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。
我认为我们将在十年内实现人工超级智能。
I think we will have artificial superintelligence in ten years.
我的意思是，一种在每个领域都比最聪明的人类还要聪明的AI。
And by that, I mean an AI that is smarter than the smartest human in every discipline.
我认为，您无疑是全球顶尖的半导体和半导体供应链专家之一。
You are, I would argue, one of the world's premier experts on semiconductors and the semiconductor supply chain.
现在，美国政府肯定会以各种方式礼貌地表示：嘿。
Now you have the US government, who I'm sure in many, many ways is going to politely say, hey.
我们知道，您更喜欢台湾生产的巧克力蛋糕，而不是英特尔在俄勒冈州和亚利桑那州生产的巧克力蛋糕。
We know that you like the, chocolate cakes made in Taiwan more than the chocolate cakes made in Oregon and Arizona by Intel.
我们非常希望您能购买这种美国产的巧克力蛋糕。
We'd really like you to buy that American chocolate cake.
在阿联酋，让我作为一名美国人感到极其尴尬的是，您在阿布扎比作为犹太人可能比在大多数美国城市还要安全。
In The UAE, and one thing that was, like, profoundly embarrassing for me as an American to realize is that you might be safer being Jewish in Abu Dhabi than you are most American cities.
我认为世界将依靠太阳能运行。
I think the world's gonna run on sunlight.
我是一名坚定的环保主义者。
I'm a committed environmentalist.
你知道，我喜欢在户外活动，但我真的相信全球变暖已经解决了这个问题。
You know, I love to be outdoors, but I genuinely believe global war amigas has solved the problem.
我深信，如果没有X，没有人会相信10月7日真的发生过。
I profoundly believe if it were not for x, nobody would believe that October 7 really happened.
我同意。
I agree.
今天我邀请到了加文·贝克，他是我多年的朋友，也是Atreides的杰出投资者，同时也是对半导体供应链和人工智能未来最了解的人。
I'm joined today by Gavin Baker, a longtime friend and incredible investor at Atreides and the most knowledgeable person on the semiconductor supply chain and the future of AI.
你绝对不能错过这一期。
You do not wanna miss this episode.
我们还在查理·柯克遇刺后的第二天录制了这一期节目。
We also recorded this episode on the day after the assassination of Charlie Kirk.
我想听听加文对这一事件的政治影响、以及社交媒体和主流媒体中正在发生的一些事情的看法。
I wanna hear Gavin's thoughts on the political implications of that, some of what's going on in social media and the mainstream media as well.
请听。
Listen in.
欢迎回到《Invested》的另一期节目。
Welcome back to another episode of invested.
我很高兴能邀请到我的老朋友加文·贝克。
I am thrilled to have my old friend, Gavin Baker.
欢迎，加文。
Welcome, Gavin.
谢谢。
Thank you.
谢谢。
Thank you.
非常高兴能来到这里。
So excited to be here.
欢迎来到特拉维夫。
Welcome to Tel Aviv.
哦，是的。
Oh, yeah.
太棒了。
It's awesome.
加文和我是在一个奇怪的情况下认识的。
Gavin and I met in a strange situation.
我们俩在一架从旧金山湾区飞往拉斯维加斯的飞机上经历了剧烈颠簸，当时我们正在谈论股市，这不仅促成了真正的友谊，还让我们和共同的朋友比尔·盖利走到了一起。
We both had our stomachs fall out on a on a plane that went from the Bay Area to Las Vegas that had a little drop in the middle of it, and we were talking about the stock market at the time that launched a a real friendship along with our mutual friend Bill Gurley.
没错。
Absolutely.
这是一段美好的关系。
And it's been a wonderful relationship.
那已经是十多年前的事了。
It's more than ten years ago.
确实已经十多年了。
It is more than ten years ago.
是的。
Yeah.
对于那些不了解的人，加文可以说是半导体供应链中最聪明的人之一，我认为他也是我们当前所经历的AI世界中最敏锐、有时又最愤世嫉俗的观察者。
And Gavin, for those who don't know, is just about the smartest guy in the semiconductor supply chain, and I think both the most astute and sometimes cynical observer of the world of AI we're going through right now.
所以，我早早就采纳了你的一句话。
And so I adopted a line from you early on.
我们一起在阿布扎比参加一个会议时，你上台说，基础模型是历史上贬值最快的资产。
We're at a conference together in Abu Dhabi where you got up and said that the foundation models is the fastest depreciating asset in history.
你到现在还这么认为吗？
You still think that's true?
嗯，我认为原话其实是，首先我们应该把功劳归于比尔的合伙人埃里克·菲什亚，我可是毫不客气地剽窃了这句话。
Well, so I think the exact line was well, first we should give credit to Bill's partner Eric Fischia, who I ruthlessly stole that line from.
但我稍作修改，说没有独特数据和互联网规模分发能力的基础模型，是历史上贬值最快的资产。
But I did modify it and say foundation models without access to unique data and Internet scale distribution are the fastest depreciating assets in history.
我的观点略有改变，我认为XAI、Anthropic、OpenAI和DeepMind已经拉开了差距，它们各自在不同程度上拥有独特数据，其中谷歌和XAI可能拥有最多的独特数据。
I would say my opinion has changed a little bit and I do think XAI, Anthropic, OpenAI, DeepMind they've they've opened up a lead and to varying degrees they all have with probably Google and XAI having the most unique data.
我认为 Anthropic 很聪明地早早专注于代码，而 OpenAI 则因为率先布局而实现了互联网规模的分发。
I think Anthropic very smartly focused on code early and then OpenAI got to Internet scale distribution just because they were there first.
但我确实认为，从根本上说，推理模型极大地提升了基础模型的商业价值。
But I do think fundamentally reasoning models have made the business of foundation models dramatically better.
原因是，对于每一个价值上千亿美元的互联网公司，每一个伟大的互联网公司来说。
And the reason is is that for every trillion dollar plus Internet company, every great Internet company.
我太喜欢这个随口一提了。
I love that throwaway.
对于每一个价值上千亿美元的互联网公司，这种事情两年前每天都在发生。
For every trillion dollar plus Internet company like this happened every day two years ago.
不。
No.
百分之百。
A 100%.
是的。
Yeah.
对于每一个主导的互联网公司，都存在一个支撑它的相同飞轮。
For every dominant Internet company, there's the same flywheel that underpins it.
这个飞轮就是，当产品变得更好时，你会获得更多的用户。
And that flywheel is that as the product gets better, you get more users.
用户会生成更多的数据。
The users create more data.
增量数据使产品变得更优，从而吸引更多用户。
The incremental data makes the product better, which attracts more users.
这个飞轮每天都在谷歌的搜索业务、Netflix、亚马逊及其推荐引擎、Facebook每天的广告定向算法以及用于选择向你展示内容的算法中运转。
And that flywheel spins at Google every day in the search business, Netflix every day, and Amazon every day and their recommendation engines, at Facebook every day, both in their ad targeting algorithm and the algorithm they use to select what content to show you.
直到最近，基础模型公司、大语言模型中才完全不存在这种循环。
That loop was not present at all in foundation model companies, LLMs, until very recently.
因此，拥有独特数据的大语言模型之所以具有优势，一方面是因为长期来看，如果你能使用别人无法获取的数据进行训练，就会获得优势。
And so the reason that LLMs with unique data had an advantage was one it would presumably lead to over time an advantage in training if you could train on data that no one else could.
其次，如果你能访问类似X或谷歌对世界的全部知识，就可以用这些信息让大语言模型的回答更牢固地立足于现实。
And then secondarily, if you had access to something like X or everything Google knows about the world, can use that to ground the LLM's answers more firmly in reality.
但推理从根本上改变了这一局面，通过在训练后和测试时引入具有可验证奖励的强化学习。
But reasoning fundamentally changed this with the introduction of kind of reinforcement learning with verifiable rewards during post training and test time compute.
突然间，拥有庞大的用户群体成为巨大优势，并以之前从未有过的方式反馈到产品品质的提升上——直到OpenAI在2024年底推出GPT-1，DeepSeek迅速跟进推出R1，随后XAI、谷歌DeepMind和Anthropic也相继推出了自己的最新模型。
All of a sudden having a giant user base becomes a big advantage and feeds back into the quality of the product in a way that was simply not present until OpenAI came out with a one in late twenty four, DeepSea quickly followed with R1, and then XAI, Google DeepMind, and Anthropic quickly followed with their own recent models.
因此，我认为在引入推理能力后，这些大规模基础模型公司成为更好的投资标的，因为如今在AI领域，‘用户越多，数据越多，产品越好’的飞轮效应终于实现了，而在此之前，至少对于大语言模型而言，这种效应根本不存在。
So I think that these scaled foundation model companies are much better investments post reasoning because this flywheel where the more users you have, the more data you have, the better the product it is, now exists in AI where it simply did not exist before, at least for LLMs.
当你说到这是更好的投资时，我想起你之前提到的那些万亿美元级公司。
And when you say it's a better investment, I go back to the comment you said before about these trillion dollar companies.
对吧？
Right?
我想你当时说的是，我们有哪些公司？
And I think it was, you know, what do we got?
英伟达、苹果、Meta、微软、谷歌。
NVIDIA, Apple, Meta, Microsoft, Google.
我有没有漏掉谁？
Am I missing anyone?
博通。
Broadcom.
现在是博通了。
Broadcom now.
是的。
Yeah.
突然间，这变成了一个万亿俱乐部。
Like, it's all a sudden become a club, the the trillion dollar club.
你看看Anthropic和OpenAI，你会觉得，哦，这些显然是要成为万亿美金公司的。
And you look at these anthropic open AI and you go, oh, those are obviously gonna be trillion dollar companies.
顺便说一下，今天我跟两个人聊了，他们告诉我，在最近的Anthropic融资轮中投入了超过一亿美元。
I sat today with somebody, by the way, to to two people who told me they put up over a $100,000,000 in the recent anthropic round.
我说，哇，这是在押注万亿美金的前景。
And I said, boy, betting on the trillion dollar outcome.
我停顿了一下，说：押注万亿美金的前景。
I stopped myself for a second and said, betting on the trillion dollar outcome.
我做这一行已经三十年了。
I've been doing this thirty years.
这种事没发生过几次。
There haven't been that many of those.
是啊。
Yeah.
嗯，
Well,
我只想说，看到Grok Code发布并占据OpenRouter上编码令牌的绝对主导份额，真是令人着迷。
I will just say I do think it's been fascinating to see Grock code release and take overwhelmingly dominant share of the coding tokens on OpenRouter.
Grok Code 是一个很棒的产品。
Grok Code is a great product.
但没错，这些数字确实很大，只是有点滑稽。
But, yeah, these are big numbers, but it's just funny.
你知道，Anthropic，那是什么来着？
You know, Anthropic, what was that?
1800亿？
180,000,000,000?
没错。
Yep.
我的意思是，他们在短短六个月里，从10亿增长到60亿，速度比历史上任何公司都快。
I mean, they went from 1,000,000,000 to 6,000,000,000 faster than any company in history in essentially six months.
所以突然间，你的月收入变成了原来的30倍。
So all of a sudden, you're 30 times current month's revenue.
我认为那里确实存在一些公司结构问题。
I think there are real corporate structure concerns there.
比如在2018年制定治理文件、微软进行投资时，人们还认为通用人工智能是个幻想。
Like in 2018 when the governing documents were written and Microsoft made its investment, people thought AGI was a fantasy.
而一旦宣布实现通用人工智能，所有来自OpenAI的未来技术都将归属非营利组织。
And as soon as AGI is declared, the all of kind of the future that technology and future technology from OpenAI goes to the nonprofit.
但无论如何，OpenAI的年收入已经达到了上百亿美元。
But nonetheless, I mean, OpenAI has, you know, 10 plus billion dollars revenue.
XAI虽然不公开，但正在迅速扩张。
XAI, it's not public, but they're scaling rapidly.
所以现在，你可以根据收入倍数来评估这些公司的价值。
So these companies now all a sudden you can value them on revenue multiples.
一旦你建立了以推理为核心的产品闭环，我认为很明显，这将形成一个难以打破的寡头垄断格局。
Now that you have that core product loop driven by reasoning in place, I think it's pretty clear that it's going to be an oligopoly that's going to be tough to break into.
但我想，你问题的核心可能是：这对现有的万亿级以上公司而言，是属于持续性创新还是颠覆性创新？
But I think maybe the heart of your question is to what degree is this a sustaining or disruptive innovation for the existing trillion dollar plus companies?
我认为答案必须是——我不想耍聪明或回避问题，它两者都是。
I think the answer has to be I mean, not to be clever or avoid the question is it's both.
我想在之前的交流中，我提到过，在人工智能出现之前，这些巨型科技公司各自都有自己的领域。
So in I think maybe in in in a prior interaction, I said, you know, pre AI, each one of these mega cap tech companies was kind of in its own swim lane.
比如奈飞，虽然和亚马逊、谷歌、苹果有些竞争，但奈飞依然占据主导地位。
Like Netflix, they competed a little bit with Amazon and Google and Apple, but, you know, Netflix was pretty dominant.
谷歌在搜索领域占据主导地位。
Google, dominant in search.
Meta，在社交领域占据主导地位。
Meta, dominant in social.
微软，在生产力领域占据主导地位。
Microsoft, dominant in productivity.
然后你还有谷歌、微软和亚马逊这三大云计算寡头。
And then you have this cloud computing oligopoly of of Google, Microsoft, and Amazon.
而且，没有人真正能在电子商务领域与亚马逊竞争。
And, you know, nobody's really competing with Amazon in in ecommerce.
在AI、LLM和Transformer出现之后，某种程度上，它们现在都处于同一场比赛中，因为很明显，Transformer能够胜任这些公司核心业务的几乎所有工作——你可以制作个性化电影、个性化歌曲。
Post AI, post LLMs, post transformers, at some level, they're now all in the same race because I think it's very clear that the transformer is gonna be able to do essentially everything that lies at the core of each of those companies for you're gonna be able to make personalized movies, personalized songs.
你将拥有一个能够为你完成所有购物的智能代理。
You're gonna have an agent that can do all of your shopping.
你知道，履约基础设施仍然具有价值。
You know, the fulfillment infrastructure will still be valuable.
它们显然也能进行搜索。
You're they can obviously do search.
然后我认为这还是一个开放性问题。
And then I think it's an open question.
在一个AI能够生成看似出自人类之手的完美内容的世界里，人类创作的内容价值何在？
What is the value of human created content in a world where AI can create perfect content that looks like it's from humans?
所以，他们其实都在同一场游戏中，但与此同时，成功的核心要素是算力和数据。
And so, you know, they're kind of all in the same game, but at the same time, the core ingredients to success are compute and data.
而这些公司拥有最多的算力和最多的数据。
And these companies have the most compute and the most data.
因此，尽管这一变革极具颠覆性且令人恐惧，但它也具有延续性，因为它契合了它们的一些优势。
So while it's very disruptive and very scary, it's also sustaining and that it plays to some of their advantages.
如果你能吸引到合适的人才，谷歌就做到了这一点。
If you can attract the right talent, Google was able to do that.
亚马逊并没有真正参与这场游戏，但它们投资了Anthropic，而后者无疑拥有顶尖人才。
Amazon, not really in the game, but they bet on Anthropic, which for sure has the right talent.
XAI已经成功做到了这一点，但苹果甚至都没有尝试过。
XAI has been able to do that, but Apple has not even tried.
到目前为止，Meta基本上失败了。
And Meta has until now essentially failed.
我们看看他们的策略第三版或第四版是否能奏效。
We'll see if, you know, v three or v four of the strategy works.
自由球员市场是有效的。
The free the free agency market works.
自由球员市场。
The free agency market.
你知道，她非常幽默。
You know, she's very funny.
我曾在一家大型金融机构工作了很久，我认为大多数大公司最终都会意识到，这非常困难。
Like, I worked for a long time at a large financial institution, and I think most big companies, they eventually learn that it's pretty hard.
你培养内部人才，却引进昂贵的外部人才，这会让忠诚的内部员工感到被冷落和沮丧，你肯定在Meta看到了这种情况。
You have a culture of growing your own talent to bring in expensive external talent because it really disenfranchises and upsets the loyal internal employees, and you've for sure seen that at Meta.
他们有很多优势，但这也可以说是他们一生中最艰难的斗争。
They have a lot of advantages, but it's also kind of the fight of their lives.
这就是为什么我认为他们全都陷入了一个囚徒困境，你知道，资本支出将攀升到难以想象的水平。
And that's why I do think they're all they're in a prisoner's dilemma where, you know, the amount of CapEx is going to a to an unimaginable place.
我觉得，你可以看到这些公司先发行债务，然后取消股息，再停止股票回购，因为他们都感到，在这场所谓的大型语言模型竞赛中获胜是关乎生死存亡的。
Like, I think, you know, I think you could see some of these companies, you know, first issue debt, then cancel their dividends, then stop the buybacks because they all feel that winning in this, we'll say, winning in LLMs is existential.
他们认为，另一端的奖赏是人工通用智能，而他们几乎都怀有近乎宗教般的信念，认为它将创造数万亿美元的经济价值。
And they feel like the the prize on the other side is artificial superintelligence, which they all have a near religious belief is gonna create tens of trillions of dollars of economic value.
是近乎宗教般的信念，还是完全宗教般的信念？
Near religious or fully religious belief?
也许介于两者之间。
Maybe even somewhere in between.
是的。
Yeah.
我的意思是，以当前的发展轨迹来看，我认为这有助于实现人工通用智能。
I mean, I you know, the current trajectory the current trajectory, I think, is supportive of getting to artificial superintelligence.
但我认为，这一切中最有趣的一点是，我们根本不知道超级智能的经济回报会是多少，因为从定义上讲，我们从未见过它。
But I think one of the most interesting parts of all of this is we don't know what the economic returns to superintelligence will be definitionally because we have never seen it.
如果人类已经通过像阿尔伯特·爱因斯坦这样的人将自然法则的极限推到了极致，
And if if humans, we have pushed the limits with, you know, people like Albert Einstein or whoever else, push the limits of the natural laws of the universe.
如果我们已经将物理学、生物学、化学以及支撑这一切的数学推向了极限，那么超级智能的经济回报可能并不会那么高。
If we push the limits of physics and biology and chemistry and and the math that underpins all of it, then the economic returns to superintelligence, they may not be that high.
另一方面，如果我们尚未触及极限，而超级智能正在治愈癌症、研发曲速引擎、并殖民多个恒星系统，那么超级智能的回报将极其巨大，但这完全无法预知。
On the other hand, if we haven't and superintelligences are curing cancer, invading warp drives, and we're colonizing multiple solar systems, the returns to superintelligence are gonna be really high, but it's very unknowable.
因此，我认为存在一种近乎宗教般的信念，认为我们终将实现超级智能，这种信念在数据上是有依据的；但超级智能的经济回报，从定义上讲是不可知的，而正是这一点，我认为才真正体现了某种宗教般的信仰。
So I think there's a I would say there's a quasi religious belief that we're going to get to superintelligence that I think is well founded in data, but the economic returns to that superintelligence is to me definitionally unknowable, and that is where I think there is, you know, for sure religious belief.
我一直在关注的一件事是，当你在风险投资行业成长起来时，无论是2000年，还是本世纪二十年代初，软银的孙正义都出现了，他将资本武器化了。
One of the things I I I keep looking at is as you grow up in the venture business in the year 2000 and then again, you know, in the in the late teens and early twenties of this of this century, you know, came Masa from SoftBank and he weaponized capital.
他试图将资本武器化。
He tried to weaponize.
他试图将资本武器化。
He tried to weaponize capital.
当你把资本武器化时，最终就会演变成一场你死我活的较量，就像你曾经在Lyft和Uber之间看到的那样，直到Uber胜出，但他们在私募市场中真的想要这种对决。
When you weaponize capital, you end up with this fight to the death like, yeah, which you knew were in Lyft for a while until until Uber won it, but they really want it in the private markets.
所以，他们在私募市场筹集了不正当的巨额资金来实现这一点。
So, you know, they'd raise untoward amounts of capital in the private markets to do it.
现在我们正陷入一场激烈的竞争，多家上市公司都在追逐同样的目标。
Now we're in this incredible race where you've got multiple public companies chasing the same bacon.
当然是犹太洁食培根，但追逐的还是同样的目标。
Kosher bacon, of course, but chasing the same bacon.
还有一些私营公司也在做同样的事情，它们都在争夺所谓同一个奖品。
And a couple of private companies doing the same thing, And they're all fighting over what's, I guess, notionally the same prize.
是的。
Yes.
Anthropica 专注于代码，而 OpenAI 是面向消费者的。
Anthropica kind of specializes in code and OpenAI is the consumer.
但从根本上说，它们追求的是同一个目标。
But but foundationally, they're after the same the same prize.
这会如何收场？
How does that end?
而且正如你所说，他们都怀有这种宗教般的信念，各自筹集了数百亿美元的资金。
And to your point with they all have this religious belief, they're raise hundreds of billions of dollars, maybe each.
对吧？
Right?
正如我们所相信的OpenAI的演示文稿那样，虽然我不确定是否真的如此，为了达到这个目标。
As we believe OpenAI's deck, which I'm not sure I do, to to get to this.
那这已经带来了回报吗？
And and has that ended the return?
我的意思是
I mean
到目前为止，实际上有一家著名的风险投资公司，我们在这档播客中不会提及它的名字，一家著名的美国风险投资公司。
Well, so to date, there has actually been a storied venture capital firm whose name we will not mention on this podcast, a storied American venture capital firm.
他们的一位合伙人写了一篇题为《2000亿美元的问题》和《6000亿美元的问题》的文章。
One of their partners wrote an essay called the $200,000,000,000 question and then the $600,000,000,000 question.
这篇文章的核心观点是：第一个2000亿美元的问题是，这2000亿美元的GPU支出将从何处获得经济回报？
And the point of the essay was the first $200,000,000,000 question is where's the economic return on this 200,000,000,000 in GPU spend gonna be?
那么，6000亿美元的投资回报在哪里？
And then where's the return on the 600,000,000,000?
对我来说，有趣的是，有时候我觉得，风险投资家们完全沉浸在自己的风险投资世界里。
And the funny thing to me about that, you know, sometimes I think, you know, VCs, they're they're they're they're very in their own world of venture.
对我来说，有趣的是，除了X AI之外，所有在GPU上花费最多的公司都是上市公司。
The funny thing to me about that was all of the biggest spenders on GPUs with the exception of x AI are public companies.
是的。
Yes.
而上市公司会公布一种叫做季度财务报告的东西，它们需要提交一份名为10-Q的文件，这份文件必须准确无误。
And public companies report something called quarterly financials and they file a document called the 10 Q, which has to be accurate.
你可以利用这些文件来计算一种叫做投资资本回报率的东西。
And you can use those documents to calculate something called a return on invested capital.
这其实很容易计算。
It's really easy to do.
到目前为止，所有这些支出的投资回报率都显著为正。
And to date, the ROI on all of this spending has been strongly positive.
是的。
Yes.
其中一部分你可以说，他们在差不多同一时间开始重视成本控制。
And some of that you could argue was that they got religion about expenses at around the same time.
但在人工智能的世界里，从本质上讲，经济回报大多将来自用GPU替代或增强支付流程。
But in a world of AI, fundamentally, the economic return, a lot of it will come from either replacing or augmenting payments with GPUs.
劳动力替代。
Labor replacement.
或增强。
Or augmentation.
我们走着瞧。
We will see.
但最激进的观点认为，肯定是劳动力替代。
But the maximalist, I think, sure would say labor replacement.
所以我认为这是一种合理的计算方式。
So I think that's a fair way to calculate it.
而在过去六个月里，这一点变得尤为明显，你开始看到收入的显著加速。
And then I would just say it's become really clear over the next over the last six months in particular, you started to see big accelerations in the top line.
这其中很大一部分是通过使用大语言模型、人工智能和GPU来改进支撑这些公司业务的推荐算法。
And a lot of that has been in using LLMs and AI and GPUs to improve the recommender AI algorithms that power a lot of these companies.
但即便如此，尽管回报不错，问题在于下一代NVIDIA GPU——Blackwell的支出规模如此巨大，这些500亿到1000亿美元的数据中心，ROI还能继续保持良好吗？
But nonetheless, I mean, the returns have been good, but just the question is the quantum of spend with Blackwell, which is the next generation NVIDIA GPU, is so big, you know, these 50 to $100,000,000,000 data centers, just will the ROI continue to be to be good?
因此，对于2000亿美元的问题，答案是肯定的，ROI非常出色。
So while the answer to the $200,000,000,000 question was yes, the ROI was awesome.
对于6000亿美元的问题，答案同样是肯定的，ROI一直非常出色。
The answer to the $600,000,000,000 question was yes, the ROI has been awesome.
现在我们面临的是万亿美元级别的问题，而ROI实际上依然保持良好。
Now we're at the trillion dollar question and the ROI has actually still been good.
我们正迅速迈向2万亿到3万亿美元的规模。
We're rapidly heading for 2 to 3,000,000,000,000.
那么，ROI还能继续维持良好吗？
And will the ROI continue to be good?
未知。
Unknown.
如果不是，是经济规律决定的，还是囚徒困境决定的？
And if it's not, do economics dictate, or does the prisoner's dilemma dictate?
是的。
Yeah.
那些优秀公司可以持续投入。
That's a great companies could keep spending.
这是个很好的问题。
That is a great question.
我想说，我认为你是全球顶尖的半导体和半导体供应链专家之一，对此有极其深刻的理解。
I wanna think, you know, you are, I I would argue, one of the world's premier experts on semiconductors and the semiconductor supply chain and understand this incredibly deeply.
如今主导的架构是GPU。
So the dominant architecture today is the GPU.
这种状况会持续下去吗？
Does that continue?
我认为这非常困难。
I think it is very hard.
要取代GPU总体上将非常困难。
It will it's gonna be very hard to unseat GPUs in general.
特别是英伟达，我认为我有幸投资过的两位最杰出的CEO就是詹森和埃隆。
And and I would say NVIDIA in particular, I think the two most exceptional CEOs I've ever had have the privilege of, I guess, investing in are Jensen and Elon.
我刚在富达工作时，作为一名半导体分析师才六个月，英伟达就是我最好的投资建议。
I was when I was a very young man at Fidelity, six months into being a semiconductor analyst, NVIDIA was my best idea.
当时它的市值可能只有20亿美元。
It was maybe a $2,000,000,000 market cap.
因此富达持有英伟达15%的股份，而作为一名年轻分析师，我有幸每周与詹森交谈一两个小时，因为我当时是最大的股东。
So Fidelity owned 15% of NVIDIA and I had the privilege as a very young man of speaking to Jensen essentially an hour or two every week because I was the largest shareholder.
稍后在2012年，我管理的基金成为特斯拉的最大股东，那时特斯拉市值还很小，但他们确实是极其杰出的CEO。
And then a little bit later in 2012, the fund I managed became the largest shareholder of Tesla at a small market cap, but they're exceptional, exceptional CEOs.
詹森状态绝佳，非常专注。
And Jensen is super on his game, super dialed in.
他能如此接近所有实验室，这是一个巨大的优势。
It's a big advantage to be as close to all of the labs as he is.
但我认为问题在于，英伟达今天的市值，我想说是4万亿美元，或者说就是4万亿美元。
But I think the issue is that NVIDIA's market cap today is I'm gonna say, is it 4,000,000,000,000 or I'm gonna say it's 4,000,000,000,000.
我的意思是，它是4万亿美元，4.5万亿。
I mean, it's 4,000,000,000,000, 4 and a half trillion.
假设GPU的市值是4.3万亿美元。
And, like, let's just say that the GPU market cap 4,300,000,000,000.0.
太惊人了。
Amazing.
GPU的市值，如果展望三到四年，我们称之为加速器的市值将达到10万亿美元。
That the GPU market cap, maybe if you look out three or four years, our we'll call it accelerator market cap is is 10,000,000,000,000.
到那时，我认为你确实会开始看到一些细分领域出现。
And at that point, I do think you start to see niches develop.
所以你永远不可能在正面交锋中击败英伟达。
And so you're never going to beat NVIDIA in a head on assault.
尤其是当你有AMD这样一家卓越的公司，它们在制造相当不错的芯片，虽然可能在系统和软件方面略逊于NVIDIA，但要超越NVIDIA真的非常困难。
And that's what and particularly when you have AMD, who's an exceptional company, you know, making pretty good chips, maybe lacking some of the systems and software that NVIDIA has, but it's gonna be very hard to out NVIDIA NVIDIA.
我认为，这正是这些超大规模云服务商内部ASIC项目试图实现的目标。
And that is what I think a lot of the internal ASIC programs at these hyperscalers are trying to do.
我觉得这有点像是徒劳无功。
I think that's kind of a fool's errand.
TPU花了三代产品才取得成功。
It took the TPU three generations to be successful.
我认为，任何超大规模云服务商中，最好的芯片团队无疑是AWS团队。
I think for sure the best silicon team at any hyperscalers, the AWS team.
我认为，要让他们的Trainium芯片达到卓越水平，可能需要第二代半，甚至第三代产品。
And I think it will really take, you know, their, call it, second and a half generation of Trainium, maybe the third for that to be amazing.
但毫无疑问，这些ASIC项目最终都会取得成功。
But for sure, those are going to be successful ASIC programs.
我认为亚马逊团队里有不少人都是做这个的。
I think a bunch of them are here on the Amazon team.
当然。
Absolutely.
是的。
Yeah.
它来自安纳普尔纳。
Came comes from Annapurna.
安纳普尔纳。
Annapurna.
是的。
Yeah.
比利和纳法·巴沙拉。
Billy and Nafa Bashara.
是的。
Yeah.
那是一个了不起的团队。
So that's an incredible team.
但我觉得其他人都很难在GPU性能上超越GPU。
But I think everybody else is gonna struggle to out GPU the GPU.
但如果你能选择一个细分领域，你知道，因为在某种程度上，所有设计都是关于权衡的。
And but then if you can pick a niche, you know, because at some level, all designs are about trade offs.
就像在坦克战中，他们谈论铁三角。
Like in in tank warfare, they talk about the iron triangle.
你必须在进攻、防御和机动性之间做出权衡。
You have to make a trade off between offense, defense, and mobility.
由于每个人都使用相同的原材料，你强化防御的同时，机动性就会下降，因为坦克变得更重了。
And because everybody is dealing with the same raw materials, the the stronger you make the defense, the less mobile it is because the more the tank weighs.
所以你必须做出权衡，半导体领域也是如此。
And so you you have to make trade offs, and it's the same thing in semiconductors.
因此，这些细分领域已经变得如此庞大，我认为有机会做出与GPU不同的权衡，走一条不同的路。
And so has these niches become so big, I think there is an opportunity to make trade offs that are different from what the GPU is doing, to do something different.
然后你可以从一个只有0.5%的细分市场开始，但这已经是数十亿美元的收入和数百亿美元的市场价值，之后你就可以继续发展了。
And then you can maybe start with a niche that's 0.5%, but that's, you know, billions in revenue and tens of billions in market value, and then you can go from there.
因此，最近你开始看到我的公司——我应该说，我们是Cerebras的投资方。
And so recently, you started to see my my firm, I should say, is is an investor in Cerebras.
Cerebras、Grok，我认为还有其他几家采用了SRAM架构，这能实现与GPU根本不同的功能。
And Cerebras and Grok, and I think a couple of others have an SRAM architecture, and that can do something fundamentally different from a GPU.
在一块大芯片上。
On a big chip.
在一块大芯片上。
On a big chip.
特别是Cerebras。
Shrebras in particular.
Cerebras做了一件非常激进的事情。
Shrebras did something very radical.
他们没有把300毫米的晶圆切割成许多小芯片作为芯片的基础，而是直接使用整个晶圆。
Instead of cutting up a 300 mill 100 millimeter wafer into lots of little die and making that the basis for a chip, they use the whole wafer.
这是一种非常激进的方法，对他们来说也非常困难。
Very it's a really radical approach, and it was really hard for them.
他们花了第三代芯片才取得成功。
It took them their third generation of their chip before they saw success.
但现在他们和Grok、Giro Q——另一家SRAM公司，可能采取了没那么激进的方法——都取得了实际的商业进展。
But now they and and Grok, Giro Q, was another SRAM company, maybe less radical approach.
但我认为它们都获得了真正的商业认可，因为它们能做些不同的事情，并且能在令牌上实现极高的吞吐量。
But I think they're both seeing real commercial traction because they could do something different and that they can generate extremely high throughput on tokens.
所以每秒处理的令牌数真的很重要。
So tokens per second really matters.
但即使你使用一个推理模型，当它在思考时，实际上也是在生成令牌。
But even you use a reasoning model, you know, when it's thinking, it's literally generating tokens.
根据你使用的模型，你可以大致看到它的思维链条。
And depending on who you use, you can kind of see its chain of thought.
一个原本需要三十秒才能给出的答案，在Cerebras上只需三秒。
And, you know, an answer that might take thirty seconds takes three seconds on Cerebras.
所以这是一个细分市场。
So that's a niche.
下一代硅芯片。
Next Silicon.
我们的共同投资。
Mutual investment of ours.
我们的共同投资。
Mutual investment of ours.
他们也在做一些根本不同的事情。
They're also doing something fundamentally different.
我不太清楚NextSilicon有多少是公开的，但他们确实在做些不同的事。
I don't I don't I don't know how much is public about NextSilicon, but they're doing something different.
我知道，我看到很多，我可以选一个细分领域。
And I would you know, and I I saw a lot, and I could pick a niche.
你知道的？
You know?
选一个细分领域。
Pick a niche.
而且，你知道，我认为这是一个不争的事实：在许多半导体领域，真正成功需要经历三代产品。
And, you know, and I think it's it's a truism that in a lot of semiconductors, it takes three generations to really, really be successful.
为了明确一下，是两代，也就是说，你需要三次流片，直到第三代芯片才最终成功。
Two generations, just to be clear, therefore, three generations of tape outs, you need the third generation chip, and you finally make it.
人们不记得英伟达是一个三十年磨一剑的‘一夜成名’案例。
People don't don't remember that NVIDIA is a thirty year overnight success.
百分之百。
A 100%.
没错。
Yeah.
没人意识到这一点。
Nobody realizes.
一个三十年磨一剑的‘一夜成名’案例。
A thirty year overnight success.
贾斯汀·陈并不是一夜之间就赚到一万亿美元，或者43000亿美元的。
Trillion dollars or $4,300,000,000,000 did not come to Jensen overnight.
这是三十年的毕生使命。
It was thirty years of a life's mission.
绝对如此。
Absolutely.
而且，每个芯片的周期大约是两到三年，这真的很难。
And just in case that each chip is kind of like a two to three year cycle, and it's really hard.
把芯片取出来。
Take the chip out.
你把一个大型数字文件发送到台湾的台积电。
You send off this big digital file to Taiwan Semi in Taiwan.
你焦急地等待三到四个月，让他们制造芯片。
You anxiously wait for three or four months while they make the chip.
有没有办法
Is there a way
加快这个过程？
to speed up that process?
我的意思是，我们正在加速一切。
I mean, we're speeding up everything.
简直难以置信。
It's, like, unbelievable.
不。
No.
我觉得这非常有趣。
I do think it's very interesting.
特斯拉在德克萨斯州与三星达成了巨大的合作关系，其中一项条件是埃隆能够对晶圆厂的设计和布局产生影响，就像他重新构想电动汽车、火箭和数据中心一样。
Tesla has made this giant partnership with Samsung in Texas, and one of the conditions of that partnership was that Elon would be able to have influence over the design of the fab and the way it's laid out and in the same way he kind of reconceptualized electric vehicles and then rockets and then data centers from scratch.
我认为，几乎每一位AI架构师和数据中心架构师都普遍认为，在Hopper世代，不可能构建超过3万台的训练集群，这一点被严重低估了。
I think it's very underappreciated that, you know, the consensus belief of essentially every AI architect, data center architect was with the Hopper generation.
你无法构建超过3万台的连贯训练集群。
You could not get more than 30,000 coherent and training cluster.
他从第一性原理出发审视数据中心，做了完全不同的事情，最终实现了20万台的连贯集群。
You know, he looked at a data center from first principles, did something very different and got 200,000 coherent.
他会从第一性原理出发来看待半导体制造。
He's gonna look at semiconductor manufacturing from first principles.
所以我认为这可能会带来一些有趣的结果，但没错，制造这么一小块半导体确实很疯狂，尤其是像GPU这种并不小的芯片。
So I actually think that is something interesting may come of that, but no, it's kind of wild to make one little semiconductor, you know, which in the case of a GPU is not that small.
GPU被称为全掩模芯片，而现在是两个这样的芯片。
GPU is what's called a full reticle chip, and then now it's two of them.
想象一下，它的大小大约是iPhone的一半。
Think of it as being about half the size of an iPhone.
制造它需要四个月的时间。
That takes four months to make it.
然后你拿到芯片，那一刻紧张得不得了。
And then you get the chip back, and it's, you know, an incredibly anxious moment.
你真的会把它插上，然后祈祷它能工作。
You you literally plug it in, and you hope it works.
这叫上电，但有时候它就是不工作。
It's called power on, and sometimes it doesn't.
然后你得做一些调整，或者如果你真的搞砸了，就得从头再来。
And then you have to make some tweaks, or if it you if you really screwed up, you have to start again.
所以我不记得我们是怎么说到这个的，
So I don't I don't know how how we got on that, though,
两年。
two years.
我想从这里转个话题，因为你提到了台积电。
I I wanna segue from that because, you you mentioned TSMC.
所以，大约百分之九十
So, like, ninety percent
哦，英伟达，一夜成名。
Oh, NVIDIA, overnight success.
是啊。
Yeah.
三十年。
Thirty years.
是的。
Yeah.
所以台积电正在打造一个围绕它的完整生态系统。
So one of the things that TSMC is going for is that there's a whole ecosystem built up around it.
对吧？
Right?
所有东西实际上都集中在台湾，围绕着它发展。
Everything is literally in Taiwan built up around this.
而半导体已经成为一个国家安全问题。
And semiconductors have become a national security issue.
所以我要问的第一个问题是：在你看来，这真的是一个国家安全问题吗？
So first question I ask is, actually, is it a national security issue in your view?
我觉得这一点被夸大了。
So I think it is a little overstated.
所以台积电。
So Taiwan Semi.
台积电比英特尔和三星这些其他竞争对手领先大约十八个月，可能长达两年。
So the Taiwan Semi is is eighteen months ahead of kind of Intel and Samsung, who are the other competitors, maybe two years.
两年可是很长的时间。
Now two years is a lot.
这就是摩尔定律。
That's Moore's Law.
所以当你获得一颗芯片时，不管你怎么称呼它，性能能提升50%，功耗降低30%。
So you are getting a chip, you know, however you wanna call it, you know, 50% more more performance, 30% less power.
这很重要。
It matters.
但如果世界不得不倒退到两年前，那会对全球造成巨大冲击，但世界并不会因此停止运转。
But if the world had to roll back to two years ago, that would be a big shock to the world, but the world would not stop spinning.
有很多产能实际上比台积电落后两年。
There's a lot of capacity that's essentially two years behind Taiwan Semi.
别提中国了，以色列也有很多产能，没错。
Forget China, but there's a lot of capacity in Israel Yep.
在韩国、日本和美国。
In South Korea, in Japan, and in America.
尺寸更大。
In larger in larger sizes.
是的。
Yeah.
两年前。
In the in two years ago.
比如，高一个音阶。
Like, you know, one note up.
对。
Yep.
比台积电落后两年。
Two years behind where Taiwan Semi is.
对。
Right.
所以整个世界都得倒退。
So the world would just have to roll back.
嗯。
Mhmm.
那现在推出的是哪一代iPhone？
There wouldn't be what what generation of iPhone just came out?
17代。
17.
对。
Yeah.
好的。
Okay.
如果真发生了，iPhone 17的预测就会停止。
So if that happened, there'd be the iPhone 17 prediction would stop.
嗯。
Mhmm.
展开剩余字幕（还有 480 条）
iPhone 16，你就得回退到 iPhone 16。
The iPhone 16, you'd have to roll back to the iPhone 16.
嗯。
Mhmm.
而且你还得去，我不知道。
And you'd have to take I don't know.
我的意思是，半导体行业的人都会生我的气，但你知道，要把那些芯片的设计迁移到旧工艺上，得花六个月的艰苦工作。
I mean, semiconductor guys get mad at me, but, you know, it'd be six months of really hard work to port the design of the chips that are in that To the old process.
迁移到三星或实际上现在亚利桑那州的台积电晶圆厂的设计规则，那个规模可真大。
To the design rules for the Samsung or actually now the Taiwan Semi fab in Arizona, which is really big.
但要回退一代。
But a generation back.
嗯。
Yeah.
你就是得回退。
You just gotta roll back.
嗯。
Yeah.
嗯。
Yeah.
但越来越多的是台积电。
But increasingly, we have Taiwan semi.
他们在亚利桑那州所做的是非常接近技术前沿的。
What they're doing is Arizona is pretty close to the bleeding edge.
嗯。
Mhmm.
你去过那里吗？
Have you been there?
我没有去过亚利桑那州。
I have not been I have not been to Arizona.
他们并不让投资者参观亚利桑那的晶圆厂。
It's not something that they let kind of investors tour the the Arizona fabs.
如果我是台湾，比如，马特，你来过吗？
If I was Taiwan, say, Matt, have you come?
嗯。
Yeah.
其实，我有一个反直觉的看法。
Well, it's actually I kinda have a counterintuitive take.
台湾的政客们，嗯。
The politicians in Taiwan Mhmm.
他们认为，不应该让先进制程在台湾以外的地方生产，因为这关系到他们的生存。
Believe that they should never let the leading edge process be manufactured outside of Taiwan because that ensures Their survival.
美国会保护台湾。
That America will protect Taiwan.
我觉得，以一种奇怪的方式，可能地缘政治上的紧迫性会更小。
I think in a weird way, it might be there might be less of a geopolitical imperative.
如果你是中国，也许当这种制程也位于美国时，会有一种民族自豪感的驱动力。
If you're China, maybe there's a national pride imperative if that process was also in America.
因为很明显，你知道，这一点在台湾媒体上曾短暂讨论过，但如果中国入侵台湾，中国是得不到那些晶圆厂的。
Because for sure, you know, and this was briefly discussed in the Taiwanese media, but, like, if China invades Taiwan, China is not getting those fabs.
中国得不到那些晶圆厂，也得不到那些工程师。
China is not getting those fabs, and they're not getting those engineers.
如果你看过《帝国反击战》，我的意思是，我觉得，嗯，好吧。
You know, if you've seen, like, the empire strikes back, like, I mean, I I think it's you know, okay.
你是一名关键的台积电工程师。
You're you're a key TSM engineer.
有一架飞机正飞往日本、韩国、以色列、亚利桑那、俄勒冈和德克萨斯。
There's a plane going to Japan, South Korea, Israel, Arizona, Oregon, and Texas.
每架飞机都将由六架美国战斗机护航。
Each one will be escorted by six American fighter jets.
你希望去哪里？
Where do you wanna go?
你知道的？
You know?
所以中国得不到那些晶圆厂，我认为这实际上可能正在趋于稳定。
So China's not getting those fabs, and I think it actually might be stabilizing.
但另一件我认为重要的是，台积电领先不过才六年。
But the other thing that I think is important to realize is Taiwan Semi has only been ahead for six years.
没错。
Yep.
五十年来，英特尔一直处于领先地位。
For fifty years Intel was ahead.
我去过科学园区。
And, I've been to Science Park.
而传奇的董事长兼首席执行官张忠谋曾将超越英特尔视为一个梦想，一个美好的梦想。
And Morris, the legendary chairman and CEO, he described catching Intel as a a dream, a beautiful dream.
他不确定自己有生之年能否实现，但那确实是一个美好的梦想。
And he wasn't sure it would happen in his lifetime, but it was a beautiful dream.
而很难追赶领先者的原因在于，半导体制造有点像烘焙，即使给两个人同样的配方，味道也不会完全一样，而且大家都在使用相同的设备。
And and the reason it's very hard to catch the leader is semiconductor manufacturing is is a little bit it's a little bit like baking where you can give two people the same recipe and it doesn't taste quite the same, and everybody's using the same equipment.
但如果你领先了，而前沿客户在使用你的配方，这就能让你在下一个配方上占得先机，比如如何将所有原料——面粉、糖、盐——结合起来。
But if you were ahead and the leading edge customers are using your recipe, it gives you a start on the next one and how you kind of combine all the ingredients, you know, the flour, the sugar, the salt.
在这种情况下，你可能用的是光刻机、计量设备、刻蚀和沉积设备等等。
And in this case, you know, maybe it's the scanners and the, you know, the metrology machines and the and, you know, the the etch and depots.
是的。
Yeah.
但你从根本上使用的是相同的设备。
But you're fundamentally using the same machines.
你有一套配方。
You have a recipe.
因为你比其他人早十八个月开发出上一套配方并找到了有效的方法，这有助于你更快地制定出下一套配方。
And because you developed the last recipe and figured out what worked eighteen months before anyone else, that helps you figure out the next recipe.
台积电之所以能赶超英特尔，是因为英特尔犯了两个关键错误。
The reason Taiwan Semi caught Intel is Intel made two crucial errors.
第一，英特尔有一段时间基本上是由三位CFO主导的。
One, Intel for a while was run essentially by three CFOs.
董事长、首席财务官和首席执行官，一位知名的半导体分析师称他们为‘三傻’。
The chairman, the CFO, and and the CEO, a well known semiconductor analyst called them the three idiots.
他们犯了一个巨大的错误。
And they made a really big mistake.
因为这是一个试错的游戏，几乎既是艺术也是科学，你们已经长期烘焙这些配方了，但他们不明白制造部门最重要的人员是所谓的‘白发元老’。
Because this is a game of trial and error, and it's almost as much art and science, and you've been baking these recipes for a long time, they did not understand that the most important people in the manufacturing division were what were called the gray beards.
那些五六十岁、长期把手指插进蛋糕里、舔一舔、说‘我们需要更多糖’的人。
The people in their fifties and sixties who had been, you know, sticking their finger in cakes for a long time, licking it, being like, we need more sugar.
明白吗？
Okay?
他们觉得：‘这太聪明了。’
They're like, oh, this is so smart.
我们请来这些三十岁的年轻人，可以裁掉这些五十五岁的老员工来省钱，这些年轻人精力充沛，这就是第一个灾难。
We got these really these 30 year old guys, and we can nuke these 55 year old guys and save some money, and these guys are energetic, and that was disaster number one.
行吗？
K?
台湾人永远不会犯这个错误。
Taiwanese is never gonna make that mistake.
不会再有人犯这个错误了。
Nobody's gonna make that mistake again.
他们犯的第二个错误，我认为商业和生活中的大多数错误都源于傲慢。
The second miss mistake they made, I think most bad decisions in business and life come down to arrogance.
他们相信自己如此优秀，以至于能够跳过所有人，直接转向所谓的EUV（一种光刻技术）。
And they believed that they were so good that they could move to what's called EUV, which is a kind of lithography, one node after everyone else.
尽管他们是ASML的最大股东，但他们对ASML收取的高价感到愤怒——这些价格其实非常合理，台湾企业不得不购买，但他们却认为自己可以不靠它们也能做到。
And this was despite being the largest shareholder of ASML, but they were outraged at the prices that ASML was charging that were so good that Taiwanese, they have to buy those, but we can do it without them.
使用所谓的DUV进行双重甚至三重图案化。
Double triple patterning with something called DUV.
这让我想起了我祖父的一句至理名言：给的工资少，就只能雇到猴子。
That was a reminder of my grandfather's great axiom, which was you pay peanuts, you get monkeys.
这太好笑了。
That's hilarious.
当然。
Absolutely.
是的。
Yes.
没错，马雷斯。
That's right, Marjes.
威尔·达诺夫，我那位了不起的富达投资组合经理，也是我的好朋友和导师，他管理着大约3000亿美元的资产，数额惊人。
Will Danoff, my the great Fidelity portfolio manager and a good friend and mentor of mine who runs, I think, close to $300,000,000,000 in incredible numbers.
他的父母曾在纽约的服装区工作，常对他说：价格会被遗忘，但品质永存。
His his parents were, you know, in the Garment District of of, I guess, New York and would say to him, price is forgotten, but quality remains.
真棒。
Oh, good one.
现在价格也很重要。
Now price is important.
是的。
Yeah.
在我们的业务中，价格确实很重要。
Price does matter at our business.
但即便如此，有时支付高价是值得的。
But nonetheless, sometimes prices are worth paying.
所以那是一个严重的错误。
So that was a terrible mistake.
然后，台积电拥有的另一大优势是整个半导体设备行业。
And then, the last thing Taiwan Semi had going for them was the entire industry of semiconductor capital equipment.
因此，ASML、科磊、应用材料、兰姆研究等公司都派出了他们的顶尖团队为台积电服务，因为他们对英特尔在尖端技术上的垄断感到不满。
So ASML, KLA ten core, Applied Materials, Lam Research, they had their a teams working at Taiwan Semi because they were unhappy that Intel effectively had this monopoly on the leading edge.
那么现在发生了什么？
So now what's happened?
台积电已经做出了这一赌注。
Taiwan Semi has made the bet.
我们如此优秀，以至于能在英特尔之后率先实现高数值孔径EUV技术。
We're so good that we can do high NA EUV, a node after Intel.
英特尔率先采用了这个节点。
Intel went to that node first.
其次，这些KLA和ASML的A团队现在都去了英特尔，因为台湾积体电路制造现在占据了垄断地位。
Second, the a teams from these the KLAs and the ASMLs have been at Intel because now Taiwan Semi has a monopoly.
我认为你们那里有一支团队，由传奇半导体投资者、杰出高管李斌领导，他深知需要做什么。
And I think you had you have a team there, led by Lip Bu, who's a legendary semiconductor investor, great executive, you know, who understands, what needs to be done.
实际上，我认为帕特·盖尔辛格在媒体上受到了非常不公平的对待。
And I actually think Pat Gelsinger was has been very unfairly treated in the press.
你知道，他推行了一项大胆的战略。
You know, he embarked on an audacious strategy.
这几乎是他们唯一的机会。
It was kind of their only chance.
现在，美国政府无疑会在许多方面礼貌地表示：嘿。
And now you have the US government who I'm sure in many, many ways is going to politely say, hey.
我们知道，你们更喜欢台湾生产的巧克力蛋糕，而不是英特尔在俄勒冈州和亚利桑那州生产的巧克力蛋糕。
We know that, you know, you like the, chocolate cakes made in Taiwan more than the chocolate cakes made in, Oregon and Arizona by Intel.
我们真的很希望你们能买下这款美国巧克力蛋糕，然后找点别的东西。
We'd really like you to buy that American chocolate cake and, you know, just find something.
你觉得美国政府收购英特尔10%的股份怎么样？
What do you think about the US government taking 10% of Intel?
用华尔街的话来说，就像拉里·威尔德曼开着我的玛莎拉蒂冲下悬崖一样，心情复杂。
To to quote from Wall Street, mixed emotions like Larry Wildman going off a cliff in my Maserati.
我不知道是否有人知道这句话，但拉里·威尔德曼是迈克尔·道格拉斯的商业死敌，总之就是心情复杂。
I don't know if anyone knows that quote, but Larry Wildman was Michael Douglas' great great business rival, but mixed emotions.
我认为，让美国政府明确表示英特尔是美国的国家资产，这一点很重要。
I think having the United States government very clearly say, you know, that Intel is a national asset for America.
我们要持有其一部分股份。
We're going to take a stake in it.
我认为这很重要。
I think that's important.
我不知道这是否能通过旧的《芯片法案》或现在的方案实现，我不确定。
Whether whether you could have done that with the old CHIPS Act or this, I don't know.
但我认为，无论好坏，这一届政府通过持股，将会对美国半导体公司，包括英伟达，发表比以往政府更积极的声音。
But I do think this administration, better or worse, with the equity stake, they're going to be vocal with American semiconductor companies, NVIDIA included, probably in a way that prior administrations would not.
为英特尔带来业务。
To bring business to Intel.
为英特尔带来业务。
To bring business to Intel.
即使这会增加这些商业公司的成本，迫使它们落后半代甚至一代。
Even if it costs those commercial companies more money and they have to kind of stay a half a generation or generation behind.
是的。
Yeah.
我认为，在可预见的未来，最先进的产品可能仍会在台湾制造。
I think the the the cutting edge products will probably be still be made to Taiwan for the foreseeable future.
我的意思是，我们将看看英特尔的18A和14A工艺节点表现如何。
I mean, we'll see how Intel 18 a and 14 a nodes pan out.
但你知道，英伟达现在有七款非常重要的芯片，也许其中一些可以由英特尔生产。
But there's, you know, you know, NVIDIA now has, yeah, I think seven chips that are very important, and maybe some of them you can make it Intel.
但另一件令人着迷的事情是，如果英特尔能好好整顿自己，台积电可能会非常惊讶。
But the other fascinating thing is if Intel can just get its act together, Taiwan Semi, I think they're very surprised.
你知道，他们一开放亚利桑那的工厂，订单瞬间就被抢光了。
You know, they open up the Arizona fab, sold out in milliseconds.
没错。
Yep.
然后他们提价20%，还是卖光了。
Then they raised price 20%, still sold out.
需求简直不可思议。
There's unbelievable demand.
没错。
Yeah.
我觉得美国晶圆的价格完全可以溢价接近100%。
I think you could probably charge close to a 100% premium for American wafers.
真的吗？
Really?
是的。
Yeah.
就像人们会希望在
Like, people will would have Made
美国制造。
in America.
为什么？
Why?
不是因为间谍技术？
No spy technology?
不是。
No.
我只是觉得地缘政治的担忧是真实的。
I just think the the geopolitical fears are are real.
我需要问你一下，抱歉。
I need to ask you sorry.
说吧。
Go ahead.
嗯。
Yeah.
我只是觉得，这个政府的许多政策都源于新冠期间美军意识到的一件事：天啊。
I just this administration, I think a lot of their policy is about I think during COVID, what the US military realized was, wow.
我们现在对中国的依赖太深，根本无法与他们开战。
We are too dependent on China to fight a war with them right now.
没错。
Yep.
一场持久战。
An extended war.
如果他们知道，如果我们知道，他们也知道。
And if they know that, if we know that, they know that.
这意味着我们无法阻止他们对台湾采取行动。
And that means that we are not deterring them from Taiwan.
所以我认为，他们确实相信‘以实力求和平’，能够可信地威慑中国，保护台湾——这个小小的民主国家。实际上，我觉得在很多方面，它和以色列很相似，人口大约是其四倍。
And so I think, you know, they they do believe in peace through strength and to credibly deter China and to keep Taiwan, which is a democracy, you know, a small democracy that, you know, I actually think in a lot of ways it's kind of parallels Israel, you know, maybe four times the population.
但这是一个小规模的民主国家，而以色列如今在地理上基本被朋友包围，但过去并非如此。至于台湾，取决于台湾执政者是谁，中国可能更像朋友，也可能更具有对抗性。但我认为，如果我们能有效与中国脱钩，就能实现持久和平，避免在西太平洋爆发一场中国与对手正面交锋的大型战争。
But, you know, it's a small democracy and, you know, Israel is now essentially geographically surrounded by friends but that was not always the case and you know Taiwan, know depending on who's in power in Taiwan China is more of a friend or maybe a little more adversarial But I think they believe that if we can effectively decouple from China, then that can bring durable peace and to and avoid having a giant war in the South Pacific where essentially China faces off.
我认为人们都忘了这一点。
I think people forget it.
就做美国吧。
Just be America.
是的。
Yeah.
做日本吧。
Be Japan.
那就做韩国吧。
It'd be South Korea.
做澳大利亚吧。
Be Australia.
菲律宾。
Philippines.
会是菲律宾。
It would be The Philippines.
很少有人了解这个世界这一地区的地理情况。
Not many people know the geography of that part of the world.
我发现当你身处纽约和洛杉矶之间时。
I've discovered when you when you when you find yourself between New York and Los Angeles.
是的。
Yeah.
是的。
Yeah.
当然。
For sure.
是的。
Yeah.
我的意思是，日本肯定会参与其中。
And I mean, Japan would absolutely be involved.
是的。
Yes.
嗯。
Yeah.
所以，说句显而易见的话，我的iPhone对美国消费者来说会贵20%。
And so, I mean, just to state the obvious, so my iPhone would cost the American consumer 20 percent more.
嗯。
Yeah.
我的iPhone，我的意思是，成本会增加50%。
The iPhone I mean, that then my cost 50% more.
嗯。
Yeah.
因为正如你所说，这不仅仅是晶圆的问题。
Because to your point, it's not just the wafers.
还有这么多其他东西
There's all these other things
都是在那里生产的。
that are made there.
是的。
Yeah.
但其中很多正迅速转移到墨西哥。
But all a lot of those are rapidly being relocated to Mexico.
没错。
Yep.
印度。
India.
是的。
Yeah.
绝对如此。
Absolutely.
而且，你知道，英伟达、苹果、亚马逊和超大规模云服务商正在推动这一切。
And, you know, NVIDIA and Apple and Amazon and the hyperscalers are driving a lot of this.
实际上，大型中国公司正在美国设立子公司，这些子公司可以合法分离，并将他们最好的工程师带到美国或墨西哥建立制造工厂。
Actually, big Chinese companies are opening up American subsidiaries, that can be legally separated, and they're bringing their best engineers to stand up manufacturing plants in America or Mexico.
就像日本汽车制造商那样。
Like the Japanese auto manufacturers
三十年前就是这样。
did thirty years ago.
百分之百。
100%.
是的。
Yeah.
很有趣。
Fascinating.
是的。
Yeah.
我觉得我有必要问这个问题。
I I feel need to ask this question.
我们现在在特拉维夫。
We're in Tel Aviv now.
所以美国政府刚刚拿走了英特尔10%的股份。
So the United States government just took 10% of intel.
在过去二十五年里，以色列政府通过补贴和税收优惠，投入了数百亿谢克尔，以吸引英特尔在以色列设立晶圆厂。
The Israeli government has spent tens of billions of shekels over the last twenty five years in both subsidies and tax incentives to have intel set up their fabs here in Israel.
如果你是以色列政府的领导人，面对美国刚刚拿走英特尔10%股份的情况，你会如何应对，如何提升以色列的立场？
If you were the government of the state of Israel, what would you do in reaction to, in response to, to level up regarding Intel after The US just took 10% of Intel?
你是半导体供应链方面的首席顾问。
You're you're the chief adviser on semiconductor supply chain
在以色列政府中。
in the Israeli government.
我的意思是，什么是公平的，什么是可能实现的。
I mean, there's there's what's fair, and then there's what's possible.
所以在一个公平的世界里，是的，如果我是内塔尼亚胡，我肯定会提出要求。
So in a fair world, yeah, Israel I would go if I were Bibi, yeah, I would absolutely ask.
是的。
Yeah.
我不知道5%、2%、10%的英特尔股份，哪个数字才合理。
I don't know what's what's a reasonable number for 5% of intel, 2% of intel, 10% of intel.
但对我来说，内塔尼亚胡和以色列整体似乎还想要美国的其他东西。
But it also feels to me like maybe Bibi and Israel in general want want other things from America.
所以你必须把这些因素也纳入考量。
And so you have to put that in the calculation.
但在一个公平的世界里，以色列绝对应该分得英特尔的一部分。
But in a fair world, absolutely, Israel should have a a cut of intel.
厉害的是，以色列完全可以这么做。
The great thing is Israel can do that.
这是美国政府的行为，他们可不是以优惠价格拿到的。
It's the American government, it's not like they got a got a good price.
他们得到了市场价格。
Like, they got they got the market price.
对。
Right.
所以，以色列也可以做同样的事。
So, you know, Israel can go do the same thing.
他们可以打电话给情报机构说，而情报机构不会拒绝。
They can call up intel and say and I'm and Intel's not gonna say no.
他们需要资金。
They need capital.
他们可以为任何他们想要的金额达成同样的协议，也许他们就应该这么做。
They can make the same deal for however much they want, and maybe they should do that.
有意思。
Interesting.
你觉得呢
And do you think
有没有可能把这看作一个由盟友组成的保障性供应链，比如英特尔、台积电和英伟达，它们在这里拥有数千名设计人员。
is there anybody like to think about this as a secured allied supply chain whereby, you know, Intel and TSMC and NVIDIA, which has here thousands of design Yeah.
你知道，还有成千上万的员工和众多设计人才，以及苹果公司，它在这里拥有难以置信数量的芯片设计师。
You know, and thousands of employees and many design people and Apple, which has here, you know, unbelievable numbers of chip designers.
是否存在这样一个盟友供应链呢？
Is there like an allied supply chain?
百分之百。
A 100%.
我之前说的就是台湾半导体工程师。
That's what I said about the Taiwan semi engineers.
没错。
Yeah.
你可以去日本，作为盟友。
You can go to Japan, ally.
你可以去韩国，作为盟友。
You can go to South Korea, ally.
你可以去以色列，作为盟友。
You can go to Israel, ally.
你可以建立台积电以色列分公司，或者去英特尔以色列工作，也可以去亚利桑那州、德克萨斯州或俄勒冈州。
And, you know, stand up TSM Israel, or you can go work at Intel Israel, or you can go to Arizona or Texas or Oregon.
很有趣。
Interesting.
这些地方是地球上制造领先数字芯片的仅有的地点。
Those that those are the places on planet Earth where those leading digital chips are made.
别无他处。
Nowhere else.
因此，以色列绝对是这一联盟供应链中极其重要的组成部分。
So Israel is 100% a very strategic part of this allied supply chain.
你之前提到过，我觉得这一点很引人注目：台积电希望只提供一代稍旧的技术，或者台湾认为，如果他们只向美国转移一代半导体制造技术，就能确保自身安全。
You mentioned before, which I thought was striking that TSMC wants to do kind of one generation back, or Taiwan thinks that if they give only one generation back of technology, semiconductor manufacturing technology to The United States, that will that will keep them safe.
讽刺的是，美国政府对英伟达与中国的关系正是这样做的。
Ironically, that's exactly what The United States Government did with NVIDIA vis a vis China.
100%。
100%.
我们的共同朋友比尔·古尔利认为这是一个糟糕的决定，因为中国由于受到限制，要么会开发出更好的技术，要么围绕现有技术打造更先进的软件，从而绕过这些限制。
Our mutual friend Bill Gurley thinks that's a terrible decision because, you know, the Chinese because of the constraints on them would either develop something that'd be much better or build what's called better software around this, you know, to kinda leapfrog the constraints.
你认为美国的政策是正确的吗？
Do you think The US policy was correct and think
你觉得比尔是对的吗？
that Bill's right?
我 respectfully 不同意比尔的看法。
I respectfully disagree with Bill.
首先，美国及其盟友，包括台湾和以色列，在半导体制造方面的优势在于这是一种配方，无法被复制。
So, one, the the advantage that America and its allies, Taiwan included, Israel have in semiconductor manufacturing, because it's a recipe, you can't copy it.
你能拿到F-35的图纸吗？
You get the plans for the f 35?
你能制造出F-35。
You can make the f 35.
F-35中唯一类似于半导体的部分是喷气发动机。
The only part of the f 35 that is kind of like semiconductors is the jet engine.
而在中国追赶西方航空技术的过程中，喷气发动机是耗时最长的领域。
And that was the area where it took China the longest to catch up to the West in terms of aviation technology.
因为喷气发动机涉及大量技艺和反复试验。
Because the jet engine, there's a lot of art and trial and error.
即使你有了设计图纸，情况也是如此。
Even if you have the plans sometimes.
半导体制造也是如此。
And semiconductor manufacturing is the same thing.
我认为，如果我们早在拜登政府初期就采纳了这些政策的话，
And I think had we adopted these you know, we really adopted these policies in you know, early in the Biden administration.
中国在制造和设计上原本需要六到八年才能完全赶上，但我觉得我们确实阻止并延缓了他们。
I think China was six to eight years away of absolutely catching up in manufacturing design, but I think we really did stop them and slow them down.
而我们现在拥有数十年的领先优势。
And we have a multi decade lead.
数十年的领先优势。
Multi decade lead.
哇。
Wow.
所以今天，我们可以说领先四年，但这是因为四年前、六年前，他们几乎能接触到一切。
So today, we'll call it it's a more of a four year lead, but that is because four years ago, six years ago, they had access to essentially everything.
对。
Right.
现在他们不能了。
Now they don't.
所以我们将会看到这四年的差距会发生什么变化。
So we will see what happens to that four year gap.
我认为这个差距可能会进一步扩大，但中国工程师确实非常优秀。
Like, I think it probably widens, but, I mean, the Chinese are really good engineers.
他们非常聪明，极其勤奋。
They're very smart, extremely hardworking.
是的。
Yeah.
实际上，我觉得中国和美国在国家精神上非常相似，只是中国的教育体系更好，平均而言人们可能更勤奋，而且无疑更专注于科学和工程——这是他们的优势，也是美国的劣势。
It's actually kind of like, I think China and America are actually very similar as as countries in terms of the ethos, except except the Chinese have a better educational system and I think are probably harder working on average and certainly are are more focused on science and engineering to their credit and the detriment of America.
但我认为这会很难。
But I think it's gonna be hard.
我认为这四年的领先优势会扩大。
I think that four year lead is gonna grow.
有意思。
Interesting.
在这种背景下，你给了他们一款比国内替代品领先一年、但比以色列这样的美国盟友以及美国自身使用的芯片落后一年到一年半的芯片。
And then in that context, you are giving them a chip that is, call it, a year ahead of the domestic alternatives, but a year to a year and a half behind what firms in Israel as an American ally get to use, what America gets to use.
我认为这是一项相当明智的政策。
And I think that I think that's a reasonably smart policy.
我认为，完全不给他们芯片是个大错误，而给予他们完全相同的芯片也可能是个错误，因为即使存在明显的硬件劣势，目前唯一能挑战美国封闭式实验室的前沿实验室，仍然是那些处于巨大劣势的中国实验室。
I think it's, like, the the position of giving them nothing is a big mistake, and then maybe the position of giving them the exact same chips is a mistake because, I mean, even with a pretty big hardware disadvantage, you know, the only frontier labs that are challenging the American closed source labs are the Chinese labs with a big disadvantage.
我的意思是，他们非常优秀。
I mean, they're very good.
而且，你知道，人工智能有着真正的地缘政治影响。
And, you know, there are real geopolitical ramifications of AI.
如果你看看我们的行业
If you look at our industry
广义上讲，科技领域的投资者，我以比尔为例，但不说具体金额，因为我对他在这个问题上的立场持不同意见。
kind of broadly, investors in technology, I take Bill as an example without saying the amount because I also respectfully disagree with with his position on it.
历史上，他们并没有太多参与地缘政治，但地缘政治已经成为人工智能投资环境的重要组成部分，比如从安德里森到哈德良，再到X，没错。
Historically, they haven't been that involved in geopolitics, but geopolitics has become a meaningful part of AI of by the way, the investing environment, you know, from Anderill to, you know, Hadrian to x Yeah.
绝对如此。
Absolutely.
还有X AI。
And x AI.
绝对如此。
Absolutely.
还有特斯拉，因为电动汽车和能源。
And Tesla because of electric vehicles and energy.
我的意思是，
I mean,
确实如此。
it's Absolutely.
这变得越来越有趣了。
It's becoming fun.
我认为，这以前从未属于硅谷投资者的视野，直到这个时代。
That hasn't been like in the in the purview of Silicon Valley investors, I think, ever until this until this era.
我觉得这是对的。
I think it's right.
你会怎么对待一个只是来向你介绍一家芯片公司的人，说嘿。
How would you approach somebody who's just bringing you a chip company and say, hey.
这里实际上存在地缘政治影响，会影响投资回报，会影响你未来制造这种芯片的地点，可能会影响零部件的供应。
There's actually geopolitical ramifications here that'll impact investment returns, that'll impact where you can manufacture this chip over time that may impact availability of, you know, parts.
是的。
Yeah.
嗯，我会说，我的意思是
Well, I would I I would say, I mean
对。
Right.
所以，里伯斯，你提到他们的一个核心客户是阿联酋人。
So Riebers, you mentioned one of their core customers was the Emiratis.
是的。
Yes.
他们最初唯一的客户就是阿联酋人。
Their only customer initially was the Emiratis.
他在那里待了很久。
He's there for a long while.
他曾在媒体上被报道过。
He's reported in the press.
嗯。
Yep.
这可能影响了IPO，也可能没有。
That may or may not have held up the IPO.
我的意思是，这简直是个大问题，是个地缘政治问题。
I mean, that that's like a big that's a geopolitical issue.
美国还没决定是否完全信任中东国家，你知道的，比如在先进AI方面。
America hasn't decided yet to fully trust, you know, Middle Eastern countries Yeah.
在先进人工智能方面。
With advanced AI.
嗯。
Yeah.
看来这一届政府已经决定，确实信任阿联酋，而且我认为他们也信任海合会国家。
It seems like this administration has decided that they do trust, for sure, The UAE, and I think they trust the GCC countries.
嗯。
Yeah.
幸运的是。
Thankfully.
是的。
Yeah.
对。
Yes.
而且我认为，你和我都花了很多时间在阿联酋。
And I think, you know, you and I have both spent a lot of time, you know, in The UAE.
有一件事让我作为一名美国人感到非常羞愧，那就是在阿布扎比作为犹太人可能比在美国大多数城市更安全。
And one thing that was, like, profoundly embarrassing for me as an American to realize is that you might be safer being Jewish in Abu Dhabi than you are in most American cities.
是的。
Yeah.
这让我作为一名美国人感到羞耻。
Which is, like, shameful to me as an American.
但我认为美国已经决定他们是可靠的盟友。
But I think America has decided that they are reliable allies.
但毫无疑问，据报道，Cerebras 的首次公开募股受到了 CFIUS 的影响。
But for sure, I mean, it was publicly reported that the Cerebras IPO was caught up in CFIUS.
但我只想说，我永远不会押注美国会失败。
But I would just say, I mean, I would never bet against America.
我永远不会押注以色列会失败。
I'd never bet against Israel.
我会站出来，支持所谓的西方民主国家。
Like, I would line up, like like, with the you know, I I would line up with the Western democracies.
这就是我的观点，也是我投资的方式。
And that would be my perspective, and that that would be the way that I invest.
是否存在所谓的主权供应链之类的东西？
Is there such a thing as a sovereign supply chain or something like that?
我的意思是，美国现在正在全力推进，但即便你努力推进，世界依然高度互联，你仍需要在墨西哥或印度制造，需要从这里、乌克兰或中国获取稀土资源。
I mean, America's making a good run at it right now, but it feels like even if you make a good run at it, the world is so interconnected, and you need to manufacture whether it's in Mexico or India and, you know, need rare earths from here or whether it's the Ukraine or China.
主权供应链真的现实吗？
Is sovereign supply chain even realistic?
嗯，如果‘主权’包括盟友的话，我认为这是可能的。
Well, sovereign, but, like, emphasis on if you include sovereign to include allies, I think it's possible.
所以如果你把日本、韩国、以色列和欧洲也算进去，他们确实还在制造一些东西。
So if you include Japan and South Korea and Israel and Europe, like, do still make a few things.
你知道的。
You know?
欧洲正逐渐变成一个仅供全球富人游玩的主题公园。
Europe is, like, on its way to just being a theme park for, you know, wealthy people from around the world.
但欧洲除了奢侈品之外，仍然生产一些有用的东西。
But Europe still makes some useful things outside of, like, luxury goods.
意大利烘焙用的食品原料可是个大生意。
Food ingredients apropos baking in Italy is a big deal.
我说的是认真的。
I'm not kidding, by the way.
是的。
Yeah.
欧洲并没有完全没落。
Europe is not totally lost.
所以，如果你把欧洲、美国、墨西哥、以色列、海湾合作委员会、韩国、日本算在一起，我认为这可以构成一条供应链。
So, you know, if you have Europe, America, Mexico, Israel, the GCC, South Korea, Japan, I think that can be a supply chain.
在当今世界，你能在美洲或中国完全制造出所有东西吗？
Can you make it all in today's world even in America or all in China?
我觉得这很难。
I think it's hard.
所以，像芯片主权这样的东西根本不存在。
So there's no such thing as chip sovereignty, for example, or anything like that.
但存在盟友芯片。
But there's chip Allied chip.
我会把这称为盟友芯片主权供应链。
Allied chip sovereign supply chains is the way I'd frame it.
你如何让人关注这一点？
And how would you get someone to focus on that?
我的意思是，你觉得政府正在采取行动。
I mean, I You think the administration is It's happening.
是的。
Yes.
没错。
Yeah.
你觉得安心吗？是的。
You feel comfortable that Yeah.
这100%正在发生。
It is it is 100% happening.
我想你曾经在某处发过推文提到过。
I think you said somewhere you tweeted that.
我们还能说‘发推文’吗？
Can we say tweeted anymore?
发帖。
Posted.
发布。
Posted.
发布。
Posted.
埃隆，是的，你提到了发布。
Elon Yea, you mentioned posted.
我问他，你对转推怎么办？
I said to him, what do you do about retweet?
他说重新发布。
He said repost.
是的。
Yes.
嗯。
Yeah.
这场人工智能竞赛是历史上最重要的竞赛。
This AI race is the most important race to win ever.
当然。
For sure.
昨天，前天，我在这里的一个会议上参加了一个小组讨论。
Yesterday, two days ago, I did a panel at one of these conferences here.
我请了八位非常聪明的人上台，让他们预测一下十年后人工智能会发生什么。
Tried eight very smart people up there and asked them to make one prediction about, ten years from now, what will happen in AI.
还有Dream Security的创始人Shalayv Julio，这是一家人工智能公司。
And Shalayv Julio, who's the founder of Dream Security, which is AI.
他之前在播客中谈到过用人工智能保护国家，他说，十年后，将有三件事定义超级大国。
He was on the podcast before AI to protect nations, said, in ten years, there are three things that are gonna define super nations.
如果上个世纪是核武器，那么十年后将是人工智能、量子计算和网络技术。
If in the previous century it was nuclear, ten years from now it's gonna be AI, quantum computing, and cyber.
因为你可以通过网络攻击让一个国家瘫痪，还需要破解密码等等。
Because you'd be able to take down a country with cyber attacks and you need to break codes, etcetera.
你认为谁会赢得人工智能竞赛？
Who do you think is gonna win the AI race?
第一个是，谁会率先成为拥有AI的超级大国？
Is the first one, like, who gets to super nation status with AI?
你认为呢
And do you
你同意沙利夫的说法吗？
agree with what Shalif said?
我同意沙利夫的说法。
I agree with what Shalif said.
我对量子计算可能稍微持怀疑态度，尽管它在网络安全方面无疑会非常重要。
I may be a little more skeptical on quantum, although it for sure is gonna be very important in cyber.
你和我想法一样。
You and me both.
但话说回来，我认为很难押注于美国及其盟友之外的国家。
But, again, I think it's tough to get to bet against America and friends.
就像，虽然可能九个月前看起来美国疏远了很多国家。
Like, even though, like, it just you know, I think it seemed maybe nine months ago that America had alienated a lot of people.
我认为这些联盟仍然稳固，美国及其盟友非常难以被超越。
I think, like, these alliances are holding, and America and friends is very tough to bet against.
所以我认为美国及其盟友会一起达成目标。
So I think American friends get gets there together.
而中国总是紧随其后。
And then I think China is always right behind.
像深海项目所做的那些事情真的很令人印象深刻。
Like, what what deep sea did was impressive.
是的。
Yeah.
你相信所有这些数据吗？
Do you believe all the numbers?
不相信。
No.
然后还有一些
So and then some of the
这些数字显然是编造的。
numbers were were were, you know, clearly made up.
你知道，这里面有一条线。
You know, there's a line in there.
这个项目花费了600万美元，还不包括消融和实验的费用。
This cost $6,000,000, you know, excluding the cost of ablations and experiments.
我还发了条推文，类似说：除了这个，林肯夫人，戏演得怎么样？
And, you know, I I posted on x kind of like saying, other than that, missus Lincoln, how's the play?
嗯。
Yeah.
毫无疑问，许多先进的英伟达GPU被走私进了中国。
And for sure, lots of advanced NVIDIA GPUs are smuggled into China.
就像我在其他场合说过的，我们根本拦不住毒品，而我们还有缉毒局专门负责这事。
As I've said in other forums, like, cannot keep drugs out, and we we have the DEA devoted to keeping them out.
我们甚至刚炸了一艘毒品船，可还是拦不住毒品。
And we literally just bombed a drug boat, and we cannot keep drugs out.
我的意思是，唯一有价值的东西，比如不管我们怎么称呼它，每公斤的这些毒品、钻石或GPU，中国都希望它们能进来。
Well, I mean, the only thing, like, has valuable per, like, you know, whatever whatever we're gonna call it, kilogram has some of these drugs or diamonds are GPUs, and China wants them to come in.
所以，如果美国都无法成功阻止那些美国民众群体想要的东西，而中国这个效率极高的政府又想把这些东西带进来，并且在全球拥有各种各样的前台公司来操作，那就更难了。
So if America cannot successfully keep out something equally valuable that the segments of the American population want, It's hard when China, the government, which is very effective, wants to bring these in and has, you know, all sorts of, you know, front companies all over the world to do it.
所以中国拥有大量先进的GPU，只是没有原本可能拥有的那么多。
So China has a lot of advanced GPUs, just not as many as they would have otherwise.
你一直是埃隆背后的大投资者。
You're you're you've been a big investor behind Elon.
这么说应该没错。
I think that's fair to say.
当然。
Absolutely.
你提到你认为全球变暖和气候问题确实是个大问题，但你觉得它已经被解决了。
You mentioned that you think global warming and climate is actually a big issue, but you think it's solved.
我同意。
I agree.
是因为太阳能吗？
Because of solar?
是因为太阳能和电池。
Because of solar and batteries.
还有，你
And, you
要知道，万一这些核聚变公司中有一家成功了呢？看起来确实有一家有可能成功。
know, god forbid, one of these, you know, fusion companies works, and it seems like one of them Might.
很可能成功。
Probably will.
是的。
Yeah.
我不知道是哪一家。
I don't know which one.
但我觉得，我的意思是，当我提到相信全球变暖是个问题时，这已经变成了一种意识形态。
But, yeah, I think it's I mean, I don't say this when a I I think believing in global warming is a problem became part of an ideology.
如果你只是看看关于太阳能和电池正在发生的事情的事实和数据，我会觉得，在我有生之年，世界将完全依靠自然阳光运行。
And if you just look at the facts and the data around what is happening with solar and batteries, I get to like, the world is gonna run on natural sunlight in my lifetime.
我的意思是，在美国
I mean, in America
大胆的主张。
big statement.
在你有生之年，世界会完全依靠自然太阳能运行？
The world is going to run on natural solar fully in your lifetime?
也许是在我孩子们的有生之年。
Maybe maybe my children's lifetime.
我的意思是，如果
I mean, if
我想打个预防针。
I wanna hedge.
但我的意思是，这纯粹是经济问题。
But, I mean, it's just it's economics.
是的。
Yep.
这纯粹是经济问题。
It's just economics.
每年它都变得更便宜，而其他所有能源形式却越来越贵。
It's every year it gets cheaper, and essentially every other source of power gets more expensive.
而且，确实，所有的电力都来自太阳。
And and and, definitely, all power comes from the sun.
化石燃料只是储存的阳光。
Fossil fuels are just stored sunlight.
核能只是人工的阳光。
Nuclear is just artificial sunlight.
实际的阳光加上电池，真的能够——我认为也必将——助你达成目标。
Actual sunlight plus batteries, it really can and I think will get you there.
我的意思是，在美国，我们的人均排放量又回到了20世纪20年代的水平。
I mean, in America, we are back to, I think, like nineteen twenty levels of emissions on a per capita basis.
我认为，只要看看太阳能在成本和部署方面的数据图表，就很难理解为什么还会对全球变暖感到担忧。
I think it's very hard to understand math, look at the charts of what's happening with solar from a cost perspective and a deployment perspective and be really worried about global warming.
你知道，中国在其中扮演了重要角色。
You know, China was a big part of that.
我觉得情况非常相似。
Think it's very similar.
百分之百。
100%.
完全是经济因素。
100 economics.
因为经济原因。
Because economics.
没错。
Yeah.
他们需要，所以中国没有国内的化石燃料。
They needed so China does not have domestic fossil fuels.
对。
Right.
所以他们从中东进口大量石油。
So They import a lot of oil from the Middle East.
是的。
Yeah.
还有俄罗斯。
And Russia.
绝对如此。
Absolutely.
所以，这就是为什么他们大力推动太阳能和电池，同时也大力推动电动汽车。
So, yeah, that's why, a, they went so hard on solar and batteries and b, went so hard on EVs.
因为对他们来说，这是国家安全问题——依赖从其他地方运输来的石油。
Because for them it was a national security issue, their dependence on oil shipped from other places.
所以，我认为世界将依靠阳光运行。
So, yeah, I think the world's gonna run on sunlight.
我是个坚定的环保主义者。
I'm I'm a committed environmentalist.
你知道，我喜欢在户外活动，但我真的相信全球变暖已经是一个被解决的问题。
You know, I love to be outdoors, but I genuinely believe global warming is a solved problem.
在你有生之年
In your lifetime
或者在你孩子的有生之年去防范。
or your kids' lifetime to hedge.
而且趋势是，就像一块巨石已经开始从山坡上滚落。
And that the trend is it's just it's like the boulder has started rolling down the hill.
它正沿着正确的方向滚落，最终会带你到达目的地。
It's rolling down the right path, and it's going to take you there.
仅仅是经济因素。
Just economics.
但自然地，我觉得埃隆一个人在减少碳排放方面为这个星球所做的贡献，比所有环保活动家和格蕾塔·通贝里加起来还要多。
But, naturally something I want something just Elon did more for this planet in decarbonizing it than all of the activists and the Greta Thunbergs combined.
那些疯狂的人，比如往画上泼油漆，你知道吗？他们反而让人想买一辆大卡车。
Those crazy people who were, you know, splashing paint on paintings, like, you know, they make people, you know, wanna go buy a big rig.
你知道吗？
You know?
他们什么有用的事都没做。
They don't do anything useful.
埃隆做了些有用的事。
Elon did something useful.
如果你关心地球，关心全球变暖，你就必须支持埃隆。
Like, if you care about the planet and you care about global warming, you just have to be pro Elon.
很多年前，范·琼斯写了一本书叫《绿色经济》，基本上是在嘲讽所有的环保主义者。
Some some significant number of years ago, Van Jones, I think, wrote a book called the green color economy where he basically lampoons or kinda pans all the environmentalists.
他说归根结底，一切都只是经济问题。
He just says some it's all economics at the end of the day.
解决经济问题，气候问题自然就会解决，同时我们还会拥有绿色职业，我认为埃隆已经证明了这一点。
Fix the economics and the climate problem will get fixed, and we'll have green collar jobs too, which I think Elon has proved.
是的。
Yeah.
比如太阳能安装工之类的。
You know, with solar installers, etcetera.
归根结底，经济和金钱驱动着世界。
At the end of the day, the economics and the money moves the world.
但你需要大规模投资，才能降低生产成本，我认为特斯拉能做到这一点。
But you need to kinda get massive investment into this to get production cost down, which I think would happen with With Tesla.
在中国也是如此。
With China here.
但电池储能问题真的解决了吗？
But is the battery storage thing really solved?
比如，在国家层面，我曾与能源战略负责人交谈过。
Like, on a national level, I I spoke to the head of energy strategy.
她也在我主持的以色列政府小组讨论中出现过。
She was on my panel also for the Israeli government.
你知道，我们这里有天然气。
You know, we got natural gas here.
我问她：那么，你们打算如何为要在荒漠中建造的大型数据中心供电？
And I said to her, well, how are gonna run the big data center that we need to build in the negative desert?
她对我说：我们现在能源是够的，但如果真的要建设百万级GPU集群，我们可能会遇到问题。
And she said to me, well, we got enough energy right now, but if it really goes a million GPU cluster, we may have a problem.
我说：那太阳能呢？
I said, well, what about solar?
我说：那我们就需要储能了。
I said, well, then we need storage.
你能在国家层面实现储能吗？
And can you get the storage on a national level?
这很难。
It's tough.
我们需要在全球范围内，尤其是在美国和法国，大幅提升电池制造能力。
We need a lot more battery manufacturing capacity globally and in particular in in in America and France.
有人关注这个问题吗？
Is anyone focused on that?
特斯拉非常关注这个问题
Well, Tesla's very focused
是的。
on that.
对。
Yeah.
对。
Yeah.
这又回到了埃隆身上。
It also is back to Elon.
是的。
Yes.
他是你愿景中的单一故障点。
He's a single point of failure for your vision.
我认为也许十年前，他确实是这个领域的关键人物，但他现在已经启动了这些进程，我认为这些事情一定会发生。
I think maybe ten years ago, he was a single point of failure for this, but he's what he's now set in motion, I don't I I think this is gonna happen.
他可能是我有生之年实现登陆火星并成功殖民火星的关键瓶颈。
He might be a single point of failure for getting to Mars and and successfully colonizing Mars in my lifetime.
你想去吗？
You wanna go?
我绝对觉得，看到蓝色的日落和蓝色的日出会非常酷，因为那里的极性是相反的。
I absolutely I think it would be really cool to see a blue sunset and a blue sunrise because the polarities are reversed.
有人告诉我，如果我想去，那里会给我留一个位置。
I have been told that there will be a spot for me if I want one.
你之前还说过，最糟糕的结果是单一AI模型给出唯一答案，这非常像奥威尔式的情景。
So I wanna something else you've said is that the worst outcome is one AI model giving one answer, which is very Orwellian, by the way.
对吧？
Right?
没错。
Yeah.
在很大程度上。
In a big way.
因此，要获得多种答案，需要多个AI系统和多套价值观。
So multiple AI systems to give multiple answers need multiple sets of values.
我认为，关于这些模型的一个被忽视的话题是，它们所吸收的价值观是什么？
And I think one of the under talked about topic around these models is what are the values they're ingesting?
你知道，现在整个维基百科都在被审视，它是一个被AI大量吸收的巨大语料库。
You know, right now, there's an entire scan about Wikipedia, which is a giant ingestion giant corpus that that's been ingested by AI.
Reddit是另一个例子，而你作为投资者，投资了Reddit，因为它们被AI吸收了。
Reddit's another one, and you're an investor in astute investor in Reddit as they were ingested by the AI.
但你知道，垃圾进，垃圾出，这正是驱动这些系统的东西。
But the you know, garbage in garbage out, that's what's fueling these things.
首先，如果所有人都吸收了相似的语料库，就会得到相同的答案。
If everyone's ingesting kind of the same corpus is number one, you get the same answer.
其次，那些操纵维基百科的人如今正受到国会调查，这非同小可。
Number two, whoever's crafted those things with manipulations of Wikipedia now being investigated by congress, that's a big deal.
我们会得到一套价值观。
We're gonna get one set of values.
我们如何才能得到不同的答案，如何才能获得不同的观点，尤其是当人们得到答案后，不再进行深入探究时？
How are we gonna get to different answers, and how are we gonna get to different viewpoints, particularly as people get answers and rather rather than kind of further inquiry?
极端主义总会自我毁灭。
Extremism always kind of burns itself out.
我认为，有时在造成巨大破坏之后才会如此。
And I would say circa, you know, sometimes after doing great damage.
但在2023年，以及我相信2024年初，发生了一件巨大的丑闻：如果你问OpenAI、Claude和谷歌的Gemini，是冒犯一个人的性别认同更严重，还是杀死一千万人更严重？
But, you know, in 2023, there was and and I believe early twenty four, you know, there was, you know, a giant scandal because if you asked OpenAI and Claude and and and Google's Gemini, is it worse to misgender one person or kill 10 million people?
它们会回答：这是个艰难的问题，但毫无疑问，冒犯一个人的性别认同更严重。
It would say, it's a tough question, but it's for sure worse to misgender one person.
我不是说冒犯性别认同不重要，但杀死一千万人显然更严重。
And I'm not saying misgendering is not a big deal, but killing ten million people is obviously worse.
但这些AI被灌输了如此奇怪的美国‘觉醒’单一文化，而这种文化实际上在世界其他地方并不存在。
But these AIs had been so imbued with this, you know, strange American woke monoculture, which really, you know, doesn't, you know, exist in the rest of the world.
你知道，在法国，他们曾发表过一篇大文章，讨论这件事有多荒谬。
You know, like in in France, you know, they had a big article about how crazy it was.
这是法国，没错。
This is France Yeah.
说这件事有多荒谬。
Saying how crazy it is.
我认为这件事自己就消退了。
I think that burned itself out.
我认为，还有许多这样的情况，比如你根本无法让AI生成一张白人男性的图片。
I think, you know, and then there are all these instances of, you know, you essentially could not get the AIs to generate a picture of a white man.
没错。
Yeah.
你知道，让我画一个典型的维京人，但画出来根本不像维京人。
You know, draw me, you know, draw me a picture of, you know, the prototypical Viking, but, you know, it didn't look like a Viking.
所以我认为这种现象有助于推动天平的摆动，就像埃隆在X和XAI上所做的那样。
And so I think that kind of that was helpful to moving the pendulum as was, you know, what Elon has done with X and XAI.
我认为他打破了奥弗顿窗口，恢复了言论自由。
I think cracking the Overton window for bringing back free speech.
我认为，拥有中国的人工智能也会有所帮助。
And something that I think is and then having the Chinese AIs is also gonna be helpful.
我认为被低估的关于马斯克成功的一点是，他的公司始终目标明确。
And something that I think is underrated about Elon's success is his companies are always very mission driven.
所以，在特斯拉，你问任何人为什么来这里？
And so anyone at Tesla, you ask them why are you here?
为了脱碳世界，让世界变得更绿色。
To decarbonize the world and make the world green.
目标非常明确。
Very mission driven.
SpaceX可能是目标最明确的公司。
SpaceX, maybe the most mission driven company.
你问那里任何一个人，你为什么来这里？
You ask anyone there, why are you here?
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。