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。
利奥波德·阿申布伦纳的态势感知对冲基金爆雷，清仓了全部股票组合。
Leopold Aschenbrenner's Situational Awareness Hedge Fund blows up and sells off its stock portfolio.
OpenAI 将最新模型的价格最高下调了 80%，而科技巨头的财报让萨提亚·纳德拉高兴得不得了。
OpenAI has cut prices on its latest models by as much as 80%, and big tech earnings leave Satya Nadella a very happy man.
接下来，我们会连线 Semafor 的里德·阿尔伯戈蒂，为大家带来最新动态。
That's coming up with Reed Albergotti from Semafor right after this.
面对持续的变革与机遇，TMT 领导者需要拿出实实在在的成果，而不仅仅是想法。
In the face of ongoing disruption and opportunity, TMT leaders need to deliver tangible results, not just ideas.
当速度和执行力变得至关重要时，普华永道结合市场洞察、行业深度经验，以及人工智能、云计算和新兴技术，加速您的转型，从战略到落地实现可衡量的投资回报。
When pace and performance matter most, PwC combines market insights and deep sector experience with AI, cloud, and emerging tech to accelerate your transformation and drive measurable ROI from strategy to execution.
普华永道能帮您预见未来，超越颠覆，赢得竞争。
PwC can help you anticipate what's next, outpace disruption, and compete.
更多信息，请访问 pwc.com。
For more information, visit pwc.com.
欢迎收听《大科技》播客，这里探讨科技世界及其更广阔领域的理性与深度对话。
Welcome to Big Technology Podcast, a show for cool headed and nuanced conversation of the tech world and beyond.
今天给大家准备了一档很棒的节目。
We have a great show for you today.
我们要好好聊聊利奥波德·阿申布伦纳的“情境意识”对冲基金崩盘的事，不过它还在挣扎求生。
We're gonna talk all about the implosion, but it's still kicking, of Leopold Aschenbrenner's Situational Awareness Hedge Fund.
看来问题出在基金规模太大，而套期保值做得不够。
Too much fund, too little hedge seems like to be the problem there.
我们还会讨论OpenAI下调最新模型价格的事情。
We'll also talk about OpenAI cutting prices on its latest models.
有的降幅高达80%。
Some of them by as much as 80%.
这说明了AI价格战什么情况？
What does it say about the AI price war?
当然，现在是大科技公司财报周，我们会梳理一下主要看点，特别是从大科技公司的表现中，看看跟AI投资热潮有什么关系。
And, of course, it's big tech earnings week, so we'll just go through the big takeaways, from what we saw from big tech, especially as it relates to the AI trade.
今天做客的是来自Semafor的里德·阿尔贝戈蒂，他可是我们的老朋友了。
So joining us today is Reed Albergotti, returning champion from Semafor.
Reed，很高兴见到你。
Reed, great to see you.
欢迎回来。
Welcome back.
能来这儿真的很棒。
It's awesome to be here.
被说成是回归冠军，我倒真不知道这还是个比赛。
Glad to be introduced as returning champion, but I I I didn't know this was a a contest.
所以
So
确实是。
It is.
没错。
It is.
当然了。
Definitely.
关键绩效指标？
KPIs?
我怎么才能赢？
How do I win?
我怎么才能保持领先？
How do I stay on top?
我觉得就开心地玩就行。
I think just have a good time.
做你平时做的事。
Do what you do.
这个没什么好担心的。
There's no worries on this one.
所以，里德，不知道你有没有关注过那个‘情境意识’基金，就是那个由莱奥波德·阿申布伦纳搞的耳机对冲基金，他以前是OpenAI和FTX的员工，那应该是全球表现最好的对冲基金之一。
So, Reed, I don't know if you've been following the blow up of situational awareness, which has been this, this headphone, this headphone, this headphone hedge fund from Leopold Aschenbrenner, who's a former OpenAI employee, a former FTX employee, It was sort of the highest performing hedge fund in the world, I think.
我的意思是，它从几百万美元的管理资金，一路涨到了两百亿美元的资产管理规模。
I mean, it went from millions of dollars in funds to $20,000,000,000 in assets under management.
然后它实际上，你知道的，我觉得‘爆仓’这个词可能有点重，但这周它确实崩盘了，不得不抛售。
And then it effectively, you know, I think blow up is maybe a little too strong, but it effectively blew up this week and it had to sell.
它把股票投资组合里很大一部分，甚至全部，都卖给了由肯·格里芬管理的 Citadel。
It's a large part of its stock portfolio or all of its stock portfolio to Citadel run by Ken Griffin.
我当时就在想，这话题会不会太冷门了，不适合拿来开场？
I I was like, I was thinking, is this too niche to start the show?
但这个故事实在太吸引我了，我根本等不到节目后半段再讲。
But I I it's just one of those stories that I'm just too fascinated by to let wait until the second half.
所以很好奇你对这里发生的事有什么看法。
So curious what your perspective is on what's happened here.
显然，他还是会保留自己资产的一部分，包括在 Anthropic 的大额持股。
Obviously, he'll still continue with the percentage of his assets that he's kept, including a large stake in anthropic.
但我相信这里面肯定有不少值得总结的经验教训。
But I'm sure there's some big lessons to draw here.
我也很想知道你从中得出了哪些结论。
And I'm curious which ones you've drawn.
是啊，从表面上看，这件事根本就不是什么科技故事。
Yeah, in a way, like if you just look at what happened on its face, it is just not a technology story at all.
对吧？
Right?
就是一个 guy 杠杆加得太猛，市场押注出问题，没法覆盖短期亏损，只能被迫平仓。
It's a guy who got over levered, you know, with his with his, you know, market bets and had to sell his position because, you know, he couldn't he couldn't cover the the short term losses.
对吧？
Right?
其实要是看他的投资组合，那基金表现可能还不错。
He's probably actually, the fund was actually doing well if you look at like his bets, right?
但你要是杠杆太高，根本扛不住这种短期波动。
Like, but, you know, you you can't you can't you can't handle these short term dips if you're if you're over levered.
所以呢，谁在乎呢？
So it's like, oh, who cares?
这跟科技有什么关系？
Like, why is this important for technology?
我觉得最有趣的是，你观察一下大家的反应。
I what what but what I think is so fascinating about it is that it's it you watch the reaction to this.
当然，很多人是幸灾乐祸。
And, like, of course, there's a lot of schadenfreude.
你知道的，这人之前可是天之骄子，《华尔街日报》还专门给他做过专访，而且他还是个有效利他主义者。
You know, this guy was like the golden boy, you know, the Wall Street Journal was profiling him and, you know, he's this like, you know, EA effective altruist person.
我觉得有些人看到他落难还挺开心的，但这背后反映出的问题是，现在科技行业内部存在巨大的分歧。我认为主要是传统派科技人士和那些自称为加速主义者的人之间的对立，后者只关心新技术的开发和应用。
And I think like, people sort of I think there's some joy in him going down, but what it but what it it like from some people, but what it gets at is like, there is such a divide in the tech industry right now between I think they're the sort of more traditional tech people who, you know, some might call themselves accelerationists, they just want to see, you know, this new technology being built.
然后这种说法，我甚至都不叫它有效利他主义了。首先，自从FTX崩盘之后，我就再也没听人提过这个词了，大家也不再这么自称了。
And then this kind of like, I don't even call it EA because I first of all, like since FTX blew up, like, I don't even hear EA, like people don't describe themselves that way anymore.
但确实还残留着一些有效利他主义文化的痕迹，这些人跟传统的科技圈人士不太一样。
But there is this sort of remnants of the culture of EA of these people who are just not like your traditional tech people.
哪怕你退一步想想，科技圈里没人会吹嘘这个。我在硅谷这么多年，从来没人跟我炫耀过他们在公开市场的股票交易成绩。
And if you even if you just step back and think about it, like, no one in tech would would brag like no one in Silicon Valley, I don't know about you, I've never had anyone brag to me about their public market stock trading.
在硅谷，大家根本不会把这种事当成值得炫耀的资本。
Like, that is, like, not what people value in Silicon Valley.
他们看重的是创造。
They value building.
所以对我来说，这恰恰凸显了这种新的文化分歧，我特别感兴趣。
And so in a way, it's like this to me, it just highlights this this sort of new cultural divide, which I'm fascinated by.
我真的不知道这会走向哪里。
Like, I just I don't know where where it's gonna go.
它可能会消失，也可能演变成更大的裂痕，本质上现在有两个不同的科技圈。
It could disappear or, you know, it could turn into an even bigger rift and there's like two there's two tech industries essentially.
明白了。
Okay.
你这话有点过头了。
This taking that too far.
我觉得你并没有。
You know, I don't think you are.
这个话题特别值得深挖。
And this is a very interesting thread to pull.
我觉得我们可以先聊聊这个话题，然后再稍微说说利奥波德手里拿的到底是什么，因为这本身也挺有意思的。
And I think we should talk about it for a moment and then talk a little bit about what Leopold was actually holding because that's pretty interesting in and of its own right.
不过有效利他主义这个概念挺有意思的，因为它有几个特点，我在这里就不展开细说了。
But the EA thing is interesting because there's a couple things that sort of characterize effective altruism and I'm not going to do it justice here.
但这个运动的一个显著特点就是：尽量多赚钱，然后尽可能多做善事。
But one of the things that characterizes the movement is sort of you try to make as much money as you can and then do as much good as you can.
所以以前的利他主义者可能直接说我要去给有需要的人做志愿者，而有效利他主义者可能会说，我先找份最赚钱的工作，攒够了钱再捐出去。
And so whereas previous altruists might have just been like, I'm going to go volunteer for people in need, effective altruists might say, I'll take the most capitalist job I could find and then eventually donate.
这就跟像GiveDirectly这样的运动有关，他们基本上是不附带任何条件的捐赠，顺便说一句，我觉得这是个很棒的计划，向世界各地的地方提供援助。
And that's associated with movements like GiveDirectly where they give basically no strings attached donations to and I think that's a great program by the way, to places all over the world.
所以另一方面，我就说这一件其他的事情吧，因为它结合的方式非常有趣，就是有一种很大的信念叫做期望值，我不认为这是他们独有的，但基本上就像……只想举个例子，这是经典的例子，对吧？
So the other side of it, I'll just say this one other thing because it's very interesting how it combines, is there's a big belief in this formula that's called expected value, and I don't think it's unique to them, but it's effectively like, want Just to give one example, this is the canonical example, right?
如果你能抛硬币，49%的概率是世界毁灭，51%的概率是创造一个乌托邦，他们会每次都抛硬币，因为乌托邦的期望值高于灭绝的期望值。
If you could flip a coin and 49% is you blow up the world and 51% is you create a utopia, they would flip the coin every time because the expected value of utopia is higher than the expected value of Annihilation.
就像，哦，如果你把五十一和四十九加起来除以二之类的，那还是五五开。
It's like, oh, if you add fifty one and forty nine up and divide by two or whatever, that's still fiftyfifty.
从数学上看，押注乌托邦显然更有价值。
The math makes sense that the better value is going to be on the Utopia bet.
这是最近几十年里，第二次出现来自这种背景的人——另一个是萨姆·班克曼-弗里德——他们下的注大到让人不禁怀疑：这种思维方式，如果用在经营企业上，是不是太危险了？
And so this is the second time in in recent history, where someone who comes from that sort of background, and the other one being Sam Bankman Fried seems to have made, you know, big enough bets that make you think is this is this a pretty, disastrous way to, to think if you're running a business.
据我们所知，这里并没有欺诈行为。
Now, no fraud here, as far as we know.
他甚至没亏钱，虽然不是完全一样的崩盘，但事情的发展轨迹确实有点相似。
He's not even negative, but as a a, you know, not a parallel blow up, but it's it rhymes in a way.
是的。
Yeah.
从长远来看，萨姆·班克曼-弗里德也没亏。
Sam Bankman Fried wasn't negative either in the long run.
他可能是有史以来最成功的投资者。
Was the best investor of all time.
有意思。
Mean
要是没逼他抛售的话。
If they didn't make him sell.
这就像另一回事了，很多人去了Jane Street Capital工作后，看世界和市场的态度就像觉得这事儿跟小孩过家家一样简单。
This is the this is like the other part of this is like a lot of these people go work at Jane Street Capital, and they they look at the world and the markets as this like, oh, it's like child's play.
他们总觉得市场就是个算法，只要破解了就能搞定，这种看法既不带感情色彩又过于简化现实。
Like, it's this algorithm that you can just crack and like, you know, it's sort of like this unemotional view of it's and it's like a simplistic view of the world.
结果最后就炸在自己脸上了。
And then it blows up in their faces.
因为如果你用这种非常简单的逻辑来看待这些问题，感觉就像是在大学宿舍里聊哲学一样。我是说，大家都想要人工智能嘛。
Because if you look at this stuff in this very simplistic way, it's like, well, you know, the, the, it's like you're sitting in your college dorm room talking about philosophy, you know, well, I mean, everyone wants AI.
当然，你也知道，要建设这些AI数据中心，肯定得购买相关的硬件和产品。
And of course, like, you know, these things that these, these products will have to be purchased to build these AI data centers.
所以，你是不是只要把所有钱都投进这些股票里就万事大吉了？完全不用去管市场上那些跟这些东西长期实际价值毫无关系的波动。
Therefore, could you just put all your money in those stocks, and you're going be fine without like, you know, really paying attention to how like there's, there's fluctuations in the market that have nothing to do with like the actual, you know, long term value of these of these things.
你知道的，FTX的情况差不多就是这样发生的。
And, you know, I mean, that this is sort of what happened to FTX.
而且我觉得，这简直是一种傲慢到极点的态度。
And yeah, I just think I think that's like, it's almost like this hubris.
我觉得这种态度特别让人反感，对吧？
And I think that's like a real turnoff to people, right?
但事实并非如此。
And it's not.
还有，很多人投身人工智能，是因为他们觉得，这是自己能做出最大贡献的领域，毕竟这项技术有潜在危险。
And the other thing is like, that a lot of these people went into AI, because they were like, well, this is where I can do the most good because this is a dangerous technology.
所以他们觉得自己有责任去引导它、确保它被妥善管理，而这本身也是一种傲慢的看法。
And so I have to go into this to kind of help steer it and make sure that it's, you know, that it's properly stewarded, which is also like a kind of hubristic way of looking at it.
这可不是硅谷的思维方式，对吧？
And it's not a Silicon Valley way of looking at it, right?
这跟传统上对技术的认知完全不一样，对吧？
That's not like the traditional way of, of thinking about technology, right?
技术本来是件好事。
Technology is this good thing.
当然，凡事都有负面作用。
Of course, there's always downsides.
但你知道，还是让人兴奋，你去动手做，是因为这太迷人了，你正在参与未来。
But you know, it's exciting and you go and you build it because it's it's fascinating and you're part of the future.
对我来说，这种思维上的分歧，才是这个故事真正想讲的。
And it's just this to me, the mindset divide there is what this story is really about.
要不是因为这个，亚历克斯，我们根本不会聊到这件事。
Because if it wasn't for that, Alex, would we be talking about we would not be talking about this.
对吧？
Right?
最近其实发生过更大的丑闻，比如华尔街有人干了更疯狂的事，亏了更多钱。
Like, there's there's been bigger blowups recently, you know, on Wall Street of people who've done crazier things and lost more money.
对吧？
Right?
但这些事儿，科技节目里从来没人提。
And we don't talk about that on tech shows.
对吧？
Right?
所以在我看来，这才是关键所在。
So to me, that's why it's important.
是的。
Yeah.
明白了。
Okay.
我对这一点有点不同看法。
I'm gonna disagree with you slightly on that.
我觉得我们还是会讨论它。
I do think we would still be talking about it.
不过我认为，这个故事之所以如此吸引人，正是因为你说的那些深层因素。
Although I think this adds I think the reason why the story is irresistible are the undertones that you bring up.
但我认为，不可能把它和科技故事割裂开来，就像你提到的傲慢例子，当时对AI即将爆发的押注，可以说是纯粹到极致。
But I think it's impossible to disassociate it with the tech story because, and this is going back to that hubris example, situational awareness was or is TBD, the most pure bet on AI taking off and taking off very soon.
简单说，他想做的就是从技术奇点中赚钱。
Effectively, if you could give a one sentence description of what he's doing, it would basically or what he was attempting to do is basically profit off the singularity.
他的想法是，奇点已经来了，一切都回不去了，只要现在押对了，就能实现指数级增长。
Effectively, the belief was we're in the singularity now, nothing's going back to the way it was, and if you make the right bets now, you can go exponential.
他确实这么做了。
He really did.
但正因如此，我觉得有必要聊聊他最大的持仓有哪些。
But this is why I think it's important to talk about what his biggest holdings were.
他最大的持仓是Nebius集团、Neo Cloud、闪迪和美光，也就是内存和存储芯片，还有另一个Neo Cloud公司CoreWeave。
Its biggest holdings were Nebius Group, right, and Neo Cloud, SanDisk and Micron, which is like the RAM and the memory, and CoreWeave, another Neo Cloud.
所以，他长期持有的都是那种‘瓶颈型’押注——就是觉得AI需求会暴涨，这些公司价值会翻好几倍。
So basically, his long holdings were effectively the bottleneck bro type of long holdings, which is like demand for AI is going to increase so substantially that these companies are going to be you know, worth multiples of what they were.
今年很长一段时间里，情况看起来确实如此。
And for a long time this year, it actually seemed like that was right.
所以他做多这些，而为了进一步支持他的观点，做空的则是软件。
So he was that that on the long side, the short side, to take his thesis even further, was software.
据说，他做空的是一些软件公司，比如Adobe。
Reportedly, of the shorts were software companies like Adobe.
对吧？
Right?
所以，如果你问，怎么把AGI这种东西放进一个对冲基金里？
So this this is basically like, if you were to ask, how do you put, like, AGI in one hedge fund?
答案就是这个。
It would be this.
但就像我开头说的，对冲基金本来就得分散风险，而他这个根本没做任何对冲。
But like I said at the top, the job of the hedge fund is you gotta hedge a little bit, and this was an unhedged hedge fund.
确实没对冲。
It was unhedged.
没错。
Right.
说真的，这操作太傻了。
I mean, it's just dumb find it.
这就是为什么科技公司都不愿意等太久才上市，因为公开市场的状况太疯狂了。
This is why this is why tech companies don't they wait so long to go public because they don't because the public markets are insane.
说实话，很多时候根本不是人在交易这些股票，你知道的。
Like, I I don't even and first of all, like, it a lot of times aren't even humans trading, you know, in these stocks.
对吧？
Right?
全是算法在交易。
It's just algorithmic trading.
而且再说一点，我其实写过很多次了。
And secondly, like, I don't actually I've written about this a lot.
我觉得华尔街根本不懂科技或者人工智能。
Like, I don't actually think Wall Street understands technology or AI.
他们只是在跟风炒作，这些交易的涨跌其实都是由网络迷因在推动。
Like, I don't I think they're just you know, it's like meme it's like memes are driving this, you know, these trades up and down.
对吧？
Right?
但你得承认，他确实有眼光。
But he's I mean, look, you have to give him credit.
他是最早一批真正押注内存领域的人，早就看出内存会短缺。
Like he was one of the first people to really like put his money down on memory and see that there were there was this memory shortage.
对吧？
Right?
从长远来看，他确实没错。
And long term, again, he's right.
这类东西，说白了，都是不错的投资标的。
Like this stuff is, you know, these are good bets.
现在聪明人做的，就是趁低价入手这些股票。
Like what smart people are doing now is they're just buying these stocks at a discount.
对吧？
Right?
现在它们都在打折，正好入手。
They're they're on sale right now, and so you buy them.
而且你知道，这些股票都出现在我们每个人的401(k)账户里。
And it's like, you know, that's what and they're in all of our four zero one k's, etcetera.
我得说清楚，我从来不买卖个股。
Like, I don't trade individual stocks just to be just to be clear.
但长远来看，大家迟早都会买这些东西。
But, like, you know, it's a it's a long term, like, people are gonna be buying this stuff.
这是一项很有价值的技术。
It's valuable technology.
至于对冲基金，如果你做短线交易，当然得对冲风险。
And, you know, you like, hedge funds, of course, if you're doing short term trading, like, yeah, you have to hedge.
你不能把自己置于这种境地。
Like, you can't, you know, you can't put yourself in this position.
但他对这个领域根本一无所知，对吧？
But he had no experience in this space, right?
这可是基础中的基础。
This is like 101.
今天我们请到了Semafor的技术编辑里德·阿尔伯戈蒂。
We're here with Reed Albergotti, the technology editor at Semafor.
我们至少从利奥波德那里收到了一条给同行或投资者的说明。
We did hear from Leopold, at least in a note to his counterparts or his investors.
他向投资者解释说，他卖出股票只是为了平掉自己的空头头寸。
So so what he basically tells investors is the investors in his fund is he sold only to the point where effectively he covered his shorts.
这是他原话怎么说的。
So so he said this is his direct words.
这个基金并没有被关闭、清算，也没有转成只对私人开放的基金。
The fund was not shut down, liquidated or transformed into a private only fund.
我们仍然像以前一样，继续以混合型公私基金的模式运营。
We are continuing to operate as a hybrid public private fund as before.
不过，我们会把公开账户的运作改为付费模式，同时从这些事件中吸取教训。
However, we will manage our public book only as a paid on on a paid for basis while we draw the lessons from these developments.
最重要的是，我们采取了必要的措施，为下一次战斗做好准备。
Most importantly, we took steps that were necessary to fight another day.
好的。
Okay.
这很重要。
This is important.
他不仅做了这个押注，就像我们说的，是对AGI的无对冲押注，而且还用了杠杆，对吧？
So not only did he make this bet, you know, like we said, an unhedged bet on AGI, he did it with leverage, right?
据说在某个时候，他的杠杆比例达到了四倍，也就是说，如果内存价格继续上涨，他的收益会成倍增长。
So there was there was, you know, side some point, the reporting is that he was like four x leveraged on this bet, which means like, you know, if memory continues to go up, then then his number goes up a lot more.
但事实是价格没涨，这就引发了连锁反应。
But if it doesn't, which it didn't, then that that sort of leads to the cascading effect here.
我上小学的时候就学过这个了，对吧？
I mean, I learned about this in like grade school, right?
这根本不是什么新概念。
I mean, this is like not this is not a new concept.
对吧？
Right?
这简直就像1929年的股市崩盘。
I mean, this is like the nineteen twenty nine stock market crash.
这都是很基础的东西。
It's like, it's very basic stuff.
我不太确定。
I don't know.
好吧。
Alright.
利奥波德继续说。
Leopold continues.
他提到，作为阶段性更新，我们目前未经审计的当月净收益预估为67%，全年净收益为+80%。
He says, as an interim update, our current unaudited estimate of net month to date performance is 67% and of net year to date performance is plus 80%.
所以本月是负67%，全年是正80%，我明白了。
So negative 67 on the month plus 80 on the year, I get.
他说最终数据会通过我们正常的报告流程公布。
And he says final figures will follow through our normal reporting process.
要是全年收益有百分之八十，我想你也会接受的。
I guess you'll take it if you're plus 80 on the year.
我不确定。
I don't know.
听起来挺不错的。
That sounds pretty good to me.
比我的401(k)今年表现好多了。
Better than my four zero one ks this year.
所以，这位先生的情况可能还没到尽头，Leopold还没完。
So so basically this guy might still like it's it's not the end of Leopold.
他手里依然管理着数十亿美元，但这次确实是个让他清醒的时刻。
He still has billions of dollars that he's managing, but certainly a humbling moment for him.
这确实让人感到谦卑。
It's definitely humbling.
确实如此。
Yeah.
他还会继续走下去。
And he'll he'll go on.
他会好起来的。
He'll be fine.
所以我说，真正吸引我的是这种文化意义，而不是别的。
That's why it's like I'm like, this is not like the it's it's this cultural significance that to me is interesting.
不是具体的交易本身，你知道的，生活总要继续。
Not the not the actual trade, you know, It's life life goes on.
我不确定你怎么想，但我觉得这些股票长期来看不会出问题。
I mean, I I don't know about you, but, like, I don't see these stocks having issues in the long run.
现在网上到处都是笑话，说这些东西太贵了。
Like, you know, there's a whole meme right now about this stuff being expensive.
我知道我们待会儿再聊这个，但说到底，我根本不觉得这事会反转或放缓。
I know we'll we'll talk about that later, but, you know, it's like, ultimately, I just don't see this this thing reversing or slowing down.
它一直在往前走。
Like it's it's moving forward.
你是怎么想的？你同意吗？
How do you I mean, do you do you agree?
不同意。
No.
我觉得算力在未来好几年里依然会有巨大需求，我觉得是这样。
I think compute there will still be a large demand for compute, at least for next important number of years, I think.
顺便说一句，这会是接下来几周节目的一个主题——期限错配，对吧？
So it's like this was sort of the whole thing, by the way, this is going be a theme on this show for the next couple of weeks is duration mismatch, right?
你可能在大方向上是对的，但在时机上却错了。
You can be right on the general thesis, you could be wrong on the timing.
一旦时机判断失误，后果可能非常严重。
And when you're wrong on the timing, that could be devastating.
这一点对利奥波德也适用。
That goes for Leopold.
对所有那些投资大型数据中心的投资者来说，也同样适用。
But it also goes for all these investors in the big data centers.
意思是，我给你建一台大型计算机，然后你在几年内把它变成利润。
Like the idea is I build the big computer for you and then you within x number of years turn that into profit.
我们知道，大笔资金已经投在了这些大型计算机上。
We know that the big computer money is being laid out.
但还不确定这些投入最终能不能赚回钱来。
We don't know if that's going be turned into profit.
这些AI公司最终能靠开发AI模型赚钱吗？
Eventually, these AI companies going to make money by developing these AI models?
很可能吧，不过现在大家对具体怎么赚还存在争议。
Probably, although it's sort of debate about how that happens now.
但他们会面临压力，必须在资金到期前实现回报。
But they they are the pressure will be to do it on a timeline that lines up with when the money comes due.
是的。
Yeah.
不过我觉得这里有个区别，你可以看看过去的技术建设热潮，都经历过这种兴衰周期。
I mean, I think the difference though here is like, you can look at past, like, tech build outs, and there have been these boom and bust cycles.
以前有过暗光纤。
You had the dark fiber back in the day.
我猜你在节目里聊过很多次这个了。
I'm sure you've talked about that a lot on the show.
现在这东西其实已经被广泛使用了。
I mean, this is, like, getting used today.
需求特别高。
Like, there's high demand for it.
所以你看像Meta或者SpaceX AI这些公司，押注算力，要是时间点没踩准，或者干脆押错了——这取决于你怎么看——他们也能转手把算力卖出去，毕竟需求这么大。
So when you see people, like, make these bets on compute, Meta or SpaceX AI, and they're off on the timing, right, or maybe off totally if depending on your opinion, they can then turn around and and sell that compute because there's so much demand.
所以我觉得他，还有其他人投资的这些瓶颈，其实还有很多其他瓶颈。
So I think these bottlenecks that he's invested Leopold invested in, and others like there are other bottlenecks too.
我觉得，这些瓶颈，其实别人也早就提过了。
I think, those actually, they're like prevent me other people have made this point too.
我不是第一个这么说的人，但这些瓶颈恰恰能防止整个局面失控。
I'm not the first person to say this, but like those actually, you know, kind of prevent this thing from going off the rails.
这些瓶颈就像一种防止泡沫失控的机制，你知道的。
It's like a it's like a bubble, you know, prevention mechanism, the bottlenecks.
对。
Yeah.
那我们不妨把这个观点跟本周的一条新闻对照一下，看看是不是完全说得通。
Well, that is so let's actually so let's run that idea next to a headline from this week and see if it's totally Loctite.
最近我们节目里一直在讨论，模型的发展正逐渐接近某种平衡，不一定是完全一致，但已经很接近了，对吧？
Because we've been talking a lot on the show recently about how models are starting to reach not necessarily parity, but maybe close to it, right?
我们刚看到的 Kim EK3 就是个例子，它虽然还比不上 Fables 或 GPT 5.6 这样的顶级模型，但已经跟最新一代的水平很接近了。
The Kim EK3 situation that we just saw was another model that sort of it's not equivalent to like, let's say the Fables and the GPT 5.6 soles, but close to the latest series of models.
当市场上不是一两个厂商，而是有好多家都在提供算力时，价格 inevitably 会往下掉，不然怎么竞争呢？
And when you have not one or two liters, but a bunch of liters, the prices will inevitably come down because how else do you compete?
CNBC 就发了一篇相关的报道。
CNBC has a story for us on this.
OpenAI 调低了两款 GPT 5.6 模型的价格，因为企业对成本越来越敏感了。
OpenAI cuts prices for two of its GPT 5.6 AI models as companies grow sensitive to costs.
来看看这篇报道。
Here's the story.
周四，OpenAI 宣布大幅下调其两款最新 AI 模型 GPT-5.6 Terra 和 GPT-5.6 Luna 的价格，距离它们公开发布大概才过了三周。
OpenAI on Thursday announced it was slashing the price of two of its latest artificial intelligence models, GPT 5.6 Terra and GPT 5.6 Luna, roughly three weeks after their public release.
公司正面临压力，需要迎合对成本更敏感的客户群体，因为企业客户在没有明确投资回报预期的情况下，不太愿意部署昂贵的模型。
The company is facing pressure to cater to a more cost sensitive customer base where enterprises have been less inclined to deploy expensive models without a clear picture on the return on their investments.
所以 Terra 的价格降了 20%，而 Luna 据说是他们最轻量级的模型，价格直接砍掉了 80%。
So Terra gets cut 20% and Luna, which I believe is the light most lightweight model that they have, is cut by 80% on the price.
现在 OpenAI 说他们在这些模型上找到了一些效率提升，我不怀疑这一点。
Now OpenAI says they've they've found some efficiencies in these models and I don't doubt it.
但是 Reed，回到你之前提到的观点，当你看到这些降价时，需求背后的逻辑一直是：那些想要抢购数据中心的公司相信，他们可以拿到 GPU，用它跑模型，然后把输出的 token 加价卖出去。
But Reed, going back to your point earlier, when you see these price cuts, the demand has always been basically that the companies that want to snatch up these data centers believe that they can take the GPU, use it, run a model through it, and then markup the tokens that they get out on the other end.
但如果加价空间越来越小，我们还能看到那种近乎无限的需求持续下去吗？
But if the markup is lower and lower, do we see that sort of unlimited demand can persist?
我觉得这里其实有两三个不同的情况在发生，对吧？
Well, I think there's two or three there's, there's actually two or three sort of different things going on there, right?
一个是，到底谁在卖数据中心的设备，比如显卡供应商。
One is, who's who's actually selling the equipment for the data centers, right, the GPU providers.
然后是模型提供商，他们开发软件和应用层，打造运行在这些硬件上的前沿模型。
Then you've also you got the the model providers, who are who are creating the software, the application layer, the frontier models for that run on these things.
还有就是数据中心服务提供商，对吧？
And then you're actually talking about the data center services providers, right?
光有数据中心，做个只会提供机架的纯基础设施商，长远来看并不是个好生意。
You don't just having a data center and being a dumb, you know, essentially like rack provider is not actually like a great long term business.
现在倒是挺赚钱的。
It's pretty good now.
所以像Core Weave这样的公司，才会想在数据中心基础上提供更多服务，对吧？
But that's why like these companies like Core Weave, they want to offer services on top of the data center, right?
实际上这里有三类不同的角色。
There's actually like three things going on.
第一类是芯片制造商和内存供应商，不管怎样他们都能卖得出去，因为总有人会用。
One, the the chip makers and the people who make the memory and all this stuff, like they're selling this no matter what, because like, someone's just gonna use it.
所以他们还行。
So they're okay.
对吧？
Right?
至于模型提供商，那就是另一个问题了，对吧？
And then I think the model providers, it's a different question, right?
你其实是在问，能不能真的花几十亿美元去训练这些模型？
Which you're asking, which is like, can you actually spend billions of dollars training these models?
然后呢，你还能高价卖它们，可有些中国公司却能从这些模型里提炼出东西，免费提供出来。
And then, you know, and then, you know, charge a premium for them while some Chinese company can essentially, like, distill from that model and offer it for free.
当然，
Of
这些中国模型还是得跑在极其昂贵的GPU上。
course, you know, those Chinese models still have to run on really expensive GPUs.
所以对英伟达来说，根本没改变游戏规则。
So like doesn't change the game for for Nvidia.
但模型提供商那边，确实有些公司根本不会直接用他们的服务。
But the model providers also like there yeah, some companies aren't gonna use them.
他们会自己做微调。
They're gonna try to fine tune.
他们会搭建本地的数据中心。
They're gonna build on premise data centers.
他们会做各种各样的事情。
They're gonna do all sorts of stuff.
对吧？
Right?
但这个市场的总规模太大了，我觉得总会有一些公司和企业，还是会用Anthropic和OpenAI的模型——顺便说一句，这些公司提供的是一整套模型，从便宜高效的到前沿的都有。
But this is like, the total addressable market is so large, that I think there will always be companies and businesses that are going to, you know, they're going to use anthropic and OpenAI models, which by the way, you know, they are these companies offer like a whole suite of models from like really cheap, efficient ones to the frontier.
这些模型在OpenAI和Anthropic自家开发的软件平台上运行得更高效、效果更好。
And those models run more efficiently and better on the harnesses, like essentially the software that OpenAI and Anthropic build.
而且这里还有一个数据飞轮效应，对吧？
And there's a data flywheel there, right?
用的人越多，模型就越聪明，效率也越高。
So the more people use them, the better they get, the more efficient they get.
我觉得这跟当年云服务的普及很像，对吧？
So there's, I think we saw the same thing with cloud adoption, right?
那时候大家觉得云服务更贵，心里犯嘀咕：干嘛非得用云？
Like, you know, people, it was cloud was more expensive, like, why would I do that?
我自己建个数据中心不就行了。
I'll just build my own data set.
但后来大家都转向云了，因为稳定可靠，省心省力。
Eventually, they went to the cloud because it was just, it was just more predictable.
长远来看反而更省钱，也不用自己操心维护。
In the long run, it's cheaper, you know, it's, it's something you don't have to worry about.
所以我觉得，这些论点确实有道理，但前沿公司也确实面临不少风险。
So I think you can kinda, you can make these arguments, I think there are definitely risks to these frontier companies.
不过我觉得，中国那些开源的免费模型，倒未必是最大的威胁。
I don't really think the, the Chinese open source, you know, the free models are the, the most important risk.
我觉得这些前沿模型公司根本不可能掌控世界，成为主宰全球90%经济的巨头，那种情况根本不会发生，但它们依然能做成不错的生意。
I think that's a that means like, I don't think these companies by any means the frontier model companies are going to like take over the world and become these huge companies that control like 90% of the global economy like that not gonna happen, but there's still good businesses.
我就是这么看的。
That's sort of how I look at it.
我们往往对这类事情抱有非此即彼的想法，而我正努力避开这种思维。
Like, we tend to have this zero sum thinking around this stuff, which is like, what I try to get away from.
是的，顺便说一句，我们特别喜欢在这档节目里深入探讨这些细微差别。
Yeah, no, it's by the way, we love talking through nuance on this show.
事实上，我觉得这得看是哪一周。
In fact, I think that it depends on the week.
有时候大家会说，这期节目完全是为AI做的。
There'll be one week where folks will say, Hey, this show is in the can for AI.
但下一周我们聊别的内容时，又会觉得，这期节目怎么成了末日论节目了。
And then there'll be a next week when we cover something else where it'll be like, This show is a doomer show.
我们只是想全面考虑各种可能的结果，并对它们进行压力测试。
And it's like, we're trying to just consider the full range of potential outcomes and pressure test them.
所以说到这点，我觉得你说得对。
And so on this note, I think you're right.
可能不是中国，但看起来实验室们正在降低模型价格，而且不只是它们在这么做。
It might not be China, but it seems like, first of the labs are lowering prices on their models, and it's not just them.
这是来自CNBC那篇报道的内容。
You know, this is from that CNBC story.
微软首席执行官萨提亚·纳德拉在周三的季度财报电话会上多次强调，公司推出的模型在成本上很有优势。
Microsoft CEO, Satya Nadella, repeatedly highlighted his company's cost effective models during its quarterly earnings call with investors on Wednesday.
谷歌这个月也推出了三款新模型，目标就是以更低的价格击败竞争对手。
Google also debuted three new models this month that aim to undercut competitors on cost.
所以咱们再回过头说这一点。
So so just to go back to this point.
对吧？
Right?
我不是说，呃，你知道的。
Because I I'm not saying it's well, you know what?
我不能完全排除这里的情况是零潜力、零机会，所以我想听听你的看法。
I won't necessarily rule out that it's a zero potential zero situation here and just going to throw it to you and hear your perspective on it.
关键问题是，是什么推动了这些数据中心的扩建？
Basically the point is like what's driving this build out of the data centers?
没错，当然是因为对人工智能的需求。
Yes, it's of course demand to use AI.
也许最终胜出的还是那些超大规模云服务商。
So maybe it is just like the hyperscalers that end up winning in the end.
但很多推动力来自OpenAI和Anthropic，他们相信自己能在算力上竞争。
But a lot of the push is coming from OpenAI and Anthropic who believe that they will compete on compute.
如果OpenAI、Anthropic、Google和Microsoft的定价都下降了，
And looking at it, if the prices come down for OpenAI, the prices come down for Anthropic, the prices come down for Google, the prices come down for Microsoft.
那他们在这些建设上的投入还值得吗？
Are those investments in all those data centers then worthwhile?
你明白我的意思吧？
You see what I'm
说什么？
saying?
我明白你的意思。
Yeah, I see what you're saying.
另外，我们也要把云服务商和芯片制造商区分开来，对吧？
I mean, and also like, let's let's like also differentiate the hyperscalers from like the chip makers, right?
还有那些生产设备的公司，比如英伟达，不管怎样他们都稳坐钓鱼台，反正东西卖得出去。
And the people who make all the equipment like, you know, I think Nvidia is just sitting pretty like they're just selling this stuff no matter what happens, right?
我觉得云服务商确实能赚钱，但他们还得卖服务。
I think the hyperscalers, you know, yeah, I think they'll make money, but they also have to sell services.
他们不能只当GPU供应商，还得提供大量配套服务。
They can't just they can't just be like GPU providers like they have to, but there are a lot of services to be sold.
这些技术本身其实没那么好用，除非你……这可是今年的主旋律，对吧？
Like this stuff doesn't really work that well, unless you this has been the story of the year, right?
这些模型确实非常强大。
Like these models were super powerful.
我们早就有了推理模型，但一直没太大用处，直到人们学会怎么把它们嵌入框架，连上工具，循环运行，诸如此类。
We had the reasoning models, but they weren't really doing all that much until people figured out how to put them into these harnesses, connect them to tools, run them in loops, you know, all this stuff.
而这一切都得靠更多GPU，需要更多token，对吧？
And all that stuff just requires more GPU just requires more tokens, right?
所以我觉得，没错，最终这些技术确实会降价，但接着就会出现贾万悖论——大家反而用得更多了。
So I think like, yeah, I mean, in the end, sure, like these things come down in price they get but then, you know, you have Javon's paradox, people just use more of it.
你说是不是？
Right?
我不确定你用这东西是什么感觉。
And it's so I don't know about you when you use this stuff.
我 personally 用它，就是想搞清楚这项技术到底怎么回事。
I I use it personally because I I wanna try to understand the technology.
我觉得特别有趣，而且真的很有用。
I find it very fun and very and actually useful.
我不仅自己用，还老跟别人聊这个。
Like I've, you know, but then I'm also talking to people about it all the time.
我看到很多人，你知道的，用这种方式比我高级多了。
And I see how people, know, there's a lot of people who are using it in obviously way more advanced ways than I am.
你会觉得这技术真有用，它不是那种昙花一现的东西，对吧？
And you're like, this is useful technology like this isn't the kind of thing it's not a fad, right?
它不像那些计步手表之类的东西，大家当时觉得挺酷。
This isn't like, I don't know, those, you know, pedal counting watches or something that people like are like, Oh, this is cool.
我还想着是不是该把钱都投到计步手表上呢。
I I should put all my money into, you know, into step counting watches.
说实话，这项技术对个人和企业来说都非常实用。
Like, this is actually like, you know, really useful technology for individuals and for businesses.
所以市场这么大，我真觉得这不会是个零和游戏，因为全世界都不会容忍某一款人人都得用的产品只有一个赢家。
So the market is so large that I just don't see I don't I can't see this being a zero sum game only because the world will not tolerate, like, one winner in a in a product that everyone in the world has to use.
那样的话，那家赢家公司就会变得过于强大。
Would create that that that company, that winner would be way too powerful.
事情根本不是那样运作的。
It just doesn't it just does not work that way.
这样说得通吗？
Does that make sense?
是的。
Yeah.
对。
Yeah.
完全同意。
Totally.
没错。
Yeah.
我觉得吧，我不该说它们会彻底消失，但在我看来，必须得有公司能靠这些显卡赚到钱，这样整个产业链才能持续下去。
I guess like, you know, I shouldn't say that they're going to zero, but to me it's like you need to have companies that are gonna make money on top of those GPUs for that build up to continue.
我看到这些价格在往下掉，就在想，这利润到底从哪儿来呢？
And I'm looking at these prices coming down and I'm like, well, where's that going to happen?
不过我觉得你已经回答了这个问题，说明了你认为会在哪里发生。
But I think you've answered the question in terms of where you think it will happen.
展开剩余字幕（还有 480 条）
我并不反对你的看法。
Don't think I disagree with you.
大家愿意为此付费，而且随着这些技术越来越有用，使用量只会继续增长。
Willing to pay for it and the usage is just gonna increase as these things become more useful.
是的。
Yeah.
还有另一件事让我觉得特别不可思议。
I mean, I this is the other thing that's crazy to me.
人们每个月居然愿意花二十美元用聊天机器人。
Like, people are paying, like, $20 a month for chatbot.
普通消费者真的在花钱，这简直让人难以置信。
Like, consumers, you know, are pay like, they're it's crazy.
我没想到大家愿意花这么多钱，因为当年我刚入行报道科技时，大家都说没人愿意为这种东西买单。
Like, I can't I'm surprised how much people are actually willing to pay because the the adage when I got out here covering tech was like, no one will ever pay for this stuff.
像谷歌这种东西，你根本不可能向用户收费。
Like, you could not charge for Google.
你根本不可能向用户收Facebook的费用。
You could not charge for Facebook.
这太不可思议了。
That's insane.
现在你在X上能看到多少个蓝标？
How many blue check marks do you see on x now?
真是让人想不到。
That's crazy.
那些都是在花钱的人。
Those are people paying money.
要是有人在2013年跟我说，人们会每个月花十块、二十块买Twitter，我肯定觉得他疯了。
Like, if you, if you put, like, people will pay $10 or whatever, $20 a month for Twitter on my bingo card in, like, you know, 2013.
我肯定会说，你疯了吧。
I would have been like, you are insane.
根本没人会为Twitter付钱。
Like, no one will pay for Twitter.
现在确实有人愿意付钱了。
And they are now.
我觉得这是一个被严重低估的变化。
And I think that's a whole change that is, like, underappreciated.
真的，大家其实都愿意为这些内容买单。
Like, people actually pay for this stuff.
对。
Yeah.
我在参加的几次活动里做过这个实验，我问大家：假设你正在用 ChatGPT，如果 OpenAI 把价格翻倍，你愿意付双倍吗？
I did this experiment in a couple of events that I was at where I was like, all right, if let's say you're using ChatGPT and OpenAI doubled the price for you, would you pay double?
结果所有人都举了手，甚至有人举手表示愿意付三倍。
And all the hands went up, triple, more hands went up.
说实话，我有点不敢大声说出来，但它对我来说真的太好用了。我现在用的是20美元的套餐，如果基础版涨到每月100美元，只要还能提供现在的服务水准，我也愿意付。
And I don't know, I feel like I shouldn't say this out loud, but it's become so useful to me that if it became a $100 a month at the base in terms of what I'm getting now, I'm on the $20 plan, I would do it.
我想他们心里也清楚这一点，但另一方面，这些公司现在都处于扩张期，对吧？
I think they know that, but they're also like, these companies are in growth mode, right?
他们愿意亏一大笔钱来抢占市场份额。
They're willing to lose a ton of money to gain market share.
这本身就是他们的玩法。
Like that's the game.
所以别一看到降价就以为是清仓甩卖，这可不是衣服鞋子那种快消品。
So just because somebody lowers prices, this is not like, Oh, okay, now it's like in the discount bin and who can't you know, this isn't fashion products, right?
他们是想冲出去彻底占领市场。
This is like, they're trying to go out and take over the market.
说得对。
And that's right.
但价格战确实是真实存在的。
But price but price wars are a real thing.
对吧？
Right?
你完全可以把机会理解成，你的利润最终会被竞争蚕食殆尽，变成一个毫无利润的普通商品。
Like you could eventually write your opportunity as my margin just kind of compete away all the profit in a commodity.
你可以这么做。
You could.
但增长型科技公司通常不是这么玩的。
I mean, but that's not typically how it works in like growth tech businesses.
对吧？
Right?
你先去抢占市场，赚钱的事儿以后再说。
Like you you go out and you and you win the market and then you worry about the money later.
当然了，这些公司都想着上市。
Of course, these companies are trying to go public.
我不太确定。
I I don't know.
关键是这么回事。
Like, this is the thing.
你真觉得这些IPO会推迟吗？
Like, do you really I mean, maybe the mark maybe these IPOs get delayed.
唉，我真的说不准。
Like, I don't know.
不过这已经超出了我平时想的范围，我还没怎么深入考虑过。
But that is that is actually beyond my like, I haven't thought that much about it.
不过我觉得，确实有个说法值得提一提。
But, like, there is an argument I think to be made.
他们其实可以再等等，别那么急着上市。
Like, they should wait a little bit to to IPO.
我的意思是，
I mean,
那你的理由是什么呢？
what's the arg what would your argument be for that?
关键是，如果你还在疯狂增长的阶段，投资者真会买账吗？
Well, it's just if you have like, if you're still in this, like, insane growth mode, like, is that gonna make sense to investors?
就像你说的，投资者都会看，你卖产品，总得赚得比成本多吧？
Like you said, investors do look at this, like, well, you should if you're selling a product, like, you should be charging more than it costs you to, you know, for that product, right?
但话说回来，亚马逊在公开市场上亏了多年，才终于扭亏为盈，所以这事儿也说不准。
And like, ultimately, but you know, on the other hand, I don't know, Amazon lost tons of money in the public markets for years before they finally, you know, turn to profit.
也许这样也没问题。
So maybe it's fine.
但你得是那种特定类型的公司，还得有个特定类型的CEO，才能赢得散户的信任。
But like, you have to be a certain type of company, a certain type of CEO to, like, gain the trust of of retail investors.
对吧？
Right?
像埃隆·马斯克就是这样的人。
Like, you know, Elon Musk is that type of person.
达里奥吗？我不太清楚。
Is is Dario I know.
萨姆？
Sam.
我不确定。
I don't know.
即使这样，也可能很艰难。
Even that it's can be tough.
确实不容易。
Tough.
可能会很吃力。
Can be tough.
有起有落。
They ups and downs.
是的。
Yep.
好的。
All right.
那我们继续聊聊这个话题——当谈到供应商融资时，一切都会好起来的。
So let's continue on this theme of it's gonna be all right when we talk about vendor financing.
这周你讲了一个关于 Safe Super Intelligence 的故事，这家公司由前 OpenAI 首席科学家 Ilya Sutskever 主持。我完全没想到，他突然说：‘我们有突破了，现在该投入大量算力了。’
So you had a story this week about Safe Super Intelligence, which is run by Ilya Sutskever, the former chief scientist of OpenAI, where I kind of out of nowhere Ilya was like, all right, we have a breakthrough and now it's time to put a lot of compute behind it.
我觉得他说的是，该造台更大的电脑了，这可是他常挂嘴边的一句话。
I think he was like, time to build the bigger computer, which is one of his favorite lines.
结果英伟达决定给他们大手笔投资，五十亿美元，而这家公司连产品都没有。
And Nvidia decided to make a big strategic investment in them, 5,000,000,000 for a company that doesn't have a product.
一年半以前，这种规模的融资绝对是史上最大的，遥遥领先。
I think before a year and a half ago, that would have been the largest venture raise in history by some margin.
可现在呢，大家就觉得挺平常的。
And now it's kind of like whatever.
不过你之前在Semafor写的那篇文章，核心观点其实是：这事儿没啥问题。
So but your your perspective on it when you wrote in Semafor was basically like, it's all good.
那你具体说说，到底发生了什么，为什么你会这么想？
So talk through talk through what happened and why you feel that
没错。
way.
是的。
Yeah.
一切都好。
It's all good.
不，我的意思是，我觉得这就像一种循环投资，对吧？
No, I mean, I think I think there's this like, it it's like this circular investing, right?
这才是真正让人担心的地方。
That's the big concern.
如果你是一家公司，比如英伟达，你本质上是在买自己的客户。
It's like, well, if you're a company, you're, you know, if you're Nvidia, you're essentially buying a customer.
这可不是什么好事。
That's not a good thing.
如果我只是付钱给你来买我的产品，那这算不上真正的客户。
Like, you know, it's not a real customer if I'm just paying you to buy my product.
但归根结底，这其实是在押注超级人工智能的安全性。
But in the end, it's like, I guess you're it's a bet on safe super super intelligence.
伊利亚·苏茨克弗是个有实绩的人。
Ilya Sutskever is a proven commodity.
我个人觉得有理由持怀疑态度，因为他们太保密了。
Like, I personally think there's reasons to be skeptical because they've they're so secretive.
而且说实话，在人工智能的发展史上，重大突破从来都不是在秘密中发生的。
And I just I just you haven't seen in the history of AI development, big breakthroughs happen in secret.
通常都是论文一发表，大家就纷纷讨论，互相启发，然后差不多同一时间冒出各种新想法。
These usually, you know, these papers come out, they ping pong around everybody, you know, it cross pollinates, and then people come up with ideas around the same time.
但也许这次真的不一样。
So but maybe there isn't.
也许他们真在秘密中找到了什么关键突破。
Maybe that maybe they've figured something out, like, working in secret.
对吧？
Right?
所以英伟达看到了，就觉得得押注这个方向。
So Nvidia sees that and they go, we wanna bet on that.
他们觉得，这可能是下一个大突破。
Like, you know, they've they've got the next big thing.
好吧。
Okay.
这是一方面。
That's one thing.
但假设他们失败了呢？
But let's say they fail.
假设所谓的安全超级智能，根本搞不定呢？
Let's say, safe super intelligence, like, doesn't get it done.
他们现在建了台超大的计算机，人人都想要这种大机器。
They built this big computer now, and everybody wants big computers.
对吧？
Right?
所以他们要么卖这台机器，要么像SpaceX AI或者Meta那样，把这台大机器租给别人用。
So they're gonna be able they'll either sell it or they'll rent out their their big computer to other people like like SpaceX AI did or Meta did.
所以如果你是英伟达，你会觉得，这其实没多大风险。
So if you're Nvidia, you're like, there's not actually like that much risk here.
其实最终也没多大风险。
Like it's not it's not in the end that much risk.
这些人里头，像伊利亚，他特别擅长的一点其实被严重低估了——就是能把这些大型计算机跑得特别好。
And like one thing these people are all good at, like Ilya, which is really underappreciated, is making these big computers work really well.
说实话，这可能是现在AI研究员最有价值的技能了，对吧？
Like, that is actually like probably the most valuable skill of an AI researcher at this point, right?
就是想办法把这么多GPU高效地连起来，让它们同时全力运转。
It's like figuring out how to efficiently string these GPUs together, make them work all at the same time.
这事儿真不是一般人能干好的。
It's really hard to do.
所以如果你是英伟达，你可能会想：就算伊利亚在AI算法上没出什么颠覆性突破，他依然是最顶尖的超大规模算力服务商之一。
So like, you're basically if you're Nvidia, you're like, well, if if Ilya doesn't have this major breakthrough breakthrough in AI algorithms, well, he's probably one of the best, like, hyperscaler providers around.
所以对你来说，要是你叫詹森，这根本算不上一个疯狂的押注。
So it's really just not like a I to me, it's not a crazy bet if you're if you're Jensen.
我觉得这里面其实有很多保险措施。
There's a lot of hedging in that one, I would say.
是的。
Yeah.
不，我是想说，如果真是这样，那为什么还要卖显卡呢？干脆直接拿股权换显卡，谁需要就给谁好了。
No, I was gonna say basically like if that's the case, why even sell your GPUs and why not just give GPUs away for a stake of anybody, of a company of anybody who wants them.
事实上，英伟达确实就这么做了。
And in fact, that is what Nvidia is doing.
他们对初创公司采取了新策略：想要显卡？可以，但得拿公司一部分股权来换，我们负责提供硬件。
They have this new approach with startups where they're like, all right, if you want GPUs, we'll take a chunk of your company and we'll deliver the hardware.
对，但他们并不是说要包揽一切。
Yeah, but they're not like, we're gonna just be that we're just gonna do everything.
这种说法不对。
That's not correct.
他们可不是打算自己去建数据中心。
They're not like, we're gonna just go build a data center.
他们确实参与得更多了，现在都是整机架卖，而不是像以前那样，你可能根本买不到单张显卡了。
I mean, they do more of it because now they're selling it in entire racks as opposed to, like, I don't think you can really just buy one of these GPUs now.
你得整套机架一起买，这其实是个挺聪明的策略。
You gotta buy the whole rack, which is probably a smart a smart move.
不过说实话，我还没听到什么有力的反对意见。
But, yeah, I mean, it it to me, it like, I I have not heard an argument.
也许有人能钻空子，但我真没听过谁提出过像样的反驳。
Maybe you can maybe you can steal man it, but, like, I haven't heard an argument that against that really.
大家只是说，万一这整个就是个泡沫，那英伟达就得自己扛着一堆烂摊子了，诸如此类的担忧。
It's just like, it's more like, well, this thing could all be a bubble, and then Nvidia is left holding the bag or, you know, something like that.
我同事莉兹·霍夫曼还把他们比作金融危机时的AIG，我觉得这个说法挺抓眼球的。
But, like, they're the like, my colleague lives Liz Hoffman compared them to AIG, right, in the financial crisis, which I think is a cool headline.
这说法确实挺有冲击力的。
Like, that's that's, like, exciting.
说起来也挺吓人的，就像恐怖片那种吓人法。
It's, like, kind of scary, you know, like, like horror movies are scary.
但我个人觉得，这跟当年的情况其实差得挺远的。
But I I also think it's like, I don't I don't I think there's so there's so many differences between that situation and this one.
好的。
Okay.
我其实希望你，如果你愿意的话，可以去抢这个风头。
I actually want you to, if you're willing to or you want to, to steal man.
确实有个不错的理由能说明为什么这种情况还会持续。
Yes, there's an argument, a good argument to be made for why this continues.
但如果真要崩盘，你觉得会在哪儿开始瓦解？
But where could you see it unraveling if it does?
这是个好问题。
Look, it's a great question.
我觉得最大的风险可能是，哪天出现某个算法上的突破，让咱们根本不再需要高性能电脑了。
I think that probably the biggest risk is that there's some algorithmic breakthrough that actually means you don't need powerful computers anymore, right?
比如有人发现，其实人类大脑就是个特别高效的计算机。
Like someone figures out, oh, actually, I don't know, the human brain is a very efficient computer, right?
它只用大概二十瓦的电力就能运行。
It runs on whatever 20 watts or something.
它不需要大型数据中心。
It's a doesn't require a big data center.
说起来，还真有一些公司正在培育人脑组织，想把它们当计算机用，就是真正的脑组织。
There are actually companies to come to think of it that are growing human brains and want to use them as computers like actual brain tissue.
所以你看，这可不是纯粹的科幻。
So you know, this isn't total sci fi.
他们说，我们其实可以在一个U盘大小的设备上实现通用人工智能。
And they're like, well, you know, we've we actually can do AGI on a, you know, on a thumb drive or something, right?
我们根本不再需要那些庞大的数据中心了。
It's like, we don't need these big data centers anymore.
那样的话，整个体系可能就崩了。
Then maybe the whole thing collapses, right?
突然间，那些数据中心就变成了空荡荡的光纤管道，因为你只需要一小块地方——比如OpenAI在德克萨斯建的巨型数据中心里的一角——就能运行所有需要的计算。
It's like, all of sudden, you just you can, you know, there's there's this the data centers become dark fiber, because, you know, all of a sudden, like, you could you can run everything you need on one tiny sliver of the, you know, the massive data center that OpenAI is building in Texas.
我不确定。
I don't know.
这是一种可能。
That's one.
我觉得这真的有可能发生。
Like, I think that's a real possibility.
不过，说实话，我也说不准。
But, like, of course, I I don't know.
那谁在研发这个呢？
Like, who's developing that?
我觉得，也许是那些搞微型脑组织计算机的人。
Like, I think, you know, maybe it's the mini brain tissue computer people.
我也说不准。
I don't know.
对。
Right.
如果真发生了，文明肯定会彻底改变。
And if that happens, then civilization definitely changes.
是的
Yeah.
我的意思是，这对文明来说不会是坏事。
I mean, it won't be bad for civilization.
我觉得这反而是对文明有利的事。
I I like that would be a good thing for civilization.
那会是件坏事。
It'd be a bad thing.
如果你是英伟达，那确实是个坏事。
It'd be a bad thing if you're Nvidia.
对你这种建数据中心的人来说，这可不是好事，对吧？
The bad thing if you're building these data centers, right?
对OpenAI和Anthropic来说是坏事，他们会被彻底超越。
Bad thing for OpenAI and Anthropic, they get totally superseded.
但人类将能获得廉价的智能资源。
But humanity has this like, access to cheap intelligence.
我们没必要建这么多新电厂，你知道吧，听起来好像挺吓人，但有些海湾地区的投资者亏点钱，大家也就继续往前走了。
We don't have to build all these new power plants like, you know, that sounds like, yeah, like, and some of these investors, you know, some of these investors in The Gulf lose lose some money and, you know, we all move on.
是的。
Yeah.
好的。
Alright.
在休息之前，我想跟你聊聊萨姆·阿尔特曼这周都在忙什么。
Before we go to break, I wanna talk to you a little bit about what Sam Altman has been up to this week.
他最近去了华盛顿，跟政府官员讨论OpenAI未来的发展方向。
He's actually been in Washington DC talking to government officials about what's to come for OpenAI.
我念一段《华盛顿邮报》的内容给你听，你帮我判断一下，这到底是新动向，还是他们早就一直在做的事。
Let me read to you a little bit from the Washington Post and you can share with me whether you think this is actually new or it's the stuff they've been doing already for a while.
报道里说，阿尔特曼在简报中提到，即将推出一款产品，能让多个AI助手——也就是所谓的智能代理——同时在后台协同工作，分工合作。
So the story says, in the briefings Altman described an upcoming product that would enable multiple AI assistants known as agents to work simultaneously in the background, dividing tasks and collaborating with each other.
他还提到，这套系统甚至能解决以前从未有人解出来的数学难题。
He also described how the system could answer math problems that have never been solved before.
他提到，这些新智能体有可能改变美国经济，让员工能完成原本需要外包给专业人士的任务。
He described how the new agents could transform the American economy, allowing workers to do tasks that typically would have been outsourced to other professionals.
有人提到，软件工程师可以用这些智能体来协助人力资源工作，作家也能用它们来做平面设计。
He described how a software engineer could use the agents to help with human resources or a writer could use them to enable graphic design, one of the people said.
你觉得怎么样？
What do you think?
这不就是老生常谈嘛。
Same stuff for new stuff.
我觉得这也没什么新鲜的，对吧？
Sounds like the same stuff to me, right?
我的意思是，这本来就是……
I mean, that's how
我看到这篇报道时，反应也是一样。
that was my reaction when I read that.
对。
Yeah.
我是说，我用 Codex 的时候，你知道，他们那个桌面端，现在应该就是 Mac 上的 ChatGPT 桌面应用。
Are you I mean, when I use, Codex, know, their their desktop, their I guess now it's just chat GPT, the desktop app on Mac.
我老是让这些智能体去生成子智能体。
I'm always asking these, these agents to spawn sub agents.
说实话，我觉得我自己没怎么玩明白，但它们确实就这么在干。
Like, it's, it's like, I don't think I'm really that great at it, but like, that is what they're doing.
对吧？
Right?
它们在互相协调。
They're coordinating.
我就跟它们说：你当产品经理，然后你去派几个子智能体去干活，我直接跟你聊就行，这样你就不用忙活了。
I'm like, you be the, product manager and then I want you to spawn sub agents to do the tasks and then I can talk to you so you're not busy.
我之前就是个傻瓜，总直接让智能体帮我做所有事。
I've That's been being a silly human and asking to the agents to do everything that I want them directly.
不过我觉得，现在它们大概会自动这么做了。
Well, I think probably they're going to automatically do that now.
没错。
That's right.
我觉得这正是他所说的那个新方向。
I think that's the new thing that he's talking about.
其实就是更好地协调这些工具。
Like, just but it's just I think that's just better orchestration of these things.
关键是，你怎么去设计它们？
Like it's, it's how do you make them?
怎么让它们在更少人为干预的情况下完成工作？
How do you get them to do their job with less human involvement?
我经常得去检查进度。
Like, I do a lot of checking in.
我不知道你怎么样，但用 ChatGPT 的时候，你总会看到那个蓝点，然后就得去查看一下。
I don't know about you, but like, I'm, you know, you get the blue dot on if you're using the chat Gpt, you get the blue dot and then you got to check-in.
要是能直接说：‘这是我的目标，你想想怎么最好地达成它’，那就太好了。
Like, it'd be great if you could just be like, look, here's, here's my goal, figure out how to the best way to get to that goal.
我过一周再跟你跟进一下，你知道的，我就
And I'll check back with you in a week or something, you know, like, I'd
没问题。
be fine.
你一般用它来做哪些类型的项目？
What sort of projects do you use it for?
我是说，我什么都用。
I mean, I just use it for everything.
不过最近一个，上星期的那个，其实不是工作上的事。
But I mean, last, like the most recent one last week was it was not a work project.
纯粹是个人的事。
It was just personal.
我们小区最近发水，我们需要收集一些数据，报给县政府，好让他们掌握洪涝情况。
Like, we were in my neighborhood, we have like flooding, and we needed to collect data for, for the county so that they can get data on flooding.
他们根本没有这方面的数据收集机制，加州连个系统的洪涝数据收集项目都没有。
They have no there's no county, like California doesn't have like a data collection project for flooding.
所以我做了这些数据收集项目，但花了我两天时间，之后我就偶尔去看看进展。
So I made the data collection projects, but it took me two days and it, you know, I just checked in every once in a while.
现在我们社区有了一个数据收集平台，还有数据库和后台系统，可以上传照片和视频把数据发给县政府。
But now our neighborhood has a data collection portal and, you know, a database and a back end and we can send data to the county with photos and videos.
说白了就是一个网页应用。
It's basically a web app.
不过这只是我上周刚做完的最新一个项目。
But that's just the most recent one that I did last week.
其实也没花多少工夫。
And, you know, it's just isn't that much work.
别人都说：天啊，你是怎么搞出这个的？
And people are like, wow, how did you do this?
我就说：这真不是我做的。
I'm like, I didn't do it.
是Codex做的。
Codex did.
是啊，没错。
Yeah, right.
我的意思是，要是能一次性搞定就好了，现在这过程太折腾了，反复检查、不断试错，要是两天后直接出成品，那该多好。
I mean, but but it'd be great if you could just one shot that like there's a lot of checking in, there's a lot of trial and error, it could be great if you like, after two days, you just get a product and it's done.
我的意思是，那才叫新鲜。
I mean, that would be new.
感觉像是
Feel like
我觉得会实现的。
I think it'll happen.
对。
Yep.
好的。
Alright.
我觉得肯定能实现。
I think it'll definitely happen.
好的。
Okay.
咱们先休息一下，回来再聊聊微软为什么可能度过了有史以来最棒的一周。
Let's go to break and come back and talk a little bit about why Microsoft might have had its best week ever.
然后再多说一点关于科技巨头的财报情况。
And then a little bit more on big tech earnings.
这些内容马上回来继续。
That's coming up right after this.
本期节目由DeepL赞助播出。
This episode is brought to you by DeepL.
最近我和DeepL的创始人亚拉·库特利亚夫斯基在YouTube上聊天时，聊到了专用AI的价值。
When I sat down with DeepL's founder, Yara Kutlyaevsky, on YouTube recently, we got into the case for specialized AI.
DeepL语音功能，就是当对话需要实时进行时的真实体验。
DeepL voice is what it looks like when the stakes are real time conversation.
说实话，我真希望以前做跨国采访的时候就有这个工具，让语言障碍完全不是问题。
And, honestly, it's something I wish I'd had for my own cross border interviews, turning a language barrier into a non issue.
DeepL Voice 能够实时翻译超过40种语言，适用于线上会议和面对面交流，帮助人们用自己最习惯的语言顺畅沟通，不丢失任何语气和细节。
DeepL Voice delivers live translation in over 40 languages for virtual meetings and in person conversations, helping people speak in their preferred language without losing flow or nuance.
无论你是和客户洽谈、与供应商谈判，还是和全球同事协作，它都能跟上你的节奏，轻松应对你行业特有的技术术语、缩写和产品名称，确保你真正想表达的意思不会在翻译中走样。
Whether you're meeting with a customer, negotiating with a supplier, or collaborating with global colleagues, it keeps pace with you in real time, easily handling the technical terms, acronyms, and product names specific to your business, so what you actually mean never gets lost in translation.
对于开发者来说，DeepL 的语音 API 让你可以直接把实时语音转写和翻译功能嵌入到你的产品里。
And for the builders listening, DeepL's voice API lets you embed real time speech transcription and translation directly into your products.
你自己去试试看吧。
So go check it out for yourself.
你可以在 DeepL.com/tryvoice 免费试用 DeepL Voice。
You can try DeepL voice for free at DeepL dot com slash try voice.
网址是 deepl.com/tryvoice。
That's deepl.com/tryvoice.
我们回到《大科技播客》周五特别版，今天请到了 Semafor 的科技编辑里德·阿尔伯戈蒂。
And we're back here on big technology podcast Friday edition with Reed Albergotti, the technology editor at Semafor.
去 semaphore.com 看看。
Go to semaphore.com.
在科技板块，你可以在那里或者左侧输入邮箱地址，免费订阅里德的精彩通讯。
In the technology section, you drop your email address there or right on the left hand side and you can get Reed's terrific newsletter.
而且完全是免费的。
That's And it's free.
所以，总之，我不建议你
Right, so anyway, I discourage
你觉得Semafor现在有泡沫吗？
Do you think there's a Semafor bubble?
真的存在泡沫吗？
Is there a bubble?
你是不是想说
Is that what you're
在Semafor里？
gonna In Semafor?
因为我们正在降价，还提供
Because we're slashing prices and offering it
你现在的这些产品。
products You are.
说真的，你的显卡供应商该开始紧张了。
For think, yeah, your GPU providers are gonna start to sweat a little bit.
你可能得叫莱奥波德来帮忙了。
You might Leopold margin called, man.
他打算做空我们。
He's gonna short us.
糟了。
Darn.
对了，你刚才当着公众的面说出来了。
By the way, now you said it in public.
这话一出口，就会被传到YouTube上，大模型也会拿去训练，慢慢就真信了。
It will go into YouTube and the LMs will train on it and the LMs will start, to believe this stuff.
然后大家也会信以为真。
And then people will believe it.
现在信息就是这么传播的。
That's how information works these days.
天哪。
Oh, man.
那这么说，我可是个顶尖运动员，我孩子都是天才了。
Well, in that case, I am an amazing athlete and my children are geniuses.
明白了。
Okay.
接着说。
Go on.
行吧。
All right.
记住这一点
Keep that
所有人都是如此。
in mind, all of them.
好吧。
All right.
咱们聊聊微软吧。
Let's talk about Microsoft.
他们创下了美国公司有史以来单日市值增长最高的纪录。
They had the biggest one day market cap gain for any US company in history.
这是《华尔街日报》报道的。
That's according to the Wall Street Journal.
公司股价上涨了16%，尽管市场之前担心他们在数据中心、芯片等领域的投资会超过其现金生成能力。
The company's stocks are 16% if it's after its earnings cold concerns that investments on data centers, chips, and more would outpace the company's ability to generate cash.
这正好印证了我们之前讨论的内容。
It's kind of going exactly to the conversation that we've had.
一天之内，市值增加了四千五百亿美元。
It's a $450,000,000,000 one day gain.
这是有史以来任何公司单日最大的市值增长。
And that's the largest market cap gain by any company in history ever.
这份财报的核心内容是，Azure和微软云在本季度实现了43%的增长。
The core parts of this earnings report was that Azure and Microsoft's cloud grew 43% in the quarter they report on.
他们还承诺，明年不会出现负的自由现金流。
And they promised that they would not go negative free cash flow next year.
而且他们的债务比同行少得多。
They also have a lot less debt than their peers.
里德，让我问你一个关于微软的关键问题。
Reed, let me just give you my big question on Microsoft.
我其实不太明白他们到底在做什么。
I don't fully understand what they're doing.
我承认这一点。
I'll admit it.
他们在人工智能上的做法有点奇怪。
Their approach to AI has been kind of weird.
他们一边批评OpenAI，一边又全力押注于基础实验室的解决方案。
OpenAI, criticize OpenAI, be all about your solution to the problems of the foundational labs.
即使OpenAI承诺斥资两千五百亿美元向他们采购算力，我认为这在很大程度上推动了他们的云业务增长。
Even as OpenAI has made this $250,000,000,000 commitment to buy compute from them, which I think is largely what's propelling their cloud growth.
市场对此非常买账。
And the market loved it.
所以你能帮我理清一下，微软到底在搞什么吗？
So help me understand what's going on with Microsoft.
但你自己刚才也说了。
But you said it yourself.
我的意思是，现在市场最想要的不就是效率故事吗？
I mean, the market wants an efficiency story right now, right?
这正是他们愿意掏钱的原因。
You know, that's what they're buying.
而微软一直在卖这个故事。
And Microsoft's been selling that.
你不是还采访过穆斯塔法吗？
I mean, you've interviewed Mustafa.
我确实采访过穆斯塔法。
I interviewed Mustafa.
我不久前去了微软的人工智能团队，和他聊了他们那些模型。
The I went down to Microsoft AI not too long ago and talked to him about their models.
他们正在打造这些前沿模型。
And they're they're building these these frontier.
他们最终确实想建这些前沿模型，但现在更注重效率，从头开始构建。
Well, eventually, they wanna build these frontier models, but they're focusing on efficiency and building them from scratch.
你明白吧？
You know?
他们的用户基础非常庞大。
And they have a massive install base.
对吧？
Right?
关键问题是，他们能不能顺利过渡到AI时代，把所有产品都升级成智能化的，保住企业市场？市场现在显然觉得，他们在这条路上走得很稳。
Like, they they like the, I think the big question though, is like, can they to getting their strategies, like, they actually transition through this AI phase and turn all of their products into, you know, intelligent, you know, basically hold on to that enterprise business and the market clearly thinks right now, they've got a great path toward it, you know, in that direction, right?
他们的模型效率超高。
They're super efficient models.
他们有这么多业务，现在大家最关心的就是怎么省钱。
They have all these businesses, which are the meme right now is like, all wanna save money.
他们花在令牌上的钱太多了。
They're spending too much on tokens.
如果微软能解决这个问题，那他们肯定能稳住局面。
And so if Microsoft is positioning themselves to be the answer to that question, like, they're gonna do fine.
他们会继续守住优势，这就是我的看法。
They're gonna keep they're gonna hold on to the that's how I read it.
除非我漏掉了什么，不然情况就是这么简单——简直离谱，居然还拿这个当头条。
Unless I'm missing something like that that's kinda it's a simple I mean, it's crazy that it's a that it's the record one day game.
我真没意识到这一点，直到你提出来我才反应过来。
Like, I actually didn't even realize that until you until you said it.
这市场真是让人震惊，因为根本没什么新东西，微软早就在这么做了，也一直有暗示，只是市场现在才反应过来。
That's the the the market's just it's just mind blowing to me because I'm like this is not there's nothing new here like this has been their strategy for a while now and they've been telegraphing it but I guess the markets just got the memo.
对，市场反应确实挺大的。我是说微软今年之前可以说是表现最差的大科技公司或者超大规模云服务商了，直到昨天对吧？
Yeah the market for has really been I mean Microsoft was like the worst performing hyperscaler or big tech company this year, I guess up until yesterday or yesterday, right?
市场对它的看法也是两极分化的。
And the market has sort of been of two opinions about it.
如果你看空微软，你会觉得他们押注 OpenAI 会失控，这项技术会颠覆他们的企业级业务。
It's like if you're if you're against Microsoft, you're just like the OpenAI bet is going to spiral out of control while that technology disrupts your enterprise business.
要想成功破局，情况必须反过来才行。
To thread the needle, it has to be the opposite.
肯定是OpenAI会赢。
It has to be OpenAI is going to crush it.
而且人工智能会继续快速发展，在这个过程中，你的企业业务也能持续增长。
And AI will continue to go apace and along the way you will continue to be able to grow your enterprise business.
看到市场在这两者之间摇摆不定，挺有意思的。
And it's interesting to see the market vacillate one to the other.
这又回到了我们之前聊到的利奥波德的话题，就是市场对人工智能的看法似乎每周都在变。
And it goes back to our Leopold discussion is that the way the market has seen AI shifts seemingly week to week.
这一周你还是王者，下一周就成笑柄了。
One week you're the king and one week you're the joke.
这种波动在局势明朗之前，恐怕会一直持续下去。
And this volatility is just gonna be the nature of the beast for a while until this settles out.
我完全同意这个看法。
I agree with that assessment completely.
市场波动大，是因为根本没什么扎实的依据可循。
It's volatile because it's based the market is not there's no fundamentals that they can really look at.
更多靠的是感觉，每周都在变，谁火谁凉，全看情绪。
It's based a lot on vibes, like week to week vibes, like who's up, who's down.
萨提是个出色的CEO。
If you're Sathi is a great CEO.
我觉得他做得非常好，早早看清了AI的趋势，果断押注OpenAI，直接冲进了这场竞赛的中心。
He's I think he's done a great job of, like, seeing AI, making this OpenAI bet, really, like, pulling themselves right into this into the thick of this race.
而且他没冲过头，没有过度投入。
And then, like, not getting over his skis, like, not overinvesting.
所以现在他们的债务比率实际上改善了不少。
So now they have this better this better, debt to, you know, the debt ratio essentially.
我觉得他肯定坐在那儿想：这真是太棒了。
And, like, I just think but he's gotta be sitting there going like, this is great.
我喜欢这种单日涨幅，破纪录的单日涨幅，但话说回来，这一切都会过去的，对吧？
Like, I love this one day gain, my record one day gain, but, like, this too shall pass.
没错。
Right?
他心里清楚，大家都知道，现在是高点。
Like, he know like, they all know, like, they're up now.
但过一阵子就会跌下去。
They'll be down later.
等跌了，他们也知道迟早还会再涨回来。
And when they're down, they'll you know, they know they'll be back up.
至于谷歌，我觉得他们现在稍微低了一点，但我觉得他们也都明白，孙达尔正在走同样的钢丝。
And, you know, Google, I think, has is a bit they're a bit down now, but, like, you know, I think they're all like, Sundar's threading that same needle.
他们都在走钢丝，一边是颠覆性创新，另一边是盲目押注犯大错，既要守住老业务，又明白这不会永远持续，他们都在想，萨提亚不是说过吗，我要让谷歌跳起舞来？
They're all they're all threading that needle of, like, disruption on one side, you know, like, big big, like, hasty bet mistakes on the other side, holding on to that old business while understanding that it's not forever and they have to be, you know, they're all like, what was it Satya said I'm gonna make making Google dance?
他们都在跳舞呢。
Like, they're all dancing.
对吧？
Right?
而且我，但我
And I but I
我觉得这确实是在跳舞。
think that's dancing.
是的。
Yeah.
我觉得他们都在跳舞，这太棒了。
I think that's great that they're all dancing.
我喜欢他们把积压多年的现金拿来做投资。
Like, I love that they're taking these cash pals that they were sitting on forever.
而且他们正在花这笔钱。
And they're spending that money.
我觉得这太棒了。
And I think that's awesome.
他们就该花这笔钱。
They should be spending that money.
最让人唏嘘的是，世界上最大、最有价值的科技公司，居然不知道该怎么处理手里积攒的巨额现金。
That was like the saddest thing that like the biggest tech companies in the world, the most valuable companies in world couldn't figure out what to do with all that cash.
你知道吧？
You know?
现在倒好，全拿去跳舞了。
And now they're just spending it on dancing.
没错。
Right.
不过现在他们也开始亏钱了。
But they're also going negative now.
对吧？
Right?
他们的自由现金流正在迅速减少。
That their free cash flow is evaporating.
你想想啊
You're like
这太好了。
It's great.
该花就花，借钱也行，冒点险也没关系。
Spend it, take the debt, take the risk.
这太好了。
It's great.
我们都该为此高兴才对。
We should all be celebrating it.
对吧？
Right?
这简直太棒了。
Like, this is awesome.
这才是你想要的。
This is what you want.
创新就是这样发生的。
Like, this is how innovation happens.
我们真没必要一副大难临头的样子。
Like, the you know, we shouldn't be like, oh, no.
这些公司正在大量投资于这种全新的创新技术。
Like, these companies are you know, they're they're spending a bunch of money on this new innovative technology.
不，我们都会从中受益。
It's like, no, we're all gonna win.
大家真该放松点，别紧张，这些公司又回到了初创状态，必须自我革新。
Like, just everybody needs to just chill out, like, move, you know, they're they're startups again, they have to reinvent themselves.
他们必须向前发展。
They have to move forward.
他们正面临来自初创公司和创新者的竞争压力。
They're facing competitive pressure, pressure from startups and innovators.
我们还没真正看到呢，对吧？
We have which we haven't even really seen yet, right?
AI这东西太新了，根本还没形成所谓的创造性破坏的应用层。
Like, the AI stuff is so new that like, there there isn't really even like this this like, creative destruction application layer yet.
所以没关系。
So it's fine.
一切都会慢慢好起来的。
It's all gonna work itself out.
明白了。
All right.
今天觉得这没什么大不了的，可能还有安迪·贾西。
Someone else who might think it's fine today is Andy Jassy.
亚马逊的AWS实现了37%的增长。
Amazon turned in 37% growth at AWS.
他们多年以来一直徘徊在百分之十七到十八之间。
They were hovering around 17%, 18% for years.
现在翻了一倍还多。
Now they're double that and more.
今天又涨了百分之十五。
They're up 15% today.
情况也一样吗？
Same story?
是的。
Yeah.
我觉得，我们之前其实就聊过这个。
I I think, look, we were talking about this earlier.
他们正在建造规模庞大的数据中心，这些设施在人工智能时代将变得极其宝贵。
They are building these massive data centers that are going to be incredibly valuable in this in this AI era.
而且不仅仅是数据集，他们还在搭建各种服务，未来这么多应用都会跑在AWS基于这些GPU构建的平台上。
And like, it's not just the data set, like they're also building all these services, like so much of this is going to run on these on whatever AWS is building on top of those GPUs.
所以呢，他们做的事情也真的不少。
So they're, you know, they're also it's a lot.
这些都是长期的事。
These are long term things.
有起有落是正常的。
They will be up, they will be down.
但归根结底，除非有人真搞出我们之前说的那种二十瓦的微型脑力计算机
But like, in the end, unless somebody builds that tiny little 20 watt brain computer thing that we were talking about earlier
没错。
Right.
或者冒出什么别的意外突破，不然在我看来，他们都在赢。
Or something else comes out of left field, like, it's all kinda they're all kinda winning in my mind.
对。
Yep.
好的。
Alright.
谷歌，这里有个挺有意思的故事。
Google, interesting story here.
上周，他们的云收入增长了82%，但表示打算再多花点钱。我当时就想，既然你的云收入都在以82%的速度增长，而且你也说这笔支出对推动云收入增长至关重要，还说要继续投入，现在你不仅有搜索业务，还有机会在云服务领域跟亚马逊和微软掰手腕，干嘛不放手去干呢？
Last week, they they grew cloud revenue by 82%, but said they were gonna spend a little bit more to which I was like, if you're growing your cloud revenue at 82% and you're saying the spending is fundamental to growing that cloud revenue and you say you're going to spend more and you now have not only search but a chance to be in league with Amazon and Microsoft on web services, why not do it?
结果市场立马就给了他们一记重锤。
The market punished them right afterwards.
市场的想法是，大家对这种大规模的AI投入开始感到警惕了。
And the notion was that the market had grown wary of all this AI spending.
不过，今天我可以告诉大家，谷歌已经弥补了之前的所有损失，而且赚得更多。
However, today I can report that Google's made up all of that loss and more.
现在的股价已经超过了财报发布后下跌前的水平。
They're now above where they were before the stock got hit post earnings.
这是为什么呢？
Why is that?
我觉得市场看到了微软和亚马逊的表现，然后直接套用了这个逻辑：既然它们能做到，谷歌肯定也没问题。
I think the market saw Microsoft, they saw Amazon, and they just sort of put a heuristic on and they said, well, if they can do it, Google can do it too.
我们几乎和另外两家同步上涨了。
We're gonna, it's almost up in lockstep with the other two.
或者可能是大量算法交易在操作，全是电脑在随机下注，但参与的人最多。
Or there's a bunch of algorithmic trading happening and it's all just, computers just making random bets, but it's the most people.
是的。
Yeah.
我的意思是，你可能说得对。
I mean, you could be right.
当初谷歌因为那些投资遭重创时，我当时觉得，这事儿也跟网络迷因有关。
Like, I I when when Google took that hit for for all those investments, I I thought my thinking was that there's this is, like, also meme based.
对吧？
Right?
这跟他们云业务的实际基本面根本没关系。
It's like, it's disconnected from the actual fundamentals of their cloud revenue.
真正关键的是，外界普遍觉得谷歌在ChatGPT爆发时落后了，但后来赶了上来，现在又站到了前沿。
And it's really about the fact that like, there's this sort of view out there that Google, you know, they were behind the chat GPD moment happened, they caught up, they were on top.
现在呢，那个‘Harness’事件一出，人人都在夸Claude。
And now, you know, the harness thing happened, everybody's, everybody's loving Claude.
谷歌反倒没怎么被人提起。
And Google's kind of like not as much in that conversation.
所以大家觉得它有点落后了——说实话，在编程方面，它确实慢了半年左右。
And so they're viewed as being a little behind, which I mean, on coding, they admittedly are, they're like six months behind on coding.
于是市场就认为：如果你在模型上领先，我们倒愿意接受这种投入，可你根本没领先啊。
So, so I think the market's like, yeah, like, you know, if you were on top with the models, then we would be cool with this spending, but you're not really on top with the models.
所以我们不买账。
So we're not.
这在我眼里完全说不通。
It's like, which is just to me, like, it's none of it makes any sense.
简直毫无逻辑可言。
It's like totally illogical.
谁在乎这些啊？
It's like, like, who cares?
其实呢，云服务这块，用谷歌云的人是越来越多了。
Like, if these are cloud services, actually, more and more people are using Google Cloud.
在人工智能时代，谷歌云其实表现挺不错的。
Like, it's actually pretty good in AI era.
不知道为啥，就是挺多人喜欢用它。
For whatever reason, people seem to like it.
这也不是什么统计数据，就是一种整体氛围吧。
I just hear don't have a This isn't a statistic, but just more like a zeitgeist thing.
现在用谷歌云的人，跟以前完全不一样了。
People are using Google Cloud in a way that they weren't before.
以前可没听说全是用AWS的。
I didn't hear it was all AWS before.
但他们是在给别的模型提供服务。
But that's they're serving other models.
不是他们自家的模型在推动这个。
That's not their models that's doing that.
在AI代理时代，这个产品非常适合人们正在开发的东西。
It's a pretty good product for what people are building in the AI agentic era.
用来托管所有应用之类的，确实挺合适的。
It's a good place to host all your apps and etc.
而且你看，
And it's like,
他们开始卖芯片了，比如TPU。
they're starting to sell chips, TPU.
对，没错。
Right, right.
TPU业务做得真不错。
The TPU business is great.
你知道吗，这也是另一个很有趣的方向，对吧？
Like, you know, that's also another fascinating one, right?
他们正在和别人谈合作，就是不想自己花钱。
Like they're, they're like making these d these partnership deals with people because they don't want to spend.
他们想把建设这些数据中心所需的巨额资本转移出去。
They want to offload some of the capital it takes like build these data centers.
他们确实在这么做。
They're, you know, they're doing that.
我不太确定。
I don't know.
但这些其实都不重要。
It's it's but it's like none of that matter.
这跟谁拥有最好的模型完全脱节了。
Like that's so disconnect from, like, who has the best models.
从这个角度看，它就跟AWS一样，但AWS并没有最好的模型。
So why is in in that sense, it's it's just like AWS, but AWS doesn't have the best models.
那为什么市场能接受AWS，却不能接受谷歌呢？
So why is the market cool with AWS, but they're not cool with Google?
我觉得这种态度其实很不理性的。
Like, it's also irrational is my point.
是的。
Yep.
好吧，也许这个更合理些，Meta因为AI成本上升，股价跌了10%。
Alright, maybe this one is more rational meta drops 10% as the AI costs increase.
我知道，你肯定又要说，建大计算机，最后一切都会好起来的。
Now I know, okay, you're gonna tell us build big computer, it's gonna be fine in the end.
但我的意思是，你根本没理解我的观点。
But the way I'm not I'm not that you're not.
不，你根本没看懂我对Meta的判断。
No, you haven't predicted me on meta.
我真的搞不懂Meta这事。
I don't get the meta thing.
我不会插嘴打断你。
I'm not gonna I'm not gonna get in your way here.
你继续说。
Go go.
我也没搞明白这个元宇宙到底怎么回事。
I didn't figure out the meta thing.
我就想，你们是想当云服务巨头吗？
I'm like, you guys want to be a hyperscaler or something?
我觉得元宇宙这事儿吧，说真的，我对社交媒体在这整个时代会变成什么样，完全没数。
Like I see meta is the thing where I think like, I don't know what's going to happen with social media in this in this whole era.
对我来说，元宇宙根本没给出一条清晰的路，他们到底怎么应对冲击？至少没看到什么真正的新点子。
Like, just like, haven't, to me, meta hasn't really shown a way through, like, how do they, like their businesses sort of getting disrupted or at least like, they don't have like, big new ideas.
他们就说，我们也在造超级计算机。
They're just like, we're building the big computer too.
可我们自己根本不知道这东西能干嘛。
And we don't really have a use for it.
那就租给SpaceX AI之类的公司吧。
So we're going to rent it out to like, like, space x AI.
但马克·扎克伯格毕竟不是埃隆·马斯克，对吧？
But like, Mark Zuckerberg is not Elon Musk, right?
他们也没有火箭，也没在造人形机器人。
And they don't have rockets and they aren't building humanoid robots.
确实没有。
They aren't.
他们的自动驾驶车队也不是路上最多的。
They don't have the biggest fleet of autonomous vehicles on the road.
说到底，他们到底在干嘛？在我看来，Meta 这个问题才是我们聊过的所有公司里最大的问号。
Like, they're not so it's like, what are they like, I just don't, to me, the meta one is the is the big, that's like the biggest question mark of all the ones that we've talked about.
你怎么看这个问题？
I don't know how you feel about that.
是的，我的意思是，电话会上那些分析师，我倒不是说他们在恳求扎克伯格，但意思就是希望他把多余的算力变成云服务。
Yeah, I mean, basically on the call, the analysts, I don't wanna say they were begging Mark Zuckerberg, but they were basically begging him to turn that excess compute into a hyperscaler.
而扎克伯格说，把所有算力都卖了，图一时利润，那太傻了。
And Zuckerberg goes, I think it would be foolish to basically just sell all the compute and take a short term profit.
可市场那边就一句话：卖了它。
And the market is just like, sell.
但说到Meta，我不确定这种说法对不对。
But with Meta, I don't know if this is right.
我并没有确凿的证据说明他们内部是怎么想的。
I have nothing to sort of say that this is the thinking inside there.
但我忍不住想，他们是不是在等，看看能不能造出一个AI伴侣。
But I can't help but wonder if they're just waiting to see if it's possible to build an AI companion.
到目前为止还没人成功，但要造出一个不会因为更新就毁掉人生活的AI伴侣——不会让人跟伴侣分手，也不会怂恿人伤害自己。
And to date it hasn't been possible yet, but to build an AI companion that won't destroy people's lives if it gets an update and won't tell people to break up with their partners or potentially harm themselves.
这正是问题所在，也解释了为什么我们还没看到情感机器人泛滥。
That has been the issue with That's kind of why I think we're not seeing the proliferation of the love bots.
OpenAI本来打算推出‘脏话模式’，但最终没发布。
OpenAI was supposed to do dirty mode, but never released that.
因为现在这太危险了。
It's because it's too dangerous right now.
但如果哪家公司真能做出一个AI伴侣，我觉得它会比Reels或TikTok更上瘾——毕竟，你等于创造了一个永远陪在你身边的数字朋友。
But if a company can figure out how to build an AI companion, which I think is going to be even stickier than Reels or TikTok, because come on, it's just like you've built a digital friend that's always there for you.
我不是说这是件好事。
I'm not saying this is a good thing.
我只是觉得，Meta 说不定在等那个时机，一旦有机会就会全力押注这条路线。
I'm just saying that potentially Meta is waiting for that opening and then is just gonna go hard on that on that route.
也许吧，但我觉得这生意不太靠谱。
Maybe, but I don't think that's a very good business though.
我就是这么想的
Like that is where I'm like
我告诉你原因。
I'll tell you why.
比如，你可以问问你的观众，是的。
Like, much you will Yeah.
比如，你可以问他们，如果让他们举手，有多少人愿意支付三倍于 ChatGPT 的费用。
Like, ask your audience, like, when you ask them to raise their hands, whether they would pay triple the chat GPD cost.
比如，你可以问问他们，愿意为 AI 伴侣花多少钱。
Like, ask them how much they would pay for an AI companion.
我觉得没那么多钱。
Like, I don't think it's that much.
Meta的商业模式主要是广告和推荐。
Well, the meta business will be ads and referrals.
现在可不会像你的伴侣说：咱们先别聊这么深了，
Now it won't be like your lover is gonna be like, why don't we pause this deep central conversation for
没错。
Exactly.
去听个赞助商Kayak的广告。
To hear from our sponsor Kayak.
没错。
Exactly.
但也许有一天你会跟它说：我真得订张机票去个地方，它就会说：好，我帮你订好了。
But maybe you'll tell it one day, I really do need a flight to somewhere and it will be like, all right, I bought it for you.
然后马达加斯加航空会给你返点，或者
Then Madagascar Air gives a kickback or
好吧，行吧。
something Okay, fine.
有人能做一个人工智能伴侣，这市场是存在的。
There's a market for someone could build an AI companion.
其实已经有公司在做这个了，而且还能赚到钱。
There are other companies doing that, by the way, and they can make some money off of it.
这根本算不了什么。
Like, it's nothing.
我觉得，跟脸书和Instagram相比，这不过是九牛一毛。
It's a drop in the bucket, I think, compared to what Facebook is and Instagram.
他们的核心业务，还有WhatsApp，都属于那一块。
Like, their their core businesses and and whatever WhatsApp, you know, fits into that.
我只是觉得，这事儿他们确实能做，但根本不算什么难题。
But I just think it's not like, yeah, they could do it, but it's like not that hard of a problem.
这问题本身也没多有意思。
It's like not that interesting of a problem.
而且我觉得，最后也赚不了多少钱，根本算不上什么大数目。
And I don't think it really makes that much money in the end, like enough money to really matter.
这个话题，咱们各持己见吧。
Agree to disagree on this one.
说得好，里德。
This is good, Reed.
你我平时在很多事上都意见一致。
You and I, we typically agree on so much.
可今天一整期，咱们居然一直有分歧。
That's we've been at odds the whole show today.
你以后不会再请我上节目了。
You're never gonna have me on.
我已经不是那个冠军了。
I'm not the champion anymore.
你会想：这人我就不想再搭理了。
You're gonna be like, I'm done with this guy.
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。