Jensen Huang's Exit Could Reshape the AI Industry Forever
打开互动全文版(中英对照 + 朗读 + 问答)→Dylan 做客本期,探讨黄仁勋若突然离场将如何颠覆英伟达、AI 融资与整个行业的权力格局。
Dylan joins to explore how a sudden Jensen Huang exit could upend Nvidia, AI funding, and the entire industry's balance of power.
我们节目上看到了一些分析反馈。想他了。我们上次和他聚已经超过一周了。
And we have seen analysis coming back on the show. Missed him. It's been over a week since we hung out with him.
这不是 Duo,是三星。
It's not a Duo. It's a Samsung.
哦,三星。活在未来啊。你用了多久了?
Oh, Samsung. Living in the future. How long have you had that?
我用折叠屏已经三四年了。
I've had foldables for three or four years.
哇。总是领先一步。
Wow. Ahead of the curve. Always.
你对 Duo 的评测是什么?那足以把你拉过去吗?你只用 iPhone Air 吗?
What is your Duo review? Is that going to be enough to pull you over? Were you only in iPhone Air?
iPhone Air。三星折叠屏。这是精英级别。
iPhone Air. Foldable Samsung. This is the elite class.
哦,两部手机。你真有品位。这简直是在投资部门和 research 部门之间筑起防火墙。
Oh, two phones. You're class. This is screaming a firewall between the investing arm and the research arm here.
一部工作用,你知道。一部工作用。一部私人用。
One for work, you know. One for work. One for personal.
好吧。这就对了。好。嗯,恐怕我得说,Jordy 和你们俩,John 和 Jordy,只有这个号码,没有那个。
Okay. There we go. Good. Well, I am afraid to say that both Jordy and you guys, John and Jordy, only have this phone number, not this one.
我们得解决这个问题。我们得解决这个问题。
We'll have to fix that. We'll have to fix that.
公平地说,我只——我在更新。嘿,我在更新我的先验。
To be fair, I only— I'm updating. Hey, I'm updating my prior.
我在更新我们对友谊的先验。我以为我一直发的是私人号码。我以为我们关系那么好。
I'm updating our priors on our friendship. I thought I was texting the personal the whole time. I thought we were like that.
这个是在这个号码上吗?好吧。注意安全。
Is this on this one? Okay. Stay safe out there.
你们会尝试预测黑天鹅吗?按定义这基本上是不可能的,但两周前我发现自己在想,真的绞尽脑汁想什么是最不可预测的异常情景,可能改变场上的游戏规则。我想我们找到了。不幸的是,我没有——
Do you guys try to predict black swans? It's effectively impossible by definition, but I found myself two weeks ago thinking, like actually trying to rack my brain on what is the most unpredictable outlier scenario that could change the game on the field. And I think we got it. Unfortunately, I didn't—
你的猜测是什么?
What were your guesses?
我不知道。我觉得很多猜测更多围绕整个贸易金融网络中的个体参与者。
I don't know. I think a lot of them were much more around individual players in the financial web of the entire trade.
嗯,大家一直在等的大事件是下一次训练运行的结果显示,除了编程之外没有新的增长点。那就是——品味是新模式。对许多白领工作来说,能坐在桌子对面和人握手确实很重要。所以我们永远不会取代律师或医生,甚至不会增强他们到那种程度——因为你甚至不需要做工作替代那件事。只是事实是,一个软件工程师一年能部署价值 10 万美元的 AI,而人们眼睛都不眨。这对营销人员、销售人员还没发生。如果发生了,收入确实会达到一万亿美元。但如果没有——
Well, the big one that everyone's been waiting for is some result from the next training run that shows that there's not another leg up beyond coding. That there is something—taste is the new mode. And for so many white collar jobs, it really does matter that you can sit across the table from someone, shake their hand. And so we're never going to replace the lawyer or the doctor or even augment them to the tune of—because you don't even need to do the job displacement thing. It's just the fact that a software engineer can deploy $100,000 worth of AI in a year without people blinking. That hasn't happened for the marketer, for the salesperson yet. And if that does happen, the revenues do go to a trillion dollars. But if it doesn't—
不过我不知道那算不算黑天鹅。那不是——
I don't know if it's a black swan, though. That's not—
不,那不是黑天鹅。那是,呃——
No, it's not a black swan. That's, uh—
我觉得那是我们可以讨论的事情,对吧?我们都有——
I think that's something we can all debate, right? We all have—
没错。不,但我想说的另一件事是,我非常确信,至少在未来一年,Jensen 不会让一家估值超过 100 亿美元的 AI 公司倒闭。就这样。就像,他不会让那种头条出现。他有能力给任何公司一轮有意义的上涨融资,让所有人留在游戏中,因为很多人想要那个头条,尽管在目前的大局中,一家 200 亿美元的公司归零也不会有太大影响。但感觉 Jensen 甚至不会容忍那种头条,因为空头会咬住它,美餐一顿。
Exactly. No, but another thing I would say is I was very confident that at least for the next year Jensen would not let a company—an AI company valued at north of $10 billion fail. Like, period. Like, he was like, I'm not going to have that headline. I have the ability to give any company a meaningful up round and keep everyone in the game because a lot of people want the headline even though in the grand scheme of things right now a $20 billion company could go to zero wouldn't really have that much impact. But it didn't feel like Jensen would even tolerate that kind of headline just because the bears would sort of sink their teeth into it and have a nice tasty meal.
是啊。
Yeah.
是啊。我是说,我觉得黑天鹅是这个,对吧?我想我们都预测过各种各样的事情。
Yeah. I mean, it's—I think the black swan is this, right? I think we've all predicted all sorts of things.
完全同意。
Totally.
嗯,即使有所有的知识、洞察和谣言。我是说,我甚至听说过一些谣言,我没太当真,但业内人说,哦对,Jensen 在 Nvidia 和某人这样做。我当时想,哇。
Um, even with all the knowledge and insight and rumors. I mean, I had even heard about rumors which I didn't put much credit to but like people in the industry like, oh yeah, Jensen does this with someone at Nvidia. I was like, wow.
Palmer 式的生活。
The Palmer life.
不,但那是我们一直在问的另一件事——预测它,对吧?你听到这种极端程度——有很多其他——这和贝索斯的故事如此不同,和盖茨的故事如此不同,和埃隆的故事如此不同。它和实际业务、商业交易、地缘政治和国内政治如此紧密地交织在一起。
No, but that's another thing we've been asking—predict it, right? And you hear to this extreme degree—like there have been plenty of other—this is so different than the Bezos story, so different than the Gates story, so different than the Elon story. Like it is so entwined with the actual business and the business dealings and geopolitics and domestic politics.
埃隆和员工睡过,但不是这种——
Elon has slept with employees, but not to this—
不是和间谍。是啊。是啊。
Not with a spy. Yeah. Yeah.
但他的免罪牌是他有大约 15 个孩子妈。所以你不能把那个归咎于——
But his get out of jail free card is that he has like 15 baby mamas. So you can't like pin that on—
是啊。而且他有更多的控制权。我想我们一直在问自己的问题是——我们在 2025 年 10 月 10 日谈过——我们特别提到了 Mega Speed。我们特别提到了 Alice Hong。我们谈过整个情况很可疑。很多细节基本上都是公开的。甚至——所有这些视频都是公开的,你知道,你可以解读这些视频,然后说,好吧,这看起来至少像是一种相当亲密的关系。
Yeah. And he has a lot more control. I think the question that we've been asking ourselves is like—we on October 10th of 2025 we talked—we specifically named Mega Speed. We specifically named Alice Hong. We talked about how the whole situation was suspect. Like a lot of the details were basically out in the open. And there was even—there was all these videos were public, you know, and you could read into the videos and be like, all right, this looks like at the very least like a pretty close relationship.
嗯。
Mhm.
我是说,其中一些视频里他的妻子和女儿在同一个房间,对吧?就是那种——
I mean, some of these videos have his wife and his daughter in the same room, right? That's the thing that's like—
是啊。是啊。是啊。是啊。
Yeah. Yeah. Yeah. Yeah.
这有点意味着基本分析是,好吧,如果全家都在那里,可能什么都没发生。也许这纯粹是——也许他纯粹是理性的经济行为者。他尽可能激进地玩,以在国际上发展业务。从芯片出口的立场,我们不喜欢那样。如果你是對华鹰派,你不喜欢那样。但法律条文在那里,我们很少——CEO 不会因为把法律条文推到极限而被解雇。所以如果你卖绝对最大值,比如 H20,那是为了绕过芯片出口而设计的,很多对华鹰派不喜欢,但他们说,啊,那是法律条文。你找到了漏洞。你找到了边缘。所以这完全是一个黑天鹅,因为它比任何人——
Which kind of meant the basic analysis was like, okay, well, if the whole family's there, probably nothing's happening. Like maybe this is purely—maybe he's the purely rational economic actor. He's playing it as aggressively as you can to grow the business internationally. We don't like that from a chip export position. If you're a China hawk, you don't like that. But there is the letter of the law, and we rarely—like CEOs don't get fired for pushing it right up to the letter of the law. And so if you're selling the absolute max, like the H20, which was designed to get around the chip exports, like a lot of China hawks didn't like that, but they were like, ah, that's the letter of the law. You found the loophole. You found the edge. And so it is just a complete black swan in that it's so much further than what anyone—
所以总的来说,完全缺乏判断力,考虑到这家伙朋友那么多,他有同样多——似乎同样多的敌人,能从这件事中受益。
So overall, total lack of judgment, just given that this guy, as many friends as he has, he has an equal—seemingly an equal amount of enemies that stand to benefit from this.
你觉得 Nvidia 的 AI 央行交易管道会怎么样?我猜如果你是这些公司中的一员,你不能指望——哦对,Jensen 想投 20 亿——而现在看来,这些投资等到真正签文件的时候,可能都要走委员会流程了。
What do you think happens to the Nvidia AI central bank deal pipeline? I assume if you're a number of these companies you can't lean on—oh yeah, Jensen wanted to do two billion—and now it seems like those investments, when it comes time to actually sign the docs, will probably be going to committee.
是的,我认为这是 Nvidia 最大的担忧:现在是一群人在委员会里做决策,这只会让决策变慢。从技术上讲,他们可能因此落后。显然有三个主要方面。技术方面,Nvidia 深深嵌入决策过程,没有一个明确的技术负责人。也许 Jonah 是,但还有很多直接向 Jensen 汇报的人在做技术工作,所以很难快速行动。供应链方面,Nvidia 想采取激进措施,锁定供应,说服供应链做他们从未做过的事情——在技术和供应上,把供应锁定给 Nvidia。考虑到无论谁当 CEO 可能都不如 Jensen 有魅力,他们不会说‘当然,Nvidia,你可以拿走我 60% 或 70% 的产能,价格还低于市场价。’这大概很难。第三个方面是,Nvidia 在支撑供应链上的很多东西,对吧?没有 Jensen,美国的开源就不会存在。Reflection 和 Thinking Machines 都在推动美国的开源前沿,两者都获得了 Nvidia 的大量投资和支持,生态系统中许多其他公司也是如此。所有这些应用公司都依赖开放模型——无论是 Harvey、Instinct、Sierra,它们都用开放模型。所有推理提供商,那些市值超过 100 亿的公司,都依赖中国的开放模型。如果我们现在更严厉地打击中国——因为限制措施的最大捍卫者是 Jensen 和他的团队——现在的情况是,鉴于他们的信誉扫地,被调查、被传唤等等,那么会有更多的限制。这样一来,中国模型还会以当前速度继续进步吗?如果它们继续进步,它们会把所有东西都开源吗?这些都是现实问题。我认为最大的输家肯定是开源生态系统。
Yeah, I think that's the biggest worry for Nvidia: now it's a committee of people doing stuff, and that's just going to make decisions slow. Technically they could fall behind because of this. Obviously there are three main things. There's the technical side—Nvidia is so ingrained in helping make decisions, there's no one clear technical lead. I mean, maybe Jonah is, but then there are a lot of people who are also directly reporting to Jensen doing technical stuff, so it's hard to move as fast. Supply chain wise, Nvidia wants to do aggressive things, lock up supply, convince the supply chain to do things they've never done before in terms of technology and supply, and lock it up to Nvidia. That's probably a lot harder to do given whoever the CEO is is probably not as charismatic as Jensen, and they're not going to be like, 'Yeah, sure, Nvidia, you can have 60% or 70% of my capacity at a price below market rate.' That's just probably tough. And then the third leg is that Nvidia is propping up a lot of things in the supply chain, right? Open source in America would not exist without Jensen. Both Reflection and Thinking Machines are pushing the open-source frontier in America, and both have massive investments and backing from Nvidia, as do many other companies across the ecosystem. And all these application companies rely on open models—whether it's Harvey, Instinct, Sierra, they all use open models. All the inference providers, which are multiple of the 10-plus billion dollar companies, all rely on open models from China. And if we're slamming the hammer down on China way harder now—because the biggest defender against restrictions was Jensen and his team—what happens now is given their credibility is out the window and they're being investigated and subpoenaed and whatever is going on on that side, then there are going to be a lot more restrictions. In which case, do Chinese models continue to progress at this current pace? And if they do continue to progress, do they open source everything? These are real questions out there. I think the biggest loser is the open source ecosystem for sure.
你如何看待 Nvidia CEO 这个职位的技术角色?你看到 Jensen,听到他做交易,看到他在活动、播客和新闻上露面。我觉得如果有一些深度的技术工作在进行,那都是在幕后。这有多真实?他参与下一代芯片的架构设计吗,还是他的技术角色更多是招聘和寻找下一代伟大的技术人才——比如激励,毫无疑问是激励人们,并把他们带进来?
How do you think about the technical role of the CEO position at Nvidia? You see Jensen, you hear about the deal making, you see him at events and on podcasts and news hits. I think if there's some deeply technical work happening, that's behind the scenes. How real is that? Is he involved in the architecture of the next chip, or is his technical role more recruiting and finding the next generation of great technical talent—like inspiring, no doubt inspiring people, and bringing them in?
他不设计架构,但当有决策时,我想所有关注 Nvidia 的人最终都会看到的一个常见现象是:你会听到关于选项 A、B、C 的真实有效的信息,Nvidia 可能同时在做这三件事,然后在最后一刻,Jensen 把所有力量都压在一个选项上。
He's not designing the architecture, but when there's a decision, I think the common thing that everyone following Nvidia ends up seeing is that you hear real valid things about option A, B, and C, and Nvidia could be doing all three at once, and then at the last second, Jensen is pushing all the weight behind one.
明白了。那么,如果有 A、B、C、D,选哪个?是委员会决定吗?谁来决定现在选 A,并让所有在 B 和 C 上的供应商不生气, actually 只是推动 A?
Got it. Now, if there's A, B, C, and D, which one gets chosen? Is it a committee? Who's deciding who's now chosen A and getting all the suppliers on B and C to not be pissed off and actually just push behind A?
哦对,这就是创始人模式那回事。
Oh yeah, that's the founder mode thing.
如果 A 启动了,你瞬间切换到 B。
And if A starts up, you switch to B on a dime.
那真的很难。这些事情。
That's really hard. These things.
是的。是的。我记得在 Meta 的某个时候,他们的虚拟现实战略有两个不同的操作系统。他们有 XR 和 AR 和 VR 之类的。就是做了很多额外的工作,然后扎克不得不进入创始人模式,说看,会有一些人被重新分配。会有一些裁员,会很激进,这就是创始人模式公司的好处。那么,潜在继任者的短名单上有谁?Colette 不能做吗?是她不想要吗?从这里会怎么发展?
Yeah. Yeah. I remember in Meta at one point they had two different operating systems for their virtual reality strategy. They had XR and AR and VR or something. And like it was just so much extra work being done and then Zuck had to go in founder mode and be like look there's going to be some people that are getting reassigned. There's going to be some layoffs like it's going to be aggressive and and that's the benefit of the founder mode company. So uh who is on the short list for potential successors? Can Colette not do it? Are there is that is she not want it? How does this develop from here?
我不确定 Colette 想要什么。我认为她很棒,但归根结底,她无法在技术上领导公司。我认为内部最合适的两个人是 Ian Buck 和 Jonah Alben。我认为 Ian 很棒。我认为 Jonah 不太在公众视野中,但他在技术上最接近领导者,技术上负责大多数技术组织。所以我认为从那个角度看,这两个人是最重要的。但你知道,是的,Ian 有魅力,可能很擅长周旋,但永远不如 Jensen,那么他在技术方面会一样好吗?在供应链方面会一样好吗?或者 Jonah,他在供应链方面会好吗?他擅长周旋吗?所以,我不知道是否有人能被推上来并以那种方式接替 Jensen 的角色。如果你这样做,比如 Jonah 或 Ian 现在负责,他们之前负责的事情会不会在某种程度上垮掉,比如他们投入的专注、不懈的努力。所以现在我最喜欢的 CEO 人选是从外部看。会是 Charlie Kawwas 和 Johny Srouji。这两个人——Charlie 是博通的二号人物,已经很久了,非常棒。Johny 是苹果的硬件负责人,他基本上领导了芯片团队和战略。我认为世界上只有这两个人能驾驭供应链和技术细节。我认为显然——如果是 Hock Tan 或 Lisa Su 会更好,但我不认为他们俩会做这个。我想他们只是——
I'm not sure what Colette wants. I think she's awesome, but at the end of the day, she cannot lead technically the company. I think internally the two people that I think would be best suited would be Ian Buck and Jonah Alben. I think that Ian is great. I think Jonah is not so much in the public light, but he's technically like the closest to the leader, and is technically in charge of most of the technical orgs. So I think those are the two that are the most important from that perspective. But like you know, yes, Ian is charismatic, probably really pretty good at wheeling and dealing, not nearly as good as Jensen will ever be, but then is he going to be as good on the technical side? Is he going to be as good on the supply chain side? Or Jonah, is he going to be good on the supply chain side? Is he going to be good on the wheeling and dealing? And so, you know, I don't know if someone can be propped up and take over the role of Jensen like in that way. And if you do, like if Jonah is now or Ian is now in charge, do the stuff that they were in charge of fall down to some extent, like in terms of the focus that they had on it, the relentless effort. So now my favorite picks for CEO would be from an external perspective. It would be Charlie Kawwas and Johny Srouji. These two guys—Charlie is number two at Broadcom and has been for a long time and is awesome. And Johny is the head of hardware at Apple and he's the one who's basically led the chips team and strategy. Those are the only two people I think in the world that can navigate a supply chain and technical details like that. I think obviously there are—it'd be better if it was Hock Tan or Lisa Su but I don't think either of them will do this. I think they're just like—
是啊,那 Trainium 组织、TPU 组织的领导者呢?我的意思是,你强调了苹果。有很多人领导着巨大的半导体项目,但他们仍然比真正的 CEO 或真正的创始人低两级。
Yeah, what about leaders from the Trainium organization, TPU organization? I mean, you're highlighting Apple. There's plenty of people that are leading huge semiconductor efforts, but they're still two rungs down from the true CEO or the true founder.
安迪·贾西是想收购 Annapurna Labs 的那个人。嗯,他不会离开亚马逊 CEO 的职位去当英伟达 CEO。你得退几步。
Andy Jassy is the one who wanted to buy Annapurna Labs. Well, he's not going to leave Amazon CEO to be Nvidia CEO. You take some steps down.
我觉得那里有非常优秀的人。我真的很喜欢亚马逊的团队。他们工作节奏超快。我觉得谷歌很棒,但我不认为他们有一个能胜任所有这些事的人。我的意思是,如果说有的话,谷歌的 Amin Vahdat 是所有这些基础设施的负责人,但我从没问过他。我只是有种感觉,他不会想做这个,对吧?我感觉 Trainium 的领导者们,我知道他们非常想待在幕后,在亚马逊所有芯片事务中,我不认为他们想做这个。
I think there are really great people there. I really love the teams at Amazon. They work and move super fast. I think Google's awesome, but I don't think they have a single person that can do all of the above. I mean, if anything, Amin Vahdat is at Google, he's like the lead of all this infra stuff, but I've never asked him. I just have a feeling he wouldn't want to do this, right? I have a feeling that the leaders of Trainium, I know they very much want to be in the background, and of all the Amazon chip stuff, I don't think they want to do this.
是啊。那黑马呢?肖恩·马奎尔提到了纳什·奥罗拉,一个已经在 CEO 位置上但没经营万亿美元公司的人。你有什么谦虚的期望吗,就几千亿?
Yeah. What about a wild card? Sean Maguire was throwing out Nash Aurora, someone who's already in the CEO seat but not running a trillion-dollar company. Do you have any humble expectations, just a couple hundred billion?
小公司,对吧?我不太了解 Nikesh。我见过他一次。我觉得他很棒,但严格来说,Palo Alto 已经交付过硬件,对吧?那绝对是他们业务的一部分,但那和英伟达在前沿做疯狂的事情是一回事吗?看看 Blackwell,它太棒了。如果他们没在 Blackwell 上做那些事,他们会失去大量市场份额。他们在最后一刻决定了要做什么。在 Volta 上,他们在流片前大约 3 个月决定加入 Tensor Core。谁会说过,‘是啊,我们就推出当前的东西。让我们把它加到当前的东西里。’他们会说,‘哦,让我们把它加到下一个东西里。’如果他们把它加到下一个东西里,他们在 AI 上会落后多少?英伟达永远不会建立主导地位,对吧?正是这些技术——而且这不像你在对抗——
Tiny companies, right? I don't know Nikesh too well. I've met him once. I think he's awesome, but technically Palo Alto has shipped hardware, right? That's definitely a part of their business, but is it the same thing as Nvidia leading edge doing crazy stuff? Look at Blackwell, it was amazing. If they did not do what they did with Blackwell, they would have lost a lot of market share. They decided at the last second what to do. With Volta, they decided to add the Tensor Core like 3 months before they taped it out. Who would have said, 'Yeah, let's just push out the current thing. Let's add it into the current thing.' They would have been like, 'Oh, let's add it to the next thing.' If they add it to the next thing, how much further behind would they have been in AI? Nvidia never would have established dominance, right? It's these technical—and it's not like you're fighting—
不不不,不是贬低 Palo Alto Networks,但他们的竞争对手没那么厉害,而英伟达是在对抗 OpenAI 芯片团队、博通芯片团队、谷歌芯片团队、Anthropic 芯片团队、Meta 芯片团队。你知道,这些家伙,亚马逊芯片团队,这些人都很厉害。AMD,他们都很强。还有所有这些初创公司,对吧?初创公司,他们都很棒。
No no no knocks against Palo Alto Networks, but their competitors are not that amazing versus Nvidia is fighting an OpenAI chip team, a Broadcom chip team, a Google chip team, an Anthropic chip team, a Meta chip team. You know, these guys, Amazon chip team, these guys are good. AMD, they're all really good. And all these startups, right? The startups, they're all good.
并不是说他们无能,你知道吗?我觉得这非常不同,对吧?莉莎·苏再厉害,她打败的是无能的英特尔,对吧?那个定义了一个品类、创造了它并抵御所有入侵者的人是什么样?这和做第二名或做出聪明的战术商业决策、在品类已经建立后创新非常不同。就是非常不同。
It's not like they're incompetent, you know? I think it's very different, right? As great as Lisa Su is, she took down Intel who was incompetent, right? What is the person who defined a category and created it and kept all the invaders at bay? It's very different than being second place or making smart tactical business decisions, innovating on something once there's already a category established. It's just very different.
你认为中国实验室正在如何反应和更新他们的路线图?
How do you think Chinese labs are reacting and updating their roadmaps?
中国实验室与英伟达合作非常紧密,这是公开的秘密。即使英伟达不直接向他们出售芯片,他们也有来自美国以及上海办公室的工程师不断前往这些实验室,与他们交谈,帮助他们优化,了解他们在做什么,并将这些优化带到英伟达的软件栈和开源中。我认为中国实验室已经开始更新了,对吧?像被制裁的智谱和 DeepSeek 这样的公司是不同类型芯片的重度用户。智谱支持大约七种不同的中国加速器。它们不如英伟达的,但嘿,它们能用。DeepSeek 与华为关系非常密切。可能有很多关于先验的更新。那些严重依赖英伟达的中国实验室现在可能比那些已经大力多元化的中国实验室更弱。我认为这同样适用于 Anthropic 与 OpenAI。显然 OpenAI 有自己的芯片并且研发时间更长,但他们 90% 以上是英伟达,而 Anthropic 只有三分之一或更少的英伟达。所以对他们影响的程度完全不同。OpenAI 与英伟达的合作比 Anthropic 更紧密。嘿,这对 Anthropic 相对于 OpenAI 来说是个积极因素。Anthropic 抵制开源,而开源现在更弱了,这对 Anthropic 来说很棒,对吧?这类事情不仅影响中国实验室,也影响美国实验室。闭源正在获胜。
It's an open secret that the Chinese labs are working super closely with Nvidia. Even if Nvidia is not selling them chips directly, they have engineers constantly from both the US as well as their Shanghai offices going to these labs, talking to them, helping them optimize, understanding what they're doing and bringing those optimizations over to Nvidia software stack and the open source. I think the Chinese labs have already started updating, right? Companies like Zhipu who are sanctioned and DeepSeek are heavy users of different types of chips. Zhipu supports like seven different Chinese accelerators. They're not as good as Nvidia's, but hey, they work. DeepSeek is super close with Huawei. There are a lot of probably updates on priors. And it's the Chinese labs that relied heavily on Nvidia probably are weaker now than the Chinese labs that relied heavily already diversifying. I think the same applies to Anthropic versus OpenAI. Obviously OpenAI has their own chip and has been working on it longer, but they're 90 plus percent Nvidia, whereas Anthropic is like one-third or less Nvidia. And so it's just a completely different level of impact on them. OpenAI works more closely with Nvidia than Anthropic does. Hey, this is a positive for Anthropic versus OpenAI. Anthropic pushing back open source and open source being weaker now, this is great for Anthropic, right? These sorts of things flow through not just the Chinese labs but also American labs in general. The closed source is winning.
是啊。是啊,这很有道理。鉴于基本上任何类型的与中国业务关系都将面临更严格的审查,你对这将如何影响 Mercor 和各种数据公司有什么预测吗?
Yeah. Yeah, that makes a lot of sense. Any predictions on how this will impact the Mercors and various data companies given that there's going to be more scrutiny on basically any type of business relationship with China?
我不太了解 Mercor 的具体情况,但我确实知道很多数据公司向中国出售大量数据。实际上,这是一种非常有趣的蒸馏方式。即使中国实验室不直接拥有闭源模型,一些最大的客户是所有数据公司,比如 Mercor 等等。所以他们生成数据,而他们生成数据的流程涉及许多来自前沿实验室的语言模型,以及提取和过滤等等。如果你用这些前沿模型制作环境或其他东西,然后卖给中国,那在某种程度上你就是在从它们中蒸馏信息。这不像传统的蒸馏。我想向中国出售芯片将面临更严格的审查。向中国出租芯片和云服务将面临更严格的审查。出售数据尚未受到任何审查,它可能会落入灰色地带。现在人们关注硬件。但在英伟达,Jensen 和所有为——Lutnick 和 David Sacks 以及所有那些某种程度上阻止对英伟达向中国出口的一些限制的人——现在在任何允许中国获得一些东西的论点中,尤其是 Lutnick 和 Jensen,都严重信誉扫地。多年来我们一直支持某些类型的限制。这是我们与英伟达最大的争议点。我想他们肯定——
I don't know about Mercor specifically, but I do know a lot of the data companies sell tons of data to China. Actually, it's a really funny way to distill. Even if the Chinese labs don't directly have the closed source models, some of the biggest customers are all the data companies like Mercor and so on and so forth. And so they generate data and their pipelines for generating data involve a lot of language models from the frontier labs and extracting and filtering and all that. And if you're making environments or whatever you're making with these frontier models and you sell them to China, that's just in some way you're kind of distilling the info out of them. It's not like traditional distillation. I imagine selling chips to China is going to be under a lot more scrutiny. Renting chips to China and clouds is going to be under a lot more scrutiny. Selling data is something that has not had any scrutiny yet and it will probably fall into the guise. Right now people focus on hardware. But there's—in Jensen and all the people in Nvidia that were defending—Lutnick and David Sacks and all these people that were sort of preventing some of the restrictions on Nvidia into China are heavily—especially Lutnick and Jensen are heavily discredited now in any argument to allow China to have some stuff. We have been pro certain types of restrictions for years. It's been our biggest point of contention with Nvidia. I think they sure—
他们真的很不喜欢那个。
They really disliked that.
你知道,当我们对 AMD 或 Google 感到兴奋时,他们不理解,他们把这看作是我们可以在工程上超越他们。但我认为当涉及中国时,情况就不同了,对吧?现在掩护已经没了。我希望政府在限制措施上能明智一些,对吧?确保某些类型的芯片不被出售,真正铁腕。确保台积电和内存公司不向中国出售。确保高端设备不流向中国用于制造芯片。也许对数据有一些限制,但不要限制芯片租赁,对吧?有一整套瀑布式的限制,或者从超大规模云厂商那里租赁芯片。也许限制像 Mega Speed 这样的随机芯片公司向中国租赁芯片,或者重新转移。但也许不是那种,嘿,如果 Google 要租给字节跳动,也许这有道理,因为 Google 总能在需要时或政府告诉他们时收回芯片,而且 Google 总能向字节跳动收取更高的利润,这样字节跳动的利润就低于 Meta。你知道,这些才是对美国竞争和领导力相关的事情。我希望所有限制不是那种——显然钟摆会摆动,会有很多鲁莽的行动,但希望焦点是如何让美国赢。
You know, they didn't understand when we were excited about AMD or Google and they viewed that as we can out-engineer them. But I think when it came to China, it was a different aspect, right? And now the cover is gone. I kind of hope that the government is smart with the restrictions, right? Make sure that chips don't get sold of certain kinds, really ironclad. Make sure TSMC and memory companies don't sell to China. Make sure high-end equipment is not going to China that makes chips. Maybe some restrictions around data, but don't restrict renting chips, right? There's sort of a whole waterfall, or renting chips from, let's say, hyperscalers. Maybe restrict random chip companies like Mega Speed from renting chips to China, or re-diverting. But maybe not the like, hey, if Google's going to rent to ByteDance, maybe that makes sense because Google can always pull away the chips when they need to, or if the government tells them to, and Google can always charge way more margin to ByteDance, and so that makes ByteDance less profitable than Meta. You know, these are the sort of things that are relevant for US competition and leadership. And that's what I hopefully all restrictions are not like, obviously the pendulum will swing and there will be a lot of brash actions, but hopefully the focus is on how do you make America win.
是的,我喜欢。我记得你把自己标榜为不是对华鹰派,而是白头海雕,我觉得这个说法很好。你如何反思《AI 2027》中的预测?
Yeah. No, I like it. I remember when you branded yourself not as a China hawk, but as a bald eagle, and I thought that was a good way to put it. How are you reflecting on the predictions in AI 2027?
你们做的视频太好了。我们得转发一下。
The video you guys made is so good. We got to repost that.
是的,我们今天得转发。你如何反思《AI 2027》中的预测?当我看到模型能力时,我觉得它非常准确。我们完全在正轨上。向那里的整个预测团队表示热烈祝贺。我仍然对两个预测有点不确定。一个是他们预测会有 10 万人的反 AI 抗议。中期选举有点像是那样。所以我认为中期选举可能会稍微推迟一些。但更有趣的问题是,就在这个时候,《AI 2027》预测在 2026 年 11 月左右中国会觉醒,我一直在质疑这一点。很难获得准确数据。你觉得我们在中国觉醒这条线上进展如何?这是中国觉醒的一个例子吗,因为他们在对我们最强大的 AI CEO 设置蜜罐,还是他们仍在沉睡?
Yeah, we got to repost that today. How are you reflecting on the predictions in AI 2027? When I look at the model capabilities, I feel like it's incredibly accurate. We're completely on track. Huge congratulations to the whole forecasting team there. I'm still a little uncertain about two predictions. One is they forecasted that there would be a 100,000 person anti-AI protest. The midterms were kind of that. So I think the midterms might kind of stave off a little bit of that. But the more interesting question is right around this time AI 2027 predicts that in around November of 2026 China wakes up and I've been sort of interrogating that. It's very hard to get accurate data. How do you feel we are progressing along the line of like China waking up? Is this an example of China waking up because they're honeypotting our most powerful AI CEO or are they still asleep?
我们必须——总有一种可能性,他们认为英伟达芯片有后门,哦,他们不想让它们在那里。
We have to—there's always a possibility that they believe Nvidia chips have back doors in them and oh, they don't want them there.
而黄仁勋是——黄仁勋是资产。
And Jensen is Jensen's the asset.
是的,有一百万个,但就像如果你说嘿香港有一堆 GPU 但让它们留在那里,或者我们不想让它们在这里,那看起来不像觉醒。
Yeah, there's a million but like it doesn't seem like you're waking up if you're like hey there's a bunch of GPUs in Hong Kong but keep them there. Or we don't want them in here.
说得好。是的。这如何让你认为有某个群体有更长的时间线?
That's a good point. Yeah. How does that make you think that there's some group that has much longer timelines?
是的,就《AI 2027》而言,我的意思是,我非常相信 AI,我认为《AI 2027》是一个很大的跨度,但嘿,结果他们今年到目前为止预测的可验证的事情中有 70% 是准确的,这很不可思议,对吧?就像哇。是的。我仍然对某些扩散方面持怀疑态度,但我认为 RSI 看起来越来越真实。中国觉醒可能是一个很大的可能性。这是另一件事,对吧?如果我们过于激进地摆动禁止中国东西的钟摆,中国现在会不会摆动他们的钟摆并阻止供应链?你知道,每台英伟达服务器都有中国内容。
Yeah, I would say as far as AI 2027, I mean, I'm very AI-pilled and I thought AI 2027 was a big stretch, but hey, turns out 70% of the verifiably things that they predicted so far this year have turned out to be accurate, which is incredible, right? Like wow. Yeah. I'm still skeptical on some of the diffusion aspects, but I think RSI is looking more and more true. China waking up is probably a big likelihood. That's the other thing, right? If we go too gung-ho on the pendulum of banning stuff in China, does China now swing their pendulum and prevent the supply chain? It's, you know, every Nvidia server has China content.
当然。
Sure.
每台 Google TPU 服务器都有中国内容。就像每一件东西都有中国内容。中国可以随时停止。当然。所以有一定程度的,实际上就像请你知道,你能给中国在 AI 和其他地方的最少东西是多少,以防止他们切断供应链。这是一个可以解释的世界观。就中国觉醒而言,这可能是可能的,对吧?我不——你提到的另一个,比如中国对黄仁勋设置蜜罐。我实际上不认为她是口语意义上的间谍。是的。我也不认为她是口语意义上的蜜罐。她在中国商界和家庭中是一个人脉广泛的人。她的家庭人脉广泛。如果你在中国人脉广泛,如果你在美国人脉广泛,当地情报机构告诉你时你会和他们交谈,你会告诉他们想听的话。我只是不认为人们会像,你知道,不,或者像,你知道,那里只是有很多压力。
Every Google TPU server has China content. Like every one of these things has China content. China can stop it if they want. Sure. So there's some level of like actually like please you know what is the amount of stuff that you could minimal amount of stuff you can give China in AI and give them in everywhere else sort of to prevent them from cutting off the supply chain. Is a view of the world that could be construed. As far as like you know China waking up it's probably possible right? I don't— Another one that you mentioned like China honeypotting Jensen. I actually don't think she's a spy in the colloquial sense. Yeah. And I don't think she's a honeypot in the colloquial sense. She is a well-connected individual in the Chinese business world and family. Her family is well connected. And if you're well-connected in China, if you're well connected in America, you will talk to intelligence agencies when they tell you to and you will tell them what they want to hear. I just don't think people are going to be like, you know, no or like, you know, there's just a lot of pressure there.
是的。CIA 向我要你的个人电话号码,我不得不让他滚开。太残酷了。
Yeah. The CIA was asking me for your personal phone number and I had to tell him to take a hike. It was brutal.
我们得弄清楚他从哪里得到那件夹克。约翰,你能给我们迪伦·帕特尔的个人电话号码吗?我说,看,我在他的世界里有点二把手。
We got to figure out where he got that jacket. John, can you get us Dylan Patel's personal phone number? And I said, look, I'm kind of second fiddle in his world.
是的。只是职业关系。
Yeah. Just a professional connection.
个人和工作电话直到几个月前。
Personal and work phone up until a few months ago.
好吧。好吧。我有点不那么受伤了。
Okay. Okay. I'm a little less hurt.
不。但无论如何,我认为只是——他们在获取一些信息,对吧?她显然在微信上向所有人展示她与黄仁勋关系密切,可以在中国拿到芯片。黄仁勋,他知道吗?可能。我的意思是,我不知道,但可能他知道。但像 Super Micro 不需要对黄仁勋设置蜜罐就能向中国运送数十亿美元的服务器,对吧?就像我认为还有更多。
No. But anyways, I think that just be—they were getting some info, right? She was clearly showing everyone on WeChat she is close to Jensen and can get them chips in China. Jensen, did he know? Probably. I mean, I don't know, but probably he knew. But like Super Micro didn't have to honeypot Jensen to ship billions of dollars of servers to China, right? Like this is like I think there's a lot more.
还没有出来。他们可能有过婚外情。我们不知道。
It still hasn't come out. They might have been having an affair. We don't know.
我们不要跳进——
Let's not jump into—
我确定。我确定。
I'm sure. I'm sure.
无论如何。这对风险投资界意味着什么?我们看到很多风投聚集在黄仁勋周围。这纯粹是因为开源生态系统吗?这对其他——我的意思是你提到了说唱公司,但风投实际上是否退出了大型实验室交易,所以这就是为什么他们比预期更多地聚集在黄仁勋周围。你知道,在我回答那个问题之前,你知道,这会是——黄仁勋有——这取决于这些事情有多糟糕,对吧?有很多信息出来。我不知道。
Anyway. What does this mean for the venture capital community? We've seen a lot of VCs sort of rallying around Jensen. Is that purely because of the open source ecosystem? Like what does it mean for the other—I mean you mentioned rapper companies but are VCs effectively out of the big lab trade and so that's why they're rallying around Jensen a little bit more than you expect. You know, before I answer that question, you know, it'd be—Jensen has the—it depends on how bad this stuff is, right? There's a lot of info coming out. I don't know.
但如果中国那边真的热火朝天,万一 Jensen 直接去中国,搞个千亿美元的融资呢?我不知道。但那样的话,他为什么不去台湾呢?
But if the China stuff is really gung-ho, what if Jensen just goes to China and starts a hundred billion dollar raise? I don't know. But why would he not go to Taiwan in that situation?
因为美国在台湾仍有利益,我想我会这么说。我的意思是,这取决于他是否违法了,对吧?他会不会被叫到国会面前,连续几个月被斥责违法?我不知道。如果他没有违法,我想他会去圣巴特岛。
Because the US still has interest in Taiwan, I guess I would say. I mean, it depends on whether he broke laws, right? Is he going to be called in front of Congress for months and months and berated for breaking laws? I don't know. If he didn't break laws, I think he's going to St. Barts.
不管怎样,就风险投资生态系统而言,你说得完全对。他们退出了实验室交易。实验室们正在上市。至少 OpenAI 可能刚完成了最后一轮融资。那轮融资并不怎么样。然后 Anthropic 也要上市。他们获得了流动性。他们中的一些会。但接下来所有下游的,对吧?比如所有这些推理云,如果没有信贷、没有融资、没有担保,他们怎么确保算力?这是一个真正的挑战。
Anyways, as far as the venture ecosystem, you're totally right. They're out of the lab trade. The labs are going public. At least OpenAI probably just did their last round. It wasn't that great. And then Anthropic is going public. They got liquidity. Some of them will. But then everything downstream, right? Like all of these inference clouds, how can they secure compute without credit and without financing and without a backstop? This is a real challenge.
是的。谁愿意这么做呢?然后 B10、Fireworks 以及其他公司,如果没有算力,没有中国模型持续更新并对外销售,他们怎么从 10 亿美元营收增长到 100 亿美元?这是一个巨大的挑战和风险。同样,像 Instinct、Harvey 和 Sierra 这些做出很棒业务的公司也是如此。他们能继续提升和改进产品能力吗?还是必须转向闭源模型?如果是后者,如果他们必须使用闭源模型,他们的产品就只是个包装。但如果他们能切换到开源模型并拥有客户关系,他们就不是包装。所以我认为这是一个非常非常大的担忧。我认为 Jensen 是风险投资生态系统最强大的盟友,他支撑了很多公司。英伟达在风险投资和投资方面非常活跃。所以我不知道他们现在是否还会那么活跃。
Yeah. Who's going to be willing to do that? And then how do B10 and Fireworks and whoever go from a billion dollars of revenue to 10 billion of revenue without access to compute and without Chinese models being continually updated and sold externally? This is a big challenge and risk. Likewise with all these companies like Instinct and Harvey and Sierra who are making awesome businesses. Can they continue to ramp and improve the capabilities of their product, or do they have to turn to closed source models? In which case, if they have to use closed source models, their product is a wrapper. But if they can switch to open source models and they own the customer relationship, they're not a wrapper. And so I think this is a big, big worry. I think Jensen is the venture ecosystem's strongest ally and he's propped up a lot of companies. Nvidia is very active in venture and investing. So I don't know if they're going to be as active now.
我还有一个最后的问题。Jordy,你有什么要问的吗?
I had one last question. Jordy, do you have anything?
我想,呃,什么... 是的。
I think, uh, what a... Yeah.
是的。抱歉。
Yeah. Sorry.
我告诉你我的电话号码。6789 8212。
I'll tell you my phone number. 6789 8212.
等等,那是什么来着?那是... 哦,得了吧。那是佐治亚州的地方广告还是什么。
Wait, what is that from? Is that... Oh, come on. That's a regional commercial from Georgia or something.
不。不。
No. No.
那是什么?
What is that?
那是... 是那首歌吗?有一首关于它的歌,对吧?
It's a... Is it that song? There's a song about it, right?
在洛杉矶,我们只知道 do accident two。
In LA, we only know do accident two.
好吧。我知道。867。Soulja Boy。
Okay. I know. 867. Soulja Boy.
哦,是的。是的。Soulja Boy,他现在是 AI 代言人。他雇起来真便宜。真的吗?
Oh, yeah. Yeah. Soulja Boy, who's now an AI spokesperson. He is so cheap to hire. Really?
我上了一艘船,某个初创公司为了他们的系列发布还是什么。他们租了一艘船,花了 2.5 万美元雇了 Soulja Boy。
I went on a boat, some startup for their series, whatever launched. They rented a boat, hired Soulja Boy 25K.
哇。也许第二天吧。
Wow. Maybe at the next day.
他表演了。他表演了大约一个小时。他带了整个团队和他自己的 DJ 来。
And he performed. He performed like an hour long. He had his whole entourage and his own DJ come.
我们需要他参加 2027 年的 SemiAnalysis 大会。嗯,说到... 我们不是一家亏损的风险投资支持的公司,但是... 呃,GDP 能... 你对进展的温和跳跃感到惊讶吗?我看到了 5.2。是 Mythos 5.2 16%。我的意思是,这说明了基准测试的质量,它没有立即达到 100%,但请带我了解一下你对那里结果的一些反应。
We need him at SemiAnalysis Con 2027. Um, speaking of... We're not a venture-backed company that loses money, but... Uh, GDP can... Were you surprised by the moderate jump in progress? I saw the 5.2. It was Mythos 5.2 16%. I mean, it speaks to the quality of the benchmark that it wasn't immediately 100%, but take me through a little bit of your reaction to the results there.
是的。是的。所以,GDPX 和我们的基准测试,我们推出了... 我们测试了 Anthropic 的模型并推出了它,运行测试花了一两天,但更像是两天。但不管怎样,它在某些领域有了巨大的改进:软件工程和芯片设计。而且有些领域根本没有改进,这真的很有趣。我的意思是,这说明强化学习环境是 Scaling(规模扩张)的设定向量。如果你在某些领域没有独特的数据和环境,你就无法在这些地方扩展模型的质量。所以芯片设计大幅改进,对吧?我认为这说明了 Anthropic 的重点是什么?嗯,我们刚雇了一堆芯片人才,我们正在制造芯片,我们正在让模型比芯片更好。过去,Anthropic 的观点是我們不打算制造芯片,因为它不是 AGI 药丸。等我们造出芯片时,就太慢了。现在模型可以制造芯片了。所以我认为他们真的在加速这方面。同样,软件工程继续疯狂增长。Anthropic 在软件工程和芯片设计方面比其他人好太多了。太不可思议了。然后其他领域他们根本没有改进。我认为这说明数据的重要性有多大。以及这些数据公司和数据供应商。但还有这些类别的持久性,对吧?其中一些类别他们根本没有改进,他们可能没有在这些类别上花费大量数据,或者基准测试太难了。它在所有领域都有轻微改进,但一般来说,Scaling(规模扩张)规模不会让你在每个基准测试中都得到改进。你真的必须扩展技术、算法和数据。
Yeah. Yeah. So, GDPX and our benchmark, we launched... we tested Anthropic's model and launched it, and it took a day or two to run the tests, but more like two days. But anyways, it was such a big improvement in some areas: software engineering and chip design. And there are areas where it just didn't improve at all, which was really interesting. I mean, it speaks to reinforcement learning environments are the set vector for scaling. If you do not have unique data and environments in certain spaces, you will not scale quality of model in those places. And so chip design drastically improved, right? And I think it speaks to what is Anthropic's focus? Well, we just hired a bunch of chip people and we're making a chip and we're making model way better than a chip. And in the past, Anthropic's view was we're not going to make a chip because it's not AGI pill. By the time we get a chip out, it's too slow. Now the model can make the chip. And so I think they're really accelerating on that front. Likewise, software engineering continues to gain like mad. Anthropic is so much better than everyone else in software engineering and chip design. It's incredible. And then other areas they just didn't improve. And I think it speaks to the volume of how important data is. And these data companies and these data vendors. But also the durability of some of these categories, right? Some of these categories they didn't improve on at all and they're probably not spending a ton of data on, or maybe the benchmark is just so difficult. It did improve slightly in all domains, but generally scaling size is not going to get you improvement in every benchmark. You really have to scale techniques, algorithms, and data.
在数据方面,有多少... 数据飞轮的价值有多大,即通过 harness 直接免费提供给公司、实验室的数据,与专门从数据经纪人那里购买的数据相比?你对每种数据的价值有平衡吗?
On the data side, how much of it is... how much value is there in the data flywheel, the data that comes through the harness directly just given to the company, to the lab for free, versus bought from a data broker specifically? Do you have a balance on how valuable each is?
是的,所以我们与许多芯片设计数据公司和 AI 芯片设计公司建立了关系。据我所知,Anthropic 买的很少。好吧。他们一直在自己构建。是的。这是唯一的答案,对吧?他们还能怎么做?但与此同时,在一些领域,比如生物,Anthropic 建立了自己的湿实验室,他们在那里收集数据,但也在其他地方购买大量数据。所以我认为其他人在内部构建什么、外部构建什么方面有不同的策略。但我认为数据供应商非常重要。而且鉴于模型变得如此之好,一个通用的大规模数据供应商,把你和一群所谓的数据标注员连接起来,那并不是有价值的,对吧?是专家。是专家。所以如果你是一名材料科学专家,你在那里制作数据,你可以做出很棒的生意。但那有多可扩展呢?
Yeah, so we've established a relationship with a lot of the chip design data companies and AI chip design companies. Anthropic, as far as I understand, buys very little of this. Okay. They've been building it themselves. Yeah. Is that the only answer, right? How else would they do this? But at the same time, some of the spaces like bio, Anthropic started their own wet lab and they're collecting data there, but also they're buying a lot of data elsewhere. So I think other folks have different strategies in terms of what do I build internally, what do I build externally. But I think the data vendors are really important. And given models are getting so good, a general large-scale data vendor where we're connecting you to a bunch of people that are like okay data labelers, that's not what's valuable, right? It's experts. It's experts. And so if you're an expert in material science and you're making data there, you can make a great business. But how scalable is that?
完全同意。完全同意。是的。好吧,下次你上节目时我们会知道的。非常感谢你来参加。
Totally. Totally. Yeah. Well, we'll find out next time you're on the show. Thank you so much for coming on.
我们就不打扰你工作了。祝你顺利。
We'll let you get back to work. Have a good one.
真是疯狂的一周。
What a week.
再见。真是疯狂的一周。足以载入史册的一周。祝你过得愉快
Goodbye. What a week. One for the history books. Have a great