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。
随着华尔街试图计算，人们对AI泡沫的担忧正在加剧。
AI bubble fears are growing as Wall Street tries to do the math.
让我们在大型科技播客中深入探讨，周五特别版，与DA Davidson科技研究主管Gil Luria共同解析AI泡沫专题报告。
Let's break it down in the big technology podcast, Friday edition, special report on the AI bubble with DA Davidson, head of technology research, Gil Luria.
广告结束后马上回来。
That's coming up right after this.
Capital One的技术团队不仅讨论多智能体AI。
Capital One's tech team isn't just talking about multi agentic AI.
他们已经部署了一个。
They already deployed one.
它叫做聊天礼宾服务，正在简化购车流程。
It's called chat concierge, and it's simplifying car shopping.
通过自反思和实时API检查的分层推理，它不仅帮助买家找到心仪的车辆。
Using self reflection and layered reasoning with live API checks, it doesn't just help buyers find a car they love.
还能协助安排试驾、获得融资预批，并估算以旧换新价值。
It helps schedule a test drive, get preapproved for financing, and estimate trade in value.
先进、直观且已投入使用。
Advanced, intuitive, and deployed.
这就是他们的技术实力。
That's how they stack.
这就是Capital One的科技。
That's technology at Capital One.
事实上，AI安全就是身份安全。
The truth is AI security is identity security.
AI代理不仅仅是一段代码。
An AI agent isn't just a piece of code.
它是您数字生态系统中的一等公民，需要得到相应的对待。
It's a first class citizen in your digital ecosystem, and it needs to be treated like one.
这就是为什么Okta正率先保护这些AI代理。
That's why Okta is taking the lead to secure these AI agents.
解锁这层新防护的关键——身份安全架构。
The key to unlocking this new layer of protection, an identity security fabric.
企业需要一种统一全面的方法，通过一致的政策和监督来保护每个身份，无论是人类还是机器。
Organizations need a unified comprehensive approach that protects every identity, human or machine, with consistent policies and oversight.
不要等到发生安全事件才意识到您的AI代理是一个巨大的盲点。
Don't wait for a security incident to realize your AI agents are a massive blind spot.
了解Okta的身份安全架构如何帮助您保护下一代身份，包括您的AI代理。
Learn how Okta's identity security fabric can help you secure the next generation of identities, including your AI agents.
访问okta.com。
Visit okta.com.
网址是0kta.com。
That's 0kta.com.
欢迎收听《大科技》播客周五版，我们将以传统、冷静且细致的方式解读新闻。
Welcome to Big Technology Podcast Friday edition where we break down the news in our traditional, cool headed, and nuanced format.
今天我们为您准备了精彩节目，因为我们将深入探讨AI泡沫。
We have a great show for you today because we're gonna go deep inside the AI bubble.
这确实是我期待已久的一期节目。
This is really an episode I've been looking forward to doing for a long time.
我们将深入分析数据，看看哪些公司负债过高，这一切是否可持续。
We're gonna look fairly deep inside the numbers, which companies are taking on too much debt, whether all of this is sustainable.
我们曾以多种形式探讨过这个问题。
We've touched on it in various different formats.
但今天，我们要介绍一位你们在播客中听过的嘉宾，因为我们研读过他的分析，现在要让他‘现身’——至少通过声音与大家交流。
But today, we're actually gonna bring a name that you've heard on the podcast before because we've read his analysis and bring him to life for you, at least via voice.
DA Davidson科技研究主管吉尔·卢里亚将与我们共同探讨这些话题。
Gil Luria, who's the head of technology research at DA Davidson, is here with us to discuss it all.
吉尔，很高兴见到你。
Gil, great to see you.
欢迎来到节目。
Welcome to the show.
谢谢邀请，亚历克斯。
Thanks for having me, Alex.
我真的很欣赏你的分析。
So I have really appreciated your analysis.
通常华尔街分析师评论趋势时，我们听到的多是他们看好的方面，而较少听到他们担忧的问题。
Oftentimes when we see Wall Street Analysts, weigh in on trends, it's typically we typically hear from them about the things they think are going well, but less about the things they think, are not going well.
你是真正敢说真话的人。
You're somebody who's, who's really called balls and strikes.
你既讨论过做法正确的公司，也点评过行事不当的企业。
You've talked about, you know, the companies that are doing this the right way, the companies that are doing it the wrong way.
这种分析极具价值，因为当下华尔街乃至所有人都在试图判断：AI投资热潮会延续当前趋势，还是说这不过是泡沫终将破裂。
And, that analysis is super valuable because we find ourselves in this moment where Wall Street and really all of us are trying to figure out whether the AI investment curve is gonna keep going the way it's been going or whether it's a bubble and everything is going to pop.
那么让我至少先提出这不是泡沫的观点。
So let me give you at least to start the argument that it is not a bubble.
这个观点来自旗语网的里德·阿尔贝戈蒂。
And this is coming from, Reed Albergotti at semaphore.
他表示AI正处于充满机遇与不确定性的市场，而非泡沫中。
He says AI is in a market of opportunity and uncertainty, not a bubble.
他写道，本周市场惩罚了CoreWeave和Palantir等AI股票。
He writes, the market punished AI stocks like CoreWeave and Palantir this week.
似乎全世界都认定AI泡沫正在破裂，没人想成为最后接盘的人。
It seems like the world is convinced that the AI bubble is deflating and nobody wants to be the last one out.
这不仅是华尔街的现象。
This isn't just a Wall Street phenomenon.
最近我参加的每场科技晚宴，大部分时间都在讨论记者们如何追问乐观的科技高管们这种情况能持续多久。
Every tech dinner I've been to lately, a good chunk of the conversation was spent on journalists peppering bullish tech executives about how long this really can last.
但高管们说的话确实有道理。
And yet what the executives are saying makes sense.
他们销售的产品让客户欲罢不能，而这个产品的潜在市场几乎是全球每个人和每家企业。
They are selling a product that customers can't get enough of and the total addressable market for this product is virtually every person and company on the planet.
本周我们看到AMD CEO苏姿丰预测AI计算市场将增长到1万亿美元，Anthropic计划2028年盈利等乐观数据。
And this week we saw AI companies touting rosy numbers from AMD CEO Lisa Su predicting the AI compute market would grow to 1,000,000,000,000 to Anthropic getting profitable by 2028.
尽管人们总在谈论AI泡沫和债务过高，但里德指出了一个关键：这些产品存在无法满足的需求，而且像苏姿丰这样的上市公司CEO发言必须严谨，就连Anthropic这样亏损的企业也计划几年内盈利。
So for all this talk of AI bubble too much debt, I think Reid is making a really good point here, which is that there is insatiable demand for the products, and you do have public company CEOs like Lisa Su, who, like, have to have some rigor behind the things they say talking about these major numbers, and even companies like Anthropic which are losing a lot of money, planning to get profitable in a few years.
吉尔，你怎么看？
So what's your read on this, Gil?
这里有很多需要解读的内容，我采用的框架是承认两者都是事实。
There's a lot to unpack there, and the framework that I do it with is to say both things are true.
因此，人工智能是我们长期以来最具革命性的技术。
So AI is the most revolutionary technology that we've had in a really long time.
无论是回溯到互联网时代还是工业革命时期，我们只能事后才能看清其影响。
Whether it's back to the Internet or back to the industrial revolution, we'll only know in retrospect.
但显然这些工具非常强大，而且正在不断改进。
But clearly the tools are very powerful and are getting better.
要认识到这一点，你只需要亲自使用它们。
All you need to know in order to realize that is just to use them.
让CHAT GPT帮你完成困难的任务——那些你通常会请别人做的事，无论是总结、写作还是提供建议，你会发现它不仅能力惊人，而且比一年前进步很多，更比两年前强得多。
Ask CHAD GPT to do things for you that are hard, that you would ask other people to do, whether it's summarizing, writing, giving you advice, and you see that not only is it incredibly capable, but it's much better than it was a year ago, and it's much better than the year before that.
所以没错，市场对这个产品的需求永无止境。
So yes, there is insatiable demand for this product.
确实如此。
That is true.
围绕这项能力有很多健康的行为，这些健康行为来自理性深思的商业领袖，比如微软、亚马逊和谷歌的决策者，他们正在明智地投资扩大人工智能的交付能力。
There's a lot of healthy behavior around that capability, and the healthy behavior are reasonable, thoughtful business leaders like the ones at Microsoft, Amazon, and Google that are making sound investments in growing the capacity to deliver AI.
他们能做出明智投资的原因在于拥有所有客户资源。
And the reason they can make sound investments is that they have all the customers.
他们拥有所有企业客户，并通过与OpenAI和Anthropic的合作关系，也掌握了所有消费者关系。
They have all the business customers, and by extension of their relationships with OpenAI and Anthropic, they have all the consumer relationships as well.
因此当他们进行投资时，使用的是资产负债表上的现金，有强大的现金流作为后盾，他们清楚这是高风险投资，并会做好平衡。
And so when they make investments, they're using cash on their balance sheet, they have tremendous cash flow to back it up, they understand that it's a risky investment, and they balance it out.
所以这些都是事实。
So all of that is true.
与此同时，我们正目睹一些不健康的行为，这正是我们这些经历过金融泡沫或科技泡沫的人所识别的模式——我们曾在过去见过类似情况，现在要说：等等，这是不健康的行为。
At the same time, we are seeing behaviors that are unhealthy, and that's where those of us that have lived through financial bubbles or technology bubbles are recognizing patterns that we've seen in the past and are saying, hold on, this is unhealthy behavior.
有些公司正在践行不健康的行为，这正是我们试图揭露的。
And there are companies that are exercising unhealthy behavior, and that's what we're trying to call out.
因此这里存在合理辩论：什么是健康的，什么是不健康的。
And that's why there is a good reasonable debate here is what's healthy and what's unhealthy.
你提到了几家公司，我想具体谈谈其中几家，因为它们代表了这种行为的某些方面。
And you mentioned several companies here, so I want to touch on a few of them because they represent some of this range.
Halantir是当今世界上最好的公司。
Halantir is the best company in the world right now.
我甚至不打算说它是最好的软件公司。
I'm not gonna even say the best software company.
它是当今世界最佳企业，因为它能做到的是深入一家公司并询问：你们当前最迫切的需求是什么？
They're the best company in the world right now because what they're able to do is go into a company and ask them, what's your biggest need right now?
你认为AI能为你做什么？
What is it that you think AI can do for you?
然后从头到尾彻底实现它。
And then do it soup to nuts.
这就是为什么你会看到他们惊人的增长率，他们在商业和政府领域都取得了巨大成功——后者只是以更隐秘的方式进行类似工作。
And that's why you're seeing extraordinary growth rates there, and they're being incredibly successful both there and on the government side where they're doing similar things in just a more clandestine way.
然后是像CoreWeave这样的公司，它正是我所指不良行为的典型代表。
Then there's companies like CoreWeave, which is the poster child for the bad behavior that I'm talking about.
我们正在讨论一家初创公司借钱为另一家初创公司建设数据中心的情况。
We're talking about a startup that is borrowing money to build data centers for another startup.
两家公司都在大量烧钱，却仍能通过举债来推进建设，尽管既没有客户保障，也看不到这些投资何时能带来回报。
They're both losing tremendous amounts of cash, and yet they're somehow being able to raise this debt capital in order to fund this build out, again, without having the customers or the visibility into those investments paying off.
健康与不健康的经营行为之间存在一系列差异，我们需要厘清这些以避免重蹈覆辙。
So there's a whole range of behaviors between healthy and unhealthy, and we just need to sort that out so we don't make the mistakes of the past.
我们可以深入探讨为何我认为举债投资数据中心是不健康的方式。
We can delve into why I think debt is an unhealthy way to invest in data centers.
这个讨论很有价值，因为当我们说'不希望债务融资'时，背后不仅有历史教训，还有更深层的原因。
I think that's a worthwhile discussion because when we just say, oh, we don't want debt financing, there's a reason for that, and it's not just our experience from the past.
好的。
Okay.
实际上，我们现在就重点讨论这个问题吧。
Actually, let's go right there right now.
债务确实是个问题。
The debt is an issue.
以甲骨文为例，它正背负巨额债务建设AI数据中心，寄希望于OpenAI未来可能（或不可能）带来的收入承诺。
I'm thinking about companies like Oracle, which is taking on the the Oracle itself is taking on a tremendous amount of debt to fund its AI data center build out, with this, like, promise that OpenAI will eventually pay them some revenue that it may or may not materialize.
在我看来，作为没有深入研究过甲骨文财务状况的人，他们实质上是在用山姆·奥特曼为OpenAI创造营收和利润增长的能力作为公司杠杆。
Seems to me, I mean, as someone who's not spent too much time digging into the finances of Oracle, that they're effectively leveraging the company on Sam Altman's ability to deliver revenue and effective maybe profitability growth for OpenAI.
再看Meta，虽然账上现金充裕，却仍选择发债来资助AI数据中心建设。
Then you have Meta, which has had a lot of cash on the balance sheet and they're now using debt to fund their AI data center build up.
对于Meta或许可以说，这种融资方式符合企业常规操作——即便持有现金，从财务角度看发债仍更合理，毕竟业务增长良好时债务总能偿还。
Now maybe with Meta you could say, you know, companies typically build this way and even though they have the cash it just makes more sense from financial standpoint, because there are arguments to say, okay, fund it with debt, who cares, you'll pay it back, everything's growing well.
但我确实想听听你的看法，为什么债务在这里是个大问题。
But I I actually will will turn it to you and and hear your perspective on why debt is such an issue here.
我们才刚刚开始看到债务问题出现在这次讨论中。
We we've just started to see debt make its way into this conversation.
那么为什么这是个问题？我们该有多担心？
So how why is it a problem and how concerned should we be?
所以我们必须回到金融基础101。
So we have to go back to finance one zero one.
对吧？
Right?
有些东西我们通过股权融资，通过所有权；有些则通过债务融资，通过长期支付利息的义务。
There are certain things we we finance through equity, through ownership, and there's certain things we finance through debt, through an obligation to pay down interest over time.
作为一个社会，长久以来我们都把这两部分放在正确的位置，对吧？
And as a society, for the longest time, we've had those two pieces in the right place, right?
债务适用于我有稳定现金流和有资产可以抵押贷款的情况，这样用未来现金流换取当前资金才有意义。
So debt is when I have a predictable cash flow and an asset or and or an asset that can back that loan, and then it makes sense for me to exchange capital now for future cash flows to the lender.
重申条件：需要有长期资产作为贷款抵押，和/或可预测现金流支持还款，对吧？
So again, the conditions are an asset that is long standing that can back the loan and or predictable cash flows to support the loan payments, Right?
这就是为什么会有房贷。
That's why we have a mortgage.
房贷就是同时满足这两个条件的例子。
A mortgage is an example of both.
对吧？
Right?
房贷就是...等等
A mortgage is wait a second.
如果我停止还贷，银行就会收回房子
If I if I stop paying my mortgage payments, the bank owns the house.
因为他们只贷给我房屋价值的80%，即使房价小幅下跌，他们也不会受损
And since they only lend me 80% of the value of the house, even if the value of the house goes down a little bit, they'll be fine.
而且他们掌握我的收入情况，即使在华尔街这种地方也相对可预测
And they have an access to my income, which is relatively predictable even on Wall Street.
所以他们确信我会按时偿还房贷
And so they know that I'll pay my my mortgage payment.
这种贷款是合理存在的
That's that's a loan that should be there.
对吧？
Right?
我们用股权来投资更具投机性的事物，用于发展壮大并拥有增长成果，但我们对现金流并不确定
We use equity for investing in more speculative things, for when we want to grow and we want to own that growth, but we're not sure about what the cash flow is going to be.
这才是正常经济的运作方式
That's how a normal economy functions.
当你开始混淆这两者时，就会陷入麻烦
When you start confusing the two, you get yourself in trouble.
正如你所说，甲骨文正在犯这个错误
And that's what, to your point, Oracle is doing.
他们说：这个初创公司承诺未来五年能带来3000亿美元的高利润收入，所以我要借钱建设基础设施来实现这个目标
They're saying, well, I I have this start up that's promising me $300,000,000,000 of revenue at a high margin over the next five years, So I'm gonna go borrow money to build out the infrastructure in order to deliver that.
等等，Oracle被揭露的本质是...稍等。
And and what Oracle has been exposed as is, hold on.
OpenAI承诺给我3000亿美元。
OpenAI promised me 300,000,000,000.
他们还承诺给微软2000亿美元，亚马逊380亿美元，CoreWeave 25亿美元，总计1.4万亿美元。
They also promised Microsoft 200,000,000,000, Amazon 38,000,000,000, CoreWeave 25,000,000,000, in total 1,400,000,000,000.
这家刚刚做出所有承诺的公司到底是什么来头？
And who's this company that just promised all that?
这家公司今年最多能有150亿美元营收，却要承受比这更大的亏损。
A company that, at best, will have $15,000,000,000 of revenue this year and will be losing more than that.
所以我们今年可能会亏损超过200亿美元。
So we'll be losing probably more than $20,000,000,000 this year.
这样的财务状况，他们凭什么让我借钱？我怎么能确定这些现金流？
So are they in a position to for me to borrow money because I have certainty around those cash flows?
不，这是恶劣行径。我们讨论的是——如果你借钱给投机性投资，而投资对象又是投机性客户，这就是恶劣行径。坦白说，正是这种行径拖累了最近几天的市场，因为人们意识到这种恶劣行为正在发生，没人愿意沾边。
No, that's bad behavior, and that's what we're talking about here is if you're borrowing money to make a speculative investment based on a speculative customer, that's bad behavior, And and frankly, that's what's dragged the the market down over the last few days is the realization that this bad behavior is happening, and nobody wants a piece of that.
好吧。
Okay.
那么如果无法...我是说，就以Oracle为例。
So what are the consequences then if you can't I mean, alright, let's say Oracle.
我们就继续用Oracle这个例子说。
Let's keep with this Oracle example.
对吧？
Right?
他们付不起钱，他们付不起。
They can't they can't pay.
你知道，他们正在建设这些数据中心。
You know, they're building these data centers.
这是那种情况，好吧，假设音乐停止，OpenAI会说，好吧，AI的进步将停滞不前，或者出于各种原因，AI发展放缓，他们实际上不会使用那些数据中心，或者没钱支付给你。
It's, one of those things that, alright, let's say the music stops, and, you know, OpenAI is like, alright, there's not gonna be any more AI improvement left or For any number of reasons, AI development slows down, they're like, actually, we're not gonna use those data centers, or we don't have the money to pay you.
这看起来只是Oracle的局部问题。
What is is it it seems like that's just a local issue for Oracle.
这是局部问题还是整个经济层面的问题？
Is that or an is it that or is it an economy wide problem?
答案取决于我们讨论的规模有多大。
The answer to that is it depends on what magnitude we're talking about.
再次强调，这是以往周期尤其是金融周期的教训。
Again, lessons of previous cycles and especially the financial cycle.
如果我们有数百亿美元债务投入的资产不再具有生产力，那么一旦出现问题，发行或持有这些债务的人会亏损，那些为这些贷款公司（如Oracle、CoreWeave等）持股的人也会亏损。
If we have tens of billions of dollars of debt into an asset that stops being a productive asset, then if there's a problem, it's the people that issue that debt or own that debt that lose money and the people that own stock in the companies that made those loans, Oracle, CoreWeave, etcetera.
如果是数百亿美元的规模，一些金融机构会亏损，主要是那些债务和股权的持有者会遭受损失。
They'll lose if it's tens of billions of dollars, some financial firms will lose, and and that's and and mostly the owners of that debt and that equity will lose.
当债务规模达到数千亿美元时，问题就开始显现了——至少几周前我们正朝着这个方向发展。
The problem starts happening when you get into hundreds of billions of dollars of debt, which is where we were headed at least as of a couple of weeks ago.
再说OpenAI这家初创公司，很棒的初创企业，优秀的产品，它向所有这些实体承诺了1.4万亿美元，这些实体和OpenAI都能出去筹集债务资本，这意味着他们及其客户正在寻求数千亿美元的债务。
Again, OpenAI, this startup, great startup, great product, a startup committed $1,400,000,000,000 to all these entities, so those entities, as well as OpenAI could go out and raise debt capital, which means they were seeking, they and their customers were seeking hundreds of billions of dollars of debt.
如果两年后我们面临这样的情况：存在数千亿美元债务，而AI需求趋于稳定，或者我们建设了足够的数据中心来满足当时的需求，租赁AI资源或GPU访问权限的价格下降，那么所有这些资产产生的收益都将不足以支付债务利息。
If we are here two years from now, and there's hundreds of billions of dollars of debt, and the demand for AI stabilizes, or we built enough data centers to support the demand we have at that point in time, and the price for leasing pieces of AI for renting access to GPUs goes down, all of those assets then can't pay enough to pay the interest expense on that debt.
所有债务同时违约。
All of that debt defaults at once.
现在我们讨论的是系统性风险。
Now we're talking about systemic risk.
这正是人们现在警告的——部分金融投资者在这里或那里损失数百亿美元是可以接受的。
That's what folks are warning about right now, is it's okay if some financial investors lose tens of billions of dollars here and there.
如果我们有数千亿美元的债务都集中在单一产品上，而这个产品价格下跌导致所有资产变得一文不值，那么整个经济都会被拖垮。
If we have hundreds of billions of dollars of debt into what is really just one product with one price, and that price goes down and all those assets become worthless, now we're gonna drag the entire economy down.
再次强调，我们都是从经验中得出这些结论的。
And again, we're all saying this from experience.
每个人都应该重温《大空头》，而不仅仅是为了看克里斯蒂安·贝尔、玛格特·罗比和赛琳娜·戈麦斯。
Everybody should rewatch The Big Short and not just to see Christian Bale and Margot Robbie and Selena Gomez.
这部电影精彩地讲述了一个小问题如何因为普遍化而演变成影响所有人的大问题。
It's a great movie that talks about how it's a little problem until everybody does it, and then it's a big problem that affects everybody.
对。
Right.
好的。
Okay.
那么关于这个还有一个问题。
So one more question about this.
他们是从谁那里借的钱？
Who are they borrowing from?
具体是哪些机构、投资者或个人在提供这些资金？
Like, who are the institutions that are giving them or the the investors or the individuals giving them this money.
我是说，吉尔，你阐述得非常清楚。
And if it I mean, Gil, you lay it out so well.
我们不知道。
We don't know.
这是推测性的。
This is, speculative.
真的不应该用债务进行投机，因为再次强调，它可能会崩盘，然后你就会遇到麻烦。
Shouldn't really use debt for speculation because, again, it could go under and then you could have a problem.
他们是从谁那里借的钱？
Who are they borrowing from?
你认为那些放贷的人是怎么计算的？他们显然明白你所说的那些事，却还是决定把钱借出去。
And what do you think the calculations were from the people lending this money that they who obviously understood the things that you're saying and said, you know what?
不管怎样，还是把钱给他们吧。
Let's give them the give them the cash anyway.
简短的答案是国内最大的金融机构。
Well, the short answer is the largest institutions in the land.
美国银行、摩根大通、三菱银行。
US Bank, JPMorgan, Mitsubishi Bank.
这些就是借钱给CoreWeave的公司。
Those are the companies lending to to CoreWeave.
而且，我们认为他们做的计算并不正确。
And and and, again, the math they're doing, we believe, isn't the right math.
让我深入解释一下我这话的意思。
Let me dig into that, what I mean by that a little bit.
重申一下，这是一种投机性资产。
Again, this is a speculative asset.
尽管我们都在使用AI并为之兴奋，正如一开始所述，确实如此，这很激动人心，我们也需要更多算力，但它仍是投机性资产，因为我们无法预知未来两到五年究竟需要多少——毕竟我们毫无经验可循。
Just because we're all using AI and excited about, which we laid out at the beginning, we are, and it is exciting, and we need a lot more compute, it's still a speculative asset in the sense that we don't know how much of it we're really gonna need in two to five years because we don't have experience doing that.
这完全是新生事物。
This is brand new.
我们无法预知五年后GPU的租赁价格。
We don't know how much a GPU is gonna rent in five years.
你是说，所以
You mean, so
我们拿到OpenAI的收入预测，他们声称能赚到——大概1000亿美元一年？我承认这说法有点夸张了。
we get the revenue projections from OpenAI that they're gonna make, like, you know, a $100,000,000,000 a year, I guess I'm being exaggerating a bit.
你根本不能...不能信任他们，因为你完全不清楚他们的依据是什么？
You can't really you can't trust you can't trust them because you just don't how do they know?
正是如此。
Exactly.
所以第一点，AI可能如我们预期般发展，但也可能不会。
So so one, AI may turn out as well as we we expect, but it may not.
第二点，OpenAI并非孤立存在。
And two, OpenAI is not in a vacuum.
他们面临竞争。
They're competing.
他们之所以过度承诺、营造'大而不倒'的假象、用'假装成功直到真正成功'的策略拉所有人下水，部分原因在于他们清楚自己正在与Meta、谷歌和马斯克等资源远胜于他们的对手竞争。
Part of the reason they're overpromising and and creating this too big to fail and and fake it till you make it and getting everybody else to have skin in the game, part of the reason they're doing that is because they know they're competing with Meta and with Google and with Elon, people that have a lot more resources than they do.
所以他们声称到2027年要达成1000亿美元营收——就像山姆·奥特曼刚说的那样——完全是在睁眼说瞎话。
So for them to say, oh, we're gonna have a 100,000,000,000 of revenue by 2027, which Sam Altman just did, is completely disingenuous.
他根本不明白。
He has no idea.
他面对的是更庞大、更有实力的竞争对手，这些公司的技术至少和他旗鼓相当。
He's competing against much bigger, more powerful companies that have technology that's at least as good as his.
基于这种预期放贷很危险，因为这些GPU设备——你建数据中心出租GPU，现在每小时租金可能是4美元。
So lending money based on that is dangerous because, again, these GPUs, you're building a data center, you're renting out GPUs, and right now maybe you're renting out a GPU for $4 an hour.
或许这样商业模式还能成立。
And maybe that way the business makes sense.
但这些GPU每年性能都在大幅提升，三年后同款GPU每小时租金可能只剩40美分，到时候数据中心就毫无价值了，因为连运营成本都覆盖不了。
But these GPUs keep getting so much better every year that that same GPU in just three years may be only renting out at 40¢ an hour, at which point the data center is literally worthless because that won't be enough to cover the expense of operating the data center.
这就是危机所在——当摩根大通、美国银行或三菱银行的风控人员忽视这点时。
So this is where we get in trouble, when somebody underwriting a JPMorgan or US Bank or Mitsubishi Bank ignores that.
回答你的问题：他们为什么这么做？
And to answer your question, why would they do that?
这些可都是专业人士。
These are professionals.
因为他们不用承担风险，懂吗？
It's because they don't have the downside, right?
他们接到的任务就是把资金投进AI领域。
They have a mandate to deploy capital into AI.
他们的上司从更上级接到指令说：我们的AI资产配置不足。
They got an order from their boss who got an order from their boss that says, We don't have enough AI in our portfolio.
去给我找些人工智能项目来投资。
Go find me AI to invest in.
于是有人找上门来说，嘿，看好了，我正在建一个数据中心。
And so somebody comes to them and says, Hey, look, I'm building a data center.
借我点钱。
Lend me money.
我会付你9%的利息。
I'll pay you 9%.
这利息可太划算了。
That's fantastic interest.
然后你就签约了。
And you sign up for it.
那年你因为签下这笔交易拿到了巨额奖金。
You get a big bonus that year based on signing that deal.
就算交易黄了，就算数据中心三年后一文不值，你也无所谓。
If the deal goes sour, if if the data center's worthless in three years, you don't care.
又不用你把奖金退回去。
You're not giving your bonus back.
这就是我们在金融危机时期的真实写照。
That's the world we had back in the financial crisis.
这就是当年我们陷入困境的原因，也是现在如果不采取措施就可能重蹈覆辙的原因。
That's how we got in trouble then, and that's how we could get in trouble now if we don't do something about it.
吉尔，这个过渡很巧妙，因为你提到了《大空头》——这周这片子特别应景，毕竟迈克尔·伯里（电影原型人物）发现我们在向根本不该获得贷款的人发放极度投机性贷款时，他做空了房地产市场。
So, Gil, it's a great segue because you brought up The Big Short, and this has definitely been a week where that is applicable because Michael Burry, basically the star of that movie, of that story, He who the guy who effectively shorted he shorted the housing market when he saw that, we were engaging in extremely speculative loan behavior to people who should not get loans.
他本应开始好转，却并未真正开始。
He has started to well, not started.
他现在已经拉响了真正的警报。
He has sounded a real alarm now.
这很有趣，因为你描述的激励机制与那些为次级贷款放贷者的动机如出一辙——这些人本不该给那些负担不起房贷的人放贷，但他们反正能拿到奖金。
And it is interesting because the incentives that you described are sound exactly similar to the incentives of the people writing loans for people for the subprime mortgages, people who shouldn't have have have have gotten that loan for a house they couldn't afford, but they were gonna get their bonus anyway.
对吧？
Right?
这简直就是那个故事的翻版。
It's the it mirrors that story.
本周巴里成为头条，因为他关闭了自己的公司，并直指当前AI领域的估值和行为问题，正如你提到的折旧问题。
And this week, Burry made headlines because he'd shut down he completely shut down his his firm, and he he basically, you know, described it to the valuations and the behavior we're seeing, with AI and for the reason that you just outlined, which is depreciation.
这是他的一条推文：
Here's a tweet from him.
通过人为延长资产使用寿命来虚增盈利的折旧操作，是现代最常见的财务欺诈手段之一。
Understanding depreciation by extending useful life of assets artificially boosting boost earnings, is one of the more common frauds of the modern era.
好。
K.
所以基本上，如果你不准确计算GPU的折旧，他实际上就是在指控这是欺诈行为。
So, basically, if you don't if you don't if you don't accurately capture depreciation of the GPUs, he's effectively calling it a fraud.
大规模增加资本支出购买英伟达芯片和服务器（产品周期仅两三年），绝不应导致计算设备使用寿命的延长。
Massively ramping CapEx through purchase of NVIDIA chips and servers on a two to three year product cycle should not result in the extension of useful lives of compute equipment.
然而这正是所有超大规模企业都在做的事。
Yet that is exactly what all the hyperscalers have done.
据我估算，2026至2028年间他们将少计1760亿美元的折旧。
By my estimates, they will understate depreciation by 176,000,000,000 from 2026 to 2028.
到2028年，甲骨文——又是甲骨文——将虚增26.9%的收益，Meta虚增20.8%等等，但情况会更糟。
By 2028, Oracle, there's Oracle again, will overstate earnings by 26.9%, Meta by 20.8%, etcetera, but it gets worse.
所以巴里可能就此基本...你知道...关张歇业了。
And so that so that may be then Burry basically, you know, closes up shop.
他的意思是，这些公司都说芯片五六年才折旧完，但就像你说的，如果英伟达芯片性能提升这么快，折旧速度会大幅加快，导致他们现在投资数十亿的数据中心变得一文不值——用你的话说。
So what he's saying is that, like, all these companies say that these chips will depreciate over five or six years, but like you said, if the if the Nvidia chips get that much better that much more quickly, we can have a much more accelerated depreciation making the data centers that they're investing billions in today worthless as you put it.
你如何评价巴里对当前局势的批评？
How do you evaluate Burry's critique of the situation?
听起来你同意他的观点。
Sounds like you agree with him.
他说得一点没错。
He's spot on.
完全正确。
He's spot on.
顺便说，《大空头》讲的就是他如何料事如神却差点没撑到最后的真实故事。
By the way, Big Short is the story of how he was spot on but he almost didn't make it.
对吧？
Right?
电影大部分都在展现预言成真需要多久，即使你判断正确。
A lot of the movie is about how long it takes to play out, and you can be right.
但如果你的正确来得太早，就...就撑不到最后。
But if you're right too early, you you don't make it.
这个故事讲述的是他和少数成功者的经历。
And and the story is about him and the handful of people that did make it.
有很多人长期做空市场，却因等待时间不足而失去一切。
There were a lot of people that were short the market for a long time and lost everything because they couldn't wait long enough.
他恰好处于能等待的位置，现在他的判断完全正确。
He was just in a position to wait, and he's spot on right now.
而且你看。
And and look.
折旧计算会变得复杂，让我从宏观层面解释，这对本次讨论非常重要。
Depreciation gets wonky, so let me just hit it at a high level because it is really important to this conversation.
对吧？
Right?
折旧基于会计准则，帮助企业声明：我拥有一项资产。
Depreciation is based on an accounting standard that helps companies say, I have an asset.
它的使用寿命有多长？
How long is it useful?
这项资产能为我创收多久？
How long can that asset generate revenue for me?
如果它能为我创收五年，那么我就应该将获取该资产的成本分摊到五年里作为会计支出。
And if can generate revenue for me over five years, then I should take the cost of acquiring that asset and spread it over five years as an expense for accounting purposes.
对吧？
Right?
这正是会计师的职责所在。
That's that's what accountants are there to do.
这些会计师在三到五年前曾与微软、亚马逊等公司合作，他们当时说，你知道吗？
And these accountants spent time three to five years ago with companies like Microsoft and Amazon and said, you know what?
基于当前的技术水平，我们评估这些芯片后认为，它们能为你创造大约五到六年的收入。
Based on where the technology is now, we're looking at these chips, and it looks to us like they can generate revenue for you for about five or six years.
正因如此，我们允许你将折旧年限延长至五到六年，这样你的支出减少，利润表现会更好。
And that's why we're going to allow you to depreciate, to extend your depreciation to five to six years, because then you have less expenses and you look more profitable.
所以我们将允许你这样做。
So we're going to allow you to do that.
但过去三年间，技术已实现巨大飞跃。
But what's happened over the last three years is that technology has taken huge leaps forward.
黄仁勋为此已筹备数十年。
Jensen Huang has been preparing for this for decades.
如今我们终于能充分利用他设计的这些卓越芯片。
And here we are, and we can make the most of the brilliant chips that he's designed.
现在他每年都能造出比前代性能强十倍的芯片。
And now he can make one every year that's 10 times better than the one he made the year before.
这非常了不起。
And that's great.
这就是我们拥有这么多强大工具的原因，也是它们进步如此神速的根源。
That's why we have all these great tools and that's why they're getting so much better.
但回到那些会计师身上，如果你现在问他们：这项资产能产生可观收入的周期有多长？
But back to those accountants, if you ask the accounts today, how long will this asset generate meaningful revenue?
他们绝不会回答五到六年。
They would not answer five or six years.
他们很可能会回答三年。
They would probably answer three years.
按照巴里先生的观点，如果你告诉亚马逊和微软，尤其是如果你告诉像甲骨文和CoreWeave这样的公司，答案是否定的。
And to mister Burry's point, if you told Amazon and Microsoft and certainly if you told companies like Oracle and CoreWeave, no.
不。
No.
不。
No.
不。
No.
这些芯片只会在三年内产生可观收入，之后盈利能力将急剧下降。
These chips will only generate meaningful revenue for three years, their profitability would decline very dramatically.
所以对微软、亚马逊、谷歌来说，我并不太担心这点。
So again, for Microsoft, Amazon, Google, I don't worry about that too much.
他们能应付。
They can handle it.
但对CoreWeave和甲骨文这类公司而言，这意味着它们再也无法筹集资金，最终都将消失。
For companies like CoreWeave and Oracle, it means they'll never be able to raise any more capital again, which would which means they would all go away.
这就是为什么这个观点很重要，尽管它有点古怪。
So that's why this point is important, even though it's a little wonky.
因为当这些公司回头告诉你'不不不'的时候——
It's in because when these companies come back and tell you, no, no, no.
我有用了五年的芯片照样运行良好。
I have five year old chips that work just fine.
这不是一回事。
That's not the same thing.
说我有一个用了五年的芯片依然运行良好，并不意味着它能产生和五年前相同的收入，这才是会计问题的关键。
Saying I have a five year old chip that works just fine doesn't mean that it can generate the same revenue that it did five years ago, which is the accounting question.
所以这些公司是在玩花招，告诉你芯片还能用。
So that's sleight of hand by these companies to tell you that the chip still works.
如果它现在的收入只有五年前的1%，那实际上已经毫无价值了。
If it's only generating 1% of the revenue it generated five years ago, it's by all intents and purposes worthless.
因此我们必须向财务部门提出正确的问题，我认为未来几年我们会这么做，并纠正这种错位——这正是巴里先生押注的。
And so that's where we have to ask the accounts the right question, and I think we will be over the next couple of years, and we're gonna be correcting this dislocation, which is what Mr.
巴里押注的就是这个。
Burry is betting on.
对了，萨提亚·纳德拉这周参加了杜阿尔克什的播客，与他和半导分析的迪伦·帕特尔进行了访谈，他基本上在讨论这个问题——大概是多久前？
Right, Satya Nadella was on, Duarkesh's podcast this week in an interview with him, and Dylan Patel from Semi Analysis, and he basically was talking about this a couple what?
一年前？
A year ago?
两年前？
Two years ago?
当时英伟达的H100芯片还是最先进的。
The the h 100 NVIDIA chip was state of the art.
现在我们已经在讨论即将部署的Blackwell架构了。
Now we're already talking about the we've Blackwell is deploying.
下一代产品也即将面世。
There's another generation coming out.
下一代产品将通过鲁宾(Rubin)平台研发，将生产这些H100芯片——虽然H100本身才问世没几年。
And the generation after that is is underway via Rubin, which is gonna make these h one hundreds, which again, just came out a couple years ago.
你知道，虽然我不会说完全没用，但几年前每块GPU要价3万美元，现在很难证明这个价值依然合理——这正是你指出的问题。
You know, not I I wouldn't say completely worthless, but you were paying $30,000 per GPU a couple years ago, it's gonna be very hard to justify, you know, having that same value, which is what you're pointing out.
但让我们看看另一方的观点，来自Semi Analysis的乔丹·纳诺斯(Jordan Nanos)提出了反驳意见。
But let's look at the other side here, which is, Semi Analysis, has the counterpoint from Jordan Nanos.
明白吗？
Okay?
乔丹表示基本没有先例表明芯片会在两三年内出现故障或性能衰减。
Jordan says there's basically no precedent to say the chip would wear would fail out would fair or fail or wear out in two to three years.
硬件制造商的标准合同期限是三到五年，他们还提供六到七年的延保服务。
The hardware manufacturers have contracts that are standard for three to five years, and they offer extended warranties for six to seven years.
巴里论点的成立前提是：英伟达在未来两三年内发布的芯片性能必须远超当前世代，使得全球所有超大规模企业都愿意启动新一轮资本支出周期。
The proof of Burry's argument would be, predicated on NVIDIA releasing chips so that that so drastically outperform the current generation in two to three years that all hyperscalers everywhere are incentivized to go through another CapEx cycle.
这意味着他们必须全部购买新芯片并替换现有设备。
They'd have to all buy new chips and rip out the existing ones.
相比'这些芯片可能在数据中心运行五到六年'的说法，前者的实现难度显然要大得多。
That seems like a much farther leap than saying we might be able to run these chips for five to six years in the data centers themselves.
对此你怎么看？
What do you think about that?
我有几点看法。
A couple of things here.
首先，我很欣赏英伟达用维拉·鲁宾(Vera Rubin)和理查德·费曼(Richard Feynman)这些有趣的名字来命名芯片。
So first of all, I think, I like that NVIDIA pick picks fun names like Vera Rubin and Richard Feynman to to name their chips.
我喜欢这个命名规范。
I like that as a naming convention.
他显然是个书呆子，但呆得恰到好处。
He's clearly a dork, right, in the best way possible.
另外值得注意的是，Dylan和Dwarkesh是室友，那间公寓真是聚会的好地方。
The other thing to note is that Dylan and and Dwarkesh are roommates, and, boy, is that a fun apartment to hang in hang out in.
对吧？
Right?
这两个人是周围最聪明的人，我很喜欢听他们讲话。
Those those two are some of the smartest people around, and I love hearing them speak.
既然说到这里，我要提醒你们注意，他们刚才说的其实是障眼法，明白吗？
Now that I've said that, I'll bring you back to the fact that what they just said is slight of hand, right?
芯片三年后还能工作，并不意味着它能产生相同的收入，对吧？
The fact that the chip works after three years doesn't mean it's gonna generate the same revenue, right?
你可以拥有一个能用的芯片。
You can have a working chip.
我可能有一台十年前的Mac或PC，现在还能开机。
I could have a ten year old Mac or a ten year old PC that still turns on.
但我不想用它，因为它无法完成我需要的工作。
I wouldn't wanna use it because it wouldn't be able to do the things I need it to do.
所以仅仅因为三年后没坏，并不意味着它能产生和三年前相同的收入。
So just the fact that it doesn't break after three years doesn't mean that it can generate the same revenue that it did three years ago.
这就是第一层障眼法。
So that's that's one level of sleight of hand.
他们指出的另一点是，哦，别担心。
The other thing they point to is, oh, don't worry about it.
我们确实有五年前的老芯片，现在仍以不错的价格出租。
We do have five year old chips that we're still renting out at decent prices.
这实际上只是反映了我们当前在扩张周期中所处的位置。
And that's really just a function of where we are right now in the expansion cycle.
我们极度缺乏处理这些AI交易、AI代币生成推理交易的芯片，人们正在租用任何能弄到手的设备。
We are so short on chips to process these AI transaction, these AI token generation inference transactions that people are renting out anything they can get their hands on.
这就像疫情期间的二手车市场。
This is like used cars during COVID.
人们宁愿溢价购买二手车也不买新车，因为当时根本买不到车。
People would pay a premium to a new car to buy a used car because there were just no cars around.
但这并不意味着二手车比新车更有价值。
That doesn't mean that that used cars was worth more than a new car.
只是说明当时极度短缺导致人们支付了过高价格，我们现在就处于这种状况。
It just means there was so much scarcity that people overpaid, and that's where we are right now.
这是不可持续的局面，因为所有人都在扩建数据中心——即便剔除那些贷款建设数据中心的投机者，亚马逊、微软、谷歌、Meta和埃隆仍在用现金流扩建数据中心。
That's not a sustainable situation because everybody is building out data centers, and again, even if you took out the bad players that are lending, that are borrowing money to build data centers, you still have Amazon, Microsoft, Google, Meta, Elon using cash flow to build data centers.
所以我们正在建设巨大的产能。
So we're building tremendous amount of capacity.
一旦产能接近饱和，那些计算能力低下的旧芯片就会变得一文不值。
Once that capacity even gets close to catching up, then the old chips that can't do as many calculations for you will be worth a fraction.
这就是市场最终会走向的局面，对吧？
And this is where the market's gonna end up, right?
市场是高效的，这就是均衡点所在。
Markets are efficient, and here's where the equilibrium is.
这就是我们将达到供需平衡的地方，以每浮点运算的美元成本计算。
Here's where we'll get to the balance of supply and demand is on dollar per flop.
每浮点运算的美元成本即每次计算的美元成本。
Dollar per flop is to say dollar per calculation.
记住，这些AI芯片的功能是生成标记（token），也就是我们所说的文字、数字或图像。
Remember, what these AI chips do is they generate tokens, which is what we call words or numbers or images.
如果一个理查德·费曼芯片每秒能生成x个标记，而H100每秒只能生成百万分之一个标记（或需要百万秒生成一个标记），那么理查德·费曼芯片的价值就是H100的一百万倍。
And if a Richard Feynman chip can generate x tokens per second, and an H100 can generate oneone million of a token per second, or take a million seconds to generate a token, then the Richard Feynman chip is worth a million times more than an H100.
即使H100能运行，我们也不会使用它，因为它跟不上需求。
Even if the H100 is working, we won't use it because it can't keep up.
这不值得，因为我们更愿意使用能生成所需数量标记的芯片。
It's not worth it because we'd rather use a chip that can generate the amount of tokens that we need.
所以这是两个完全不同的话题。
And so this is two different conversations.
这个芯片能用吗？
Will the chip work?
也许能用，但它有价值吗？
It might, but will it be worth something?
它能完成足够的计算量来产生收益吗？
Will it be able to do enough computation to generate revenue?
这是个完全不同的问题，我们认为在三年时间框架内，三年前的芯片将无法完成足够的计算量，不值得继续保留它们。
That's a completely different question, and that's where we think in a three year timeframe, three year old chips will just not be able to do enough computations to be worth keeping them on.
所以我们要么替换它们，要么将它们用于那些重要性较低、收益也大幅减少的事务上。
So either we replace them or we use them for much less important things that will generate much less revenue.
再次强调，我们必须谨慎，混淆‘它能否运作’与‘它能否产生收益’这两个概念。
Again, we have to be careful, confusing does it work with can it generate revenue.
回到巴里的观点，他的意思是这些公司将芯片折旧年限设为六年或可能是七年，但实际上它们并不会采取任何行动，这实质上会导致它们夸大利润。他说聪明的投资者会察觉这一点，并因此调低它们的估值。
And just to go back to the Burry point then, what he's saying is because these companies are writing the chip depreciation at six years or maybe, who knows, seven years, but they're not gonna actually be doing anything, that effectively they're gonna be overstating their their profits, and, you know, smart smart investors is he saying that smart investors will catch on and then ding their valuations because of it?
那么风险具体是什么呢？
Or what's the risk there?
正是如此，只需一次会计政策调整，这些公司就可能被迫报告大幅降低的利润。
That's exactly it, is that all we're one account conversation away from having all these companies have to report much lower profits.
而且我们使用利润倍数来评估公司价值。
And, you know, we use we use profit multiples to value companies.
如果一家公司不得不在三年而非五年内对大部分资产进行折旧，其盈利能力将按比例下降。在特定情况下，公司价值可能因会计师要求将折旧期从五年改为三年而下跌40%。
So if a company has to depreciate most of its assets over three years instead of five years, that means their profitability is gonna go down proportionally, and we could have in a stylized case, the value of a company declined by 40% because an accountant said, you have to depreciate this over three years instead of five years.
这就是为什么这个问题非常现实。
That's why this is very real.
展开剩余字幕（还有 338 条）
听起来可能有些晦涩，但这确实非常现实。
Sounds wonky, but this is very real.
如果将公司盈利能力降低40%，其价值也会相应下降40%。
If you reduce a company's profitability by 40%, their value will go down by 40%.
顺便说，这正是我认为有必要在节目中讨论这个话题的原因——就像债务问题一样，当涉及这类情况时事情就开始复杂化，而巴里讨论的远不止债务问题。
This is, by the way, this is why I thought it was important to have this conversation on the show because it is like when debt gets involved, that's when things start to get and and Burry's talking about more than debt.
对吧？
Right?
他提到这些折旧成本甚至会影响像微软这样没有为此承担巨额债务的公司。
He's talking about because these depreciation costs can even hit the companies like the Microsoft's that aren't taking on a tremendous amount of debt to do this.
所以这是另一个问题。
So that's another issue.
这些关于债务、折旧的讨论，正是AI泡沫话题的关键所在。
So these these conversations, debt, depreciation, this is where the rubber meets the road on the AI bubble conversation.
说估值离谱是一回事。
It's one thing to say the valuations are out of whack.
但指出这些实际压力点又是另一回事，而这些就是真正的压力点。
It's another to say here's actual the actual pressure points, and these are the pressure points.
没错。
That's right.
我再强调一次，微软、亚马逊、谷歌这些公司能够应对。
And and again, I go back to Microsoft, Amazon, Google, they can handle it.
它们业务规模庞大。
They have a big business.
而且业务多元化。
It's diversified.
这只是它们业务的一部分。
This is only one part of the business.
它们完全可以立即刹车。
They can they can literally stop on a dime.
再说它们现在动用现金，是因为这些钱躺在资产负债表上只能获得4%的回报。
Again, they're deploying cash cause right now that cash is sitting on their balance sheet and generating 4% returns.
他们说，嗯，AI这东西可不得了。
And so they're saying, Well, this AI thing's huge.
根据我们的计算，我们认为它能带来15%的回报率。
We think it can generate 15% returns based on our math.
我们就用这4%的现金在这里部署吧。
Let's use the 4% cash and deploy it here.
我们认为能获得那些回报。
We think we can get those returns.
但话说回来，一旦我们认为回报停止，我们可以立即停止资本支出，经历几年不做任何资本支出的时期来消化前期投资，这些公司照样能运转良好。
But by the way, the second we think that stops, we could stop our CapEx on a dime, go through a couple years where we don't do any CapEx while we absorb the previous investment, and those companies will be just fine.
问题出在那些要么负债累累、要么正在放贷、要么在高杠杆实体中有股权投资的公司身上。
It's the companies that either have the debt or are lending the money or have equity investments in highly leveraged entities.
这些才是能应对那种转型的公司。
Those are going to be the ones that can handle that kind of a transition.
那么，如果OpenAI真能实现Sam Altman承诺的一切，是否可以说这些都能被原谅？
As well, does this all is it all sort of forgiven to use that term if OpenAI just delivers what Sam Altman promises?
是啊，或许我们应该建立这样的认知框架：AI领域存在三大阵营对吧？
Yeah, maybe it's worth having the the our mental framework for looking at AI is that there's three camps, right?
我们需要对三种结果进行加权概率分析。
There's we have to take a weighted probability of three outcomes.
悲观派认为AI就像元宇宙或社交媒体一样——有趣但华而不实，虽然是有用的技术，但我们实在为它花了太多钱。
There's the pessimist outcome, which is AI is cute and it's useful, but it's cute like, like the metaverse, or it it it's cute like social media, in which case it'll be a useful technology, but really, we're spending way too much money on it.
现在很多人确实持这种观点。
Then there's the op and a lot of people are in that camp right now.
我不太同意这个观点。
I'm not sure I agree with that.
乐观的设想是，AI是长期以来最强大的技术。
There's the optimist scenario, which is AI is the most powerful technology in a long time.
它将强大到极大提升我们的生产力，从而推动GDP增长加速。
It will be so powerful that it will make us so much more productive that it will drive an acceleration in GDP growth.
这就是微软、亚马逊、谷歌等公司从他们角度所持的立场。
This is where Microsoft, Amazon, Google are at from their perspective.
然后还有极端的设想：我们可能距离超级智能仅剩几年时间——这种技术能在任何人类擅长的事情上超越人类。这种情况下它将大规模取代我们，为拥有这项技术的任何人创造难以估量的财富，因此我们无论如何投资都不足以实现这个目标，尤其当你相信只有最先实现的那个实体将掌控一切时。
And then there's a maximalist scenario, which is we are maybe as close as a couple of years away from superintelligence, a technology that can do anything a human can do better than any human, in which case it's going to replace us en masse, create untold wealth to whoever owns a piece of that, and therefore no investment we could possibly make is enough in order to get there, especially if you believe that only one the one entity that gets there first will own everything.
这就是马克·扎克伯格、山姆·奥特曼、埃隆·马斯克和达里奥·莫迪所持的立场。
This is where Mark Zuckerberg lives, Sam Altman, Elon Musk, Dario Mode.
他们身处这种极端阵营，认为这是一场我们输不起的竞赛，因此必须全力以赴。
They live in this maximalist camp of this is a race that we can't afford to lose, and therefore we need to build everything we can.
现在，这三种情况都有可能发生，所以你必须提前做好规划。
Now, all three of these things are possible, and so you have to plan ahead for that.
但我们大多数人属于乐观阵营，认为应该大力投资。
But most of us are in this optimist camp, which is we should invest a lot.
只是实施过程中需要深思熟虑、谨慎行事。
We just have to be thoughtful and careful about how we do it.
这样即使我们错了，面临悲观情景时，也不会导致全盘崩溃。
So if we're wrong, and it's the pessimist scenario, we don't bring everything down
经济不会因此瓦解。
with It's the economy doesn't fall apart.
知道的话，那会很好。
Know, that would be good.
顺便说一句，我们要给自己留有余地，也许最极端的设想也是对的，所以至少要在事情发生时靠近边缘，这样我们才能保持竞争力。
And by the way, let's leave ourselves room that maybe the maximalist scenario is right too, so let's at least be near the rim when that happens so we can be competitive there.
这才是看待这一切的健康方式。
That's the healthy way to see all this.
再次强调，这考虑到了人工智能非常强大这一事实。
Again, it accounts for the fact that AI is very good.
它将会大幅增长。
It will grow a lot.
顺便说一下，当我们谈论OpenAI这个实体及其所作所为时，首先要肯定他们的贡献，但也要指出一些问题。
By the way, OpenAI, when we step and talk about OpenAI as an entity and what it's doing, first of all, I wanna give them credit and put some blame as well.
我们要肯定他们在2022年11月时，谷歌已经掌握了GPT技术却将其束之高阁这件事上的贡献。
So let's give them credit for the fact that in November 2022, Google was sitting on GPT and it left it in what they call the pantry.
他们选择不推出GPT，是因为不知道如何处理它，也不确定是否应该与世界分享。
They chose not to introduce GPT because they didn't know what to do with it or if it was a good idea to share it with the world.
而OpenAI站出来说，这简直不可思议。
And OpenAI came out and said, this is unbelievable.
人们会爱上这个的。
People are gonna love this.
他们推出了一款伟大的消费产品，就是我们现在熟知的ChatGPT，如今拥有8亿周活跃用户，创造了有史以来最快的初创公司增长记录。
And they came out with a great consumer product that we now know as Chad GPT, and now has 800,000,000 weekly active users, and has driven the fastest growth of any startup ever.
为此要给予他们肯定。
Give them credit for that.
那么我们来谈谈他们最近行为的危害性——他们把野心扩张到了极点，做出种种无法兑现的承诺，当这些被揭露时，他们正拖累所有人一起沉沦。
Then let's talk about the detriment of their behavior recently, which is they have extended their ambition to a point where they've made all these commitments that they can't possibly live up to, and as that's been exposed, they've been dragging everybody down with them.
所以说OpenAI确实是个独特的存在。
So OpenAI is just this unique entity.
现在，如果你问我OpenAI应该怎么做？
Now, if you were to ask me what should OpenAI do?
确实有人问过我这个问题。
Some people do ask me that question.
我会说只需专注于ChatGPT。
I would say just focus on ChatGPT.
只需专注打造最顶尖的前沿模型，全力发展ChatGPT。
Just focus on having the best frontier model, ramp ChatGPT.
这是个了不起的产品。
It's an amazing product.
你们已经抢占了先机。
You have a head start.
人们已经把ChatGPT当作动词使用。
People are using ChatGPT as the verb.
就像过去人们说'我去谷歌一下'。
Like they used to say, I Google this.
现在人们会说'我去ChatGPT一下'。
People are saying, I ChatGPT this.
如果他们专注于此并负责任地发展，必将取得巨大成功。
If they just focused on that and grow responsibly, they will be very successful.
如果他们坚持当前路线，过度投入，执意自建数据中心、自研硬件、自产芯片，他们将无法成功。
If they go down the current path, overcommitting, deciding that they have to build their own data centers, they need their own hardware, they need their own chips, they won't make it.
因此，希望他们为了客户和股东的利益，专注于真正擅长的领域——也就是这种模式和ChargeAPT。
So hopefully, for benefit of their customers and their shareholders, I hope they focus on what they're really good at, which is this model and ChargeAPT.
好的，关于这点我还有更多问题要请教你。
Okay, I have more questions about this for you.
我们需要稍作休息，先暂停片刻，回来后再继续讨论。
We do need to take a break, so let's hop away for a moment and come back and continue.
看来我们得把这期节目称为'AI泡沫特别报告'了，因为要讨论的内容实在太多，我们会继续这个话题，并尽量涵盖一些新闻热点。
I guess we're gonna have to call this the AI bubble special report because there's so much to discuss, so we'll continue talking about that, and we will try to hit some of the news.
休息结束后我们马上继续。
So we'll do that right after this.
寻找合适的科技人才不仅困难，更是关乎企业存亡的关键——然而许多雇主仍在使用过时的招聘方式或低效平台。
Finding the right tech talent isn't just hard, it's mission critical, and yet many enterprise employers still rely on outdated methods or platforms that don't deliver.
当今市场环境下，招聘科技人才不仅是为了填补职位空缺，更是为了超越竞争对手。
In today's market, hiring tech professionals isn't just about filling roles, it's about outpacing competitors.
但面对专业技能要求、混合办公偏好和高薪资期望，要突破重围找到合适人选变得前所未有的困难。
But with niche skills, hybrid preferences, and high salary expectations, it's never been more challenging to cut through the noise and connect with the right people.
这正是Indeed的价值所在。
That's where Indeed comes in.
Indeed每月稳定发布超过50万个科技职位，使用其平台的雇主能享受精准定向投放技术，通过科技人才网络分发可使申请启动量提升2.1倍。
Indeed consistently posts over 500,000 tech roles per month, and employers using its platform benefit from advanced targeting and a 2.1 x lift in started applications when using tech network distribution.
如果我要招聘顶尖科技人才，我一定会选择Indeed。
If I needed to hire top tech top if I needed to hire top tier tech talent, I would go with Indeed.
发布您的首个职位，即可在indeed.com/techtalent享受75美元优惠。
Post your first job and get $75 off at indeed.com/techtalent.
访问indeed.com/techtalent即可领取此优惠。
That's indeed.com/techtalent to claim this offer.
确实如此。
Indeed.
为当下及未来的科技招聘需求而打造。
Built for what's now and for what's next in tech hiring.
Capital One的技术团队不仅在多智能体AI领域有所探讨。
Capital One's tech team isn't just talking about multiegenthic AI.
他们已成功部署了一套系统。
They already deployed one.
名为'聊天礼宾'，它正在简化购车流程。
It's called chat concierge, and it's simplifying car shopping.
通过自我反思与实时API校验的分层推理，它不仅帮助买家找到心仪车辆。
Using self reflection and layered reasoning with live API checks, It doesn't just help buyers find a car they love.
还能协助预约试驾、获取贷款预批及评估旧车置换价值。
It helps schedule a test drive, get preapproved for financing, and estimate trade in value.
先进、直观且已投入应用。
Advanced, intuitive, and deployed.
这就是他们的技术实力。
That's how they stack.
这就是Capital One的科技。
That's technology at Capital One.
携手AGNCY，AGNCY，共同塑造企业AI的未来。
Shape the future of enterprise AI with AGNCY, AGNCY.
作为Linux基金会的开源项目，AGNCY正在引领为智能体互联网建立可信身份与访问管理体系的潮流，这一协作层确保AI智能体能够安全地发现、连接并在任何框架下协同工作。
Now an open source Linux foundation project, AGNCY is leading the way in establishing trusted identity and access management for the Internet of Agents, a collaboration layer that ensures AI agents can securely discover, connect, and work across any framework.
通过Agency，您的组织将获得开放、标准化的工具和无缝集成能力，包括强大的身份管理系统，实现跨平台的身份识别、认证与交互。让您能自信地部署多智能体系统。与思科、戴尔科技、谷歌云、甲骨文、红帽等75家以上支持企业共同参与，为安全可扩展的AI基础设施制定标准。
With Agency, your organization gains open, standardized tools, and seamless integration, including robust identity management to be able to identify, authenticate, and interact across any platform, Empowering you to deploy multi agent systems with confidence, join industry leaders like Cisco, Dell Technologies, Google Cloud, Oracle, Red Hat, and 75 plus supporting companies to set the standard for secure, scalable AI infrastructure.
您的企业准备好迎接Vagintic AI的未来了吗？
Is your enterprise ready for the future of Vagintic AI?
立即访问agency.org探索应用场景。
Visit agency.org to explore use cases now.
网址是g n t c y.0rg。
That's a g n t c y.0rg.
欢迎回到《大科技》播客周五特别版，今天我们邀请到D集团技术研究负责人Gil Luria。
And we're back here on Big Technology podcast Friday edition with Gil Luria, the head of technology research at D.
A.
A.
Davidson。
Davidson.
Gil，我们刚才讨论了很多关于AI发展可能带来的风险，最后谈到了OpenAI。
Gil, we've been talking a lot about the potential, the potential, risks here of the AI build out, and, we ended with OpenAI.
让我们直接回到OpenAI这个话题。
Let's just, go right back to OpenAI here.
有没有可能——就像你上半场提到的——已经可以把表现良好的公司和表现不佳的公司区分开来了？
Is it possible that already, you know, you you you kind of in the first half separated the companies that are behaving well with the companies, that are not?
我只想问你这件事。
I just wanna ask you this.
有没有可能企业已经对OpenAI过度依赖了？
Is it possible that companies are already too leveraged on OpenAI?
这是《华尔街日报》的报道。
Here is, The Wall Street Journal.
科技巨头利润飙升的背后存在阴暗面。
Big tech soaring profits have an ugly underside.
OpenAI的亏损情况。
OpenAI's losses.
事情是这样的。
Here's the story.
英伟达、Alphabet、亚马逊和微软的季度利润飙升。
Quarterly profits soared at NVIDIA, Alphabet, Amazon, and Microsoft.
这些收入是否与AI相关？
Is AI revenue related or in.
现金流总体良好，不过大量资金正用于建设新数据中心。
Cash flows are mostly fine, albeit a lot is now going into building new data centers.
部分资金来自向企业实际销售AI服务。
Some of the money comes from actually selling AI services to businesses.
但多数AI相关利润来自作为供应商或投资者，支持那些开发AI聊天机器人背后大语言模型的私营公司，而这些公司烧钱速度和融资速度一样快。
But much of the AI related profits come from being a supplier to or an investor in the private companies building large language models behind AI chatbots, and they're losing money as fast as they can raise it.
OpenAI和Anthropic是AI亏损的无底洞，与上市公司利润的大幅增长形成鲜明对比。
OpenAI and Anthropic are sinkholes for AI losses, that are the flip side of the chunks of flip sides of the chunks of the public company profits.
我觉得这个故事说的是大约60，好吧。
I think this story says something like 60 okay.
这里这里，就在这里。
Here's here here it is.
OpenAI本季度的亏损相当于微软、英伟达、Alphabet、亚马逊和Meta基础盈利增长总和的65%，这还没算亚马逊从其持股的亏损公司Anthropic中录得的95亿美元季度利润。
OpenAI's loss in the quarter equates to 65% of the rise in the underlying earnings of Microsoft, Nvidia, Alphabet, Amazon, and Meta together, and that ignores Anthropic, which Amazon recorded a profit of 9,500,000,000.0 from its holding in the loss making company in the quarter.
所以我们看到的这些公司利润——虽然不是全部，但肯定是大部分——其实只是这两家公司为扩张投入的资金。
So all these profits that we're seeing from these companies, not all, but certainly the majority is just the money that these two companies are spending on the build out.
这怎么与那种观点相符呢？我是说，也许他们做得很负责任，但这些公司的股价今年确实都大幅上涨了。
How does that equate with this idea that, you know, I mean, maybe they're doing it responsibly, but certainly the the all these companies stock, share prices have have jumped dramatically this year.
那么回到我们的泡沫问题，这不也是个问题吗？
And so, going back to our bubble question, isn't that a problem too?
是的。
Yeah.
绝对是的。
Absolutely.
不过让我们稍微分析一下。
But let's let's parse that out a little bit.
首先，OpenAI是微软Azure增长的重要推手，而Azure又是微软最重要的业务。
Let so first of all, OpenAI is is a really big part of Microsoft Azure growth, and Microsoft Azure is the most important business within Microsoft.
所以我们先聚焦微软，其实亚马逊和谷歌的情况不太一样。
So let's let's focus on Microsoft and just say that this is actually less true about Amazon and Google.
它们对OpenAI甚至Anthropic的依赖要小得多。
They're a lot less reliant on on OpenAI and even Anthropic.
但让我们聚焦微软，他们约半数AI收入——这本身又是Azure增长的主要部分，而Azure又是微软整体增长的最大引擎——都来自OpenAI。
But let's focus on Microsoft and say that half of their AI revenue approximately, which is, again, a big piece of the Azure growth, which is the biggest piece of Microsoft's growth, is coming from OpenAI.
现在我们来谈谈微软AI增长的另一半。
Let's talk about the other half of the AI growth for Microsoft.
AI增长的另一半同样势头强劲。
The other half of the AI growth is very healthy.
这是因为所有企业都是微软客户，他们会向微软表示：'我们开发了这个AI工具，急需算力支持来运行它'。
That's companies, because everybody's a Microsoft customer, going to Microsoft and saying, I built this AI tool, and I really need compute capacity to be able to use it.
'我们会从Azure购买这些服务'。
I'll buy it from Azure.
微软就会回应：'这太棒了'。
And then Microsoft says, That's great.
'我们不仅会提供GPU使用权，还会配套销售数据库产品、数据仓库解决方案和数据编织服务'。
We'll sell you the GPUs, access to the GPUs, but then on top of that, we'll sell you some database products and data warehouse products and data fabric products.
而且顺便说一句，由于你们将使用Copilot，Microsoft 365的授权费用也会上涨——这对微软来说简直是锦上添花。
And, oh, by the way, your Microsoft three sixty five license is gonna go up because you're gonna use Copilot, and this is great for Microsoft.
微软。
Microsoft.
因此其他所有AI相关业务都对微软极为有利，这也是他们表现如此出色的重要原因。
So all the other AI stuff is absolutely great for Microsoft, and it's a big reason why they've done so well.
接下来我们讨论OpenAI这部分。
Then let's talk about the OpenAI piece of this.
确实，这部分存在风险，正如你所说，OpenAI目前是负毛利率业务。
Absolutely, this is the piece that's at risk, because to your point, OpenAI is a negative gross margin business.
他们声称不是这样，但实际上就是，对吧？
They claim they're not, but they are, right?
也就是说，回答你的Chad GPT问题所花费的成本比他们从你那里获得的收入还要高。
Which is to say, it costs them more to answer your Chad GPT question than they make revenue from you.
这是我们需要意识到并关注的事情，对吧？
And that's something we need to be aware of and concerned with, right?
尤其是如果这占微软收入的很大一部分的话。
Especially if it's a big part of Microsoft revenue.
就像是，等一下。
It's like, wait a second.
为公司招人时我刚刚
Getting the guys for a company losing I just
这周看到有人发推说，OpenAI的切换器要么会立即用一些垃圾内容回答你，要么会花1万美元的计算资源来思考你的查询。
saw this week there was, someone who tweeted that like OpenAI Switcher says either they will answer you immediately with some slop or they'll go out and spend $10,000 on compute thinking through, like, your your query.
因为你说得对。
Because you're right.
确实看起来，尤其是这些需要大量推理的查询，会消耗大量的计算处理能力。
It does seem like with these especially with these queries that require a lot of reasoning, it just takes a lot of computing processing power.
我不是有意打断，抱歉。
So I didn't mean to jump in, but, sorry.
所以这就是为什么它很危险。
So that this is why it's dangerous.
对吧？
Right?
但让我们思考一个类比，我认为这对理解微软视角下的合理性非常有用——那就是优步。
But let's let's think about here's an analogy that I think is very useful for understanding why this is okay from Microsoft perspective, and that's Uber.
如果你还记得，优步刚起步时，车费比出租车便宜得多。
If you remember, when Uber started, the rides were a lot less expensive than taxi.
可以说当时全城任何地方都只要10美元车费，这极具吸引力，人人都开始使用优步。
Let's say it was like a let's call it a $10 ride anywhere across town, which was so attractive that everybody started using Uber.
结果是我们使用优步的频率远超从前打车的次数，它开始替代自驾，由于价格优势，整个乘车市场的规模远超从前。
And and what what happened was we were we started using Uber a lot more than we used cabs before, and it started replacing driving, and we expanded the market for riding well beyond where it was because the price was so attractive.
但后来我们改变了行为习惯，使用频率高到让优步能逐步提价，如今30美元的车费已使其成为高利润公司。
But then what happened is we changed our behavior, and we started using it so much that Uber can then gradually ratchet up the price to a point where today Uber's a very profitable company because it's a $30 ride.
虽然有些人可能不再像10美元时期那样频繁使用，但大多数人已养成新习惯，并从中获益良多。
And and some people may may have may not be using it as much as they did when it was $10, but most of us are because we've changed our behavior, and we see a lot of benefit in using it.
现在我们支付着合理的价格。
And now we're paying the appropriate price.
合理价格从来就不是10美元。
The appropriate price wasn't $10.
它始终应该是30美元。
It was always $30.
现在我们愿意支付30美元，因为随着时间推移，我们已认识到这对我们有益。
And now we're willing to pay $30 because we've learned over time that that's beneficial to us.
我们看到了它的价值。
We see value in it.
我们愿意为此付费。
We're willing to pay for it.
这些聊天记录也应该同样处理。
The same thing should happen with these chats.
我们就用ChatGPT吧。
And let's use ChatGPT.
它依然是行业领先的。
It's still the leading one.
对吧？
Right?
目前我可能每月支付20美元。
Right now, I may be paying $20 a month.
很少有人每月支付200美元。
Very few people are paying $200 a month.
但我的邻居简拥有自己的律师事务所，她在业务中如此频繁地使用ChatGPT，以至于可以推迟甚至避免再雇佣一名助理律师。
But my neighbor, Jane, who has her own law firm, is using ChatGPT so much in her practice that she could postpone or even avoid hiring another associate.
假设那名助理年薪是10万美元。
So let's say that associate was $100,000 a year.
如果她每月支付20美元，甚至200美元，这为她创造了巨大价值——未来如果持续这样使用并意识到永远不需要雇佣那名助理，她完全可以用ChatGPT来总结证词、提取关键信息、协助制定策略和生成文件，这样就不必花费10万美元年薪，她可能愿意每年支付1万零20美元。
If she's paid $20 a month, even $200 a month, that creates so much value for her that in the future, if she continues to do this and realizes she never has to hire that associate, she could just use ChatGPT to summarize deposition, extract important information, help her strategize, and and create documents, and she doesn't need to hire $100,000 she may be willing to pay 10,020 thousand dollars a year.
因此随着ChatGPT应用范围的大幅扩展，我们愿意从每月20美元的价位转向更高的付费层级。
So as we use ChatGPT a lot more broadly, we're going to be willing to go from a $20 a month price point to a much higher price point.
所以终有一天，这会是个盈利的产品。
So at some point, this is a product that will be profitable.
我们只需要大幅扩展使用场景，让那些从中获得高价值的人愿意为其实际价值买单。
We just have to expand the usage so much that the people that are using it in a very valuable way will be willing to pay what it's worth.
这就是Uber走过的历程，也是我们聊天领域即将经历的历程。
That's the journey Uber went through, and that's the journey we're gonna go through chat.
不过我要指出的是，与Uber和网约车市场不同——那些领域容易形成赢家通吃的局面，而聊天领域完全不同。因为我可以同时与Gemini对话，Meta会给我工具实现这一点，Grok也会提供相应工具。
What I would point out, though, is that unlike Uber and the rideshare market, which lent itself to winner takes most or winner takes all, chat is entirely not like that because I could have the same conversation with Gemini and Meta's gonna give me the tools to do this and Grok is gonna give me the tools to do this.
所以这不是一个赢家通吃的市场，这意味着要达到足够有利的价格点可能需要更长时间。
So it's not a winner take all market, which means that that process of getting to a price point that is beneficial enough may take longer.
而谷歌和Meta可能会决定永远不这么做，他们愿意支付所有计算成本来把你留在YouTube和Instagram里。
And Google and Meta may decide that they never wanna do that, that they're willing to pay for all this compute to keep you in YouTube and keep you in Insta.
这就是像OpenAI和聊天领域公司面临的风险所在。
And that's where the risk is to a company like OpenAI and chat.
但如果你是微软就没关系，因为你只需用数据中心容量来托管Grok或其他有价值的聊天服务。
But if you're Microsoft, that's okay because you'll just use your data center capacity to host Grok or to host another chat that is worthwhile.
如果不行，你也可以把这些容量租给企业客户，他们用这些资源创造商业价值，愿意支付溢价购买数据库、数据仓库等等。
And if not, you'll rent that capacity out to your business customers that are using it to produce more value in their business and therefore willing to pay that premium and then buy databases and data warehouses, etcetera, etcetera.
这就是为什么当Meta承担债务并增加资本支出时市场会恐慌，因为相比微软这样拥有数据中心的公司，Meta这类公司更难让人看清其直接的发展路径是否稳妥。
And that's why the market freaked out when Meta, I guess, was taking on this debt and increasing its CapEx because it's harder to see that direct line of it's gonna be okay if you're a company like Meta versus a company like Microsoft that has the data centers.
没错。
That's right.
完全正确。
That's exactly right.
关于Meta发生了几个关键变化。
And so a couple of things happened with Meta.
因为Meta的情况很特殊，他们实际上没有企业客户可以出租这些计算容量。
Because Meta, again, is an unusual situation because they don't actually have business customers to rent this capacity to.
这其实就是扎克伯格先生想要一个越来越大的玩具罢了。
It really is just mister Zuckerberg wanting a bigger and bigger toy.
马克·扎克伯格的奇幻冒险。
Mark Zuckerberg's magical adventure.
再说一次，记住，这一切发生的原因在于他是个AI极端主义者。
And and again, remember, the reason that's happening is that he's an AI maximalist.
他认为可能再过几年就能拥有一个强大到足以统治一切的工具。
He thinks that we may be a couple of years away from having a tool so powerful he will get to rule it all.
这就是他愿意投入的原因。
That's why he's willing to spend.
他上次对投资者说的是：我正在管理这个不可思议的业务。
And what he told investors last time was, I am managing this unbelievable business.
我刚把广告收入提升了25%。
I just grew ad revenue by 25%.
这么做让我获得了难以置信的利润。
I'm unbelievably profitable doing that.
但明年我不会保持克制只增加25%支出，而是要远超这个数字，投入比25%多得多。
But instead of being disciplined and spending 25% more next year, I'm gonna go well beyond that and spend a lot more than 25% next year.
投资者对此的回应是：马克，这看起来很不负责任。
And that's what investors said, That seems irresponsible, Mark.
你今年要花的钱实在太多了。
That's a lot of money you're spending this year.
为什么明年不干脆只增加25%的支出呢？
Why don't you just spend 25% more next year?
当他说，不。
And when he said, no.
我不仅要做到那样，他们甚至抛售了股票。
I'm gonna go well beyond that, they they sold the stock.
对。
Right.
还有之前发生的另一件事。
And the other thing that happened before.
是啊。
Yeah.
他对此没有意见。
And he's okay with it.
他拥有整件事的主导权。
He owns the whole thing.
在他看来，这是他的钱，而且...没错。
He as far as he's concerned, it's his money, and and Right.
这就是他的行事方式，而且至今都行之有效，所以我不打算质疑他。
That's how he behaves, and it's worked out for him so far, so I don't know that I wanna challenge him.
Meta发生的另一件有趣的事是，当他们去借款时，并没有直接借入资金。
The other thing that happened with Meta that was interesting is that when they went out to borrow, they didn't borrow the capital.
他们成立了一个特殊目的实体。
They created a special purpose vehicle.
对吧？
Right?
他们与BlueOwl合作时表示，会投入几十亿美元。
They went out with BlueOwl, and they said, you know, we'll put a couple of billion in.
BlueOwl，你再追加一些资金，就能借到十倍于此的金额，为我们建设数据中心。
BlueOwl, you put a couple more, and then you can borrow 10 times that and build data center capacity for us.
这件事触动神经的原因是——不知你是否记得我们最初开始使用'特殊目的实体'这个术语的时候。
And the reason that hit a nerve is, I don't know if you remember when we really started using the term special purpose vehicle.
大约二十五年前安然事件时，对吧？
It's about twenty five years ago with Enron, Right?
特殊目的实体本身并不违法。
And and and now a special purpose vehicle in itself, not illegal.
隐瞒它才是违法的，安然就是这么做的，所以难以置信的是我们居然真能把人送进监狱。
Hiding it was illegal, and that's what Enron did, and that's why, incredibly, we actually got to put somebody in jail.
但特殊目的实体的本质是Meta在说：资本市场现在疯狂到愿意借钱给任何人搞Build AI项目，我们要把这当成无限金钱漏洞，让别人去借钱。
But a special purpose vehicle is meta saying the capital markets are so irrational now in their ability to lend money to anybody to do Build AI that we're going to use this as an infinite money glitch, and we're going to have somebody else borrow the money.
这笔债务不会出现在我们的资产负债表上。
It's not going to go on debt to our balance sheet.
不会计入资本支出、PPNR或我们的资产负债表。
It's not going to go as CapEx, PPNR, our balance sheet.
它会流向别处。
It's going to go somewhere else.
我们资产负债表上会有一项'经营租赁承诺'，但金额小得多，人们也不太关注这一项。
We will have a line item in our balance sheet that says operating lease commitments, but it'll be a lot smaller, and people don't pay attention as much to that line.
他们凭什么要关注呢？
And why wouldn't they?
如果他们能做到，为什么不做呢？
If they can, why wouldn't they?
这又是那种让人们感叹‘哦，这可不健康’的事情之一。
And this is, again, one of those things that got people to say, oh, this is unhealthy.
我们可不想再经历一次这种事。
We don't wanna be doing this again.
我们都知道二十五年前这事是怎么收场的。
We know how this ended twenty five years ago.
没错。
Right.
吉尔，能再占用你几分钟时间吗？
Gil, can I keep you here for another couple minutes?
我想在离开前和你聊聊这个AI囚徒困境的问题。
I wanna talk to you about this AI prisoner's dilemma before we head out.
因为问题在于——今天我们讨论了很多关于债务、贬值和OpenAI履约能力的话题，但接下来还有个核心问题：任何事物如何实现盈利？这并不简单。
Because because the question is, again, like, we've talked a lot today, I think appropriately about debt, about depreciation, about OpenAI's ability to, pay back its, or to actually meet its commitments, but then there's this question of how anything becomes profitable, and it's not so simple.
我在彭博社看到的报道显示，OddLots团队提出可能存在某种博弈论机制，会导致长期无法盈利。
And so I think what what I'm seeing here from Bloomberg is that there's a suggestion from the OddLots team that there's some game theory involved that might keep this unprofitable for a long time.
他们引用了一份分析师报告，指出推理定价中存在某种囚徒困境——推理当然是指实际使用模型而非训练它们。
So they're quoting this one report, an analyst suggested that there's a prisoner's dilemma of sorts in inference pricing inference, of course, is when you actually use the models versus train them.
如果所有推理应用都根据服务质量按使用量收费，市场或许能保持稳定。
If every inference application prices its services based on quality and charges by usage, the market might remain stable.
但由于股权投资者和风投更看重市场份额而非利润率，每家推理公司都有动机推出不限次数的统一定价，最终引发恶性竞争。
But because market share is more important than margins for the equity investors and the venture capital investors supporting these inference firms, every firm has a greater incentive to offer flat rate pricing with unlimited usage, triggering a race to the bottom.
顺便说一句，这适用于OpenAI和所有人。
And by the way, that would apply to OpenAI and everyone.
据说，所有人都在补贴重度用户，所有人都会发布曲棍球杆式增长图表，最终所有人都会发布重要的定价更新。
So he they say they say everyone subsidizes power users, everyone posts hockey stick growth charts, everyone eventually posts important pricing updates.
现在来看彭博社的消息。
Now here's Bloomberg.
人工智能不是普通技术。
AI isn't normal technology.
目前尚不清楚是否会有那么一天——当某人能够有把握地说：'你知道吗？'
It's not clear whether they will be at some point whether there will be at some point ever when someone will be in position to say, you know what?
'这就足够了。'
This is good enough.
虽然已是陈词滥调，但人们谈论AI时，要么像在创造新神，要么像在制造核弹，声称必须赶在地球上任何国家之前实现突破。
It's cliche by now, but people talk about AI like they're building a new god or they talk about it like they're building nuclear bomb, and we have to get there before any country on Earth does.
事实上，正因为这些巨大的利害关系，最近几周有讨论称美国政府可能会为某些公司提供支持，包括部分融资和债务。
In fact, it's because of these huge stakes that in recent weeks there's been talk about how a US government might backstop some of these companies, some of the companies financing and debt.
所以本质上...我的意思是他们想表达的是：这不像理性科技企业的行为，因为一方面所有人都想要市场份额，另一方面他们愿意为此投入资金。
So that that basically what there's I mean what they're saying is you you, this is not acting like a rational, technology because a, everybody wants market share and b, yeah, they're they're willing to spend, to get there.
对此你怎么看？
And so what do you think about that?
这会成为长期问题吗？
Is it is this gonna be a persistent issue?
我们该如何看待这件事？
How should we view this?
是的。
Yes.
是的。
Yes.
因为事情是这样的。
Because so here here's the thing.
这场博弈论中的玩家都是谁？
Who are the players in this game theory?
对吧？
Right?
是Meta。
It's Meta.
是谷歌。
It's Google.
是微软。
It's Microsoft.
是亚马逊。
It's Amazon.
这些公司习惯赢家通吃的市场，并把所有市场都视为赢家通吃的市场。
It's companies that are used to winner take all markets, and they think of all markets as being winner take all market.
意思是，如果我不能赢得这个市场，我将一无所获，或者至少获得的份额不足以对我产生意义。
Meaning, if I don't win this market, I'll get none of it or at least not enough of it that will be meaningful to me.
所以他们愿意不惜一切代价去赢，甚至愿意长期亏损，只为获得获胜的机会。
So they are willing to do anything to win, which to the point of that means they'll be willing to lose money for a long period of time so they have a chance to win.
于是最终只有资金最雄厚的大玩家才能胜出，因为他们可以耗到最后，或者至少向所有人表明他们愿意坚持到底。
And what happens then is it's only the biggest, most deepest pocket player that can win because they they can wait it out or at least communicate to everybody else that they're willing to wait it out.
这就是像OpenAir这样的公司毫无机会的原因——他们无法在这种支出水平下再支撑一两年，自然更不可能在这场游戏中耗过谷歌、Meta和微软。
So that's that's where a company like OpenAir has no chance because they're they can't make it through another year or two at at this level of spending, so they certainly won't be able to outlast Google Meta and Microsoft in this game.
对吧？
Right?
这也解释了扎克伯格先生的很多行为。
So and it explains a lot about mister Zuckerberg's behavior.
重申一次，他不仅是在花钱。
Again, he's not just spending the money.
他是在向我们宣告：为了胜利他愿意付出任何代价。
He's telling us he's willing to spend anything to win.
他向所有竞争对手传递的信号是：我绝不会输。
He's signaling to all the other players is I will not lose.
所以你们可以继续砸钱。
So you can keep throwing money at this.
但我会比你们砸得更久。
I'll keep throwing money at it longer.
这正是当前的局面，也是我们可能持续亏损一段时间的原因——这些巨头公司资金实在太雄厚了。
And that is exactly where we're at, which is why we may have persistent losses for a while here, because these companies have very deep pockets.
再说一次，小公司要么被大公司吞并，要么只能退出，而那些巨头坚信这是赢家通吃的游戏。
Again, the smaller ones will either get absorbed by the big ones or just have to walk away, but those big ones believe this is winner takes all.
顺便说一句，你应该能看出我并不确定这是否真是赢家通吃的局面。
By the way, you can tell I'm not sure it's winner takes all.
我认为随着时间的推移，我们可能会使用不同的聊天程序，企业也会采用不同的人工智能技术，因此我不确定赢家能通吃一切。
I think we could be using different chat programs over time and companies will be using different AI, So I'm not sure winners takes all.
我确实认为这些公司中有一些可以共同成功，但我也认为分析是正确的，他们视之为赢家通吃的局面，他们正在竭尽全力不仅留在游戏中，还要向所有人传达信号，表明他们仍在参与竞争。
I do think that some of these companies can all succeed together, but I do think the analysis is correct and that they see it as winner takes all, and they're doing what they can to not only stay in the game, but communicate to everybody else, signal to everybody else that they're staying in the game.
因此彭博社的文章提到了这种风险。
And then the Bloomberg piece brings up this risk because of that.
对吧？
Right?
所以可能会出现，不是信贷紧缩的信贷紧缩，而是抵押品紧缩。
So that there could be, not a credit crunch credit crunch, but a collateral crunch.
对吧？
Right?
抵押品紧缩是指支撑所有这些贷款的资产价值突然崩溃。
A collateral crunch is the sudden collapse in the value of assets underpinning all these loans.
然后他们引用了雷蒙德詹姆斯公司的首席经济学家杰夫·绍德在金融危机前的声明：风险在于传染效应会蔓延，并从抵押品紧缩演变成全面的信贷紧缩，这正是2008年发生的情况。
And then they quote this chief economist from Raymond James, Jeff Saud, who made this statement before the financial crisis, the risk is that the contagion spreads and morphs from this collateral crunch into a full blown credit crunch, that is exactly what happened in 2008.
所以这是否意味着，这某种程度上概括了我们一直在讨论的内容，即你可能会看到一种传染效应，因为人们在这些交易中的少数几笔上栽了跟头。
So is that is that I mean, it sort of kind of encapsulates what we've been talking about, that, you could you could see a contagion here from, you know, people being burned on a handful of, of these deals.
也许，我只是随便说说。
Maybe it's, you know, just throwing it out there.
也许是甲骨文的交易。
Maybe it's the Oracle deal.
也许是CoreWeave。
Maybe maybe CoreWeave.
然后说，好吧。
And then saying, alright.
我就是不借。
I'm just not lending.
我真的要收紧我的借贷政策，因为这次搞砸了。
I'm gonna really tighten up my lending practices because this went bad.
是啊。
Yeah.
我认为这就是我们的方向。
And I think that's where we're headed.
所以，再说一次，我们已经投入了数百亿美元债务，这意味着如果它消失，有些人会受伤，但不会影响整个系统。
So, again, we're tens of billions of in, so of dollars of debt into this, which means if it goes away, then some people get hurt, but not the whole system.
只有当我们投入数千亿美元时，我们才会受伤。
It's only if we get hundreds of billions of dollar in that we'll get hurt.
更可能的情况是，目前市场上发生的事情可能会吓到这些承销商。
Well, the more likely scenario is that what's happening right now in the market might scare these underwriters straight.
他们将停止发放这些不负责任的贷款，我们将重新通过那些拥有客户、有资金实力、并且有雄厚财力渡过难关的公司的现金流来资助这个项目。
They'll stop making these irresponsible loans, and we'll go back to funding this out of cash flow by the companies that have the customers and have the wherewithal and, again, have the deep pockets to ride this out.
因为如果你把这个情景推演下去，你会发现Meta、微软、亚马逊、谷歌完全预期那些借钱建数据中心的公司会破产。
Because if you play this scenario out, what you'll realize is Meta, Microsoft, and and and Amazon, Google fully expect all those companies that are borrowing to build data center to go bankrupt.
这对他们来说是好事，因为这意味着当价格便宜时。
This is great for them because that means that when Cheap.
他们可以以极低的价格收购资产。
They go They could buy assets at pennies on the dollar.
所以如果我是微软，我知道在两三年内，我可能不需要任何资本支出，因为我能以极低价格收购破产的数据中心。
So if I'm Microsoft, I know that in two or three years, I probably won't have to do any CapEx because I'll be able to buy data centers out of bankruptcy for pennies on the dollar.
所以我干脆让事态发展下去，如果那些不负责任的贷款方想放贷，那是他们的问题。
So I might as well let this play out, and if irresponsible lenders want to make these loans, it's their problem.
当我需要时，我就能以极低价格获得这些资产。
I'm going to be able to capture those assets when I need it at pennies on the dollar.
好的，吉尔。
Alright, Gil.
最后一个问题。
Last one for you.
我们还没讨论的潜在瓶颈是电力问题。
The one thing we haven't talked about is the potential bottleneck on power.
比如萨提亚·安德拉在布拉德·格斯特纳的播客上就谈到过这个。
Satya Andela, for instance, was talking about this on Brad Gerstner's podcast.
谈到他有些芯片无法接入，因为电力供应不上。
Talking about how, like, he has chips he can't plug in because he can't power this.
他没有可用的机架。
He doesn't have warm shelves.
他无法为这些机架供电。
He can't power, the shelves.
微软AI首席执行官穆斯塔法·苏莱曼本周早些时候在节目中提到，他们确实有训练算力，但推理算力是个问题。
Mustafa Suleyman, CEO of Microsoft AI was on this show earlier this week talking about how they do have capacity for training, but inference is a problem.
这是零对冲网站用他们特有的方式表述的。
Here's zero hedge who puts it in a zero hedge way.
有人计算过到2028年美国需要新建多少座核电站，才能支撑这些AI日常运作吗？
Has anyone done the math on how many hundreds of new nuclear power plants The US will need by 2028 for all these AI daily circle jerk deals to be powered?
你对电力问题怎么看？
What do you think about the power question?
在我看来，这确实是个日益严重的问题——如果说有什么能刹住这股AI热潮，可能就是因为电力即将耗尽。
To me, it's a it's an increasing issue that, like, if anything is gonna put the brakes on this, maybe it's just that the power will will run out.
所以电力确实是瓶颈所在。
So it it power is the the bottleneck.
但在我们这样的市场经济中，当涉及足够大的收入和利润时，我们总能突破瓶颈，现在和未来都是如此。
But what happens is in a market based economy such as ours is that when there's enough revenue and profits at stake, we work our way through bottlenecks, and that's what's happening and will happen here.
是的，电网可能无法提供足够容量来启动数据中心，因为在用电高峰时他们无法保证供电。
So yes, the grid may not be able to give us enough capacity to turn on a data center because at peak, they can't have us access to electricity.
但通过储能方案，他们可以部分满足需求。
But with storage solutions, they could give us some.
这些公司正在做的是部署所谓的'电表后端'电源——也就是发电机、涡轮机和柴油卡车，因为利润太丰厚了，值得他们停放10辆柴油卡车持续运行来给芯片供电，毕竟出租这些芯片能赚大钱。
And then what these companies are doing is putting power, what they call, behind the meter, which is to say generators and turbines and diesel trucks because it's so lucrative that it's worth it for them to park 10 diesel trucks and run them so they can power those chips because they make so much money renting out those chips.
所以我们总会找到办法。
So we will find a way.
市场自有出路，我们终将突破这个困境。
Markets find a way, and we're gonna find a way through this.
关键是要有创造力。
It's just a matter of being creative.
而且你知道，这行利润非常丰厚。
And, you know, it's very lucrative.
如果你现在是电工或暖通技术员，伙计，你可赚大发了。
If you're an electrician right now or an HVAC technician, boy, are you making bank.
你坐着私人飞机到处飞，拿着双倍薪水去安装数据中心。
You're getting flown on private jets to and making twice as much money so you can install a data center.
所以现在是当电工或暖通技术员的好时机。
So it's a good time to be electrician or HVAC technician.
吉尔，这些事总有一天能拍成部好电影。
This is all gonna make a great movie one day, Gil.
希望别像《Yes》那么惨烈。
Hopefully not as devastating as Yes.
电影里由艾德里安·布洛迪来演我。
I have Adrian Brody playing me in the movie.
好吧。
Okay.
听起来不错。
Sounds good.
这里是AI泡沫特别报道。
This has been an AI Bubble special report.
吉尔·卢里亚，非常感谢你的参与。
Gil Luria, thank you so much for joining us.
我...你知道，我觉得我需要这个。
I I, you know, I feel like I needed this.
我们需要这个。
We needed this.
这是我期待已久的深度探讨，很高兴我们终于完成了。
It's the deep dive I've been waiting to do, and I'm so glad we did it.
感谢你参加节目。
So thanks for coming on the show.
谢谢，亚历克斯。
Appreciate it, Alex.
很享受这次对话。
Enjoy the conversation.
我也是。
Same here.
好的，各位。
Alright, everybody.
非常感谢你们的收听和观看，无论你们是通过Spotify还是YouTube。
Thank you so much for listening and watching if you're on Spotify or YouTube.
尼克·克莱格，Meta全球事务前总裁、英国前副首相，将于周三做客节目，讨论我们能否信任硅谷掌控超级智能。尼克对超级智能的经济价值有些非常有趣的见解，包括拥有它是否具有经济意义。
Nick Clegg, the former president of global affairs at Meta, former deputy prime minister of The United Kingdom, is coming on on Wednesday to talk about whether we could trust Silicon Valley with superintelligence, and Nick has some really interesting thoughts about the economic value of superintelligence, whether it'll even make financial sense to own it.
希望大家届时收听。
So we hope that you tune in then.
感谢收听，我们下次《大科技播客》再见。
Thank you for listening, and we'll see you next time on big technology podcast.
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。