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。
杰森，你跟我们说说你最近最爱的播客吧？
Jason, do you wanna tell us about your new favorite podcast?
哦，真的超棒。
Oh, it's so good.
我整个信息流现在全是这种内容，你知道吗，自从取消文化把萨克斯搞垮了，现在人人都在用‘r’开头和‘f’开头的词。
My feed is now because, you know, since cancel culture ended Sacks, everybody uses the r word and the f word right now.
我Instagram上的动态，不是同志就是唐氏综合征，要不就是斗牛犬。
My entire feed on Instagram is either gay or Down syndrome or bulldogs.
反正就这三样之一。
It's one of those three.
然后我偶然刷到了一个叫‘Miss Thing’的播客，他们有个环节叫‘同志名 vs 正常名’。
And then I stumbled upon the miss thing pod, miss thing, and they do a bit called gay name straight name.
来，猜猜David的同志名和正常名分别是什么。
Here's gay name or straight name for David.
这事儿有好消息也有坏消息，弗里德伯格。
This is good news and bad news, Friedberg.
来吧。
Here we go.
是gay名还是直名？
Gay name or straight name?
David。
David.
我觉得David这名字挺直的。
David to me is straight.
明白了。
Okay.
但是——
But
他有我理想中的身材。
he has my perfect body.
这有点让人困惑，我老在想：你到底是同志吗？
It can be confusing because I'm kind of, like, are you gay?
但其实不是这样的。
And it's like, no.
我就想变成你，大卫。
I just wanna be you, David.
完全没错。
Totally.
就是像米开朗基罗的《大卫》那样，男性美的典范。
Well, it's so, like, the Michelangelo's David, the male ideal.
身材简直完美。
It's like incredible body.
对啊。
Yeah.
有点小，不好意思。
Kinda small Sorry.
是啊。
Yeah.
哦。
Oh.
有点粗糙啊。
It's a little rough.
你在看啥呢，杰伊·卡尔？
What are you
看什么呢，杰伊·卡尔？
watching there, Jay Cal?
他们这俩人选得真准。
They basically nailed these two.
不过行了，继续吧。
But, okay, keep going.
我觉得查马斯不在他们的候选名单里，但我知道杰森迟早会提到。
I don't think Chamath is on their shortlist, but I know Jason will come up at some point.
gay的名字，还是
Gay name or
正经名字？
straight name?
就是他了。
This is it.
查马斯。
Chamath.
啥？
What?
在
On
三。
a three.
对。
Yeah.
三、二、一。
Three two one.
同性恋。
Gay.
我觉得，就是那种意大利风格的毛衣，特别花哨、颜色鲜艳的那种。
I think, like, Italian sweater, like, really kinda, like, a loud, vibrant sweater.
他平时跟哥们儿去打扑克的时候就穿这件。
He, like, wears it to, like, poker night with his, like, his boys.
我不是说那种正经的扑克局。
And, like, not I'm not talking, like, straight poker.
我是说酒吧里那些同志们的扑克之夜。
I'm talking, like, gay poker nights, like, at the bar.
对啊。
Yeah.
老是聊葡萄酒。
Always talking about wine.
光会聊葡萄酒。
Talks about wine.
总是那种扭捏作态的样子。
Always sort of, like, swishes.
对啊。
Yeah.
就是嘛。
Exactly.
没错。
Yep.
还有就是，健身房里那哥们儿脱了上衣自拍的样子。
Also, it's so like the guy at the gym taking off his shirt, taking selfies.
没错。
Mhmm.
是啊。
Yeah.
别人都在心里嘀咕：哎，查莫斯，你能不能消停点？
And everyone else is kinda like, excuse me, Chamath.
我想用一下镜子。
I'd like to use the mirror.
是的。
Yeah.
我想
I'd like
看看自己。
to see myself.
明天扑克夜见。
See you at the next day, poker night.
当然。
Totally.
你带酒，我们带毛衣。
If you bring the wine, we'll bring the sweater.
好。
Yeah.
有的。
There is.
哇哦。
Wow.
他们确实提到了查马斯。
They did do Chamath.
这太棒了。
That is fantastic.
这太棒了。
That is fantastic.
给《Miss Thing》播客的小伙伴们打个call。
A shout out to my guys at the Miss Thing podcast.
哇哦。
Wow.
太精彩了。
That was awesome.
我觉得我是同性恋。
I think I'm gay.
我完全不知道。
I never knew.
你干嘛了？
What did you do?
客串了一把？
A cameo?
你付钱让他们这么做的？
Did you pay them to do that?
他们是帮我个忙，没要钱。
Did it for me as a favor.
所以我才说了句‘太棒了’。
So I did a That's awesome.
得谢谢那几位。
Well, thanks to those guys.
谢谢各位。
Thanks, guys.
感谢《Mist thing》这个播客。
To the Mist thing pod.
我以前在视频里见过他们。
I've seen those guys before in clips.
我觉得他们特别搞笑。
I find them very funny.
太棒了。
It's so great.
给我的小伙伴们点个赞。
Shout out to my guys.
那真是太酷了。
That was awesome.
好了，各位。
Alright, everybody.
说真的，欢迎回到全球第一播客。
Seriously, welcome back to the number one podcast in the world.
这里是《All In》播客。
It's the All In podcast.
今天又和大家见面的是大卫·弗里德伯格、查马斯和大卫·萨克斯。
With me again, David Friedberg, Chamath, David Sacks.
当然，我就是杰森·卡拉卡尼斯。
And, of course, I'm Jason Calakanish.
如果你是第一次听，叫我杰伊·卡拉就行。
You can call me Jay Cal if you're here for the first time.
第一个话题：OpenAI。
Topic one, OpenAI.
他们没达到ChatGPT的预期目标，弗里德伯格，无论是用户数还是收入都没达标。
They missed their targets for chat GPT, Friedberg, both on users and revenue.
咱们来聊聊这个。
Let's talk about it.
《华尔街日报》周二发布了一篇独家调查报道，称OpenAI原本预计在2025年前实现每周活跃用户10亿次互动（WOWs），但没达成。
The Wall Street Journal says in a a breaking investigative report on Tuesday that OpenAI expected to hit 1,000,000,000 WOWs, weekly active users before the 2025.
他们没达到目标，而且进入2026年都四个月了，这数字还没蹭上去。
They missed that, and they still haven't hit the milestone four months into 2026.
还有啊，Chamath，他们2025年ChatGPT的营收目标也没完成。
Also, Chamath, they missed their 2025 revenue target for ChatGPT.
具体数字没公布，但咱们之前聊过，他们现在的年化营收大概是20300亿美元。
Exact number wasn't specified, but as we've talked about here, they're at a $2,030,000,000,000 dollar run rate.
这行当里头，会计处理上还有些细节没捋清楚。
There's a little bit of accounting nuance that is yet to be worked out in the industry.
Sacks，这事有两个关键点，咱们得说清楚。
Two reasons why this matters, Sacks.
OpenAI现在签了6000亿美元的算力采购承诺，
OpenAI has $600,000,000,000 in spending commitments for compute.
你想想，这数字差不多就是他们在二级市场上目前的估值了。
Just to put that in perspective, that's about what they're trading for on secondary markets.
换句话说，OpenAI整个企业的估值，差不多就等于它明年要花的钱。
In other words, the entire value of the OpenAI enterprise equals their spend commitments in the coming year.
据说，CFO莎拉·弗莱尔对现金流问题很担心。
CFO Sarah Fryer, is coming to liquidity, is reportedly worried, hey.
收入增长根本赶不上开支，可OpenAI今年晚些时候又想上市。
That revenue isn't growing fast enough to keep up with expense, and OpenAI wants to IPO later this year.
这导致弗莱尔和阿尔特曼之间起了冲突，或者说，本来就有矛盾。
This has put Fryer and Altman in conflict or maybe there's some natural tension there.
据《华尔街日报》报道，弗莱尔觉得OpenAI还没准备好应对公众公司的披露标准。
Fryer doesn't think OpenAI is ready for public reporting standards according to the Wall Street Journal.
阿尔特曼显然想快点推进。
Altman obviously wants to move faster.
于是他们发了个联合声明。
So they released a a joint statement.
这简直太荒唐了。
This is ridiculous.
啰啰嗦嗦的。
Yada yada yada.
咱们现在听听萨克斯的看法。
Let's go to you, Sacks.
你觉得这背后到底怎么回事？
What do you think is going on here?
这些是真正的阻力，还是说这只是领跑者在我们这一代最重要的竞赛——通向超级智能的竞赛中，管理大家的预期而已？
Are these major headwinds, or is this just managing expectations as the leader of the pack in the most important race of our lifetimes, the race towards super intelligence.
其实我对这事有点不同看法。
Well, actually have a little bit of a contrarian take on this.
我知道OpenAI这周过得挺惨的。
I know that OpenAI had a really bad week.
就像你说的，那篇《华尔街日报》的文章说他们没达标，用户增长目标没达到十亿，收入也没达标。
Like you said, had that Wall Street Journal article which said that they've missed their numbers, they've missed their 1,000,000,000 user growth target, They missed their revenue numbers.
这让人开始怀疑他们能不能撑得起之前承诺的数据中心投入。
That's called into question whether they can afford the data center commitments that they've made.
而且这周他们还碰上了埃隆的官司。
And then in addition to that, they've also had the lawsuit with with Elon happening this week.
所以媒体上看来，他们这周确实挺惨的。
So in the press, it ended up being I think a pretty bad week for them.
但我有个相反的看法：我觉得过去一两周，要是看产品层面的实际进展，他们其实挺顺的。
But I have a contrarian take on this which is I think that over the past week or two, if you look at kind of what's happening at the product level, it's been a pretty good couple weeks for them.
他们发布了 ChatGPT 5.5，我在硅谷认识的那些人反馈都特别好。
They released ChatGPT 5.5 and the reviews from, you know, people I talked to in Silicon Valley have been really strong.
你去问开发者、程序员，个个都说满意。
You talk to developers, coders, they're very happy with it.
与此同时，Anthropic 最新发布的 Opus 4.7 却搞砸了。
At the same time, Opus 4.7, which is the latest Anthropic release, appears to be a bust.
大家都在抱怨它。
People are complaining about it.
很多人干脆又回退到 4.6 版了。
They're, in a lot of cases, are rolling back to 4.6.
他们说Opus 4.7在限制计算资源。
They're saying that OPUS 4.7 is rationing compute.
它把思考时间砍了一大截。
It's reducing thinking time.
效果差多了。
Not as good.
Claude里头还出了些bug。
There were some bugs in Claude.
所以要是单比ChatGPT 5.5和Opus 4.7，这俩礼拜在产品上，OpenAI确实赢了。
So if you just compare ChatGPT 5.5 to OPUS 4.7, it does appear that OpenAI has had a better couple of weeks at a product level.
我觉得他们接下来的产品改进还会继续。
And I think there's reason to believe that the product improvements will continue.
GPT 5.5是基于一个叫SPUD的新基础模型，这可是他们一年多来第一次升级基础模型。
GPT 5.5 is based on a new base model called SPUD, which is the first base model upgrade they've done in, I don't know, over a year.
有了这个新底座，以后的优化空间也更大了。
And having a new base model will pave the way for future improvements as well.
所以我觉得OpenAI现在对自己的产品挺有信心的，你也能在X上看到，开发者们的偏好正在慢慢转变。
So I think OpenAI is feeling pretty optimistic about their product right now and I think you're starting to see on X that some of the developer mojo is shifting.
我看到很多人说，他们写代码的使用习惯已经从Opus转向了GPT 5.5。
I'm seeing a lot of people saying that they are shifting their their coding usage from Opus to GPT 5.5.
我觉得萨姆最后可能是对的，但原因完全搞反了。
So I think that Sam may end up being right but for the wrong reason.
我的意思是，当初他做那些大规模算力投入的时候，是基于对消费者端用户突破十亿、达成营收目标的预期。
And what I mean by that is that when he made these big compute commitments, it was based on those estimates of hitting the billion users on the consumer side and hitting those revenue targets.
结果消费者业务表现不如预期。
The consumer business ended up being weak.
所以那些目标他都没达成。
So they missed those targets.
但与此同时，编程已经成了AI领域最重要的战场。
But in the meantime, coding has become the all important sector of AI.
因为他们提前押注算力、建了这么多数据中心，现在手里的算力比Anthropic还多。
And because they made all these compute commitments and they built out these data centers, they have more compute than Anthropic right now.
Anthropic现在受限于token了。
Anthropic is token constrained.
比如在服务Mythos时，他们的能力就被拉低了。
It's reducing their ability to serve Mythos, for example.
这让他们对Opus 4.7搞起了算力限流。
It's causing them to engage in compute gating with Opus four point seven.
我明白Dario为啥这么选。
And I understand why Dario made that decision.
我不是说这决策不够明智。
I'm not saying I mean, it was a prudent business decision.
我也没在批评他。
I'm not criticizing him for it.
但我觉得，Sam这次可能又对了，只是原因搞反了——消费端没成，可企业端却火得一塌糊涂，现在让他有底气追上来了，瞧瞧这Polymarket。
But I think, again, I think Sam may end up being right here for the wrong reason, which is he missed on consumer, but enterprise is going gangbusters and is giving him the ability now, think, to catch up on Here's your Polymarket.
这才是眼下最重要的市场。
Which is the all important market right now.
当然。
Of course.
上周我们还聊了Grok和Cursor团队联手的事。
And we talked about Grok and Cursor teaming up last week.
埃隆和他们那帮人。
Elon and the team over there.
现在Polymarket上预测OpenAI在2026年前上市的概率是32%。
Polymarket showing now a 32% chance that OpenAI goes public by the 2026.
这比去年12月的60%下降了不少。
This is down from 60% in December.
还有，查马斯，你之前提了个预警。
And, Chamath, you gave a bit of a warning.
嘿。
Hey.
钱就那么多，分来分去就这幺几份。
There's only so many dollars to go around.
SpaceX的IPO显然要率先上市了。
SpaceX IPO is obviously getting out first.
如果今年OpenAI没上市，而Anthropic先上了，那局面就挺有意思的。
And now if OpenAI doesn't go out this year, and Anthropic does, this sets up an interesting dynamic.
总的来说，你怎么看OpenAI这次这么大的投入？
What are your thoughts here, generally speaking, about the massive commitment that OpenAI has made.
他们是会一头撞上墙，还是会意外大获成功，哪怕理由根本不是当初想的那样？
Are they gonna run off the cliff, or will it wind up being brilliant, even if it wasn't strategically for the exact reasons?
我觉得他们没问题。
I think they're gonna be fine.
我觉得这公司值好几万亿美元。
I think this is a multitrillion dollar company.
我觉得Anthropic也值好几万亿美元。
I think Anthropic is a multitrillion dollar company.
现在大家对AI领域正在发生的事，完全理解错了。
I think the thing that's happening right now is a complete misunderstanding of what's actually happening inside of the world of AI.
现在卡住一切的，只有一个非常具体的关键点——就是驱动这些token所需的电力供应。
And there is one very specific choke point that is constraining everything, which is access to the power that's necessary to drive these tokens.
OpenAI错过的，其实正是没意识到当前算力根本不够，而且这问题只会越来越糟。
To the extent that OpenAI missed, I think what that is is an insight to not enough compute capacity today, and that problem is only getting worse.
Anthropic也早就遇到这情况了，他们直接想办法说服亚马逊给足算力，根本不用绕道Bedrock去调他们的模型。
You've already seen that with Anthropic as well, where they just found a way to economically induce Amazon to give them enough capacity so that you don't have to route through bedrock to get to the anthropic models.
现在他们还跟谷歌这些老伙伴谈出了新条款，不只是租算力，还搞成了利益绑定，好换来更多资源。
You're also seeing them do differentiated deals now with economic participation on top of what they already had from folks like Google to give them more capacity.
我到底想说什么？
What is my point?
这个市场里，所有事都卡在电力上。
Everything in this market is power constrained.
这些人预测不准、业绩没达标，根本不是因为需求不够。
The reason that these folks may miss a number or a forecast have nothing to do with demand.
完完全全、百分之百，是因为生成输出token所需的电力供应跟不上。
It is entirely 100% due to the supply of the power necessary to generate the output token.
今天刚公布了一个特别有意思的消息，会让这个问题变得更糟——现在不仅电力供应跟不上，连真正需要的设备也跟不上了，不只是光伏板和天然气涡轮机，现在连变压器和整个电网基础设施都缺得厉害。
There is a really interesting thing that was just announced today that will make this problem even worse, which is what you're starting to see now is backlogs build up of not just the access to the power, but then the componentry that's actually necessary, not just RESIPs and not just nat gas turbines, but now you're talking about transformers and all the actual tactical grid infrastructure.
这为什么这么重要？
Why is this important?
你看看现在正在建设的吉瓦规模，我们这儿的缺口可太大了。
If you look at the actual amount of gigawatts that are under construction, we have a huge mismatch now.
杰森，大家是宣布了一大堆项目，但实际动工的还不到一半。
People have announced all these projects, Jason, but less than half of it is actually being built.
还不到一半。
Less than half.
大部分都卡在审批流程里了。
Most of it is stuck in red tape.
主要原因就是供应链跟不上，根本没法靠谱地把这些项目真正运转起来。
Most of that is because there are these supply chain delays, so there's no credible strategy to turn any of this stuff on.
这会坑了谁？
Who will this hurt?
受冲击最大的会是Anthropic和OpenAI。
It will hurt Anthropic and OpenAI the most.
那谁会从中受益呢？
Who will this benefit?
受益的会是那些超大规模云服务商，特别是甲骨文、亚马逊、Meta、微软和谷歌。
It will benefit the hyperscalers, specifically Oracle, Amazon, Meta, Microsoft, and Google.
接下来你会看到一场来回拉锯的谈判。
And now what you're going to see is a negotiation and a trade back and forth.
我得让出多少股权？
How much equity do I have to give up?
为了获得算力，我得让出多少控制权？如果不让，又会多大程度拖慢我的增长目标？
How much control do I have to give up to get access to the compute versus how badly will I miss my growth forecasts if I don't?
这也就意味着——我们上周聊过——Grok和SpaceX将迎来绝佳机会，因为他们手上有大量闲置算力。
And now what that means is, and we spoke about this last week, that's a huge lane for Grok to just run through and SpaceX to run through because they have a ton of excess capacity.
所以我觉得，那个Cursor交易只是开胃菜而已。
And so I think the cursor deal was the appetizer.
但要是我是现在的马斯克，我早就冲进这个市场了——要是模型质量追上来了，他现在完全能对Anthropic或OpenAI干点特别疯狂的事。
But if I were Elon now, I'd be running all over this market because if the models catch up in quality, I think he could also do something really crazy with Anthropic or OpenAI right now.
可能OpenAI不行，因为
Maybe not OpenAI because of the
我们一会儿再聊那场诉讼。
We'll get into the lawsuit in
马上。
a moment.
包袱？
Baggage?
对啊。
Yeah.
但他和达里奥真该明天就谈妥这笔交易。
But, man, he and Dario should do a deal tomorrow.
所以你是想说，嘿。
So you're framing, hey.
这里真正的瓶颈是算力。
The the limited resource here is compute.
需求简直爆了。
The demand is off the charts.
不是的。
No.
真正的限制是电力。
The limiting resource is power.
电力带动算力，算力生成token，然后支撑起开发者、协同工作，还有那些消费者和企业怎么都用不够的各类项目。
Power, which then powers compute, which then provides tokens, which then services the massive developer and co work and all these other projects that consumers and enterprises can't get enough of.
明白了。
Got it.
还有，杰森，对Anthropic和OpenAI来说，另一个麻烦是，现在有大量项目就在那儿干等着，其中40%最后都会被砍掉——因为他们没把AI的正面形象搞起来，过去四年里，有四成宣布的项目最后都黄了。
And, Jason, the other factor that complicates that for Anthropic and OpenAI is all the stuff that's sort of sitting around thumb twiddling, 40% of that is gonna get canceled because they've done such a poor job of creating a good positive halo around AI that 40% of all the announced projects get canceled because 40% of all projects in the last four years have been canceled.
对。
Yeah.
而且现在人们对数据中心、AI岗位这些事挺有意见的，这给行业带来了一些阻力。
And there's there there are some bad feelings about data centers, AI jobs, etcetera, and that's causing some headwind.
有人在社会上干出暴力的事，居然还怪到数据中心和AI头上。
People are literally doing violent things in society and blaming data centers and AI for it.
我不想在这事儿上多浪费口舌。
I don't wanna give it too much airtime.
弗里德伯格，你觉得我们现在面对的这盘棋，不管是从算力、能源的角度，还是从企业上市、资本战略的角度来看，资本、算力和能源现在怎么在这么大的需求下发挥作用？
Friedberg, what's your take on the chessboard we're looking at here either through compute energy or through going public on a business level, you know, the strategic nature of capital, compute, and energy now playing a role in this massive amount of demand.
这事儿还悬着呢，谁也说不准。
Still a ball in the air kinda game.
波士顿咨询有个理论，我好像以前提过，叫‘三巨头法则’，他们反复证明，任何稳定成熟的竞争市场，最终都会自然形成4:2:1的市场份额分布，几乎占了90%的市场。
BCG had this theory, I think I talked about this once before, called the rule of three, where they've shown time and again that any stable, mature, competitive market evolves to a four to two to one ratio of market share for basically 90% of the market.
也就是说，第一名的份额是第二名的四倍，第二名又是第三名的两倍。
So there's a market leader that has four times the market share of the second place, that's two times the market share of the third place.
几乎所有成熟的竞争性市场，都是这个规律。
This is the case in pretty much every mature kind of competitive market.
所以你可以想象，AI可能会分化出消费市场和企业市场两个方向。
So you can kind of think about AI probably evolving into the consumer market and an enterprise market.
OpenAI就算还没到十亿，每周用户也快九亿了，远超Claude现在的量。
OpenAI, even if they're not at a billion, they're still at 900,000,000 weekly users, which is well ahead of whatever Claude is at.
我觉得Claude可能连一亿都不到。
I think Claude is, like, probably sub a 100,000,000.
Sacks，你可能更清楚。
Sacks, you may know.
Gemini大概在七亿到十亿之间，跟OpenAI差不多是旗鼓相当。
And then Gemini is probably closer to them at 700 to a billion, somewhere in that range, probably pretty neck and neck with OpenAI.
所以消费市场看起来，就是ChatGPT和谷歌在争第一第二，Anthropic大概排第三，说不定马斯克靠着算力后来居上。
So, you know, the consumer market looks like it's trending towards a chat GPT slash Google fight for first place and second place and then probably Anthropic in third place and maybe Elon emerges and takes off enabled by his compute capacity.
企业市场就不一样了，那是另一个战场，从Vertex的使用情况来看，实际领先的是谷歌，或者说是Anthropic。
And then the enterprise market is a little bit of a different story and that's its own market, which is kind of Anthropic or probably Google in the lead actually if you look at all the Vertex use.
谷歌说，有75%的GCP客户都在用Vertex。
Google claims that 75% of GCP customers are active users of Vertex.
所以谷歌在企业市场也拿下了相当大的份额。
So there's probably a pretty sizable market share that Google's captured on the enterprise side as well.
这大概也是为什么过去几个月谷歌股价一路飙升——他们现在在企业端和消费端都稳居第一，或者至少在争第一。
This is also probably why Google stock has absolutely ripped over the last couple of months is they're literally in first place or fighting for first place in enterprise and consumer.
但我还是觉得，就像查马斯说的，算力和能源限制是很大的瓶颈，我们在如何扩展和部署模型这方面还有 huge 的提升空间，现在才刚起步而已。
But I still think that there's a lot of opportunity to Chamath's point about the compute and energy capacity constraints in improving how we actually scale and deploy models in both the enterprise and the consumer setting and it is such early days.
我想特别提一下麻省理工两位科学家刚发表的一篇论文，他们研究的是神经网络的剪枝技术。
And I just wanna highlight this paper that came out from MIT from these two scientists and these guys published a paper on pruning techniques in neural networks.
这篇论文证明，通过把超大模型压缩，能砍掉90%的体积，但准确率一点不降，然后你可以根据需求选个合适的轻量模型来跑推理。
This paper showed that you could actually reduce the size of these networks by 90% and get the same accuracy out by pruning very large models down to smaller models, and then you can make a selection on which model to run for inference.
这么做，推理成本能直接降十倍。
And by doing this, you can actually reduce inference costs by 10 x.
每单位能耗，输出量能翻十倍，还不损失精度。
You can get 10 x the output per energy unit that goes into the data center with no loss of accuracy.
这其实是个特别有意思的算法思路，能直接用在现有的大模型上，大幅降低能耗。
And so it's a really interesting call it algorithmic technique that can be applied to the existing large models to actually make them much lower energy use.
你想啊，你用一个超大的模型去回答一个特别简单的问题。
So if you think about it, you're firing up a very large model to answer a very simple question.
其实你可以把这模型精简掉。
You can actually prune away that model.
这在AI应用里大概就跟谷歌搜索一样，情况都差不多。
Now this is probably gonna be the case in AI applications as it is in traditional Google search.
搜索里长尾需求很多，但真正占大头的就那么几类。
There's a long tail of searches, but there's a few searches that account for a large percentage of search volume.
比如，今天天气怎么样？
It's like, what is the weather?
电影几点场次？
What are the movies, times?
股票价格多少？
You know, what's the stock price?
总有一些高频问题占了大部分用户能耗，编程场景里估计也一样，就那么几类操作占了大头。
Like, there's a certain set of things that make up the bulk of consumer energy, and there's probably a certain set of things that probably make up the bulk of coding output as well.
所以如果你能通过剪枝技术，把80%的搜索、聊天请求或代码需求都交给更小的模型来处理，再动态调用一整套小型模型，把推理成本降低90%，那你在数据中心和能源容量上的使用效率就能提升十倍，远超现在的能力。
And so if you can get that 80% of searches or chat interfaces or coding requests reduced down through pruning techniques to smaller models, and then you have a whole set of smaller models that can be called dynamically and you reduce inference cost by 90%, you can make much more use, call it 10 times the use on data center and energy capacity than we can today.
我觉得我们现在才刚起步，离真正实现高效输出和节省token还有很长的路要走，而这恰恰给了像马斯克这样的人机会，去重新设计整个系统，甚至强势竞争。
So I would argue that we're still in the very early days of getting efficiency in terms of output and tokens, and we're just in the very kind of early stage of that, which also unlocks the opportunity for guys like Elon to reinvent how this is done and potentially compete pretty aggressively.
赢的方式有两种。
There are two ways to win.
要么砸算力，要么用小语言模型，也就是SLM，还有垂直领域的小语言模型，VSLM。
You could throw compute at it or you can do SLMs, small language models, and VSLMs, verticalized small language model.
比如说，要是有个专门处理天气的小模型——虽然现在还没有，但我们可以拿它当例子。
So if you had a verticalized small language model for the weather, let's say, that doesn't exist, but you can you can use it as an example.
旅行领域也会有这么一个模型。
They will have one for travel as an example.
你去谷歌查航班信息时，它肯定会把你转到一个更轻更快的系统，比如谷歌航班，而谷歌航班现在已经整合进Gemini了。
When you hit Google for flight information, it's obviously going to route you to something lighter and faster that uses Google Flights, and Google Flights has been incorporated into Gemini.
现在Gemini的性能已经接近700，快到750了。
Gemini now is right behind 700, seven hundred fifty
五千万。
Fifty million.
没错。
Yeah.
这不就是我们十八个月前在这档播客里聊过的吗，弗里德伯格？要是把AI回答放在最上面会怎样？
And it's exactly what we discussed, I don't know, eighteen months ago on this podcast, Friedberg, that what if they put it at the top?
那对他们的搜索收入会有什么影响？
And what would that do to their search revenue?
搜索收入正在飙升，他们的整体业绩也在猛涨。
Search revenue is surging, and they are also surging.
他们找到了办法，平衡了这两股对立的力量：既让搜索结果带上AI功能，又能让用户继续点击链接。
So they figured out a way to balance those two competing forces, having search results that are AI enabled, and still getting people to click on links.
他们干得相当漂亮，股价也给出了回应。
They've done it brilliantly, apparently, and the stock is rewarding.
我补充一句，杰伊，你刚才说的，其实用的是我所说的‘人类直觉’来优化小模型。
I'll just add one statement to what you said, Jay Cal, which is like you're using what I would call a human heuristic on smaller models.
我觉得我们正在走向一种状态：人类其实根本说不清这模型到底是什么。
And I think what we're evolving to is humans don't intuitively know what this model is.
它不只是个垂直领域的模型，未来还会出现一些通过自动化剪枝技术发现的模型，它们会协同工作。
It's not just a verticalized model, but there are going to be models that will be discovered through automated pruning techniques that will then be working in concert.
会有大量小模型彼此连接，而我们不再用人类的主观标准去定义每个模型。
So lots of small models that link together, and we don't define each model by some human heuristic.
比如，这个是搜索模型，那个是地图模型。
Like, this is a search model.
对。
Yeah.
这个是地图模型。
This is a maps model.
所以当这些模型被拆开来看时，我们其实搞不懂它们为啥能这么运作，但我相信，真正的进化就发生在这里。
And so we don't we don't know why these models work the way they do when they get broken down, but I do think that that's really where the evolution is happening.
换句话说，模型最终会变成一个宏观系统。
So effectively, a model becomes a macro model.
它底下藏着一堆小模型，能动态调用，这样同样的能耗下，推理能力能翻十倍。
It's got lots of smaller models underneath it that can be dynamically called and that allows you to have 10 x the inference for the same unit of energy.
萨克斯。
Sacks.
我接着你刚才说的谷歌，杰伊，补充一点。
Let me just build on your point about Google, Jay Cal.
要说OpenAI在消费者市场没达到用户和营收目标，主要原因只有一个——那就是谷歌抢走了相当大的市场份额。
I would say that if there is a single reason why OpenAI did not hit its user targets and its revenue targets, certainly around consumer, you'd have to say it's because Google managed to take meaningful share.
一年前他们还几乎不存在呢。
You know, they were basically nowhere a year or so ago.
谢尔盖都退休了，又重新出山来帮公司聚焦方向。
Sergei came out of retirement to help focus the company.
就像你说的，他们把Gemini改得特别出色，还把它无缝整合到搜索顶部，干得真漂亮。
And like you said, they did a brilliant job improving Gemini and putting it at the top of search, incorporating it.
不过话说回来，我觉得OpenAI也不全是坏消息，因为我觉得5.5版本真的挺棒的。
Now that being said, again, I don't think the news is all bad for OpenAI because I do think that the 5.5 release was great.
我们听到不少关于 Codex 的好评。
We're hearing really good things about Codex.
我觉得现在 Codex 正在抢走不少代码相关的使用份额。
I do think that Codex is taking share in coding tokens right now.
我觉得现在这几家公司在互相较劲，你追我赶，特别有意思。
And I just think we're in a really interesting place where these companies are constantly one upping each other.
两周前，大家还觉得 Anthropic 要彻底 dominate 了呢。
I mean, two weeks ago, it looked like Anthropic was gonna be completely dominant.
对吧？
Right?
Anthropic 当时的增长速度是十倍。
I mean, Anthropic was growing at 10 x.
OpenAI 才三倍，那时候看起来……
OpenAI was growing at three x and it looked like
这周 Anthropic 呢，那个
Anthropic was this week, The
服务器老是崩。
servers were going down.
我办公室里的人都在抱怨，说上不了Claude。
People were in my office were complaining we can't get on Claude.
你听我说。
Listen.
竞争才能激发出每个人最好的一面。
Competition brings out the best in everyone.
Anthropic逼得OpenAI不得不迎战。
Anthropic forced OpenAI to compete.
谷歌也让OpenAI在消费端不得不拼起来。
Google forced OpenAI to compete in consumer.
我只希望市场能一直保持这种竞争态势。
I just hope the market stays competitive for as long as possible.
我觉得这对消费者、经济，还有整个国家都是最好的。
I do think that's what's best for consumers, our economy and for our country overall.
我再提一个领域，我觉得OpenAI这周表现挺亮眼的，就是现在特别火的网络安全方向。
Let me just say one other area where I think OpenAI had a good week is in this red hot area of cyber.
显然，Anthropic用Mythos搞了个大动作。
Obviously, Anthropic made a huge splash with Mythos.
不过这玩意儿还没正式商用。
It hasn't been commercially released.
他们算力有限，但作为概念验证或者训练模型，已经在网络安全上达到了新高度。
They're compute constrained, but as a proof of concept or training model, it hit a new level of capabilities with cyber.
现在OpenAI推出了新模型GPT-5.5 Cyber，刚做完一连串测试，据AI安全研究院的测试结果，GPT-5.5是第二个完整跑通他们多步骤网络攻击模拟的模型。
But now OpenAI has released a new model called GPT 5.5 Cyber, which has just been through a bunch of tests and they've shown this was testing done by the AI Security Institute that GPT 5.5 is the second model to complete one of their multi step cyber attack simulations end to end.
所以它的能力跟Mythos不相上下，而且看起来已经准备好商用了。
So it has the same level of capability as Mythos and it does appear to be commercially ready.
你知道的，他们有足够算力来支撑它运行。
You know, they've got the compute to serve it.
我觉得这确实是个了不起的成就。
So I do think that that's a big accomplishment.
我的意思是，我们早就知道其他网络安全模型迟早会出现。
I mean, look, we knew that other cyber models were coming.
不可能只有Mythos这一家独大。
It wasn't just gonna be Mythos.
事实上，未来六个月内，所有前沿模型都会达到Mythos级别的网络安全能力。
In fact, within six months or so all the frontier models are gonna have Mythos level cyber capability.
但OpenAI这么快就推出GPT 5.5 Cyber，确实让人印象深刻。
But it's impressive that OpenAI got this GBT 5.5 cyber out so quickly.
我觉得5.5可能是第一个真正能让网络安全防御方用上的模型。
And I think 5.5 might be the first cyber model that cyber defenders actually get to use.
因为说实话，他们不像Anthropic那样受算力限制。
Because, again, I don't think they're as compute constrained as Anthropic is.
这对CrowdStrike和Palo Alto Networks这样的公司来说，简直是天赐良机，它们都早就参与了这个项目。
And this is an incredible opportunity, you know, for the CrowdTrikes and Palo Alto Networks of the world, both of which have been on the program.
现在它们一出来，就开始在这个领域大展身手。
They come out and they start attacking this space.
天啊，你真的能感觉到一切都变得紧绷起来了。
Man, you could really see everything get tightened up.
这对所有拥有Cursor、Claude、OpenAI或者Gemini的人来说，都是个巨大的赚钱机会——不光是防攻击，更重要的是把系统加固得更牢。
And this could be an incredible revenue stream for everybody who's got, whether it's Cursor, Claude, or OpenAI or Gemini, this is an amazing opportunity to tighten up as much as it is to get attacked.
我能说点看法吗？
Can I make a point about that?
因为现在大家对Mythos的恐惧太夸张了，简直像要末日了一样。
Because, look, there is so much fear right now, almost a level of panic about mythos.
人们把它当成什么末日武器似的。
People are treating it like a doomsday weapon or something like that.
其实根本不是这么回事。
It's not.
只是前沿模型已经发展到能自动执行网络操作了，就像它们现在能自动写代码一样。
It's simply that the frontier models have reached the point where they're capable of automating cyber activities, just like they're capable of automating coding.
这意味着，模型能帮网络攻击者，也能帮防御者，就像帮程序员一样，让他们发现更多漏洞。
But that means that a model could power up a cyber attacker or cyber defender the same way they can power up a coder and allow them to discover a lot more vulnerabilities.
所以这确实有风险。
So there is obviously a risk there.
但我觉得重要的是明白，Mythos或者GPT 5.5并不会制造漏洞。
But I think it's important to understand that mythos or GPT 5.5, it doesn't create the vulnerabilities.
它只是把漏洞找出来。
It just discovers them.
那些bug本来就在代码里。
The bugs are already in the code.
它们一直躺在那儿，等着黑客去发现。
They were sitting there waiting for some hacker to discover.
如果我们现在能用AI提前发现这些漏洞并修复，那实际上就是在加固我们的基础设施和安全体系。
If we can now use AI to find these bugs in advance, these vulnerabilities, and patch them, then you actually harden our infrastructure and and you harden our security.
我还觉得，从AI之前的网络攻防，到AI之后的网络攻防，这会是一次巨大的一次性升级，因为你会一下子把所有沉睡的漏洞都挖出来。
I also believe that this leap from, let's call it pre AI cyber to post AI cyber, it's gonna be, I think, a big one time upgrade cycle because, again, you're gonna find all these dormant bugs and vulnerabilities.
但我觉得，一旦熬过这个升级期，AI驱动的网络攻击和AI驱动的网络防御就会达到一个新的平衡。
But I think that once we get past that upgrade cycle, you're going to reach a new equilibrium between AI powered cyber offense and AI powered cyber defense.
展开剩余字幕（还有 480 条）
这会变得越来越平常。
It's going to become a lot more normal.
不会再让人觉得是天翻地覆的冲击了，也就是说，我觉得大家现在把它当成某种生存威胁，其实没必要。
It's not going to feel like this huge disruption, which is to say, I think, you know, people are treating this as like some existential threat.
只要大家都做好该做的事——赶在黑客掌握这些能力之前，用新工具加固代码、基础设施和安全体系，就没那么可怕。
I don't think it is as long as everyone does what they're supposed to do, which is use the new capabilities to harden their code bases and infrastructure and security before the hackers get ahold of these capabilities.
对。
Yeah.
而且，查马斯，如果你顺着萨克斯的说法往下想，全世界大概有五百万名安全专家。
And if, Chamath, if you were to look at this to build on Sacks' point, there are about 5,000,000 or so security experts in the world.
我们之前聊过token的成本。
And we talked about token cost.
每周花四十小时的token去反复测试，就能搞定。
Forty hours of tokens just pounding it, you know, a week.
只要每个安全专家的成本控制在一百美元，你就能再培养出五百万名。
You could create another 5,000,000 for a $100 per chief security officer per security expert.
所以，就像你提到的，安全专家代理的数量才是关键。
So it's the volume of security expert agent Sacks to your point.
对。
Yeah.
你完全可以拥有五千万、一亿个这样的代理。
You could have fifty fifty million of them, a 100,000,000 of them.
它们根本不用去寻找
They're not finding
没错。
Right.
你需要的东西。
Something you need.
它们只是不停地工作。
They're just Yeah.
它们从不休息。
They never sleep.
他们对这些问题的追查是永不停歇的。
They're relentless in their pursuit of these problems.
这真是个特别棒的观点。
It's a really great point.
能不能再把这一点说得更清楚点？
Just kinda just refine that.
没错，网络安全行业可能有五百万人，但真正顶尖的黑客可能就几千个。
So, yeah, there's probably 5,000,000 people the cyber industry, but there's probably only a few thousand really elite hackers.
当然。
Sure.
这些顶尖黑客根本没时间去扫描每一个可能的攻击面。
Those hackers didn't have the time to go after the entire surface area of every possible attack vector out there.
所以如果你训练一个模型来模仿他们的行为，就像你说的，它能以人类根本达不到的规模和速度运作。
And so if you train a model to do what they do, obviously, like you said, it can operate with a scale and speed that a human hacker can't.
所以很明显，你应该把这些工具交给白帽黑客，让他们主动去模拟攻击，提前发现漏洞并修补，别等黑帽黑客先拿到这些能力。
So obviously, you know, what you need to do is get these tools in the hands of the white hats, let them do the cyber attacks themselves to then find the vulnerabilities and patch them before the blackhats get ahold of these capabilities.
不过我想再最后提一点，说完我就停。
But I think it's just just one last point on this, I'll stop.
关键是得明白，中国模型大概六个月之内就能具备这些能力。
It's just it's really important to understand that the Chinese models are gonna have these capabilities within approximately six months.
哦，他们现在在深海四号上已经有了。
Oh, they have them now in Deepsea four for sure.
至少已经有一定水平了。
They've got some level.
深海四号确实挺厉害，但还达不到最前沿。
Deep sea four I mean, deep sea four is impressive in a lot of ways, but its capability is not at the frontier.
大概能达到美国前沿水平的百分之八十五左右吧。
It's maybe eighty eighty, 85% of, let's call it, the American frontier.
查马斯，你不是想插几句吗？
Chamath, you wanted to get in on this.
来，让查马斯说说。
Let's get Chamath in.
有两点。
Two things.
这事之所以可能，是因为人总会犯错。
The reason that this is even possible is because humans are error prone.
人写代码的时候，总会留下漏洞。
And when humans code, they create holes.
长期以来，我们都是人在利用人。
And so humans exploiting humans is where we've been for a long time.
现在呢，是电脑在利用人——因为电脑会主动去找出人类写下的所有漏洞。
Now we have computers exploiting humans because the computers go and seek out all these bugs that humans wrote.
下一阶段，就是机器对机器了。
In the next phase, it'll be machines versus machines.
我觉得网络攻防的本质，马上就要彻底变了。
And so I think the nature of cyber is gonna completely change.
未来五六年里，全世界的软件有很大一部分非得重写不可，一方面是因为大家要求更高的运营效率和收入增长，另一方面呢，过去那些手工写的系统，根本就太不安全了。
Probably in the next five or six years, there'll be so much reason to rewrite all of the software that runs the world in one part because you're gonna be asked to show more operating leverage and revenue growth, but in another part because everything else that was handmade in the past is just fundamentally insecure.
不管怎样，所有支撑世界运行的操作软件迟早都会被重写。
Either way, all roads will lead to all the operational software that runs the world will get rewritten.
越来越多的部分会由机器来编写。
More and more of it will be written by machines.
结果就是，它们会变得越来越牢不可破。
More and more of it will be impregnable as a result.
但这样一来，网络威胁反而会加剧，因为你得想办法让一台机器去往另一台机器里植入东西，让某个自主循环注入恶意代码或者坏令牌。
But then the cyber threat actually will only increase because then you're gonna try to figure out how to use a machine to inject something into another machine so that some agentic loop injects some malware or injects a bad token.
我觉得这事儿复杂得不得了。
And I think that's a very complicated thing.
我跟你说个事儿，说不定我都不该说——但全球最顶尖的网络安全公司，由全球最顶尖的CEO之一掌管，他可能来也可能没来参加Liquidity大会，他们自己都承认，已经能渗透并操控所有模型。
What I will tell you is I'm not even sure if I'm allowed to say this, but a very good, probably the best cybersecurity company in the world run by one of the very best CEOs in the world, who may or may not be speaking at liquidity, would tell you that they have penetrated and can essentially manipulate every model.
我就这么大概说一下吧。
Let me just let me just say it roughly that way.
明白了。
Okay.
完美。
Perfect.
对。
Yeah.
在突破奖颁奖现场——我们四个人中有三个都在那儿——我跟乔治·库尔茨聊过。
And I'm, at the Breakthrough Prize, which three of the four of us were at, I talked to George Kurtz.
你刚才说的那个人其实不是。
The other person you were kind of describing was not
不是他。
that person.
尼凯什。
Nikesh.
是的。
Yeah.
我说的就是尼凯什。
I'm talking about Nikesh.
嗯，说的是Nikesh
Well, Nikesh
George和Nikesh是这俩人领头的。
and George are the two guys leading this.
Palo Alto Networks和CrowdStrike。
Palo Alto Networks, CrowdStrike.
他们明白，George跟我讲过，现在想用这个产品或服务的人多到排长队。
And they understand the the what George told me was there is just a line out the door of people who want this product or service.
你要是看数据，比如凶杀率，我们现在正处在人类历史上最低的凶杀率水平。
And if you look at it, Friedberg, like the murder rate, like, we're sitting here with the lowest murder rate in the history of humanity.
我们这一代人亲眼看着它大幅下降，从历史长河来看更是如此。
It has gone down massively in our lifetimes, but massively over the arc of history.
我觉得网络攻击也会是同样的走势。
I think that's what's gonna happen with cyber.
攻击面就那么些，剩下的就全是人为漏洞了。
There is only so many attack vectors, and the remaining attack vectors are just gonna be human factors.
对吧，弗里德伯格？
Right, Friedberg?
一直都是这样。
That's always been the case.
随着软件越来越安全，最薄弱的环节就成了那些秘书把密码写在便利贴上，或者会计用狗的名字加123当密码之类的情况。
And as we make the software more resilient, then the the weak link is, you know, the secretary who puts her Post it note, you know, with the password there or the accountant who, you know, uses their dog's name Plus123 for their password.
对吧？
Right?
这一直是老问题了。
That's the the historic one.
明白了。
Okay.
我们快结束了，弗里德伯格，你还有什么想补充的吗？
Let's any anything you wanna add, Friedberg, as we wrap there?
哦，这有点为什么
Oh, that's a bit Why
对了，你床怎么这么乱？
is your bed so messy, by the way?
你干嘛不叫客房服务来收拾一下呢？
Why can't you just ask the room service to come in and
我能告诉你发生了什么，我帮你收拾都行。
clean your I could tell you what happened.
你听我说。
Listen.
我现在在亚特兰大。
I'm here in Atlanta.
而且你干嘛不订个套房呢？
And, also, why don't
套房不是有两间房吗？
you have a suite, like, where there's two rooms?
就只有一间房？
Like, is it just one room?
这说明我就住一间房。
This will tell me I one room.
对。
It Yes.
你懂的吧？
You know?
你要么抠门，要么穷。
Either you're cheap or poor.
到底是哪一种？
Which one is it?
我抠门。
I'm cheap.
我就是抠门。
I'm cheap.
我跟你说说实际情况。
Here's what I'll tell what here's the situation.
我今晚在亚特兰大看尼克斯队的比赛。
I'm in Atlanta for the Knicks game tonight.
你一个人住一间房。
You're in one room.
我来跟你说说我怎么做。
Here's what I do.
我想跟你讲清楚，什么叫物有所值。
I just wanna explain to you value for value.
有些人花钱买私人飞机，飞趟亚特兰大就花三万块。
Some people spend their money on private jets, and they spend $30,000 flying to Atlanta.
我呢，三万块买的是场边座位。
I spend 30,000 on courtside seats.
我不想要
I don't want
套房。
the suite.
我想把钱花在座位上。
I wanna put it into the seat.
这两样怎么兼顾？
How do do both?
你
You
可以两样都做到。
can do both.
我想我也可以两样都兼顾。
I guess I could do both too.
我正在学会接纳自己的富足。
I don't I'm I'm in the process of becoming
我不明白。
I don't understand.
拥抱自己的富足。
Of embracing my richness.
行吧？
Okay?
你都 convincing 自己了，觉得花三万块买球场边的票值，我觉得这太离谱了，但好吧。
If you've already convinced yourself that you should spend $30,000 for courtside tickets, which I think is outrageous, but okay.
你早就说服自己了。
You've already convinced yourself
每张一万块，不过呢。
10 k each, but yeah.
对啊。
Yeah.
一间有两个房间的酒店。
A hotel room that has two rooms.
明白吗？
Okay?
我打算
I'm taking
大概比你现在付的贵百分之十五。
Probably cost 15% more than what you're paying.
是两倍，不过你说得对。
It's two x, but, yes, you're right.
我来订酒店房间。
I'll get the I'll get the hotel room.
贵百分之二十，但你只是订了房间。
20 or 20% more, but you did just room
一晚大概两百块。
like like $200 a night.
那你就得一晚付四百。
So you pay 400 a night.
这样能多一间房。
You get in more room.
我的意思是，这可是亚特兰大啊。
I mean, it's it's Atlanta.
我住的是最好的酒店。
The most I'm in the best hotel.
在亚特兰大，最贵的酒店一晚要500块。
The most expensive hotel is 500 a night in Atlanta.
这没什么大不了的。
It's no big deal.
但所有房间都订满了，因为尼克斯队的人都来了。
But everything's sold out because all the Knicks people are coming.
所以你是想让我花双倍的钱，也就是一千块，你居然不肯花一千？
So you're selling me double that would have been a thousand, and you couldn't spend a thousand?
房间全订光了，因为尼克斯队现在在城里。
Is sold out because the Knicks are here.
所以我们得看看你那脏床。
So we have to look at your dirty bed.
太脏了。
It's gross.
床也没那么脏啊。
The bed's not that dirty.
拜托了。
Come on.
哎呀，别纠结了。
Oh, just deal with it.
行吧？
Okay?
拿出来发一下。
Take it out and post.
我有个关于飞去亚特兰大的私人飞机故事。
I have a private jet story about flying to Atlanta.
哦，快说说。
Oh, tell us
来，讲点什么。
something.
哦，我们懂你意思了
Oh, We got You
让我想起来一件事
reminded me.
行吧
Okay.
所以呢，那儿确实有个活动
So, yeah, there was some event there.
我就带着团队飞过去了
So I I flew my team there.
我飞机上就那么几个人
You know, there's a few people on my plane.
飞行时间还挺长的
And it's kind of a long flight.
大概四个小时左右吧
Was like four hours or something.
哦，是从海湾那边，对。
Oh, from the bay, yeah.
是的。
Yeah.
所以我到后面去睡觉了。
So I went in the back to to sleep.
一开始，你知道的，飞机刚起飞时，我带了几瓶帕皮·范温克尔威士忌。
Well, first, you know, we we started the flight and I had a few bottles of Pappy Van Winkle on on the plane.
于是我们先喝了点，然后我就去后面睡了，一觉醒来差不多就是落地的时候。
And so we started off with like a drink and then I went in the back and and fell asleep and I woke up basically when we landed.
我出来一看，那三瓶帕皮·范温克尔，全没了。
So I come out and like all three bottles are basically cashed
都是帕皮·范温克尔。
of like Pappy Van Winkle.
哎呀。
Oops.
这些瓶子才两美元一瓶。
Those are like $2 a bottle.
不是的。
No.
不是的。
No.
还有更多呢。
There are more.
这些是古董瓶子。
These are like antique bottles.
瓶子。
Bottles.
其中一个
Like one of them
是的。
was Yeah.
拿一个
Have one of
你飞机上的那种。
those from your plane.
我有一个
I have one
那种
of those
从老款猎鹰上来的。
from the old Falcon.
是啊。
Yeah.
对。
Yeah.
它们就像这种复古的
They were like these vintage
我记得你那些瓶子。
bottles of I remember you.
反正现在这东西根本找不到了。
Anyway, you can't even find this anymore.
所以这些人，我们一落地就问我：‘萨克斯，送我们去这个活动得花多少钱？’
So these guys, they asked me like when we landed, it's like, Sacks, how much should it cost for you to fly us to this event?
我就说：‘油钱大概八千，再加一万二的快乐范温克尔酒。’
And I said, well, about $8,000 in jet fuel and about $12,000 of happy Van Winkle.
飞机得加油，气氛也得伺候好。
You gotta you gotta fuel the the vibes as well as the plane.
这事儿吧。
It's
亚特兰大怎么样？
Is Atlanta nice?
我从来没真正去过。
I've never really
那些人现在还为你工作吗？还是说……
And do those people still work for you, or are they are they Yeah.
他们在亚特兰大给我打的电话。
They they called in Atlanta.
亚特兰大怎么样？
Is Atlanta nice?
听我说。
Listen.
去年我去看了底特律的比赛，那座城市正在复苏。
Last year, went to Detroit games, and that city was on the rebound.
亚特兰大其实有绝佳的机会能翻身。
Atlanta has an incredible opportunity to rebound.
我就这么说了。
I'll say it that way.
他们完全有机会把这座城市升级得更好。
There's a great opportunity for them to upgrade the city.
昨晚半夜我去了华夫饼屋。
I I went to Waffle House at midnight last night.
那晚一点枪击事件都没发生。
There was no shootings.
明白了。
Okay.
咱们继续吧。
Let's keep moving.
对了，你在亚特兰大的最佳西方酒店住能攒积分吗？
By the way, do you get loyalty points at the Best Western Atlanta or no?
我因为用的是……
I get double points because
我用的是我的最佳西方维萨卡。
I use my Best Western, Visa card.
是的。
Yeah.
哪儿都能用上。
It's everywhere you want it to be.
行吧。
Alright.
用促销码 JCAL，再送你一千积分。
Use the promo code JCAL and get a thousand extra points.
再说点 OpenAI 的新闻，马斯克和阿尔特曼那场世纪大审判，说不定是十年一遇的，开庭了。
In other OpenAI news, Musk versus Altman, the trial of the century, maybe the decade, has started.
马斯克当然在控诉 OpenAI 违反慈善信托，不当得利。
Elon is, of course, accusing OpenAI of breach of charitable trust, unjust enrichment.
他说 OpenAI 根本就是把个非营利机构偷偷改成了赚钱的公司。
He's accusing OpenAI of essentially flipping a nonprofit into a for profit.
他要求赔一百五十亿美元，还得把 OpenAI 打回非营利原形，把阿尔特曼和布罗克曼全撤了。
He's seeking a 150,000,000,000 in damages that they revert back to a nonprofit, that Altman and Brockman be removed.
马斯克和 OpenAI 的律师们当场干起来了，火药味十足。
And there were some fireworks between Elon and the OpenAI lawyers.
埃隆把这场讨论提升到了新高度。
Elon kinda leveled up the discussion.
他说：‘如果我们容许有人掏空慈善机构，那美国整个慈善捐赠的根基就完了。’
He said, quote, if we make it okay to loot a charity, the entire foundation of charitable giving in America will be destroyed.
这正是我担心的。
That's my concern.
当然，这里面有很多有意思的细节，比如格雷格·布罗克曼居然写日记，把他的计划事无巨细地记下来，简直像反派大boss一样。
Obviously, there's a ton of interesting nuances here, specifically Greg Brockman keeping a diary where he was journal maxing his plans, like a Bond villain here.
他的日记摘录里还提到一个结论：
And, the excerpts from his diary include conclusion.
我们真心想要成为B型公司。
We truly want the b corp.
但说实话，我们就是想让埃隆滚蛋。
The true answer is that we want Elon out.
如果三个月后我们真成了B型公司，那纯粹就是骗人。
If three months later, we're doing b corp, then it was a lie.
咱们要是想把这事变成盈利项目，没一场硬仗是别想的。
Can't see us turning this into a for profit without a nasty fight.
我就在想，咱们都在办公室里，这事传出去，大家肯定会说咱们对他不诚实。
I'm just thinking about the office, and we're in the office, and this story will correctly be that we weren't honest with him.
归根结底，不就是想做盈利公司，只是不想让他掺和罢了，诸如此类吧。
In the end, it's still about wanting a for profit just without him, yada yada yada.
弗里德伯格，你对这事儿怎么看？
Friedberg, your thoughts on this case.
埃隆会赢吗？
Is Elon gonna win?
我真搞不懂，格雷格·布罗克曼干嘛非得写个日记，还事无巨细地记下来？我是挺喜欢他的，但他这脑子到底在想啥？
I just don't know why Greg Brockman's got a frigging diary where he's, like, literally documenting I mean, I love the guy, but what the fuck is he thinking?
你在家闲着没事，就想着：‘来，我得把这桩罪行写下来’，‘来，我得记下来，还得留着’？
Like, you're just sitting here at home and, like, let me write about the the crime I'm committing or let me write it, like, and let me record it.
对了，还千万别删，永远留着。
And by the way, let me never delete it.
这我真搞不懂。
I don't understand this.
这可不是光写日记那么简单。
It's not just journal maxing.
这是在疯狂收集证据。
It's discovery maxing.
这完全是……
It's
证据收集到极致了。
discovery maxing.
这是在制造铁证啊。
It's a smoking gun maxing.
我还是不明白。
I don't get it.
我真的不明白啊，哥们。
I don't get it, man.
我的意思是，你们还记得《火线》里那个场景吗？有人问：你是在记录一场犯罪阴谋吗？
I mean, do you guys remember from The Wire in that scene where the guy's like, is you taking notes on a criminal conspiracy?
我完全记得，这事儿把所有人都惊到了，咱们能放一下那段视频吗？
I totally This has got everybody in the Can we play that clip?
你这是在干嘛啊，格雷格？
It's like what are you doing, Greg?
尼克，你是在记录一场犯罪阴谋吗？
Nick, is you taking notes on a criminal conspiracy?
什么
What
搞什么鬼
the fuck
你到底在想啥啊，兄弟？
is you thinking, man?
你要真想干坏事，就别把时间和日期都写下来
If you're going to commit a crime, you do not write down the date and time of
你别把犯罪的时间地点写进日记里。
the crime in your journal.
你听我说。
Well, look.
咱们还不确定这是犯罪呢。
We don't know it's a crime.
别急着下定论。
Let's not
行吧。
Okay.
当然了。
Sure.
你平时会写日记吗？
Text, do write a diary?
犯罪？嗯，算是吧。
A crime, but Well, yes.
你这事儿有点不地道啊。
You're keeping shenanigans.
查马，你写日记吗？
Chamath, do you keep a diary?
你觉得呢？
What do you think?
JKL，你写日记吗？
JKL, do you keep a diary?
我觉得，老是纠结没用。
I I believe ruminate no.
我直接跟你说吧。
I'll tell you right now.
老想着那些事，只会让自己更不开心。
Rumination is the path to unhappiness.
没人真在乎你那些情绪。
Nobody gives a about your feelings.
把情绪憋在心里只会让你更痛苦，跟配偶聊聊你的感受吧。
Hiding your feelings down is only gonna make you miserable talking to your spouse about your feelings.
去吃顿漂亮的晚餐，下次比赛坐到前排去，就像我这三十年来一直做的那样。
Just go to a beautiful dinner, sit courtside at the next, and do what I've been doing for thirty years.
疯狂试错。
Retard Maxing.
疯狂试错。
Retard Maxing.
疯狂试错。
Retard Maxing.
随着你不断尝试，只要埋头做事，启动新项目，十次有九次会失败，下注十次有九次会输，但只要有一次赢了，你就赢了。
And as a register goes up, all you have to do is work, Start new projects, nine out of 10 foul, place nine out of 10 bets, one wins, and you're golden.
下次比赛就去前排坐着。
Go sit courtside at the next game.
去吧。
Going.
人生太短了。
Life's too short.
就一直往前走吧。
And just keep moving forward.
别写任何东西，别写。
Don't write anything down, period.
就这么定了。
Full stop.
这建议真不错。
That's good advice.
对。
Yeah.
最让我惊讶的是，这家伙居然写日记。
I just the the biggest surprise to me was this guy's got a diary.
我真的不知道还有谁会写日记。
I just I don't know anyone that has a diary.
我从来没听说过这种事。
I've never heard of this.
总之，这事儿真够让人震惊的。
So, anyway, that was shocking.
至于这案子以后会怎样、法官会怎么判，我真没什么看法。
Besides that, I've I've no view on what's gonna happen with the case or what the judge will do.
我对此案也没啥可说的。
I have no comment on the case either.
我觉得挺奇怪的，这么多证据都公布了，Polymarket居然一点动静都没有。
I think it's weird that Polymarket hasn't budged even as all of this discovery has been published.
现在押伊隆赢的赔率还稳在42%到43%之间。
It's effectively at 42 or 43% that Elon wins.
我们群里的一个朋友说，搞不好伊隆名义上赢了，但最后只是把四千万美元还回去而已。
So one of the friends in our group chat said what may just happen is that Elon technically wins, and he's just credited back the $40,000,000.
也许这个盘口就是在赌这件事儿。
And so maybe that's what this poll is front running.
不过说点别的，杰森，我知道你这话是开玩笑，但‘别总纠结，一直往前走’这个道理，其实对每个人来说都是很好的生活建议。
But on a totally separate note, I think, Jason, I know you say it as a joke, but this idea of just keep moving forward, don't ruminate, I think is very good general life advice for everybody to follow.
我觉得现在的心理咨询产业和药物产业，
The modern day therapy industrial complex and the medication industrial complex, I believe,
本质上就是围绕着反复琢磨这件事运转的。
is It pivots around rumination.
没错，确实就是围着反复琢磨转的。
Well, it does pivot around rumination.
对。
Yes.
这可是通往所有这些问题的敲门砖。
That is the gateway drug to all these things.
没错。
Yep.
老是谈你的烦恼。
Talk about your problems.
你有没有听过那些去接受心理治疗的人？
You know, when these people go to therapy, you ever hear these people?
霍华德·斯特恩说，他每周找同一个治疗师聊两三天，一聊就是四十年。
Howard Stern's like, I've been in therapy with the same person two or three days a week for forty years.
我听了就只能说：好吧。
I'm like, okay.
那治疗师有什么理由停止收你每小时一千二百美元的费用？
What's the incentive for the therapist to stop charging you $1,200 an hour?
根本没理由。
There is none.
他们一停，收入就断了。
Then they lose a revenue stream.
客户也跑了。
They lose a customer.
这整个就是一场巨大的骗局。
It's all a giant fucking fraud.
萨克斯，就这个案例来说
Sacks, in terms of this case
我还不至于这么说。
I wouldn't go that far.
我觉得，把人因为成长经历而打结的心理死结解开，确实很有价值。
I do think that there's a lot of value in kind of untying some of these Gordian knots that people have because of how they grew up.
但那和具体分析、有针对性地梳理不一样，不是随便瞎琢磨，因为我觉得——
But there's a difference between that and being specific and just randomly ruminating because I
你要是真有急性创伤问题，这种漫无目的的反思没啥用。
don't think there's a lot of productive You've got an acute issue, like, in trauma in your life?
是的。
Yeah.
当然。
Sure.
好好理一理。
Unpack it.
自己想清楚。
Figure it out.
我就是说这种没完没了的自我提升、老是反复琢磨的事儿。
I'm just talking about this never ending self improvement, you know, ruminating thing.
咱们还是回到正题吧，Sacks，我们刚才说的是奥克兰的一个陪审团，对吧？
But getting back on topic here, Sacks, what's the and we're we're talking about a jury, I believe, in Oakland.
不对。
No.
但这是法官独任审判。
But it's a bench trial.
这点很重要。
This is important.
这是陪审团只提供参考意见的法官独任审判。
It's a bench trial where the jury is advisory in capacity.
但最终，还是由那位法官来做决定。
But, ultimately, that judge, she will make the final call.
赔偿金额也是她说了算。
And she'll do the damages.
所以这是不是意味着，咱们有个湾区的法官陪审团，还有埃隆——大家都知道他有点右翼，但在他的政治立场上，人们看法不一。
And so is this a Capex of, like, we've got a Bay Area jury judge, and we've got Elon who's considered, you know, a bit right wing and people don't all agree in that area in terms of his politics.
然后还有萨姆·阿尔特曼那篇《纽约客》的报道，结果发现好多人觉得被他坑了。
And then you have this Sam Altman New Yorker story and people finding out that so many different people feel he they got screwed by him.
把这两件事放一块儿，根本没法猜最后会怎么收场。
You put these two things together, it's impossible to handicap where this turns out.
萨克斯，你怎么看？
Sacks, your thoughts?
嗯，我觉得这事跟政治没关系。
Well, yeah, I don't think this is about politics.
我的意思是，埃隆想保护慈善机构，这反而更像是一种左翼立场。
I mean, I guess you could argue that what Elon is seeking, which is to protect the charity, is if anything a left coated sort of principle.
没错。
Sure.
不过我觉得这根本不是左右派的问题。
Although I don't really think it's left versus right.
听我说，我不想在这场官司里站队。
Look, I don't want to take sides on this trial.
我就跟大家一样，纯粹看看热闹。
I'm just watching like everyone else.
上次我插嘴谈过一回埃隆的官司，结果被传去录了六小时口供。
The last time I weighed in on some Elon litigation, I got deposed for six hours.
你还记得吗？
Remember that?
他们就是觉得我肯定知道点什么。
Because they just assumed that somehow I know something.
对啊。
Right.
我从来没跟埃隆聊过这个案子。
I've never talked to Elon about the case.
我对这事一无所知。
I don't know anything about it.
是的。
Yeah.
我就跟大家一样，等着看结果吧。
I'm gonna see what happens like everyone else.
我刚看了些相关报道，有一件事我想说：据说这家公司曾经想给马斯克一些股份，但他觉得这事儿有点别扭。
Now, one thing I I will say, having just read some of the coverage, is that apparently the company at some point did offer Elon shares in the company, but he thought that there was something kind of icky about it.
你记得这事吗？
Do you remember this?
对。
The Yes.
当初这场纠纷刚冒头、还没闹上法庭的时候，我在节目里说过：如果OpenAI那时候觉得自己的结构有问题，干脆就该主动找马斯克谈，让他成为股东，上资本结构表。
Because at one point, I said on our show, when this dispute started happening before it became a court case, I said, look, if OpenAI at a certain point decided they had the wrong structure, they should have just gone and done a make right with Elon and he should have been a shareholder on the cap table.
我没想到的是，他们其实真试过这么干，但马斯克拒绝了，因为他希望这个机构能继续保持公益性质。
What I didn't know is that apparently they did try to do something like that, but Elon turned it down because he did want the entity to remain a charitable entity.
你明白我的意思吧？
In other see what I mean?
根据报道，他对此有坚定的立场，直接拒绝了。
Had a principled view of it according to the reports and was like, no.
我们可是想拯救人类，你倒好，把王国的钥匙交给了微软。
We're trying to save humanity, and then you're giving this keys to the kingdom to Microsoft.
这些事儿现在都曝光了。
That's all come out.
我也没跟埃隆聊过这些事。
And I also have not talked to Elon about any of this.
但我猜，像这种事最后多半会和解，不过他也可能真跟他们死磕到底。
But my guess is, like most of these things, there'll be some sort of settlement or something here, but maybe he takes it to the mat.
谁知道呢？
Who knows?
法官罗杰斯，这位61岁、由奥巴马任命的法官，政治因素显然起了作用，萨克斯。
Judge Rogers, who's doing this 61 year old Obama appointee, politics has played a role, Sacks.
他们得告诉陪审团，不管你们对这些人有什么政治上的看法，都请先放一边。
They have had to tell the jury, like, however you feel about these individuals, politically, whatever, please put that aside.
值得注意的是，她之前主持过Epic Games诉苹果的案子，当时法院裁定苹果不构成垄断，但也有一些限制条件。
But of note is that she oversaw the Epic Games versus Apple trial over apps were exclusively ruled in favor of Apple with some caveats, that they don't have a monopoly, etcetera, etcetera.
所以这次案子肯定会特别有意思。
So this is gonna be a really interesting one.
我觉得最坏的情况是，OpenAI得想办法拆解现在的结构，那肯定会拖慢上市进程。
I think the worst case scenario is OpenAI for for OpenAI is they have to unravel this somehow, and that would delay the IPO.
那样一来，股东们就乱套了。
That would cause chaos in shareholders.
而最好的情况呢，就是双方能达成和解。
And I guess the best case is some sort of settlement.
如果埃隆真投了四五千万美元，那在股份稀释后，他应该能拿到百分之十、二十，甚至三十的股份。
And if Elon put the first 40 or $50,000,000 in, he's he's due ten, twenty, 30% of the company after dilution.
行了。
Alright.
咱们继续往下看案情进展吧。
Let's keep moving through the docket.
还有很多事要聊，祝大家打官司顺利，也祝在Polymarket上押注的各位好运。
Lots more to discuss, and, good luck to everybody in their lawsuit and those of you betting on Polymarket.
All in Summit这边卖得可真快。
All in Summit, selling up fast.
我们第五期洛杉矶站，九月份见。
Our fifth edition Los Angeles, September.
去allin.com/events看看，嘉宾阵容绝对顶配。
Go to allin.com/events, and speakers are gonna be top tier.
听说Friedberg现在把这当成了他主要的创作出口。
Apparently, Friedberg is having this as his major creative outlet.
我今天听到了一些内部消息，说Chamath打算去演百老汇的魔术剧。
I heard some back channel Chamath today that he's gonna be doing Broadway musical illusionists.
我这儿倒是有个踢踏舞的安排。
I got a tap dancing situation.
他这是要彻底变身成艺人了。
He's literally going full on entertainer.
这简直要变成杂耍剧场了，萨克斯都得收场。
This is gonna be vaudeville, Sacks wrapped up.
他打算把这事玩到一个全新的高度。
He's just gonna take it to a whole new level.
来点歌舞表演。
Musical numbers.
就像内森·莱恩那样？
Like Nathan Lane?
如果你在写代码，那这肯定就是他的大型酷儿峰会了，没错。
I think if you're coding that it's gonna be his big gay summit, yes.
这还真可能是个大型酷儿峰会。
It could be a big gay summit.
各位，这可能是我们在洛杉矶的最后一年了。
Might be our last year in LA, guys.
哇哦。
Wow.
有可能。
Might be.
他的。
His.
可能是他的。
Might be his.
哦，大家都想
Oh, everybody wants
去拉斯维加斯吧。
to go to Vegas, apparently.
他拿到了玫瑰。
He got rose.
骨头在里面。
Bone in
他的骨头，宝贝。
his bones, baby.
你能想象吗？从峰会上溜出去吃午饭，跑去玩骰子，查马斯？
Can you imagine leaving the summit for lunch and going and playing craps, Chamath?
洗个痛快澡
Get a fresh shower
就在那儿。
in there.
当然可以。
Yes.
我能想象。
I can.
当然可以。
Yes.
我能想象。
I can imagine.
我弄到一些砖头了。
I got some bricks.
哎，我就在这儿弄到一些砖头。
Oh, I got some bricks right here.
走吧。
Let's go.
真香。
Yum.
真香。
Yum.
这样Sacks就能来了。
That way Sacks can come.
对啊。
Yeah.
Sacks说他要
Sacks is like, I'm
我从来没踏足过加州，但这次我一定去。
never setting foot in California, but I will.
所以你来参加我们的活动吧，我们搞了几场现场活动。
So you come to our You know, we're doing a couple live events.
你会来吗？
Are you coming to them?
流动性或者波动性，有啥区别？
Liquidity or swing different?
流动性是一方面，还有九月份的全栈峰会。
Liquidity and then there's the the all in summit happens in September.
对。
Yeah.
我也会去参加那些活动。
I'm gonna do those too.
好嘞。
Alright.
周四，科技巨头们的财报全线爆表。
Big tech smashed their earnings on Thursday.
谷歌、微软、亚马逊和Meta全都发布了业绩。
Google, Microsoft, Amazon, and Meta all reported.
我不明白他们为啥总挑同一晚发，但就是这么干了，结果简直惊艳。
I don't know why they do this on the same night, folks, but they do, and performance was spectacular.
表现太棒了。
It was great.
不过，真正值得关注的其实是他们的资本支出计划。
However, the Capex announcements were really the story here.
我先调出图表给你们看。
Let me just queue this up and show the chart.
仅这四家公司——亚马逊、微软、谷歌和Meta——就计划在2026年投入7250亿美元的资本支出。
$725,000,000,000 in Capex guidance in 2026 from but four companies, Amazon, Microsoft, Google, and Meta.
亚马逊一马当先，砸了2000亿；微软和谷歌各1900亿；Meta也投了1450亿。
Amazon leading the pack with 200,000,000,000, 190,000,000,000 each for Microsoft and Google, 145,000,000,000 for Meta.
你再把Grok、OpenAI，还有其他一些玩家加进来，苹果新CEO还没发声，但接任后肯定也有大动作，我敢肯定。
You add Grok, you add OpenAI, and some other players to these plans, and we haven't heard from the new Apple CEO yet, but he's gonna be taking over, and he's gonna have some plans here, I'm sure.
明年咱们就要看到上万亿美元的投入了。
We are going to see the a trillion dollars a trillion dollars in build out over the next year.
这事儿真能成吗？我也说不准，但这一切全靠AI和云计算在推着走。
I don't know if this is even possible, but this is all being driven by AI and cloud computing.
谷歌云，包括谷歌套件，去年同比增长了63%。
Google Cloud, which includes the Google Suite, that grew 63% year on year.
你品，你细品。
Let that number sink in.
200亿美元的收入，涨了63%。
63% on 20,000,000,000 in revenue.
就一个季度的事儿。
That's in a quarter.
微软云，包括Azure、Windows Server、SQL Server这些。
Microsoft Cloud, that includes Azure, Windows Server, SQL Server.
他们把几项业务打包在一起，好让数字看起来更高。
They bundled some things together there to get the number to go up.
这347亿美元的收入，增长了30%。
That grew 30% on 34,700,000,000 in revenue.
亚马逊云服务，最早的云平台，收入376亿美元，增长了28%。
Amazon Web Services, the original cloud, that grew 28% on 37,600,000,000 in revenue.
这个算是比较纯粹的云业务了。
That's a bit of a pure play.
就单算亚马逊的云服务部分。
Just counts Amazon's web services.
显然，这些都在往新型云平台转型。
Obviously, these are all moving to neo clouds.
现在它们全都在跑AI任务和处理token。
These are all serving AI jobs and tokens now.
它们客户基数庞大，从最小的初创公司到最前沿的大模型，谁都嫌算力不够，这直接拉高了利润。
They have a massive customer base, and the customers from the smallest startups all the way to the biggest frontier models cannot get enough compute, and it is going to the bottom line.
但这让查马斯的现金流大幅缩水。
But this is shrinking Chamath cash flow massively.
以前这些公司可是印钞机，人类历史上最厉害的自由现金流机器，现在却放弃自由现金流、股票回购和分红，转而全力投入基础设施建设。
These were free cash flow machines, the largest money printing machines in the history of humanity, but they are giving up on free cash flow, stock buybacks, and dividends, and the focus on those three to invest in infrastructure.
亚马逊的自由现金流暴跌了97%。
Amazon's free cash flow down 97.
谷歌、微软和Meta的自由现金流分别下降了12%、12%和8%。
Google, Microsoft, and Meta down twelve, twelve, and 8% respectively.
查马斯，你怎么看这种自由现金流的枯竭？还有我们现在看到的巨额资本开支投入？
Your thoughts on this free cash flow, the end of the free cash flow deluge, and the massive, massive investment we're seeing in Capex, Chamath?
我觉得资本市场正在经历一场非常重要的结构性转变。
I think we're seeing a very important structural shift in the capital markets.
过去二十年，那七大科技巨头一路狂奔，越做越大，把所有投资资金都吸走了。
I think the last twenty or thirty years, well, twenty years, it's been that the Mag-seven just kind of ran away with it, that these big companies got bigger and bigger and it absorbed all of these investment dollars.
最主要的原因是，它们的商业模式都特别轻资产。
And the biggest reason was that it had these very asset light business models.
对吧？
Right?
你刚开发了点新软件，它自带巨大杠杆，一切顺风顺水，就亚马逊例外，人家得搞实体仓库和物流，搞不了轻资产。
You just built some more software, and it just has all this leverage, and it all just kind of worked, except maybe for Amazon because they needed physical infrastructure for warehouses and delivery and whatnot.
但总的来说，那会儿全是轻资产投资的天下。
But by and large, it was a very asset light investment cycle.
现在可好，天平猛地往另一边甩过去了。
Now all of a sudden, the pendulum is swinging violently in the other direction.
我觉得很多人搞错了一点：当投资重新转向重资产基建时，这些超大规模云服务商开的支票，表面上看他们能兑付，但其实也存在兑不了的可能。
And there's something that I think people misunderstand, which is as it moves back to these asset heavy infrastructure investments, The hyperscalers are signing checks that, I mean, I suspect their body can cash, but there's a world in which they can't.
我举个例子。
I'll give you an example.
你记得微软当初说动了三哩岛的业主，让他们重启核电站吗？
You know, when Microsoft convinced the owners of 3 Mile Island to turn their nuclear sight back on?
是啊。
Yeah.
你知道他们签的长期购电协议是什么价吗？
Do you know what their forward purchase agreement was?
比当前的现货电价高出两倍多。
It was for more than two x the prevailing spot rate for energy.
高出两倍不止。
More than two x.
问题是，这还占不了他们总用电量的很大一部分。
The problem is that's not for an enormous percentage of their overall energy needs.
你要是把这事儿推演下去，就会发现这五六个巨头，杰森，他们每年的资本开支已经高达七千亿了，更别说运营现金流上，他们还得按两倍于现货的价格买电，就为了确保未来有稳定的电力供应。
So if you play that out and you think these five or six companies all of a sudden are not just spending, Jason, 700,000,000,000 a year of Capex, which they are, but then from an operating cash flow, they're gonna be spending two x the prevailing spot rate because they just want guaranteed demand into the future.
这么多钱，到底会流向哪儿？
Where's all this cash gonna go?
不会给股东分红，也不会留在资产负债表上。
It's not gonna go to the shareholder, and it's not gonna stay on the balance sheet.
这些公司现在要开始加杠杆了。
These companies will now get levered.
他们会把金融操作玩得越来越精妙。
They're gonna get highly sophisticated around the financial engineering.
债务会越来越多。
They'll have more debt.
各种融资工具、定期贷款、循环信贷全都会用上。
They'll have all kinds of different vehicles and term loans and revolvers and all of this stuff.
五年后，他们看起来就会像庞大的传统工业巨头。
And so they're gonna look like this big bulky industrial business in five years.
那时候，真不一定能说清楚他们的估值到底该咋算。
And I'm not sure that there's a good valuation case to be made at that point.
所以我觉得，不如简单点——就像我推文说的，盯着钱走就行了。
And so I think it may be simpler, and this is what I tweeted, to just follow the dollars.
每年一千亿美元从这些超大规模云服务商手里流出去。
Like, a trillion dollars a year going out of the hyperscalers.
这些钱最终都流向了哪儿？
Where is it going?
你就跟着钱走，买这些公司，它们现在股价明显被低估了。
Just follow those dollars and buy those companies because those companies are already underpriced.
这明显让我们想起了当年大家都经历过的事。
This is, obviously reminiscent of something we all experienced.
尼克，能把你刚发我的思科图表调出来，放大到最大吗？
Nick, can you pull up the Cisco chart I just sent you and put it at max?
九十年代末到2000年，互联网基础设施经历了大规模扩张。
We had a massive build out of the infrastructure of the Internet in the late nineteen nineties and into 2000.
这导致很多激进的公司疯狂砸钱，散户也像现在追捧私企一样，疯狂追捧这些股票。
And what that caused was a lot of aggressive companies to do massive amounts of spending, a lot of retail investors to embrace these stocks like we're seeing with people trying to get into these private companies.
萨克斯，你看看思科2000年的高点。
And, Sacks, look at the 2000 peak of Cisco.
这简直是史上最夸张的走势。
This is the most extraordinary chart ever.
它花了整整二十五年才回到那个高点，整整二十年都原地踏步。
It took them twenty five years to get back to that peak, and they had a lost two decades.
当时我们买了巨量的光纤。
And we had a massive amount of fiber that wound up getting bought.
几年前我们在节目里还聊过这事。
We talked about that a couple years ago on the program.
但你看这庞大的基础设施，里面其实有东西可以让你接着往下琢磨。
But there's something for you to build off of here when you look at this massive infrastructure.
你觉得这会重演Cisco、WorldCom那些老故事吗？
You think it's gonna be Cisco Systems all over again, WorldCom, etcetera?
不会。
No.
我真不觉得会。
I really don't.
2000年那会儿的问题是，光纤铺了一大堆，却没人用。
The issue we had in 2000 was dark fiber.
当时建了这么多基础设施，结果全闲置了。
You had all this infrastructure being built out and it wasn't being used.
现在根本不存在闲置的GPU，就像布拉德·格斯特纳常说的那样。
There's no dark GPUs today as, you know, Brad Gerstner likes to say.
所以现在推动资本支出的，是对算力和令牌的惊人需求。
So what's driving the Capex now is the voracious demand for compute, for tokens.
这种需求正在提前拉动对基础设施的额外投资。
And the demand is now pulling forward this additional investment in infrastructure.
我觉得，AI的看涨逻辑就在一个下午被彻底验证了。
So I think what's happened here is that the bull thesis for AI just got validated in a single afternoon.
我的意思是，微软Azure、谷歌云、亚马逊AWS、Meta，它们的云收入全都超出了预期，超出了指引，自然也就加大了资本支出的投入。
I mean, again, you got Microsoft Azure, Google Cloud, Amazon AWS, Meta, they're all basically exceeding expectations, exceeding guidance in terms of where their cloud revenue would be, and therefore how much they're gonna reinvest in Capex.
今年我们原本预计超大规模云服务商的资本支出是6600亿美元，去年是3500亿。
This year, I think we were supposed to have $660,000,000,000 of hyperscaler Capex up from $3.50 last year.
现在新的预测是，会超过7000亿。
I think there's now the new estimate is is gonna be over 700.
这也就是说，占到了GDP的2%以上。
So this is, you know, again, it's more than 2% of GDP.
这对GDP来说是个巨大的推动力。
This is a huge tailwind to GDP.
还有篇文章说，上个季度AI贡献了GDP增长的75%。
There's another article saying that I think in the last quarter AI was 75% of GDP growth.
顺便说一句，这还只是资本支出的部分。
And by the way, this is just the Capex part.
这说的是实体基础设施的建设。
This is the physical infrastructure.
还不是那些在‘算力工厂’里生成的token带来的经济影响。
This is not the economic impact of the tokens that are generated inside the token factory.
这纯粹是建工厂的钱。
This is the building of the factories.
那这些token到底怎么用的？
How do those tokens get used?
我们现在看到的，它们不只是用来搞研究或回答问题，还能写代码。
Like we're seeing, they're being used not just to do research or to answer questions, but to create code.
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。