真正的事情正在发生，越来越多的人在关注加密货币，而关注的人越多，参与的人就越多，资金流入这个领域的现实就越明显。
For Bitcoin, I expect it to have its next move up to the highs, but I'm going to say again that my belief is that next year, because of all of the things that are happening on tokenization between now and the end of the year, real stuff is happening, more and more people are paying attention to crypto, and the more people that pay attention to it, the more people that get involved in it, the more the reality becomes that there's money flowing into the space.
如果资金流入这个领域，比特币也会表现良好。
And if money flows into the space, Bitcoin's gonna do well as well.
当我们坐下来做这个节目时，我们尽量一周都不交谈。
Now when we sit down for this, we try our hardest not to talk all week.
我们在这方面做得很好。
We do a good job of that.
虽然这有点难过，但我们尽量不交谈。
Although it's sad, we we try not to.
我们会列一个小清单。
We make a little list.
这就是我们想聊的所有话题。
Here's all the topics we'd wanna talk about.
这周的清单上没有CPI报告。
This week's list does not have the CPI report on it.
CPI数据出来了，可以说是比较“热”。
CPI came in, quote, unquote, hot.
大家都慌了。
Everyone was freaking out.
但你和我似乎都不在意，也不觉得这是什么大事。
You nor I seem to care nor think it's a big deal.
你在谈通缩，而我进来前三分钟看清单时突然意识到，我们俩都没打算聊CPI，尽管传统金融界大多数人会觉得这是本周最大的新闻。
You're talking about deflation, and it struck me as I was reviewing the list three minutes before I came in here, that neither one of us were planning to talk about CPI, even though most people in the traditional world would think that is the biggest story of the week.
对。
Yep.
你为什么不在意？
Why do you not care?
现在核心CPI是2.6。
So right now, core CPI is 2.6.
不管是2.6%、3.2%还是2.2%，都不会帮你赚钱。
Whether it's 2.6% or 3.2% or 2.2%, it's not gonna help you make money.
这不重要。
It doesn't matter.
所以如果我在这里说，利率水平不会成为你赚钱的因素，这就是现实情况。
So if I come in here and say that interest rates where the level is, is not gonna be the thing that makes you money, it's just the reality of the situation.
我其实已经对这些东西麻木了，因为它们对AI交易的现实影响微乎其微。
I've actually become numb to all of these things because they're so inconsequential to the reality of the AI trade.
这就是我想让大家明白的一点。
And that's the thing I just want people to understand.
如果只有20%，实际上是24%的标普1500成分股，在ChatGPT出现以来创造了超额收益，而那时候通胀率实际上接近两位数。
If only 20%, it's really 24% of the stocks in the S and P 1,500 have created alpha since ChatGPT, a time when inflation was actually up at close to double digits.
这已经变得疯狂了，我觉得美联储让这一切变得疯狂。
This has become insanity, and I think the Fed makes it insane.
我觉得历史也让它变得疯狂。
I think history makes it insane.
通胀是2.3%还是3.2%，真的不重要。
It really does not matter whether inflation is 2.3% or 3.2%.
在市场的历史上，唯一真正出大事的时候，是大量失业、国家因为信贷开始崩盘而陷入衰退的时候。
The only time in history of markets that something really bad happens is when we lose lots of jobs, and the country goes into a recession because credit starts to unwind.
这种情况不会发生。
That is not going to happen.
所以我完全确定这不会发生。
So I have complete certainty that that will not happen.
如果CPI意外上涨到3.5%，股市一年内下跌10%可能吗？
Can the stock market be down 10% over a year because the CPI surprises and goes up to 3.5%?
可能。
Yes.
确实可能。
It can.
那是不是世界末日了？
Is that the end of the world?
不是。
No.
我想强调一下，这个周末，从今天往回看，过去十年，标普指数每年大概给你带来17%的回报。
I'm gonna highlight this weekend that from today, if you look back over the last ten years, the S and P has given you about 17% a year.
如果看过去二十年，回报率超过10%。
If you look back over the last twenty years, it's greater than 10%.
过去三十年，也超过10%。
Last thirty years, greater than 10%.
过去四十年、五十年、六十年、七十年、八十年、九十年，回报率都超过10%。
Last forty years, last fifty years, last six years, last seventy years, last eighty years, last ninety years, greater than 10%.
你知道大多数人是怎么变富的吗？
Do you know how the majority of people get wealthy?
通过投资中资金的复利增长。
Through compounding of money in investments.
标普500指数，10年、20年、30年、40年、50年、60年、70年、80年、90年，整个时间段的总回报率都是两位数。
The S and P 500, that is 10, 20, 30, 40, 50, 60, 70, 80, 90, double digits in total return for the entire time period.
你为什么坐在这里担心别人说泡沫要破裂呢？
Why are you sitting here worried about people telling you that a bubble is gonna unwind?
为什么？
Why?
如果你90岁了，我能理解。
Now if you're 90, I get it.
如果你在市场里有钱，而且你90岁了，你可能活不到享受留在市场里的好处的时候。
If you've got money in the market and you're 90, you may not live long enough to get the benefits of staying in the market.
如果市场下跌，就像我跟我孩子们说的，我说，那就是你停止花钱，然后把更多的钱投入复利机器的时候。
If the market goes down, as I told my kids, I'm like, that's where you stop spending money and you put more money into the compounding machine.
你咬紧牙关，然后说，好的。
You buckle down and you go, okay.
我有工作。
I have a job.
在这个国家，只要想工作的人都有工作。
Everyone has a job in this country who wants a job.
每个人都是。
Everyone.
你可能不是做着你想要的工作，但你确实有工作。
You may not have the job you want, but you have a job.
所以失业救济申请人数才会创历史新低。
That's why the jobless claims are at all time low.
说到历史新低，大家可能知道每周四公布的那个数字，就是每周失业救济申请人数，这个数字是没有根据人口调整的。
And when I say all time low, so people know that number that comes out every Thursday, the weekly jobless claims, that is not adjusted for population.
所以它维持在20万，也就是说有20万人去申请了失业救济。
So it being at 200,000 and sitting at 200,000, that is the amount of people that have had to go claim insurance for being unemployed.
这基本上是有史以来最低的数字了。
That is the lowest number basically in history.
如果按人口比例来算，比人口只有现在一半的时候还要低。
And if you adjust it for the population, it's lower than it was when the population was half as much.
这比以往任何时候都低，所以没有失业问题。
It's lower than all of these other times, so there's no job loss.
我奶奶教得很好，只要人们没失业，他们就会继续花钱。
And my grandmother taught me very well, like, unless people are losing their jobs, they're still spending.
他们可能花的比挣的多，还会刷信用卡。
And they might be spending more than they have, and they're taking out credit cards.
但真正的问题出现时，是当人们开始还不上……
But when the real problems come up, it's when people start paying...
因为失业而停止偿还债务。
Stop paying their debt because they're out of work.
如果他们失业了，就会出现房地产泡沫。
And if they're out of work, we start having housing bubbles.
电信公司也会出问题。
We start having the telecom companies.
因为电信公司建了那么多设施，却没有收入进账。
Because the telecom companies built stuff, they didn't have any revenues coming in the door.
我知道像Core Weave这样的公司确实有收入进账。
I know the revenues are coming into companies like Core Weave.
我知道收入在进来，因为Anthropic是历史上增长最快的公司。
I know the revenues are coming in because Anthropic is the fastest growing company in the history of the world.
那人们都在聊什么呢？
And what do people talk about?
我觉得可能就慢了一个月吧。
I think it may have slowed down for a month.
我是说，你担心吗？
I mean Are you worried?
担心什么？
About what?
哦，我……
Oh, I...
这正是我跟大家说会发生的。
This is exactly what I told everyone was gonna happen.
今天的节目由investorhealth.com赞助播出。
Today's episode is brought to you by investorhealth.com.
你已经建立了投资组合。
You've built the portfolio.
你已经做了长期投资。
You've made the long term bets.
但你还得活着，才能让财富复利增长。
You still have to be around though to compound.
你的身体就是资产负债表，但大多数人从不像投资其他东西那样投资它。
Your body is the balance sheet, and most people never invest it in a way that they invest in everything else.
Investor Health是我们创建的一家公司，提供由临床医生开具处方的方案，涵盖体重、代谢、能量和长寿，包括GLP-1、NAD+和B12，全部由认证临床医生审核和开具，而不是像点菜一样随便选。
Investor Health is a company that we've built that is a clinician prescribed protocol for weight, metabolism, energy, and longevity GLP-1s, NAD plus B12 all reviewed and prescribed by board certified clinicians not picked off a menu.
流程很简单：在线填写一份简短的表格，临床医生审核，然后药物配制好直接送到你家门口。
The process is a short online intake form, a clinician review, then medication compounded and shipped straight to your door.
大多数患者在48到72小时内就能通过审核并收到药物。
Most patients are approved and shipped within forty eight to seventy two hours.
不用跑诊所，不用转诊，也不用排队等。
No office visits, no referrals, and no waiting rooms.
第30天、60天、90天都有定期跟进，还有20位临床医生的支持热线。
Ongoing check ins at thirty, sixty, and ninety days, and a 20 clinician support line.
健康指导也包含在内。
Health coaching is included as well.
套餐每月149美元起。
Plans start at $149 a month.
这些费用符合HSA和FSA报销条件，全美50个州以及华盛顿特区都能用。
They're HSA and FSA eligible, and it's available in all 50 states plus Washington DC.
没有合同约束，随时可以取消。
There's no contracts and you can cancel anytime.
今天就访问investorhealth.com，填写你的申请表吧。
Go to investorhealth.com and complete your intake form today.
Investorhealth.com，为生命做长远投资！
Investorhealth.com Go Long on Life!
投资者健康是一个远程医疗平台，不是医疗服务提供方。
Investor Health is a telehealth platform, not a medical provider.
医疗服务由合作的持牌临床医生提供。
Medical services are provided by affiliated licensed clinicians.
复方药物未获得FDA批准。
Compounded medications are not FDA approved.
个人效果可能有所不同，治疗需要持牌医疗提供方的评估。
Individual results may vary, and treatment requires evaluation by a licensed provider.
适用条款和条件。
Terms and conditions apply.
详情请访问investorhealth.com。
See investorhealth.com.
Anthropic的增长速度会继续放缓，我们可以对此有不同意见，因为人们开始意识到……
Anthropic's growth rate is gonna continue to decelerate, and we can disagree on it because people are realizing that there is...
他们处于商业中最糟糕的位置。
They they have the worst position in business possible.
准备好了吗？
Ready?
好的。
Okay.
他们面临的竞争越来越激烈，对手都是资金充足、很有创业精神的公司。
They have increasing competition from well funded, good entrepreneurial competitors.
我不同意。
Disagree.
请继续。
Go ahead.
等一下。
Hold on.
所以以前是
So it used
OpenAI和Anthropic。
to be OpenAI and Anthropic.
现在呢，是OpenAI、Anthropic、Grok、Gemini、Muse等等。
Now, it is OpenAI, Anthropic, Grok, Gemini, Muse, etcetera.
现在大家都开始进场了。
Everyone is now getting in the game.
不同意。
Disagree.
继续说。
Keep going.
而且，那还只是主要的实验室。
And then, that is just the major labs.
但接着，所有这些公司都开始冒出来说，等等，先别急。
But then you have all of these companies that are starting to come out and say, wait a second here.
Coinbase出来说，嘿，我们在用开源模型，我们在训练自己的东西。
Coinbase comes out and says, hey, we're using open source models and we're training our own stuff.
我们不再需要那些家伙了。
We don't need those guys anymore.
我们自己能做得更好。
We can do it better ourselves.
我们站出来这么说。
We come out, we say that.
Harvey也站出来这么说。
Harvey comes out and says it.
所以基本上大家都在施压，说，嘿，我要对开源模型进行微调。
And so you have pressure from basically people saying, hey, I'm gonna fine tune open source.
如果你看看OpenRouter，你会发现所有开源项目都在飞速增长。
If you look at OpenRouter, you see massive acceleration in all of the open source stuff.
所以像Anthropic这样的公司，以前靠寡头垄断受益，但现在这种垄断已经不存在了。
And so a company like Anthropic, which used to benefit from an oligopoly, that oligopoly doesn't exist anymore.
现在正在变得去中心化，
It is now becoming decentralized,
而且人们
and people
选择很多。
are having lots of options.
我不同意。
Disagree.
而且……
And...
好吧。
Okay.
与此同时，你面临来自开源、自己掌握智能以及其他模型的竞争，可以说是千刀万剐式的打击，而token成本也在急剧下降，因为那边也有巨大的压力。
And what ends up occurring at the same time that you're getting competition, death by thousand cuts from open source, owning your own intelligence, any other models, is token costs are falling off a cliff because there's massive pressure there as well.
所以你看Grok、OpenAI等等。
So when you see Grok, OpenAI, etcetera.
所以基本上我用的公式是：处理的token数量乘以每个token的价格等于收入。
So what you basically do is the equation that I use is number of tokens processed times price per token equals revenue.
按百分比计算，每个客户通过Anthropic处理的token数量，我猜是在下降的。
The number of tokens processed on a percentage basis going through Anthropic on a per customer basis, my guess is is falling.
Ramps AI指数就能看出来。
Ramps AI index shows this.
开源的兴起也说明了这一点。
The rise of open source shows this.
各大实验室竞争加剧，这也印证了这一点。
The increase of competition from major labs shows this.
然后你的单token价格也在下降。
Then your price per token is falling as well.
所以它们还会继续增长。
So they will continue to grow.
我并不是说它们的增长会停止。
I'm not here to say that they are going to stop growing.
但增长速度会从那种令人惊叹的状态，
But the growth rate is gonna go from, oh my god.
逐渐从非对称式增长转变为越来越线性的增长，因为它们所做的事情并非独一无二。
This is the greatest thing ever to it will go from asymmetric to it slowly become more and more linear because not anything that they're doing.
其实我觉得他们还在持续创新。
I actually think they're continuing to innovate.
只是市场竞争压力最终对我们消费者有利，但那是完全不同的商业模式，我们也卖这个。
It's just the market pressures end up being good for us as consumers, but, like, that's a very different business, and we sell that.
他们的估值，现在那些原本可能达到两万亿的公司，看起来会变成十一万五千亿。
Their valuation, people that are gonna come out at 2,000,000,000,000 now, it looks like $11,500,000,000,000.
潜在估值下降了25%。
It's a 25% decrease in potential valuation.
他们在六到八周内损失了五千亿。
$500,000,000,000 that they've lost in six weeks, eight weeks.
所以你不同意？
So You disagree?
请讲。
Go ahead.
教授，请发言。
Have the floor, professor.
那么，Jordi教授。
So Professor Jordi.
我们，我们只有，我觉得……
We we've only had, I think...
所以，我们得把对接下来一两个月这类事情的看法，和更长期的看法分开来看。
So let's separate when we have different views on things that are related to the next month or two or things like that and longer term views.
我们意见不一致的一点，在我看来其实是完全一样的情况，就是你当时认为预算会平衡的那个时候。
One of the ones that we disagreed on, which to me is the exact same situation, is when you thought the budget would be balanced.
对吧？
Okay?
这家伙又提那次他赢了我的事。
Guy goes back to the one time that he got me.
不是。
No.
不是。
No.
不是。
No.
不是那一次，而是当你对某件事做长期判断时，你是在根据一个月的趋势来做出长期结论。
Not the one time, but when you're when you're taking a long term view of something, you're making a long term statement based on one month trend of something.
我们一致的地方是，绝对不可能……
And where we agree is there is no way...
所以Anthropic有10倍的增长。
So Anthropic has 10 times.
到今年年底，他们的年度经常性收入（ARR）将大约是去年同期的10倍。
At the end of this year, they will be around 10 times where they were a year ago in ARR.
这一点毫无疑问。
No question about it.
他们仍然保持这个速度。
They're on that pace still.
即使他们只是重复从第二季度到第三季度的表现，到今年年底，他们的ARR也将接近1000亿美元。
Even if they do what they did from q two to q three, they will be close to a 100,000,000,000 of ARR at the end of this year.
据说是这样。
Allegedly.
好的。
Okay.
所以好处就是他们增长了十倍。
So the beauty of that is they've 10 times.
如果你问我明年他们会怎么样，我不知道，可能是三倍，三倍半吧。
If you ask me where I think they'll be next year, I don't know, three times, three and a half times.
所以我们一致认为它不会以同样的速度增长。
So we agree on the fact that it won't grow at the same pace that it has.
当然。
Of course.
公司更大了，各方面都不一样了。
It's a bigger company, everything.
而我不认同的地方，也是最重要的一点，我们还没聊过这个，所以我最后会通过把Sylvia和摩根大通做对比，来谈谈他们到底会怎么用这些技术。
And and where I disagree, and this is the most important thing, we haven't talked about this, so I'm gonna I'm gonna I'm gonna get to it in the end by comparing Sylvia to JPMorgan, k, in terms of what they're gonna use.
对吧？
K?
你是说每个人都会……
You're you're saying everyone's gonna...
我跟你说，我在银行工作过。
I'm telling you, I've worked at a bank.
当你每年花160亿美元或者更多在技术上时，你不会把开源作为你的重点。
When you're spending 16,000,000,000 a year or whatever they're spending now in their technology, you are not putting open source as your big thing.
那是不可能的。
That is not happening.
我来告诉你为什么。
I'll tell you why.
不。
No.
不。
No.
不。
No.
不。
No.
这甚至不算……
It's not even...
我……
I...
这个，我甚至不会让你说太多。
This one, I won't even, like, let you go too far.
如果你要告诉我完全是……
If you're gonna tell me exclusively...
不。
No.
不。
No.
好的。
Okay.
就是……
Just so...
也许我该先退一步说。
And maybe I'll back up for second.
先别管那些比赛了。
Forget all the competitions for a second.
我就跟你说说我的亲身经历吧。
I can just tell you from my experience.
当初我们跟西尔维亚开始用这些模型的时候……
When we started with Sylvia using the models...
其实我有很多问题想问。
And I've got a lot of questions, actually.
也许这样能帮我更好地理解这件事。
Maybe it's helpful for me to understand it.
我来解释一下。
I'll explain this.
我们一开始用的是ChatGPT的封装。
We start out ChatGPT wrapper.
基本上，我们开发了一系列软件。
So, basically, we build a bunch of software.
你连接你的账户。
You hook up with your accounts.
连接账户后，你可以和AI对话，它会把你关于如何减少税务的问题以及你的个人财务背景一起输入到上下文窗口中。
When you hook up your accounts, you can talk to AI, and it feeds in the context window your query, how to get my taxes down, along with the context of your personal finances.
然后它会逐项资产分析，并说，好的。
And it goes asset by asset and says, okay.
这里有税收亏损收割，那里有奖金折旧，等等。
This is a tax loss harvest here, bonus depreciation here, etcetera.
所以这比谷歌好用。
So it's better than Google.
这比普通的语言模型要好，因为你不仅有上下文，还有那个额外的信息。
It's better than just a regular LM because you get the context plus that.
我们大概五个月左右，从零做到了每月55万美元的计算成本。
We went from zero to $550,000 per month of compute cost in, like, five months or something.
简直是爆发式的增长。
Like, it was explosive.
那就像警报响了一样，让人惊呼，天哪。
That was like an alarm going off saying, oh my god.
每次我在推特上让大家去用Sylvia，钱就从我口袋里往外流，是的。
Every time I tweet and tell people to go use Sylvia, money is coming out of my pocket Yes.
因为他们用了更多的计算资源。
Because they're using more compute.
我们做的第一件事，甚至都没碰模型那部分。
Now the first thing we did was we didn't even touch the model stuff.
就是从架构上，如果你刷新页面，它会调用模型，但当我们刷新页面时，我们不应该去调用模型。
It was just like architecturally, if you refresh a page and it hits the model, well, like, when we refresh the page, we shouldn't hit the model.
所以有很多人都在琢磨怎么用这种模型计算来构建技术。
So there's a bunch of, like, people were just trying to figure out how do you build tech...
今天这种模型计算下的架构技术。
Architectural technology today with this whole model computing.
等我们把那些容易摘的果子都摘完了，才去看模型本身。
Once we got all this this low hanging fruit stuff, then we looked at the model.
我一开始并不相信要从前沿模型转向开源模型，直到我意识到这样延迟更低、性能更好。
I was not convinced to switch from the frontier models to open source until I realized that it was gonna be lower latency and better performance.
成本低得惊人。
The cost is incredible.
如果我们用Anthropic处理5亿个token，相比用我们自己的硬件和软件，我们少付了97%的钱。
If we process 500,000,000 tokens on Anthropic versus our own hardware and software, we paid 97% less.
光这一点就足以让人动心了，但真正让我下定决心的是，等等，你是在说答案会更准确吗？
That alone should be a big incentive for people to do But the thing that got me convinced was, wait a minute, you're telling me that the answers will be more accurate.
你是在说响应会更快吗？
You're telling me that they'll be quicker.
对用户来说，体验会更好。
It'll be a better performing experience for the user.
哦，顺便说一句，还便宜97%？
Oh, and by the way, it's 97% cheaper?
那当然，我们肯定走这条路。
Of course, we're gonna go this way.
所以我觉得有一个很大的阻力，大家现在还在摸索，就是很难想象摩根大通会把所有信息交给Anthropic，然后说，请别偷看。
So what I think ends up being a huge headwind, which again, people, they're still trying to, like, kinda feel in the dark on this, is I find it very hard to believe that JPMorgan is gonna take all of their information, and they are going to send it to Anthropic, and they're gonna say, please don't look.
请别拿这些数据去训练。
Please don't use this for training.
请别做任何这类事情。
Please don't do any of this stuff.
相反，他们可以选择用开源软件，然后在自己内部的数据架构上训练模型，等等。
And instead, they have the option to go and take open source software, and then they can train it on their own internal data architecture, etcetera.
现在还有监管、法律、数据隐私和安全方面的顾虑。
There are still regulatory concerns, legal concerns, data privacy, security.
就是这些乱七八糟的东西。
Like, there's all this kind of stuff.
但我其实觉得这种体验更安全、更私密、更符合监管要求，比我们用那些开源的...
But I actually think that this experience is more secure, more private, more regulatorily friendly then we're gonna use this opens...
我们用Anthropic、OpenAI这些。
You know, we're gonna use the Anthropic OpenAI, etcetera.
所以对我来说，不只是成本问题。
So to me, it's not just the cost.
而是控制权。
It is controlling.
你可以把它看作主权。
Think of it as sovereignty.
摩根大通想要对它们的人工智能拥有主权。
JPMorgan wants sovereignty over their AI.
它们不想依赖Dario。
They do not wanna be reliant on Dario.
你同意还是不同意这个观点？
Do you agree or disagree on that?
你看，问题在于你的想法是来自你在Sylvia的经历。
See, the problem is your your your thought is coming from your experience at Sylvia.
是的。
Yes.
好的。
Okay.
也许是我没看到。
Maybe I'm blind.
不过，事情是这样的。
Well, here's the thing.
你肯定希望Sylvia的估值能跟摩根大通的技术预算一样高。
You would die for the valuation of Sylvia to be equal to the tech budget of JPMorgan.
当然。
Of course.
好的。
Okay.
摩根大通刚刚
JPMorgan Just
只是时间问题，Jordi。
a matter of time, Jordi.
没错。
Exactly.
所以这就涉及到两种经济之间的差异，而这就是我想说的部分。
So Getting getting But this gets into the difference between the two economies, and this is the part that I wanna say.
摩根大通没有时间等待。
JPMorgan doesn't have time to wait.
你提到了OpenRouter。
You mentioned OpenRouter.
谁刚刚收购了OpenRouter？
Who just bought OpenRouter?
Stripe。
Stripe.
Stripe的估值是多少？
What is Stripe's valuation?
大概2000亿美元。
$200,000,000,000 about.
上一轮是1600亿，但据我了解，考虑投资的人说更接近2000亿。
Like, 160 is the last one, but the people telling me who are thinking investing is closer to 200.
花旗银行的市值是多少？
What's Citibank's market cap?
2000亿美元。
200,000,000,000.
Stripe有多少员工？
How many people does Stripe have?
1万人。
10,000.
花旗银行有多少员工？
How many people does Citibank have?
二十万。
200,000.
所以Stripe收购了OpenRouter。
So Stripe bought OpenRouter.
为什么？
Why?
他们不是一家路由公司。
They're not a they're not a routing company.
摩根大通的净利息收入……
JPMorgan's net interest income...
你看到他们最新财报上的数字了吗？
Did you see the number on their most recent earnings report?
几百亿。
Tens of billions.
是1070亿美元。
It was 107,000,000,000.
好的。
Okay.
我想把这个数字……
I wanna put that numb...
1070亿美元。
107,000,000,000.
所以摩根大通会拼命保住这笔钱，因为Stripe能给你什么呢？
So JPMorgan is gonna be in a speed race to not lose that money, because what does Stripe offer you?
就是不用……
The ability of not...
你把钱存进去，能拿到收益，而我把钱存进摩根大通，一分钱收益都没有。
You put your money in, you're getting paid yield as opposed to I put my money into JPMorgan, I get zero yield.
他们以零利率借钱，然后以四、五、六、七的利率贷出去，赚取巨大的利差。
They're borrowing money at zero, and they're lending it out at four, five, six, seven, and they're making a huge spread.
银行有两部分业务。
There's two parts of a bank.
一部分是……
There's the...
把钱存进去放着，另一部分是……
Let me put it in the bank and keep it there, and then there's the...
对。
Okay.
银行靠借贷和利差赚钱。
The bank is making money borrowing and spread.
所以我相信托尔斯滕的观点……
So I believe what Torsten's lock...
AI代理会把那些钱转走。
That AI agents will move that money out.
我公开说过这件事。
I've said it publicly.
我再重复一遍。
I'll say it again.
从2022年到现在，货币市场基金增长了70%。
If you go from 2022 and you look at money market fund growth, it's been 70% growth since 2022.
银行存款只增长了5%到6%。
Bank deposits have grown by five, 6%.
资金从银行流出的现象已经发生了。
The the the taking money out of the banks is already happening.
所以利差扩大了，他们也能把钱贷出去。
So the spread has gotten wider, and they've been able to lend it out.
我提这个的原因是，说到智能方面，未来三年对摩根大通来说唯一重要的就是智能体工作流。
The reason I bring that up is when you talk about the intelligence side, the only thing that matters to JPMorgan over the next three years is the agent workflow.
他们基本上需要应对客户，因为利差会收窄。
They need to basically deal with people because their spreads are gonna be going down.
所以利润率的游戏在于，AI如何让财富500强公司，比如花旗银行……
So the game for profit margins is how does AI in the Fortune 500 companies, the Citibank...
花旗银行要想跟Stripe竞争，而Stripe少了19万人，他们就得裁员，但还得继续做生意，保持利差。
The only way Citibank can complete with Stripe who has a 190,000 less people, they need to get rid of people and still do the business and keep their spreads.
这是一场非常复杂的博弈，一方面智能体在侵蚀你的净利息收入，让你头疼。
It is a very complicated game to do that with agents on the one side that are killing you because they are eating at your net interest income.
但与此同时，你又得想办法用它们来替代员工的工作流程。
But at the same point, you have to find a way to replace the people's workflows.
所以你得在没有员工的情况下应对这一切。
So you're dealing it from a position of no employees.
你希望用户有更好的体验，同时公司也能找到更便宜的方式来实现这种好体验。
You want the user to have a better experience, and you want the company to find a cheaper way to get that experience to be good.
我完全同意这个观点。
I completely agree with that whole thing.
对于这种趋势，我没有任何异议。
I have no problem with the fact that that'll go.
但在财富500强公司里……
But in a Fortune 500 company...
那我们把消费者分开来看，也把我认为西尔维亚属于的那部分分开来看。
So let's separate the consumers, and let's separate what I consider Sylvia as being part of...
我认为这就是所谓的代币化数字经济。
I consider this the tokenized digital economy.
数字经济，比如亚马逊对零售商场，谷歌对报纸，所有这些崛起都带来了最大的利润。
The digital economy, the Amazon versus retail malls, the Google versus newspapers, all of that rise has led to the most profits.
所以我同意的是，未来十年的赢家是那些只用开源的人。
So the part that I agree with is the winners of the next ten years are the people that only use open source.
这一点我们是一致的。
So we agree on that.
但企业还是会付钱给Anthropic和OpenAI，因为它们的智能更先进。
But the enterprises will pay Anthropic and OpenAI because of the advance...
那些最有钱的公司，应该把最多的钱花在AI上，而不是去冒险。
The smarter intelligence, and the the companies that have the most money, they should be spending the most money on their AI and not taking a risk.
黄仁勋经常说，他在BG2上也说过，人们不想冒险去买那些更便宜的芯片，而我可以用一千亿美元交付一个完整的系统，保证能用，因为我有几十年的经验，至少过去五年是这样做的。
Jensen Huang talks about all the time, and he did on BG two, that people don't wanna take the risk of buying these chips that are cheaper when I can deliver them an entire system for 100,000,000,000 that will work because I have decades of of doing this or at least the last five years.
同样的事情也会发生在Anthropic和OpenAI这些更高层面的公司上，但渐渐地，每家公司都会像你说的那样去做，就是，好吧。
The same thing will happen with Anthropic and OpenAI at the higher levels of things, but gradually each company will do what you're saying, which is, okay.
我得找到更好的办法，因为不能让成本一直涨下去，而我的收入增长得更快。
I gotta find a better way because I can't have these costs keep going up because my income is going at a faster pace.
所以在这点上，我和你的看法很相似，就是开源会赢。
So I do view this in a very similar way to you that the open source win.
但如果你听黄仁勋说的，他说得很到位。
But if you listen to Jensen Huang, he said it perfectly.
他说，我觉得把目光放在Anthropic和OpenAI的收入上是个错误。
He goes, I think it's a mistake to view Anthropic and OpenAI and focus on their revenues.
他们做的事情会非常成功。
They will be very successful of what they're doing.
他们有最好的模型。
They have the best models.
他们有最高的智能。
They have the highest intelligence.
他们会主导研究和科学领域。
They will dominate research and science.
这一点毫无疑问。
No question about it.
同意。
Agreed.
他们不再让任何人看到他们的模型了，这意味着蒸馏会减少，我们已经在中国数据中看到了这一点。
They're not letting anyone see their models anymore, which means distillation goes down, and we're already seeing that in the China data.
是的。
Yep.
所以那些开源人士认为他们有智能模型，那很好，但不是在数学、科学或重建这些领域。
So all the open source people that think they have a smart model, that's great, but not for math, not for science, not for rebuilding things there.
我认为这一切都会开始成形，我们正处于非常早期的阶段，但我认为人们怀疑是错误的。
I think all of that's gonna start to take shape, and we're in the very early innings, but I think it's a mistake for people to doubt.
收入会来的。
The revenues will come in.
我觉得他们的收入会来自很多不同的渠道，而Anthropic和OpenAI可能永远达不到人们原本对这家公司估值的预期。
I think they're gonna come in from so many different sources, and Anthropic and OpenAI may never meet the valuation of the company of what people thought.
如果五年后我们坐在这里，Anthropic和OpenAI加起来价值四万五千亿美元，那之前讨论他们能不能成功、会不会倒闭，或者是不是最好的投资机会，就都白费了。
If we sat here in five years and Anthropic and OpenAI combined were $4,500,000,000,000, It was a waste of money to talk about the whole thing both from are they gonna make it or go extinct, or is it the greatest investment opportunity?
我们已经错过了最好的投资机会，
We already missed the greatest investment opportunity with
就是他们。
them.
我其实觉得他们会继续增长，人们都在问，他们会成为伟大的公司吗？
I I actually think they will continue to grow, and and people are like, are they gonna be a great company?
他们已经是伟大的公司了。
They're already great companies.
所以，这绝不是我以任何方式贬低他们。
So, like, this is not me, like, shitting on them in any form or fashion.
我觉得的是……
What I think is...
而且可能说得更明确一点，每个公司或每个客户的token消耗量会随着时间不断下降，因为他们……
And maybe to be explicitly clear, the per company or per customer token consumption is gonna keep going down over time because what they...
OpenAI和Anthropic只是公司接触AI的入门选择。
OpenAI and Anthropic are the starter pack for a company to get involved in AI.
你一开始会先采用它们，因为你什么都不懂。
You go and you adopt them first because you don't know nothing.
你需要，嘿，
You you need, hey.
怎么才能让这个跑起来？
What is the thing to to get this going?
我要开发软件，然后我需要接入AI。
I'm gonna build software, and then I need to plug in AI.
它们有智能。
They have intelligence.
随着公司逐渐进阶，开始需要定制化、改进、优化等等，你就会从通用模型转向专用模型，进行微调，使用开源方案，诸如此类。
As companies move up the curve of I need to now customize it, improve it, optimize it, whatever, then you start to move off of the general purpose models into the specialized models to fine tuning the open source, like all that kind of stuff.
所以对我们来说，六个月之内搞定这件事非常容易。
And so for us, it's super easy to do that in a matter of six months.
但对摩根大通，甚至大型律师事务所或那些大公司来说，可能得花六到十年。
For JPMorgan or even large law firms or these big companies, maybe it takes six to ten years.
但最终结果就是，他们会想尽可能多地掌控这些，因为这是自主的、更便宜、更好，等等。
But the end result is they're gonna end up wanting to own as much of that as possible because it's sovereign, it's cheaper, it's better, whatever.
我觉得有一段时间窗口，在这期间……
I think that there is a window of time where...
我们还没讨论的另一个方面是，Anthropic有销售团队。
The the other aspect we haven't talked about is, like, Anthropic's got a sales team.
开源没有销售团队。
Open source doesn't have a sales team.
所以如果你是坐在摩根大通的那个人，你有权决定怎么做，那，我不知道。
And so if you're the guy sitting at JPMorgan and you get to make the decision as to what you're gonna do, well, I don't know.
也许那个带我去看尼克斯比赛的人，嗯。
Maybe the dude taking me to the Knicks game Mhmm.
会拿到一些生意。
Is gonna get some business.
所以，这其实还有另一层因素，不只是什么对公司长期最好。
And so, like, there is a whole other element of this that isn't just, like, what is the long term best thing for the company?
还有人际关系，还有所有这些。
There's human relationships and and all of this.
所以我并不是说它们一文不值，或者不是好的投资，或者那些东西。
So I'm not arguing that they're zeros or they're not bad investments or any of that stuff.
我只是说，如果快进到十五年后，我觉得大多数公司都会想拥有自己的完整技术栈。
I'm just arguing that if I fast forward fifteen years from now, most companies, I think, are gonna wanna own their full stack.
大公司会慢一些。
Big companies are gonna be slower.
小公司已经在这么做了。
Small companies are already doing it.
所以在我看来，这个趋势面临的阻力远不止是……
And so the trend to me is, like, that there is way more headwind here than just, like...
我们在这个播客里聊过胖协议理论吗？
Have we ever talked on this podcast about the fat protocol thesis?
没有。
No.
没有吗？
No?
肯定没有。
Definitely not.
好吧。
Alright.
让我先回顾一下。
Let me go back for a second.
我对AI的大部分看法都来自我在加密货币领域的经验。
Most of my view of AI comes from my experience on crypto.
如果你看看那些说AI会毁灭我们的人，他们之前也告诉我们政府会关闭加密货币。
And if you look at all the people who are saying AI is gonna kill us, they were telling us that the government was gonna shut down crypto.
所以，新技术之间有很多相似之处，诸如此类。
So, like, there's a lot of parallels between new technology, whatever.
而我，只是有幸亲身经历过一次这样的技术从零到一的时刻，现在我又看到了类似的情况。
And I was, only now fortunate enough to have lived through one of these, like, zero to one moments in technology, and now I'm seeing it.
我就想，哦，这肯定就是每一项技术都会经历的。
I'm like, oh, this must be, like, every technology.
但在加密货币领域，如果你回到大约十二年前，有一个非常著名的胖协议理论。
But in crypto, if you go back twelve years ago or so, there was this very famous fat protocol thesis.
这个理论的核心思想是，所有的价值都集中在协议层。
And the whole idea was all of the value accrues to the protocol.
所以应用层、基础设施层，所有这些，都不会有什么价值。
So the apps, the infrastructure, like all of that, there's gonna be no value there.
所有的价值都会流向比特币、以太坊、索拉纳等等。
It all goes to Bitcoin, Ethereum, you know, Solana, whatever.
结果这个理论并不成立，因为很明显，应用层、基础设施层和协议层都有大量的价值。
Ended up not being true because, obviously, there's lots of value at the app layer, the infrastructure layer, and the protocol layer.
AI也很类似，虽然还没人这么叫，但基本上有个所谓的“胖模型”理论，很多人已经这么想了很久……
AI is very similar in that no one has called it this yet, but there's basically this, like, fat model thesis that people have had for a lot...
好几年了，对。
For a couple years Mhmm.
就是说所有价值都会集中在OpenAI或Anthropic身上。
Which is like all of the value is going to accrue just to OpenAI or Anthropic.
我觉得现在很多UI公司和其他人都开始意识到，嘿。
I think now, you know, UI and many others are realizing, hey.
不对。
No.
会有应用层。
There's gonna be apps.
会有基础设施层。
There's going to be infrastructure.
模型层本身也会有价值。
There's going to be, also value at the model layer.
所以那个胖协议理论是错的。
And so the fat protocol thesis was wrong.
我觉得胖模型理论也是错的。
I think the fat model thesis is wrong as well.
但这并不意味着它一文不值。
But it doesn't mean that it's a zero.
只是说它不会拿走百分之百的价值。
It just means that it doesn't take a 100%.
可能它只拿走百分之四十的价值。
Maybe it takes 40% of the value.
其他层面依然有价值。
There's still value in these other layers as well.
所以公司必须选择自己在整个技术栈中的位置。
And so companies have to choose where do they play in the stack.
而且，我打算投入资金。
And, like, I'm gonna put dollars out.
我是把钱投给模型公司吗？
Do I put it to the model companies?
还是投给基础设施？
Do I put it to the infrastructure?
还是投给开源项目？
Do I put it to, you know, open source?
我的钱在这个技术栈里应该流向哪里？
Where does my money go in that stack?
不会全都流到模型公司那里。
It's not all gonna go to the model companies.
而且我……
And I...
你刚才提到的这一点，我觉得每个听众都应该注意，就是我们太纠结于某一点了，这也是我们之前聊到的，如果你只盯着经济收益，那你就是在……
So you brought up the point that I think everyone listening should should be paying attention to, which is we're getting caught too much, and I think this is where our original conversation went on this, which is if you focus on the economics of something, you're just...
浪费时间。
You're wasting your time.
意思是，觉得Anthropic和OpenAI会把所有东西都输给开源，那真是太蠢了。
Meaning, to think that Anthropic and OpenAI are losing everything to open source is just stupid.
同意。
Agreed.
这完全说不通。
It it doesn't make any sense.
我是不是觉得智能的TAM这块蛋糕大到每个人都有机会，不是赢，而是生存？
Do I think that the pie of TAM of intelligence is so big that everyone has a chance not to win, but to survive?
是的。
Yes.
我觉得关键就在于，当Jensen谈论这个的时候，他明确说了有九万亿美元会投入到计算和AI上。
I think that's the key thing is that when Jensen talks about it, he literally says there's $90,000,000,000,000 going into compute and AI.
他几乎每件事都说对了，但他还是有所怀疑，这就是关于BG two的那部分。
He's been right about almost everything, and yet he still doubted, and that was the the part about BG two.
如果你去听Zuckerberg和Alexander Wang的谈话...
If you go listen to Zuckerberg and Alexander Wang talk with...
在Cheeky Pint上，也是一样的道理。
On Cheeky Pint, it's the same thing.
去听听他说Meta相对于Neo Clouds有什么好处。
Go listen to what he said the benefits are for meta over the Neo Clouds.
他并没有说Neo Clouds不重要。
He didn't say the Neo Clouds weren't necessary.
他和CoreWeave关系密切，NVIDIA也是。
He's heavily involved with CoreWeave, so is NVIDIA.
NVIDIA提供GPU。
NVIDIA supplies the GPUs.
Jensen出来说他们的资本成本实际上在降低。
I mean, Jensen came out and said their cost of capital is actually getting lower.
如果你去看看他们的结构是怎么设置的，就会发现这话很有道理。
And if you go read on how their structure is set up, there's a lot of truth to that.
所以我觉得，人们往往把企业的经济状况和公司本身分开了。
So I just think there's a separation for people from the economics of a business and the companies.
Anthropic 十八个月前估值五百亿美元，现在估值两万亿。
Anthropic was a $50,000,000,000 valued company eighteen months ago that is now valued at 2,000,000,000,000.
不管是两万四千亿还是一万九千亿，这都不重要。
Whether it's 2,400,000,000,000, 1,900,000,000,000 is irrelevant.
我唯一确定的是，指望未来三年还能获得同样的回报，那是不现实的。
The one thing I know for sure is to bet that you're gonna get the same kind of return on it over the course of the next three years is silly.
这太傻了。
Like, that's stupid.
我们时间快不够了，但我必须得问你。
We're we're running out of time, but I have to ask you.
你看到埃隆的演示了吗？
Did you see Elon's demo
就是那些新车的演示。
of the of the new cars?
你有什么看法？
What are your thoughts?
我……
You know, I...
我一直避免谈论SpaceX，但我现在要谈它是有原因的。
I've avoided talking about SpaceX, and I'm I'm driving this into SpaceX for a reason.
因为埃隆也说过，他认为SpaceX的规模会超过全球经济。
Because Elon has also said that he sees SpaceX as being bigger than the global economy.
我想让你明白，我的想法是怎么一步步变成现在这样的。
I wanna put that in perspective of how my brain finally got to the place that it could be right.
明白吗？
Okay?
明白。
Okay.
那我觉得这会发生吗？
Now do I think that's gonna happen?
不会。
No.
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。