AGI或破产，OpenAI的万亿美元豪赌，苹果下任CEO是谁？

我们是否已经到了AI投资狂潮意味着要么是通用人工智能（AGI）要么是市场崩溃的境地？OpenAI的1万亿美元基础设施投资可能会影响我们所有人，而他有望成为苹果下一任CEO。我们稍后将全面报道。Capital One的技术团队不仅在多智能体AI上讨论，他们已经部署了一个。

Are we now at the point where the AI investment frenzy means it's AGI or market collapse? OpenAI's $1,000,000,000,000 infrastructure investment may impact all of us, and he's the front runner to be Apple's next CEO. We'll cover it all right after this. Capital One's tech team isn't just talking about multiegenthic AI. They already deployed one.

它被称为聊天礼宾服务，正在简化购车流程。通过自我反思和分层推理结合实时API检查，它不仅帮助买家找到心仪的车辆，还协助安排试驾、获得融资预批准和估算置换价值。先进、直观且已部署。这就是他们的实力所在。

It's called chat concierge, and it's simplifying car shopping. Using self reflection and layered reasoning with live API checks, it doesn't just help buyers find a car they love. It helps schedule a test drive, get preapproved for financing, and estimate trade in value. Advanced, intuitive, and deployed. That's how they stack.

这就是Capital One的技术。欢迎来到《大科技播客》周五版，我们将以一贯冷静、细致的方式解析新闻。今天节目很精彩。我们将探讨所有这些疯狂的AI投资，并自问：我们没有敲响警钟，说情况在好转前会变得更糟，是不是疯了？大量资金涌入AI，是时候看看这些钱实际流向哪里、基于什么前提，以及是否对OpenAI押注过大，也许是把整个美国甚至全球经济押在一家公司的命运上。

That's technology at Capital One. Welcome to Big Technology podcast Friday edition where we break down the news in our traditional cool headed and nuanced format. We have a great show for you today. We're gonna cover all of this crazy AI investment and ask ourselves, are we insane for not sounding an alarm and saying this is gonna get much worse before it gets better? A lot of money is going into AI, and it's time to look at where that money is actually going, what it's premised on, and whether it is putting too large of a bet on OpenAI, maybe betting the entire US, maybe the global economy on one company's fortune.

我们还将讨论谁可能是苹果下一任CEO，以及是否现在就该他上位。周五照常加入我们的是《Margins》的Ranjan Roy。Ranjan，很高兴见到你。

We'll also talk about who might be the next Apple CEO and whether if it's time for him to just step up right now. Joining us as always on Friday is Ranjan Roy of Margins. Ranjan, great to see you.

很高兴见到你。我承认，我将成为苹果下一任CEO。消息刚刚爆出。他可能是下一任苹果CEO的热门人选。

Good to see you. I'll admit it. I am gonna be the next Apple CEO. It's breaking right now. He might be the front runner to be the next Apple CEO.

如果你能修复Siri，我想很多人会——所以，没压力。这显然是件容易的事，没多少人尝试过，而且做起来很简单。

If you fix Siri, think a lot of people would So be no, no pressure. It's obviously an easy thing that not many people have tried and it's quite simple to do.

那是我向Tim推销的点子。他买账了。他买账了。他听了一段时间，意识到修复Siri的唯一方法就是让我加入。

That was my pitch to Tim. He bought it. He bought it. He's been listening for a while and he realized the only way to fix Siri is to bring me in.

嗯，我期待着能在库比蒂诺那个巨大的UFO里与你一起录制下一期节目。当然，这前提是等到你接手时全球经济还能正常运转。我现在开始觉得这事并不那么确定了，因为我们目睹了两件事：一是过去两周——自从我们上次对话以来发生了太多事——AI基础设施的投资出现了更大幅度的增长；二是甚至业内都开始出现越来越多的质疑声：这一切到底有没有意义？

Well, I, look forward to being able to do our next set of episodes with you in the big UFO in Cupertino. So that is assuming, of course, that we still have a global economy by the time you take over. And I'm starting to think it's not such a sure thing because, we've seen two things happen. One is an even greater increase in investment in AI infrastructure over the past, let's say, two weeks, and it's been two weeks since we last spoke and there as there's so much that's happened. But as that has come up, there has been an increasing chorus from people even within the industry that's starting to say, does this make any sense at all?

这会是个问题吗？让我们回到红杉资本合伙人戴夫·卡恩，他在一两年前曾写过一篇精彩文章，探讨生成式AI面临的6000亿美元问题——即最终利润是否能真正支撑投资。他现在表示这个问题已经过时了，因为当前建设阶段远超当时。他说：有一点已经很清楚，除非实现通用人工智能（AGI），否则根本无法证明未来十年规划中的投资合理性。这一切都发生在AI潜力逐步显现的同时。

Is this going to be a problem? And let us return to Dave Kahn, the Sequoia partner who wrote, of course, a great piece a couple years ago or a year ago even about the $600,000,000,000 question around generative AI and whether there's going to be enough profit to actually justify the investment. He now says that that question is quaint because we are at a stage of the build out, that is much further than that. So he says this, one thing has become clear, nothing short of AGI will be enough to justify the investments now being proposed for the coming decade. This is happening even as AI's potential is being realized.

ChatGPT年化收入已突破120亿美元持续飙升，Anthropic也以惊人速度达到50多亿美元年化收入，还涌现出一批从零快速成长至1亿美元营收的新公司。存在这样一种世界图景——微软和亚马逊似乎越来越倾向于此——下一个前沿是AI应用落地。模型已被证明非常出色，现在是时候变现投资并推动改变世界的技术革命了。但这种观点远非主流，在这些巨头之外，一场债务驱动的二次推进正在发生。

ChatGPT has continued its epic rise to north of $12,000,000,000 in run rate revenue, Anthropic has reached 5 plus billion dollars in run rate revenue in a meteoric rise, and there's a new club of companies scaling quickly from zero to a $100,000,000 in revenue. There's a version of the world, and this is the version that Microsoft and Amazon increasingly seem to be pursuing, where the next frontier is AI adoption. The models have proven themselves to be great, and now it's time to monetize these investments and drive a world changing technology evolution. But that point of view is by no means widespread. Outside of these giants, a debt fueled second push is happening.

实验室正将所有利润和资本重新投入新数据中心，而Oracle、Meta和CoreWeave等新兴公司更是毫无保留地全力押注。鉴于投资规模，唯一能解释这种策略的目标就是AGI。所以我认为卡恩的观点很有道理：两股势力在同时行动。一方面是像微软这样的公司——他们上周还来节目讨论如何更理性控制AI支出，亚马逊也可归为此类。

Labs are taking all their profits and capital and plowing them right back into new data centers, and a new breed of companies, namely Oracle, Meta, and CoreWeave, are going all in, no holds barred. Given the scale of these investments, the only objective that can be that can explain this strategy is AGI. So I think Khan is making a really good point here. Two things are happening. One is you have companies like Microsoft, which came on this show last week to talk about how it was basically being more rational with AI spending, and and Amazon, you could put them in that bucket.

另一方面是OpenAI、Oracle等推动的疯狂建设浪潮——这些数字只有实现AGI、只有造出能治愈癌症的AI时才说得通，而其中太多都是推测性的，这可能正是我们陷入危险区域的开始。那么兰詹，我想听听你的看法，你怎么想？

And then you start to see this crazy build out that OpenAI, Oracle, and others are driving, that the numbers really only make sense if you get to AGI, if you get AI that cure cancer, and so much of that is speculative, and that is where we might be getting into a danger zone. So, Ranjan, I'm just gonna turn it over to you. What do you think about that?

是的。我很高兴他区分了两种立场：一种是鼓吹‘现在就该变现投资并推动技术革命’，另一种是债务驱动的二次推进。最近几周我密切关注Oracle与AMD等交易，这些交易都聚焦资本支出而非产品。当然过去几周产品端也有惊艳时刻（包括Sohr，我们稍后会聊），我一直期待再和你探讨这个问题。

Yeah. I'm really glad that he's making this distinction between those who are kind of just pushing the idea that it's time to monetize these investments and drive a world changing technology evolution versus this debt fueled second push. I've been looking at this a lot over the last couple of weeks around the Oracle deal, the AMD deal. All these deals are focused on CapEx and not product. Again, I mean, there's been some incredible moments in product over the last couple of weeks, Sohr included, which we're gonna talk about, and I've been waiting to talk to you again.

但总体而言，这种‘必须实现AGI才能证明所有资本支出合理性’的理念，正是Oracle、Meta、CoreWeave等公司押注的方向。但没有任何实际迹象表明我们正朝这个方向迈进，所以开始显得愈发不确定和非理性。目前虽无阻碍力量，但能公开讨论这件事本身就是好事。

But I think overall, this idea that, you know, it has to be AGI. It has to be something that justifies all of this CapEx, you know, investment is what Oracle, Meta, Core Wave, all these companies are betting in. And there's been absolutely nothing that shows us that we're actually headed in this direction. So it starts to feel more uncertain and irrational. Nothing stopping it right now, but I think it's a good thing that we're talking about it.

没错。过去几周我一直在问自己，我是不是疯了才会认为这些基础设施支出中的一部分根本没有遵循你在AI研究中看到的数据？我们在节目中与业内人士讨论这个问题已经不知道多少个月了，几个月，甚至一年，我们一直在讨论规模带来的收益正在趋于平缓，这似乎是某种共识，在AI领域能达到的接近共识的程度，因为总会有些人，比如Anthropic的Dario Amodei，会说规模是通往AGI的途径，通过扩大LLMs，但其他所有人都在说我们看到的是边际收益递减。所以一方面你有这种看似共识的观点，另一方面却有投资在不断增加，仿佛这种共识并不存在。投资的行为就好像你可以通过扩大规模来实现AGI。

Right. And over the past couple weeks, I've been asking myself, like, am I a lunatic for thinking that some of this infrastructure spending is just not following the data that you're seeing in AI research? And it's something we've talked about on the show with people in the industry for I don't know how how many months now, months, a year, where we've talked about how the gains that you're seeing from scale are leveling off and that's something that seems to be somewhat consensus, as close to consensus as you're going to get in AI because there'll be some folks, maybe like the Dario Amundes of Anthropic who are saying that scale, you know, is a way to get to AGI scaling up LLMs, but everybody else is saying we're seeing, diminishing marginal returns. So you have that seeming consensus, on the other side you have investment that's building as if that's not true. That's building as if you can just scale your way to AGI.

让我们回到Dave Kahn这里。他很好地阐述了这一点。让我惊讶的是，这种对资本支出的加倍投入正在发生，即使AGI的梦想似乎正在冷却。发生了两件事。首先，尽管训练集群规模大了很多，但新模型的进展已经放缓。

Let's go back to Dave Kahn here. He really makes this point well. What's surprising me is that this doubling down on CapEx is happening even as the dream of AGI seems to be cooling off. Two things have happened. First, new model progress has tapered off despite much larger training clusters.

其次，作为一个可能的结果，AI领域的杰出人物已经开始收回他们对AGI时间表的预测。十二月，Ilya Sutskever说预训练已经死了。六月，Sam Altman说AGI将更像是一个温和的奇点，同一个月Andrei Karpathy预测了十年的智能体发展，而不是2027年的AGI。领域内人士所说的与华尔街和投资者所购买的东西之间存在如此惊人的分歧。你怎么看

Second, as a likely consequence, AI luminaries have started to walk back their AGI timelines. In December, Ilya Sutskever said that pre training is dead. In June, Sam Altman said AGI will be more of a gentle singularity, and that same month Andrei Karpathy forecasted a decade of agents rather than AGI in 2027. It's such an amazing divergence between what the people in the field are saying and what Wall Street and the investors are buying. What do you think I about

我喜欢我的奇点是温和的，所以我很高兴Sam还在这么说，但是不。不，我同意，因为这有两个部分。一个是，这些资本支出投资？这些对数据中心的投资真的会是必需的吗？

like my singularities to be gentle, so I'm glad I'm still glad Sam's saying that, but no. No. I agree because there's kinda two parts of this. It's one, you know, like, is this CapEx investment? Are these investments in data centers actually gonna be required?

并且这种计算需求会是，我们是否会达到AGI，仅仅通过这些非常重的计算过程来解决我们所有的问题？但我认为Dave Conn没有真正深入探讨的另一部分是，它被呈现为一种二元选择，即重型计算导致AGI，但以更计算高效的方式做这些事情仍然是第三条路，我认为。我的意思是，你在台上听谷歌讲算法不仅仅是原始计算。我认为我们在这方面看到了更多。我看到一些论文，我想是叫微型递归模型实际上可以达到与深度求索非常相似的结果。

And is this need compute gonna be like, are we gonna get to AGI and just these very heavy compute processes that solve all of our problems? But then the other part that I actually think Dave Conn didn't really get into is it kind of present was presented as binary that, like, heavy compute leads to AGI, but also the idea that doing these things in a more compute efficient way is still another third path, I think. And we I mean, you on stage with Google hearing about how its algorithms not just raw compute. I think we've been seeing a lot more around that. I saw some paper around, I think it was called like a tiny recursive model can actually achieve very sim similar results as deep seek.

这个想法是，如果你真的以更计算高效的方式做事，那对公司和人们来说成本效益会高得多，人们会更喜欢这种方式，而不是重型GPT-5会对你给出的每一个问题，甚至是非常简单的查询，都进行漫长而深入的思考，只是为了驱动计算使用。所以我认为，整体上，这一切唯一合理的解释是，如果我们实现了这种重型计算AGI的愿景，但目前并没有真正的迹象指向这一点，至少我看不到。正如你所说，即使Sam、Karpathy和Ilya都在这么说，这与实际进行的投资之间存在如此大的脱节。

The idea that if you actually do things in a more compute efficient way, that makes things a lot more cost efficient for companies and people will much prefer that to the heavy GPT five is gonna think hard and long about every single problem even very simple queries that you give it just to kind of just drive compute usage. So I think that the overall, the only way any of this makes sense is if we realize this vision of just, like, heavy compute AGI, which there's no real signs pointing to, at least that I can see. And as you said, even when even Sam and Karpathy and all of the and Ilya are all saying this as well, I it's such a disconnect from the actual investments that are being made.

没错。我们会探讨一下即使我们不会仅仅通过增长这些数据中心和模型来获得AGI，其背后的逻辑可能是什么，但我想我们都同意，就目前看到的投资方向与研究指向完全脱节而言，这在某种程度上是令人疯狂的。

Right. And we'll get into sort of what the logic might be even if we're not gonna get AGI simply from growing these data centers and models, but I think we can both agree that it's crazy making in a way what we're seeing right now in terms of the the investment and where the research is pointing completely disconnected.

嗯，我有个问题。除了Waymo在纽约市到处行驶之外，你目前对AGI的定义是什么？但是在这个语境下，是的，

Well, I have a question. What's what's your current definition of AGI other than Waymo's driving around New York City? But in this Yeah,

语境下，显然那是第一个定义。

context and obviously that's the first definition.

我的意思是，那是官方的行业标准，但在所有这些语境中，我很好奇，就像你提到的AI治愈癌症算是其中一种高层次的解读，但在这个背景下，你怎么看待什么可能是AGI？

I mean that that's the the official industry standard but but in all the in all these contexts, I'm curious like you mentioned AI curing cancer is kind of one like high interpretation of this, but but how do you look at what what could be AGI in this context?

我们就采用我认为这些公司正在考虑的定义，即AGI是能够完成当今50%以上白领工作的技术。

Let's just go with the definition that I think these companies are thinking about and that is that AGI that can do more than 50% of white collar work today.

好吧。我的意思是，但我想这是我仍然难以理解的部分，因为我真的相信你可能不需要极其庞大的计算能力就能完成50%的白领工作。有点像微软和亚马逊阵营的观点，认为模型已经足够好。显然，长期听众知道我在产品与模型上的立场。但我认为，存在一个世界，你可以构建这些非常复杂的工作流程，完成这些工作，而这并不需要AGI。

Okay. I mean, which but I guess that's the part where I still have trouble kinda squaring this because I really believe you can probably do 50% of white collar work without incredibly heavy compute. Kind of in that Microsoft and Amazon camp that the models are good. And obviously, the long time listeners know where I stand on product versus model. But like, I think there's a world where you can build these very complex workflows and you can do this work and it doesn't require AGI.

它需要当前的技术状态和所有那些数据中心。我们会深入探讨这些数据中心从芯片投资角度的可行性是否合理？但我认为，你不需要那种AGI的解读来真正实现AI的潜力。

It requires the current state of technology and all those data center. And we'll get into the actual like, does the feasibility of these data centers from kind of an investment in chip standpoint even make sense? But but I think you don't need that interpretation of AGI to actually make this make AI realize its potential.

好的。所以你现在所做的就是给出了为什么这种建设有意义的完美理由。因为我认为实验室会向投资者论证的是，即使我们今天停止，我们拥有的技术通过正确的编排可以自动化50%的白领工作，因此这项投资将是值得的。你实际上会听到像Dario这样的人不经意间提到，50%的入门级白领工作可能在几年内实现自动化。如果你不需要巨大的技术进步就能达到这一点，那么这就是这种建设并使投资值得的逻辑。

Okay. So what you're doing right now is you're giving the perfect rationale for why this build out makes sense. Because I think what the what the labs would argue to their investors is even if we stop today, we have technology that can with the right orchestration automate 50% of white collar work and therefore this investment is going to be worthwhile. And you actually kind of hear it slip from people like Dario who says that 50% of entry level white collar jobs may be automated within a couple of years. If you don't need a massive technological advance to get there, then that would be the logic for this build out and make the investment worthwhile.

我想知道，你是否认为向投资者提出这种级别的案例是必要的？比如，你觉得那个推销相当不错。我有点被说服了，但你认为他们现在真的达到了那个水平吗？或者你觉得这些对话现在看起来是怎样的？我认为

Do you think I'm I'm curious. Do you think that level of a case even has to be made for to investors? Like, you think that was a pretty good pitch. I'm I'm halfway in there, but but do you think they're even getting to that level right now? Or what do you think these conversations look like between I think

其中一些，是的。你会谈到像OpenAI和AMD这样的公司，我们稍后会深入讨论。是的，马特·莱维的那篇文章

some of them some of them, yes. You're gonna talk about like OpenAI and AMD and we're gonna get into it in a moment. Yeah. The Matt Levy piece

非常出色。

was extraordinary.

是的。其中一部分是，没错，那需要被提出。我觉得那就像是向Lightspeed推销的一个版本，因为我和那位向Anthropic投资了十亿美元的Lightspeed投资者聊过，他基本上在为Dario的简介中为我做了计算，并以类似你刚才解释的方式进行了说明。所以我确实认为，是的，在某些领域，集体对话会达到那个层次，但我也认为还有其他人只是想着，让我们赶紧做笔交易吧，拜托。我们需要OpenAI的品牌光环，而这正是最终的结果。

Yeah. Some of it is like, yes, that needs to be made. I feel like that was a version of the pitch made to Lightspeed for instance because I spoke with the Lightspeed investor who wrote a billion dollar check into Anthropic and he basically, for the Dario profile, did the math for me and sort of explained it in a way similar to the way that you just explained it. So I I do think that yes, that's where the congregate conversation gets in some areas, but I also think there are others who are like, let's just, let's just do a deal, please. We need the OpenAI brand shine, and and that's where where it nets out.

是的，我能理解。我能看到事情可能有两种走向？关于AGI的问题，因为我觉得像戴夫·康奈特甚至指出了三个潜在的、让事情更不可能的因素。我觉得这很有趣，因为我们都在谈论AGI，把它当作实验室会带我们达到的一个模糊概念。

Yeah. I I can see that. I I can I can see that it can go both ways? And and on that question of AGI, because I think like Dave Connett even kind of pointed out three thing three kind of like underlying factors that make things even less likely. And and I thought this was interesting because we all talk about AGI as this kind of like vague concept that the labs will get us to.

但实际上，第一个重要的事情，我觉得很少被提及，是实验室正在剥夺博士项目的人才。所以，现在真正的基础研究实际上只流向实验室，而不是大学和更传统的研究机构。尽管这一切都起源于那里，我认为这是一个完全被忽视的动态，可能会在这方面产生更持久的后果。然后，另一个我喜欢的点是，公司政治往往倾向于共识性的想法，而不是更激进、不受欢迎的想法。我觉得可以公平地说，尽管萨姆和达里奥仍然表现得像叛逆者，你知道，就像我们对抗世界，迎接这些挑战，但这些公司正在变成企业。

But in reality, like the first big thing which I don't think I hear very much about at all, is that the labs are starving PhD programs of talent. So, like, now, actually, that really kind of foundational research moves only towards the labs away from universities and kind of more traditional research. Even though that's where all of this started, is actually I think a dynamic that is totally overlooked and could have kind of longer lasting consequences around this. But then and then one of the other ones I liked was corporate politics tends to favor invoke consensus ideas over more radical unpopular ones. I think it's fair to say that, like, even though the Sams and the Darios still present themselves as kind of renegade, you know, like us against the world taking on these kind of challenges, these companies are becoming corporations.

我的意思是，估值当然如此，但你得开始想象，OpenAI的内部政治已经是传奇了，但总的来说，他们是否还能真正以那种强烈的创新方式运作，而不是开始变得有点谷歌云化，我认为这是另一个风险。

I mean, valuations, certainly, but you you have to start to imagine I mean, OpenAI's internal politics politics are are the stuff of legend, but but overall, the idea that they're gonna still be able to operate truly in that kind of, like, intense innovation way versus they're starting to get a little Google Cloudified, I think, I think is another risk to this.

是的。我认为Khan提出的这一点非常到位，基本上是说，你们在语言模型开发上投入了巨额资金，就好像把扩展大语言模型视为通往通用人工智能的直线路径一样在投资，但如果事实并非如此，你们实际上反而会拖慢这一进程，因为你们过于专注这个方向。我觉得这是一个非常好的观点。确实会听到像Demis这样的人说，我们需要在大语言模型之外取得几项突破才能实现通用人工智能，当然Jan Lakoun也一直在大力宣扬这一点。但我认为这确实是一个实实在在的风险。

Yeah. I I think this is a great point that Khan, brings up, which is basically you're you're putting so much money towards LM development, and that you're investing as if scaling LLMs is a straight shot to AGI that when you get to the if it's not, you're actually gonna slow down that pursuit because you're so focused this way. I think it's a great it's a great point. I think that you do hear from folks like Demis saying that we need a couple breakthroughs beyond LLMs to get to AGI, and of course Jan Lakoun has been loudly talking about that. I think this is a real risk though.

我觉得他说得完全正确。如果你被投入数百万美元加入，比如说Meta，从事大语言模型工作，而你原本可能正在追求一种非大语言模型的解决方案，那种跳出框架、也许只有10%成功率但一旦成功就会带来重大突破的方案，那么从整体来看，现在发生的情况很可能正在拖慢人工智能领域的发展，这真的很有意思。更不用说如果这一切真的失败会发生什么。

I think he's totally right. If you if you're being thrown millions of dollars to join, let's say, Meta to work on LLMs, and you would have otherwise been pursuing a non LLM solution that, you know, sort of out of the box and maybe had a 10% chance of working, but if it worked would be a big breakthrough, then in aggregate what's happening now is probably slowing down the AI field, which is really interesting. And let alone what would happen if this actually goes bust.

没错。而且，我的意思是，在这方面，你必须想象背后实际的人际动态是怎样的。就像，这是一个小群体。很多人以前都一起工作过。所以你不得不想象，群体思维真的会渗透到他们处理或思考任何问题的方式中。

No. And then, I mean, on that, like, you have to imagine what are the actual human dynamics underlying this. Like, this is a small group of people. A lot of these people work together. So you have to imagine the group think, like, really has to pervade the way they're approaching or thinking about anything.

显然，我认为这就是为什么深度求索（DeepSeek）的出现如此重要，因为它就像是，好吧，完全独立的团队和人实际上可以在这个领域发挥作用并且拥有不同的思维方式。但这越来越清楚地表明，这一小群人拥有相同的心态和思维方式。而且这就是他们正在让我们所有人，甚至如你所说可能让整个全球经济都卷入的赌注。

And and obviously, I think that's why deep seek was such a moment because it was like, okay, completely separate teams and people actually are can play in this realm and have a different way of thinking. But it it it really just becomes more and more clear that this smaller group of people have the kind of same mindset and think the same way. And and this is the bet they're putting us all in and the entire global economy potentially as you said.

然后最后一点是，他谈到了这些组织内部的激励机制如何驱动了短期思维，时间范围在一到三年，我认为这一点也非常重要。就是说，你知道，即使你是——嗯，特别是如果你是OpenAI，你最初是作为一个专注于长期研究的实验室成立的，但现在你需要在未来几年内回报多少，比如一万亿美元的投资？你不会说‘让我们研究前沿和实验性的东西’。你基本上会找到一个你喜欢的方法然后坚持下去。所以Khan在他的文章最后问道，新的计算能力是为了什么。

And then one last part of this is that he talks about how, the incentives inside these organizations drive short term thinking on the order of one to three years, think that's really important as well. It's just that, you know, even if you're well, especially if you're OpenAI, you're this you were founded as a lab on long term research, now you have to return what, like a trillion dollars in investment over the next couple years? You're not gonna be like, let's work on the frontier and experimental stuff. You would like basically have found a method that you like and you're you're going for it. So so Khan ends his piece asking what the new compute is for.

我就读一下这部分。‘如果新的计算能力投资并不能让我们更接近通用人工智能，那么意义何在？一种论点认为，计算能力是未来的大宗商品，无论怎样，囤积这种资源很可能都是有价值的。’我觉得这某种程度上呼应了Ranjan的观点。撇开贬值问题不谈——这让这个论点充其量也只是站不住脚——更大的问题变成了金融市场愿意为这种囤积行为提供资金支持多久，以及投资者是否甚至理解他们正在做的事情就是这个。

I'll just read read this part. If new compute investments aren't getting us closer to AGI, then what's the point? One argument is that the compute is a commodity of the future and that stockpiling this resource is likely to be valuable regardless. I think that sort of goes to Ranjan's point. Setting aside the issue of depreciation, which makes this argument tenuous at best, the bigger question becomes how long financial markets will be willing to underwrite such stockpiling and whether investors even understand that this is what they're doing.

我的感觉是，虽然研究人员越来越不确定计算能力如何转化为能力提升，但华尔街还没有完全意识到这一点。所以我们就这么说吧，我们就谈谈这个，你知道，给它打个结。基本上，我认为我们都有点担心，即使你达到了能用现有系统创造真正有经济价值的工作的阶段，也许目前这种建设热潮也过于狂热，并且容易受到a) 技术没有进步，或者b) 效率提升的影响，以至于存在一种非零的可能性，他们实际上是在——我不想说把钱烧掉，但也许正在发生的就是这个。

My sense is that while researchers are increasingly uncertain about how compute translates into capability improvements, Wall Street hasn't fully woken up to this. So let's just say let's just talk about this, you know, wrap a bow on it. Basically, I think we're both a little bit concerned that even if you get to a place where, with the current systems, you can create real economic valuable work, maybe this current build out is so overenthusiastic and is vulnerable to, a, the technology not improving, or, b, efficiency improvements that there's a nonzero chance that they're light they're effective I don't wanna say lighting this money on fire, but maybe that's what's happening.

我认为，这几乎是一种慷慨的解读。是的，就这对经济的影响而言，我们确实需要深入探讨，因为有多少资金是真实的？有多少正在被花费？如何花费？实际上，我听了一个很好的播客，谈到数据中心，比如用无人机飞过看到有数百人在工作，规模相当于下曼哈顿，或者我认为是从中央公园到苏豪区那么大。所以确实有东西在建，这至少提醒了我，这一切并非完全是凭空捏造的。

I think, it's a generous interpretation almost. I think, yeah, it's in terms of where this takes the economy, I think we we definitely need to get into that because like how much of this money is real? How much of it is being spent? How and and I actually, there's like a I listen to this really good podcast talking about the data centers and like flying a drone over and seeing, you know, like actually hundreds of people working and it's the size of Lower Manhattan or I think it was like Central Park down to Soho. So there's stuff being built, which I think is at least a good reminder for me that this isn't all just kind of like completely made up.

但我认为，我们确实还没有对资金及其流向进行真正的讨论和调查。多年来我们一直在谈论这个问题，比如哪些投资只是计算资源，哪些是实际现金，哪些地方在建设，哪些还没有被真正深入挖掘并以传统的金融意义进行分析，也许现在正是开始的时候。

But I think, yeah, that that we haven't had a genuine discussion investigation around the dollars and how they flow. We've been talking about this for years around like what investments are just compute, what investments are actual cash, where are things being built, what's you know, like it hasn't really been dug into and kind of really analyzed in a traditional financial sense, and maybe this is the moment that that starts.

好的。那我们为什么不现在就开始做一些这样的分析呢？《金融时报》有一篇文章提到，OpenAI的计算协议总额超过1万亿美元，然后质疑这是否合理，因为最终必须有人来资助所有这些建设。故事是这样的：OpenAI今年签署了价值1万亿美元的计算能力协议，以运行其AI模型，这些承诺远超其收入，引发了如何资助它们的疑问。

Alright. And why don't we start doing some of that right now? So the Financial Times has a article about it, saying that OpenAI's computing deals top $1,000,000,000,000 and then sort of asking whether this makes sense because ultimately someone has to fund, right? All this all this build out. Now here's the story, OpenAI has signed $1,000,000,000,000 in deals this year for computing power to run its AI models, commitments that dwarf its revenue and raise questions about how it can fund them.

这里有一些协议细节：与NVIDIA和AMD的协议可能分别耗资5000亿美元和3000亿美元。OpenAI与Oracle的协议可能再耗资3000亿美元。CoreWeave披露了与OpenAI价值超过220亿美元的计算协议。OpenAI还与软银、Oracle等发起了名为“Stargate”的计划，承诺为OpenAI在美国的基础设施投入高达5000亿美元。目前尚不清楚NVIDIA和AMD的协议如何融入Stargate计划，尽管我们认为它们被包括在那5000亿美元之中。

Here's some of the deals, the deals with NVIDIA and AMD could cost up to 500,000,000,000 and $300,000,000,000 respectively. Oracle's deal with OpenAI could cost another 300,000,000,000. CoreWeave has disclosed computing deals with OpenAI worth more than 22,000,000,000. OpenAI has also launched an initiative with SoftBank, Oracle, and others known as Stargate and pledged up to 500,000,000,000 in, US infrastructure for OpenAI. It's not clear how the NVIDIA and AMD deals would fit into the Stargate plans, although I think we do we do believe that they're including that as part of the 500,000,000,000.

这些协议将使OpenAI获得超过20吉瓦的计算能力，大致相当于未来十年内20座核反应堆的发电量。每吉瓦AI计算能力按当前价格部署需500亿美元，总成本约为1万亿美元。然后他们引用了DA Davidson的分析师Gil Uriah的观点：OpenAI根本没有能力做出这些承诺。硅谷“假装直到成功” ethos的一部分是让他人参与其中承担风险。

The deals would give OpenAI access to more than 20 gigawatts of computing capacity, roughly equivalent to the power from 20 nuclear reactors over the next decade. Each gigawatt of AI compute capacity costs $50,000,000,000 to deploy in today's prices, making the cost total cost about $1,000,000,000,000. And then they go to this analyst Gil Uriah at DA Davidson. OpenAI is in no position to make any of these commitments. Part of Silicon Valley's fake it until you make it ethos is to get people to have skin in the game.

现在很多公司在OpenAI上押了重注，正如我们之前提到的，OpenAI预计从现在到2029年将亏损1200亿美元。那么，Ranjan，这笔账怎么算呢？

Now a lot of companies have a lot of skin in the game on OpenAI, and as we've mentioned, in the past OpenAI is expected to lose a 120,000,000,000 between now and 2029. So how does this math work, Ranjan?

我的意思是，算不通。根本算不通。我认为，从任何标准理性分析来看，都不成立。但我一直在想的是，这是否会迫使OpenAI和其他参与这场游戏的公司再次推动，回到更重计算解决方案的话题，比如几周前我们讨论过的Pulse。

I mean, it doesn't. It doesn't. I think, like, from again, any kinda, like, standard rational analysis, it doesn't. But I the thing I keep thinking about is, do does this kind of, like, does this force OpenAI and others who are playing this game to push again, going back to the topic of kind of like heavier compute solutions. Like Pulse, we talked about a couple of weeks ago this.

你用过了吗？或者你认识谁用过吗？

Have you used it yet or do you know anyone who has?

我还没用过。但我确实认识一个人，丹·希珀，他对其评价不错。你用过了吗？

I have not used it. I do know, someone that has Dan Shipper, from every he he has, some good things to say about it. Have you used it?

我我我还没有。但再说回来，这种想法就是它会整夜消耗计算资源，只为早上给你一些更新，然后可能还有广告，就像我们讨论过的，我越来越确信这就是它的发展方向。但Sora本身，甚至这是你在GPT五发布时提出的一个大问题，就是强行将模型和平台推向更重度的计算思考和推理，即使并不需要。感觉就像，他们构建公司的方方面面都在激励他们推行最低效的解决方案，以使自己的经济模式可行。所以这部分，我认为，他们必须朝那个方向走，而且他们确实在朝那个方向走。

I I I have not. But but again, this idea that it's just gonna be like sucking up compute all night long just to give you some some updates in the morning and then potentially ads as we talked about and I've kind of like come to definitely believe that's the direction it's going. But Sora itself or even that this was one of the big issues that you brought up around the GPT five launch, but like kind of pushing the model and the the the platform into much much heavier compute thinking and reasoning when it's not required. It feels like that is I mean, everything around how they're building this company is incentivized to push the absolute least efficient solutions possible to actually make their own economics work. So that that part, I think, like, they're they have to go in that direction, and it they're they're definitely going in that direction.

但除此之外，是的，这一切对我来说都毫无意义。

But otherwise, it just yeah. None of this makes any sense to me.

解释一下推行最低效的计算项目如何让OpenAI的经济模式可行。

Explain how pushing least efficient compute projects make the economics of OpenAI work.

嗯，最终总得有人买单。所以假设你会开始支付200美元而不是30美元来获得GPT Pro的脉搏更新。到了某个时候，他们将不得不向我们收费，以支付制作我们的Cameosura所需的计算资源。但基本上，如果我们被困在当前关于需要什么模型和什么计算资源的范式中，这些计算资源就不会被利用或使用。所以你必须，就像我们之前讨论的，能否基于现有技术创建这些工作流，并使其越来越高效？

Well, someone has to pay for it in the end. So assuming that you'll start paying your $200 instead of $30 for GPT Pro to get your pulse updates. And at a certain point, they're gonna have to charge us for making our cameosuras that to actually account for the amount of compute it's requiring. But but it's basically that people are not that compute's not going to be leveraged or used if we're just stuck in the current paradigm of what models are needed and what kind of compute's needed. So so you have to, again, like what we were talking about earlier, can you kinda create these workflows based on the technology that exists today and kind of make it more and more efficient?

实际上是为了展示我们正在利用这些计算资源，这项相当于20吉瓦和所有核反应堆的投资。很快就会非常清楚，除非他们能展示出计算利用率——目前这对他们来说和收入一样重要——否则这将是一个糟糕的投资和想法。我敢肯定内部他们必须进行这些对话，因为否则，投入了这么多钱，很快人们就会说，嗯，下一笔资金我们不需要真正发放，因为没人用这些东西。

It's to actually show that we are utilizing this compute, this investment in 20 gigawatts and, like all these nuclear reactor equivalents. It's it's gonna be very clear very quickly that it's a bad investment and idea unless they can actually show like just as important as revenue is actual like compute utilization for them right now. And I'm sure internally these are conversations they have to be having because otherwise, you're putting all this money in and very quickly people be like, well, that next tranche of money, we don't need to actually release because no one's using this stuff.

我明白了。对他们来说，他们希望激励大规模计算资源的使用，因为当他们向投资者汇报时，他们想用计算使用量作为这项技术价值的代理指标。是的，没错。他们是在说这就是我们技术价值所在。

I get it. So for them, they want to incentivize massive compute usage because as they go to their investors, they want they're using that compute usage as a proxy for value of this technology. Yeah. Exactly. They're saying this is why our technology is valued.

我们无法，我们没有足够的计算资源。如果他们能讲好这个故事，那么他们可能会获得更多资金。所以这激励他们为并不真正需要的事情使用大量计算资源。

We can't we don't have enough compute. And if they're able to tell that story, then, then they might get more money. So there that incentivizes them to use a lot of compute for stuff they don't really need.

正是如此。而我们作为消费者目前正在受益，因为实际上没有人需要为此付费。这让2010年代风投补贴的优步打车看起来像，你知道，一段美好的回忆，我们作为消费者能够获得所有这些好处，生成我们的Sora视频，但实际上，目前没有人为此买单。

Exactly. And and us as consumers are benefiting right now cause no one actually has to pay for it on the other end. And this kind of makes the like 20 tens VC subsidized Uber rides look like, you know, a quaint memory where we're just able to get all this benefit as as consumers, and generate our Sora videos, but in reality, like, no one's paying for that right now.

你担心即将出现的债务问题吗？当然，很多这些都是由风投资金资助的。现在开始转向债务融资。这是来自The Information的报道。本月估值5000亿美元的OpenAI正准备筹集数百亿美元债务来资助基础设施。

Do you worry about the debt that's coming in to the picture? So, of course, a lot of this has been funded by VC money. Now it's starting to move toward debt. This is from the Feet story. OpenAI valued at 500,000,000,000 this month is preparing to raise tens of billions of dollars of debt to fund infrastructure.

这也是来自华尔街日报的报道。债务正在推动AI热潮的下一波。这就像是Khan谈到的第二阶段。一些较小的公司，最突出的是CoreWeave，一段时间以来一直依靠创新融资方式将自己推向AI前沿。Oracle也参与其中。

This is also from the Wall Street Journal. Debt is fueling the next wave of the AI boom. And again, this is like the phase two that Khan talked about. A few smaller companies, most prominently CoreWeave, have been relying on creative financing to vault themselves to the AI forefront for a while. Oracle is also part of this.

为了履行与OpenAI的合同义务，Oracle必须在OpenAI全额付款之前先投资基础设施。KeyBank Capital Markets的分析师在最近的一份报告中估计，Oracle未来五年每年需要借款250亿美元来资助这些承诺。当然，私下说，很多这些都是基于对OpenAI收入呈指数级增长而非渐进式增长的预测。Oracle已经高度杠杆化。该公司在8月份的长期债务约为820亿美元，其债务权益比约为450%。

To make good on its end of the contract with OpenAI, Oracle has to spend on infrastructure before it gets fully paid by OpenAI. Analysts at KeyBank Capital Markets estimate in a recent note that Oracle would have to borrow $25,000,000,000 a year over the next five years to fund these commitments. And, of course, just, you know, talking between us, a lot of this is based off of revenue predictions that are exponential increases really for OpenAI, not just incremental increases. Oracle is already high highly leveraged. The company has a long term debt of about long term debt of about 82,000,000,000 at the August, and its debt to eight equity ratio was about 450%.

相比之下，谷歌母公司Alphabet的债务权益比为11.5%，微软约为33%。当这些泡沫最终承担债务却无法偿还时，我们不会陷入麻烦吗？

By contrast, Google parent Alphabet debt to equity ratio was 11.5% and Microsoft's was about 33%. Don't we get into trouble when these these bubbles end up taking on debt and they can't pay it back?

嗯，从更宏观的经济角度来看，我认为目前还不太明朗，因为这仍然非常集中。你看，这里只是一家公司，负债820亿美元，负债权益比高达450%。这不像全国各地的房主背负着不合理的家庭债务。所以这种溢出效应会是什么样子，我认为还相当不清楚。但最让我觉得有趣的是，就在这次录音前几个小时，我最喜欢的软银宣布，他们正在以其芯片部门ARM Holdings为抵押，获得一笔50亿美元的保证金贷款。

Well, I think from like a larger economy standpoint, it's still a bit unclear because this is still very concentrated. So, you know, it's one company here with $82,000,000,000 of debt and a a 450% debt to equity ratio. This isn't, you know, like homeowners across the country taking on unreasonable debt against their households. So so what that kind of spillover looks like, I think it's still pretty unclear. But but my favorite was actually just a couple of hours ahead of this recording, my favorite SoftBank, they announced they're taking a $5,000,000,000 margin loan that secured against its chip unit against ARM Holdings.

他们实际上已经以ARM股票为抵押获得了185亿美元的保证金贷款。我的意思是，这就是孙正义（Masa）的风格，愿上帝保佑他。但当你回顾这些事情时，你会觉得这感觉不对或不合逻辑。

They've actually taken out $18,500,000,000 of margin loans against ARM shares. Like, I mean, this is the stuff that is Masa's son, God bless him. But this is the kind of stuff that I feel you when you look back on and it just doesn't it doesn't feel right or make sense.

你是否担心会出现真正的股市回调？显然，这种可能性是存在的。但你是否担心像软银这样的公司会因为这样的事情而破产？

Are you worried that there would I mean, obviously, there's a chance of a real equity pullback. But are you worried about something like a SoftBank going insolvent from something like this?

我的意思是

I mean

那么这会带来什么样的连锁反应呢？

And what the ripple effect is there?

我是否担心软银破产？我觉得这是一个...我不知道。孙正义总能找到出路，但我不太担心直接的溢出效应。我真的认为这仍然是一个相对封闭的群体，他们从中赚了最多的钱。真正从中受益的公司相对较少。

Am I worried about a SoftBank going insolvent? I think it's a I don't know. Masa will always make his way back, but I I don't worry around the direct spillover effects. I really think this is still the it's been still like a relatively contained group of people that have made the most money off of this. It's been relatively few companies that have truly benefited from this.

我认为，是的，如果出现股市回调，那在地缘政治上意味着什么？世界显然并不稳定，任何额外的不确定性都不利于维持稳定。但总体而言，我不知道。我还没有看到一个令人信服的论点，说明这除了股市回调之外还会如何真正溢出。我的意思是，如果这种情况消失，全球经济就没有增长故事了。

I think, yes, if there's an equity pullback, what does that mean, like, geopolitically? And I think there's the world is certainly not in a stable place, and any kind of additional uncertainty does not help, you know, kind of maintain any kind of stability. But but overall, I don't know. I I still haven't I I've been I haven't seen a compelling argument about how this really spills over other than just an equity pullback. I mean, there's no growth story for the global economy if this goes away.

它驱动了我们过去几年看到的全部增长。但这只是让一切回归理性吗？还是说实际上，我不知道。你提到，全球经济下周还会存在吗？所以，让我们听听你的看法。

It's driving the entire growth that we've seen over the last couple years. But is is that just kind of bringing back to rationality? And is that or is it actually I don't know. You you mentioned is will the global economy still be here next week? So let let let's hear your take on this.

我们那个。所以可以开始，让我们这么做，然后我们会再谈谈AMD的交易。但是，这是来自《金融时报》的消息。美国现在就是一个对AI的大赌注。公司们现在投资AI的数千亿美元，今年占了美国GDP增长的惊人40%份额。

We that. So can get let's do this, then we're gonna touch on, maybe the AMD deal again. But, this is from the Financial Times. America is now one big bet on AI. The hundreds of billions of dollars that companies are investing in AI now account for an astonishing 40% share of The US GDP growth this year.

40%。AI公司占了2025年迄今为止美国股市涨幅的80%。这正在资助美国的增长，因为AI驱动的股市从全球吸引资金，并推动了富人的消费繁荣。在某种程度上，除了AI领域之外，美国已经变成了一个对AI的大赌注。甚至欧洲股市在这个十年里表现一直优于美国，而现在差距开始扩大。

40%. AI companies have accounted for 80% of gains in The US stocks so far in 2025. This is helping fund US growth as the AI driven stock market draws its money in from all over the world and feeds a boom, in consumer spending by the rich. In a way, America has become one big bet on AI outside of the AI plays. Even the European stock markets have been outperforming The US this decade, and now the gap is starting to spread.

2025年迄今为止，从公用事业和工业到医疗保健和银行的每一个主要行业，在世界其他地区的表现都比美国好。你知道，如果AI不能为美国带来回报，美国及其经济将失去它们现在赖以支撑的唯一支柱。你的想法。

So far in 2025, every major sector from utilities and industrials to health care and banks have fared better in the rest of the world than The US. You know, if AI doesn't deliver for The US, The US and its economy will lose the one leg they are now standing on. Your thoughts.

还是回到你这边。你认为那实际上意味着什么？如果，比如说突然有预测说算力不会被有效利用，我们陷入GPU贬值，有人做了计算，你认为最坏的情况是什么？

Still back to you on this one. Do you think what does that actually mean? If that let's say suddenly there's like predictions that the compute is not gonna be effectively utilized, we get into GPU depreciation and people someone does the math, what do you think is the worst case scenario?

可能，我认为也许全球经济爆炸是一个夸张的说法。好吧。既然我们在固执地谈论它，并且

Probably was probably, I think maybe global economy blowing up is a was an overstatement. Okay. Now that we're talking about it in our bullheaded and

不是爆炸，是消失。我想你是这么说的。

Not not blowing up, disappearing. I think you said.

是的。嗯，你知道，那可能是我内心的戏剧播客人格在作祟。但不过不，你看，我认为情况会很糟糕。我我我觉得，你知道，显然全球经济会保持完整，你会这么想。我认为你指出债务和投资确实是被控制住的这一点非常正确。

Yeah. Well, you know, that might have been the, you know, drama podcaster in me. But but no, it it look, I think it would be bad. I I I think, you know, we're obviously, the global economy would be intact, you would think. I think you're really right in pointing out that the debt and the investment is really contained.

但如果它消失，如果这个AI时刻消失，你会看到你会看到美国出现一些非常负面的经济后果，尤其是在国外可能更严重。这里，这是德意志银行的说法。他们说他们说除非科技支出呈抛物线式增长，否则AI繁荣是不可持续的，而这极不可能。一位德意志银行分析师表示，AI正在拯救美国经济。如果没有科技相关支出，美国今年将接近或陷入衰退，这是贝恩公司的观点。

But when it if it were to go away, if this AI moment were to go away, you would see you would see some really negative economic consequences in The US, especially it may be outside. Here, this is from Deutsche Bank. They say they say the AI boom is unsustainable unless tech spending goes parabolic, and it's highly unlikely. AI is saving The US economy right now, says a Deutsche Bank analyst. In the absence of tech related spending, The US would be close to or in a recession this year, and this is from Bain.

到2030年，需要每年2万亿美元的收入来资助满足预期AI需求所需的计算能力。然而，即使有AI相关的节省，全球仍短缺8000亿美元才能跟上需求，而且这种繁荣是不可持续的。是的。我不知道。在我看来，总会有一个关键时刻到来，事情会变得糟糕。

$2,000,000,000,000 in annual revenue is what's needed to fund computing power, the computing power needed to meet anticipated AI demand by 2030. However, even with AI related savings, the world is still, 800,000,000,000 short to keep pace with demand, and the boom is not sustainable. Yeah. I don't know. It seems to me that there there's gonna have to come a point where the rubber is gonna meet the road here, and something's gonna go bad.

我不知道具体会在哪里发生或者影响范围有多广，但即使有AI的所有承诺，这也是我们开头读到的Khan文章中所指出的，这里将会出现一次调整。

And I don't know exactly where that's gonna be or how widespread it will be, but even with all of AI's promise, and this is what Khan was getting at in the piece we read at the beginning, a correction is there there will be a correction here.

我我觉得有一点仍然让我觉得有趣的是贝恩公司报告谈到的方式，是的，到2030年需要2.2万亿美元的年收入来资助满足预期AI需求所需的计算能力。然而，即使有AI相关的节省，全球仍短缺8000亿美元。这再次将其置于那种范式，就像你需要那笔收入来覆盖预期AI需求所需的计算能力。但但我认为这些数字是如何计算和 extrapolated 的，让我震惊。让我困惑不已。

I I think one thing that's still kind of like it was interesting to me the way the Bain and Company report had talked about, yeah, the $22,000,000,000,000 in annual revenues needed to it's what's needed to fund computing power to meet anticipated AI demand by 2030. However, even with AI related savings, the world's $800,000,000,000 short. It's still, again, putting it in that paradigm of like, you need that revenue to to cover the computing power that's gonna be needed for anticipated AI demand. But but I think it how these numbers get calculated and extrapolated, I it's it blows my mind. It boggles my mind.

这就像是在尝试做直线插值吗？是是指数级的吗？但就像，鉴于事情变化如此之大，试图预测2030年的AI需求。再次，就像我在谷歌看到的，Gemini现在利用了1.8个tokens。数字相当惊人，但还要将其 extrapolated 到未来五年，并试图理解它，真正尝试用数字来支撑。

It's like like trying to is it just a straight line interpolation? Is it a is it I mean, it's it's exponential, but, like, trying to forecast AI demand in 2030 given how much things have changed. And again, like, I think what I saw was at Google. It was like 1.8 tokens now being leveraged by Gemini. Like, the numbers are pretty spectacular, but still extrapolating that out for the next, you know, five years and trying to make any sense of it and really try to put numbers behind it.

我不知道。尤其是当机器之神降临，AGI和超级智能出现时。我不我不明白你怎么能一本正经地做这件事。

I don't know. Especially when the machine god's gonna come and AGI's and super intelligence are gonna come. I don't I don't see how you do that with a straight face.

好吧，Ranjan，说实话我刚看到这些数字时心情真的很低落，但你恰到好处地把我从悬崖边拉了回来，我很感激。好了，我们先简单探讨两家公司——甲骨文和AMD，然后进入休息时间。首先，你对甲骨文这件事怎么看？他们显然在大规模采购计算资源。

Well, Ranjan, I think I came in here really down in the dumps looking at these numbers, but you've talked me off the ledge appropriately, so I appreciate that. Alright. Let's do two two small dives into a couple companies, Oracle and AMD, and then hit the break. First of all, what do you think about this Oracle story? So they're buying obviously a ton of, compute.

根据信息显示：甲骨文成为2025年表现最佳的超大盘股，因其高管上月表示，这家曾经沉寂的数据库公司将在未来五个财年通过向OpenAI等AI开发者出租专用云服务器获得惊人的3810亿美元收入。但他们的利润率平均只有16%，在某些情况下，甲骨文在出租少量新旧版本英伟达芯片时甚至出现巨额亏损——在截至8月的三个月里，仅Blackwell芯片出租就亏损近1亿美元。一家习惯70-80%利润率的软件公司，现在利润率堪比零售业，虽然销售额上升了。但他们的股价今年上涨约80%，市值达8540亿美元，市盈率69倍，这合理吗？

This is, from the information. Oracle became the best performing mega stock of 2025 after its executives said last month that the once sleepy database firm will generate an astounding $381,000,000,000 in revenue from renting out specialized cloud servers to OpenAI and other AI developers over the next five fiscal years, but the margins they're getting on those, are averaging 16%, and in some cases Oracle is losing considerable sums on rentals of small quantities of both newer and older versions of NVIDIA chips. In the three months that ended in August, Oracle lost nearly 100,000,000 from the rentals of NVIDIA's Blackwell chips. So you have a software company that's used to 70 or 80% margins, now they're they have margins of a retail business, obviously their sales are up. Does it make sense that their stock is up about 80% this year and their market cap is at 854,000,000,000 with a 69 PE ratio?

不，不，我觉得这个问题提得非常好。大家真的应该开始关注这类数据了。因为我们之前从未讨论过这类业务的毛利率问题。长期以来我们理论上都说生成式AI的财务特征与传统软件不同——它不具备无限或近乎无限规模经济效应，而是存在实实在在的底层成本。

No, no, I think this is a really good point here. Like and and I think like everyone should start really getting into these kind of numbers. Because again, we have not talked about like, you know, gross margin on these businesses. We've all said theoretically for a long time that, you know, like generative AI has a different financial profile than traditional software. And and again, it's not something that has like infinite economies of scale or near infinite, but instead has a real cost underlying it.

看到这种变化确实令人震惊，就像你说的，利润率从70%暴跌至平均16%。或许可以辩解说这只是业务起步阶段的暂时现象，但事实上利润率结构不会随时间根本改变。可能会逐步改善，但只能是渐进式的——或许能到25%。

And to see that is actually kind of mind blowing like, again, as you said, 70% margins down to averaging 16%. And maybe you can argue that this is just at this point as they're kind of kind of like getting this business up and running and scaled. But in reality, like, there's no reason that the margin profile should change over time. Like maybe it starts to improve but it's incremental. Maybe you get to 25%.

或许能勉强达到30%。但这本质上已不同于那个长期稳定产出现金流的沉寂数据库公司。我认为这确实应该引发思考：所有这些公司的经济模式将会如何演变？它会不同，或许也能存活——可能变成巨型运营型零售式企业，但绝不再是70%利润率的软件业务。

Maybe you you squeeze out 30%. But in reality, like this is a different business than the sleepy database company that was just churning out cash for so long that, it it really should call into question, I think, like, how what the economics of all these companies are going to look like. And it's gonna be different and maybe it'll be fine. Like, maybe these will be retailer style businesses that are gigantic and operational, but it's not gonna be 70% margin software businesses.

没错。我记得分析师Il Luria（我们之前引用过他的观点）在CNBC也提出：甲骨文估值高于微软是荒谬的——前者在搭建数据中心却利润微薄，后者既建设基础设施又从中盈利。我认为AI浪潮从长期看确有价值，但当前利润率方面存在泡沫，终将消退。说到泡沫，OpenAI与AMD的交易也很蹊跷——周一MJ Siegler来做客时我们就讨论过这可能不合理，现在Matt Levine还发布了一段疑似AMD同意出让10%股份前与OpenAI的虚构对话，非常有趣。

Right. I think Il Luria was and the analyst that we quoted before was on CNBC making this point, basically saying like, it doesn't make any sense for Oracle to be valued, as at a more to be for Oracle to be more expensive, than a Microsoft, where Oracle is basically, like, setting up these data centers and not pulling in margin, where you have a Microsoft that's actually setting up the infrastructure and making money off of it, and making profit. So, I don't know. What I think sort of what we're seeing is, this push towards AI might make sense overall over time, but there is some silliness around the margins and that will shake out. Of course, there's been some silliness with the OpenAI AMD deal, MJ Siegler was here on Monday, we were talking about how, you know, maybe it didn't make sense, and now there's a really interesting fake conversation that Matt Levine published that might be what OpenAI and AMD discussed before AMD agreed to give OpenAI 10% of its company potentially.

广告回来后我们马上讨论这个。节日总是悄然而至，但现在开始购物为时不早反而能享受乐趣。Uncommon Goods让节日购物变得轻松愉快，提供数千款独一无二的别处难寻的礼物。我已经入手了——买了酷炫的护林熊卫衣和优胜美地滑雪帽，为漫长舒适的冬季做足了准备。

Let's talk about that right after this. The holidays sneak up fast, but it's not too early to get your shopping done and actually have fun with it. Uncommon goods makes holiday shopping stress free and joyful with thousands of one of a kind gifts you can't find anywhere else. I'm already in. I grabbed a cool Smokey the Bear sweatshirt and a Yosemite ski hat, so I'm fully prepared for a long cozy winter season.

这两件商品看起来都很棒，完全没有那种随处可见的量产感。网站上还有很多其他好东西。从父母到孩子和青少年，从书迷、历史爱好者、铁杆足球迷到美食家、调酒师和园艺爱好者，你会发现成千上万别处找不到的新颖礼物点子。所以早点购物，享受乐趣，今天就划掉清单上的一些名字吧。要获得下次购物15%的折扣，请访问 uncommongoods.com/bigtech。

Both items look great and definitely don't have the mass produced feel you see everywhere else. And there's plenty of other good stuff on the site. From moms and dads to kids and teens, from book lovers history buffs, and diehard football fans to foodies, mixologists, and avid gardeners, you'll find thousands of new gift ideas that you won't find elsewhere. So shop early, have fun, and cross some names off your list today. To get 15% off your next gift, go to uncommongoods.com/bigtech.

那就是 uncommongoods.com/bigtech，享受15%折扣。不要错过这个限时优惠。Uncommon Goods，我们都与众不同。用 Agency（a g n t CY）塑造企业AI的未来。现在作为开源Linux基金会项目，Agency正在引领建立物联网代理的可信身份和访问管理，这是一个协作层，确保AI代理能够安全地发现、连接并在任何框架中工作。

That's uncommongoods.com/bigtech for 15% off. Don't miss out on this limited time offer. Uncommon goods, we're all out of the ordinary. Shape the future of enterprise AI with agency, a g n t CY. Now an open source Linux foundation project, Agency 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，您的组织可以获得开放、标准化的工具和无缝集成，包括强大的身份管理，以便能够在任何平台上识别、验证和交互。让您能够自信地部署多代理系统，加入像思科、戴尔科技、谷歌云、甲骨文、红帽以及75家以上支持公司的行业领导者行列，共同为安全、可扩展的AI基础设施设定标准。您的企业准备好迎接Vagintic AI的未来了吗？立即访问 agency.org 探索用例。那就是 a g n t c y 点 0 r g。

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 seventy five plus supporting companies to set the standard for secure, scalable AI infrastructure. Is your enterprise ready for the future of Vagintic AI? Visit agency.org to explore use cases now. That's a g n t c y dot 0 r g.

我们回到Big Technology Podcast，我想有人说我们正在——他们喜欢这个节目变成了AI泡沫的侦探秀。我喜欢那个评论。让我们继续吧。所以，当然，OpenAI和AMD本周达成了这笔大交易，OpenAI和AMD基本上花费了数百亿美元进行数据中心开发，这样OpenAI就可以使用AMD芯片进行推理。这笔交易有个奇怪的部分，AMD说如果达到某些里程碑，OpenAI将有机会以每股一美分的价格获得AMD公司10%的股份。

And we're back here on Big Technology Podcast, examining I think someone said we were, they love how this show has become a detective show for the AI bubble. I love that comment. Let's let's keep at it. So, of course, OpenAI and AMD had this big deal, this week where, OpenAI and AMD basically spent tens of billions, working on, data center development, so OpenAI can use AMD chips for inference. And there was this weird element of the deal where AMD said if certain milestones are hit, OpenAI will have the opportunity to get 10% of AMD's, company's stock, basically for a penny a share.

所以彭博社的Matt Levine推测了这可能发生的原因以及它是如何达成的。以下是模拟对话。OpenAI：嗯，我们在想，我们宣布这笔交易会给公司增加780亿美元的市值，这基本上就够付芯片钱了。AMD……

So Matt Levine at Bloomberg surmised why this might have happened and how this might have come together. Here's the fake conversation. OpenAI. Well, we were thinking that we would announce the deal and that would add 78,000,000,000 to the value of the company which should cover it, basically paying for the chips. Amd...

OpenAI…… AMD：不。我很确定你得付钱买芯片。OpenAI。

Openai..dot. AMD. No. I'm pretty sure you have to pay for the chips. OpenAI.

为什么？ AMD：我不知道。我觉得不付钱好像不对。OpenAI。

Why? AMD. I don't know. I guess it seems wrong not to. OpenAI.

好吧。那么，我们为什么不按芯片价值付你现金，然后你把股票还给我们？当我们宣布这笔交易时，你的股票会上涨，我们就能收回780亿美元。是的，我觉得这办法可行。

Okay. Well, why don't we pay you cash for the value of chips and you give us back stock? When we announce the deal, your stock will go up and we'll get our $78,000,000,000 back. Yeah. I guess that works, though.

我觉得我们应该获得一些价值。好吧，你可以拿一半。你给出价值约350亿美元的股票，剩下的你留着。莱文说OpenAI和AMD的交易显然会创造大量股市价值。交易宣布后预计会提升AMD的市场估值，而且这又不会降低OpenAI的市场价值。

I feel that we should get some value. Okay, you can have half. You give a stock worth like 35,000,000,000 and then you keep the rest. Levine says the deal between OpenAI and AMD was obviously going to create a lot of stock market value. The announcement of the deal would predictably increase the value the market value of AMD, and it's not like it decreases the market value of OpenAI.

为什么不利用这种股市增值来补贴交易呢？拉杰，你怎么看？

Why not use that stock market increase to subsidize the deal? What do you think about this, Raj?

我的意思是，这其实就是我之前问你的，你觉得这些幕后对话实际上是什么样的？我很喜欢这个例子，因为我觉得这种对话现在确实可能正在发生，并非完全不可能。就像——显然，当我们宣布这个消息时，显然会提振你的股价。这绝对应该覆盖交易总成本的某些部分。这种情况我们已经见怪不怪了。

I mean, this is actually this what I asked you earlier about, like, what do you think these conversations really look like behind the scenes? I mean, I loved this because I I don't think it's completely out of bounds that this is actually some of the conversation that's happening right now. Like that, well, obviously, when we announce this obviously, when we announce this, it's gonna boost your stock. That should definitely cover some element of the overall cost of the deal. Like, it's and and we've been seeing this forever.

我是说，不是一直如此，但近几年都有。这和想象中谷歌或亚马逊与Anthropic的交易没太大区别——比如我们会给你50亿，但其中30亿是算力资源。不过这会让你的估值提升这么多。所以，这种类似虚拟货币的元素已经存在一段时间了，只是规模更加庞大。

Like, I mean, not forever, for a few years. This is not that different than imagine a Google or like Amazon and Anthropic and it's like, well, we'll give you 5,000,000,000, but it's gonna be like 3,000,000,000 in compute. But it's gonna raise your valuation by this much. So, you know, like, overall, this kind of funny money esque element of it has has been there for a while. It's just at a much grander scale.

还有一点我想提的是，就像我们刚才讨论的，过去十年我的工作一直处于零售、软件、媒体的交叉领域——零售商的平均市盈率在行情好时也就20倍左右，软件公司平均甚至像甲骨文现在都有60倍。当你开始尝试用理性、纪律和严谨性来分析时，这些公司会更像零售商吗？还是……我不知道，也许会更像工业设备商。

And one thing I I kinda wanna bring up like, I think what we were talking about a second ago, the average like my life for the last decade has kinda been at the intersection of retail, software, media, and like average retailer PE ratio is around 20 on a good day. Average software company, even Oracle right now is sitting around 60. Do you, when you start to actually try to bring some rationality to this and some discipline and rigor, are these companies gonna be more like retailers? Or what I don't know. Like, maybe it's gonna be industrial equipment.

我注意到很多说法认为星际之门（Stargate）不是AI项目而是能源项目，这门生意不错也很有意思。但它们不应该更像传统能源和能源基础设施公司那样估值，而非按照AI和软件公司来估值吗？我不知道。我觉得我们会越来越多看到这种现象，人们试图为AI公司创造不同于软件公司的新估值指标。就连AMD和OpenAI这种循环论证的财务分析方式，也开始显得荒谬了。

I've seen a lot around how Stargate is not an AI play, it's an energy play, which is a good business and can be and is interesting. But shouldn't they be valued more like a traditional energy and energy infrastructure company rather than AI and software? I don't know. I think that is something that we're we're gonna be seeing a lot more of and people trying to come up with some new metric for an AI company that's very different than software. And and even all these AMD OpenAI kind of circular ways of approaching financial analysis, I think start to look ridiculous.

没错。我认为只要市场行情好，AI公司就会获得极高的估值倍数。但一旦出现增长放缓的迹象，这些估值就会迅速缩水。

That's right. I think as long as the you have good times, then there'll be massive multiples that will be attached to AI companies. But second we see a sign of a slowdown, think that will deflate exceptionally fast.

是的。各位观众都听到了吧。去创造适用于这类新型公司的财务指标吧。它们不是软件公司，也不完全是零售业，介于两者之间。

Yeah. I think you've heard it here, folks. Go out, come up with your new financial metrics for this new breed of company. It's not software. It's not quite retail, somewhere in between.

我们需要新的衡量标准来评估这些企业。

We need we need new ways to measure this stuff.

当然这不是投资建议。所以请自行判断。

And this of course is not investment advice. So take that.

不。我们是建议大家去做调研。是的。这是在告诉我们的观众。没错。

No. We're we're telling you to go out and do the research. Yeah. Telling our audience. Yeah.

对。

Right.

好了。让我们结束这个已经占据几乎整个节目时长的话题，最后看看我们俩的感受。你的担忧程度从1到10分是多少？1表示完全不担心，10表示非常担心。

Alright. Let's let's finish this very long, already spanning almost our entire show with but just to look at at how we we both feel about this. I mean, what's your scale of concern here from, like, one being not concerned to 10 being concerned?

我还是会把它设定在335，确切地说，是为了准确表达那种担忧的感觉。常听我们节目的听众都知道，我常常对很多事情感到忧虑。我真的认为这是一次市场洗牌，有点像回归现实、脚踏实地的时刻，但我不认为基于当前情况的任何理性分析会产生大规模的溢出效应，投资者只是开始稍微冷静下来。我觉得一切都会好起来的。

I'm still gonna put it, and and regular listeners know that I am often concerned about many things. I'm still putting it at 335 to be exact, to get to the exact, feeling of concern. I really think this is a shakeout. It's kind of like a back to reality come down to earth moment, but I don't think there's gonna be massive spillover effects from any kind of any kind of rational analysis on what's really happening right now, and investors just start to start to come back down a little bit. I think it's gonna be okay.

你呢？

What about you?

我给五分。我确实认为存在一种非零的可能性，即进展停滞，围绕人工智能的许多炒作只是逐渐消退。人们意识到实际应用比炒作中描述的更难，耗时也更长。时间线比预期的要长，也许这会导致一段时间的停滞。我不确定。

I'm at a five. I do think that there's a non zero chance that, progress stalls, and a lot of this hype around AI just translate, it just fades. People realize that it's harder to implement than a lot of the hype was making it out to be, and it takes longer. The the timeline is longer than anticipated, and, and maybe that just leads to stagnation for a while. I don't know.

我并不确信这会发生，但我确实承认这种可能性。好吧，我们来谈谈OpenAI本周的公告。他们举办了这次开发者日活动。似乎每个人都只想构建东西。

I'm I'm not convinced that's gonna happen, but I I definitely appreciate the possibility. Alright. Well, let's talk a little bit about OpenAI's announcement this week. They had this Dev Day, Developer Day. It seems like everybody just wants to build.

你以前听过这个吗？OpenAI周一举行了年度开发者日，公司公布了将应用程序集成到ChatGPT的计划。演示展示了如何在不离开ChatGPT窗口的情况下调用或发现像Spotify和Figma这样的程序。随着科技界大量涌向人工智能集成，OpenAI的演示迄今为止最好地描绘了以人工智能为先的互联网可能的样子，像ChatGPT这样的界面直接清理信息并执行命令。我不知道为什么我对这个兴奋不起来，或者也许我知道——如果你觉得我从谷歌、亚马逊、苹果，现在又是OpenAI那里听过这个，我不会对ChatGPT的平台化过于狂热。

Have you heard this before? OpenAI held its annual dev day on Monday where the company rolled out its plan to build apps into ChatGPT. The demo showed how programs like Spotify and Figma can be called or discovered without leaving the ChatGPT window. With so much of the tech world barreling towards AI integration, OpenAI's demo was the best picture yet of what an AI first Internet might look like with interfaces like ChatGPT clearing information and executing commands directly. I'm I I I don't know why I can't get excited about this, or maybe I do if you feel like I've heard this from Google, from Amazon, from Apple, and OpenAI again, and, and I'm not, like, gonna lose my mind over the platformization over of ChatGPT.

我是不是低估了这个？

Am I underplaying this?

不，不。直到我实际尝试使用了ChatGPT的Figma应用，作为一个非UI设计师但充满好奇心的人，我想，好吧，我真的能开始制作模型并构建自己的应用界面吗？实际上，你可以在FigJam中制作流程图，效果还行。好吧。

No. No. I until I so I actually tried I used the Figma app for ChatGPT a bit, And I and as as a non UI designer, but someone who's curious, was like, okay, can I actually start to make mock ups and start to, like, build out my own app interface? And in reality, you can make like a flowchart in FigJam and it was okay. Okay.

这并没有什么突破性的启示。但说实话，对我来说，整个应用集成方面，我认为它最终应该会实现。但让我困惑的是，我现在仍然会直接使用Google Flights，而不是通过Gemini集成到Google Flights——毕竟它们同属一家公司。实际上，至今还没有人成功展示过这会是什么样子。但我确实认为，这给实际的应用公司提出了一个有趣的问题：到底是你数据库的价值更重要，还是你的实际用户界面和交互体验更有价值？

It wasn't anything revelatory. But honestly, to me, like, that whole app integration side, I think it should work at some point. But to me, like, the fact that Google Flights, I still go to as opposed to Gemini integrated directly into Google Flights, which it is it owns and is the same company. Like, I no one has actually shown what this can look like successfully yet. But I but I do think, like, using it it it poses an interesting question for the actual app companies themselves because at what point is the value your database basically or is the value your actual UI and interface?

所以，如果我能用ChatGPT来为我创建Spotify播放列表，那我在Spotify上的所有DJ功能是不是就都没用了？这些公司会允许ChatGPT实际上夺走它们的用户界面层吗？它们肯定会抵制，所以我不认为这会进展得很快。你试过这个吗？

And so like if I can go use ChatGPT to create my Spotify playlist for me, then all my my DJ on Spotify, does that all become useless? And do these companies let ChatGPT actually kind of take the UI layer away from them? And they're gonna push back, so I I don't see this moving that quickly. Did you try on that?

我还没试过。我就是……我实在兴奋不起来。我真的听腻了这个故事，也许我有点封闭，这样不好，因为作为报道这些的记者，我应该保持开放心态。但最近我设置了Alexa Plus，我当时觉得，哦，挺酷的。

I haven't tried it yet. I just I can't. I really can't get excited about this. I I'm I'm sick of hearing this story, and maybe I'm shut off in a way that's, like, bad because I should be open as a reporter covering this stuff. But, like, I set up Alexa Plus recently and I was like, oh, it's cool.

你可以通过Echo设备叫Uber。但我当时就想，我还是用手机叫吧。我不知道。我还没被说服这会成为未来的平台。

You can call an Uber from, you know, your Echo device. And I was just like, I'll just do it on my phone. I don't know. I'm not sold yet that this is gonna be the platform of the

哦，等等。Alexa Plus。你的看法是什么？我也设置好了。你觉得怎么样？

Oh, wait. Alexa Plus. What are your eyes? I did set it up as well. What are your thoughts?

还没用够，没法真正评价。但我跟亚马逊设备和服务的负责人Panos Penne聊过，他说到十月份，所有人都应该能用上。所以不再需要早期访问了。我觉得，如果这没实现，我们就看看会发生什么吧。但这是最新的承诺，我很期待播出那期节目，几周后就会上线。

Not enough use to really, review it yet. But I did speak with Panos Penne, the head of devices and services of at Amazon, and he said by the October, everybody should have it. So, no more early access. And I think, you know, if that doesn't pan out, we'll we'll see what happens. But, there's the latest promise, and I'm very excited to air that episode, which is coming up in a couple weeks.

我得说我还挺喜欢的。它有点接近ChatGPT语音的水平了。比如我在做饭的时候能问问题。我厨房里有个Alexa Echo Show，问更详细的问题时它表现不错。但最喜欢的是上周，好像是周一，我在查NFL比分，看着它们。

I'll say I'm liking it. It's getting kinda like somewhat on par with ChatGPT voice. Justin being able to ask questions while I'm cooking. I have a Alexa Echo Show in my kitchen and like asking more detailed questions and it works pretty well. But my favorite was last week or I think on Monday I was asking for like NFL scores and I was looking at them.

完全编造喷气机队赢了比赛，我觉得在我们的语境下这既惊人又滑稽。亚历克斯是喷气机队的粉丝。

It completely made up that the Jets won, which I thought was kind of amazing and hilarious in our context. Alex is a Jets fan.

喷气机队能赢的唯一宇宙就是AI在幻觉中编织他们的胜利。

That's the only universe where the Jets can win is AI hallucinating their victories.

而且你应该用Sora创造整个平行宇宙，让喷气机队在那里赢得超级碗等等一切。

And and you should just create your own entire Sora parallel universe where the Jets are winning, Super Bowls, all of the above.

我这周确实在喷气机队Discord里看到一个Sora视频，教练在输球后还在为球员加油，因为他们能获得更好的选秀顺位，这其实就是组织唯一关心的事。太贴近现实了。就像我们喜欢的那样——

I did see a Sora video, think probably in my Jets Discord this week where the coach was cheering on the players after a loss because they were in line for a better draft pick and that's really all the organization cares about it. It's really like too close to home. It's like we love That's to

这就是AI能做到的极限了。

as far as the AI can go.

话说回来，虽然有点跑题，但比尔·贝利奇克和北卡罗来纳大学是怎么回事，爱国者队粉丝先生？

Know, this is a bit of a diversion, but what's the deal with Bill Belichick and UNC, Mr. Patriots fan?

我我在想上周日夜赛的德雷克·梅和比尔斯的比赛。贝利奇克，我唉。这真是太难评了。我的意思是，年轻女友，糟糕的开局。我能理解如果你成为了最伟大的——你知道的，作为最顶尖的独立记者和媒体人之一，用几十年建立起的声誉就这么抛弃了，到底是什么驱使人这样做？

I I'm just thinking about Drake May and the Bills game on last Sunday night. Belichick, I yeah. It's such a tough one. I mean, the younger girlfriend, the the the terrible start. I I get if you became one of the greatest, you know, as one of the greatest independent journalists and media personalities around and build up your legacy over decades to just throw it away, like what drives someone to do that?

那就是

That's

我不知道。这就像是，好比说，成为最伟大的记者之一，然后去当大学校报的主编却抄袭。这就是比尔·贝利奇克

I don't know. It would be the the, the parallel would be, yeah, becoming one of the greatest journalists and then going to become the editor in chief of a college newspaper and plagiarizing. That is what Bill Belichick

正在做的事情

is doing right

。天哪。好了。我们来谈谈，说到Sora，我们来聊聊Sora。我的意思是，贝利奇克的事可以说个没完，但我们还是留给帕布罗吧。

now. Lord Almighty. Alright. Let's talk about, about speaking of Sora, let's talk about Sora. I mean, could go on the Belichick thing forever, but we'll leave that to Pablo.

所以，是的，The Verge有这篇报道，显然你我上周没机会讨论Sora，当时有马克斯·泽芬在，所以我想和你简单聊聊Sora，关于版权和使用的问题。OpenAI说，山姆·奥特曼基本上表示他对反应感到惊讶，或者说OpenAI没有完全预见到反应会如何。他说，我认为理论上人们会有什么感觉，然后实际看到这个东西时，人们的反应各不相同。与图像相比，感觉不同，超出了人们的预期。这当然是关于版权和权利的问题。

So, yeah, the the Verge has this story, so obviously you and I didn't get a chance to speak about SOAR last week, had Max Zephin, so I want to speak with you about SOAR briefly, the copyright thing, and the usage thing. OpenAI said it wasn't Sam Altman basically said he was surprised at the, reaction or that OpenAI wasn't fully, aware of what the reaction would be. He said, I think the theory of what it was going to feel like to people and then actually seeing the thing people had different responses. It felt different, to images than people expected. This is of course about copyright and rights.

你惊讶于权利持有人对Sora复制了他们的内容并给了他们退出选项而不是加入选项感到愤怒。这似乎很疯狂，而且不太诚实。你怎么会没预料到呢？

You're surprised that rights holders were sort of up in arms about the fact that SOR had copied their stuff and gave them an opt out as opposed to an opt in. That seems like crazy and not exactly truthful. How do you not anticipate that?

嗯，好吧。我确实想提这件事。上周我度假的时候，就特别想上播客聊聊Sora，因为我给自己做了个视频，在布鲁克林街头和马里奥兄弟电影里的库巴打架。显然，我六岁的儿子很喜欢。但当我看着它时，我觉得这太疯狂了。

Well, okay. I wanted I definitely wanted to bring this up. Last week, while I was on vacation, I was itching to be on the podcast just to talk about Sora because so I made video of myself in the Mario Brothers movie fighting Bowser on the streets of Brooklyn. And like, obviously, my six year old son loved it. But as I'm looking at it, I'm like, this is insane.

就像，我简直不敢相信这居然是可以接受的。两天后，OpenAI确实推出了更严格的内容指南。但让我震惊的是，Sam Alman表现得好像对此很惊讶。我想读一下OpenAI在48小时后发表的声明，当时大家都在创作新的《南方公园》剧集，大意是：人们渴望通过自己的想象力以及他们喜爱的故事、角色和世界与家人朋友互动。我们看到创作者有新的机会加深与粉丝的联系。

Like, I cannot believe that this is okay. Now two days later, OpenAI did introduce, like, more strict content guidelines. And it still blows my mind that Sam Alman acted like he was surprised by this. But I wanted to read the statement from OpenAI forty eight hours after, like, everyone was creating new South Park episodes was basically, it was people are eager to engage with their family and friends through their own imaginations, as well as stories, characters, and worlds they love. And we see new opportunities for creators to deepen their connections with their fans.

这是OpenAI媒体合作主管Varun Shetty。他确实说我们会应权利持有人的要求，屏蔽Sora中的角色，并响应下架请求。我觉得这太疯狂了，而且是个大问题，因为他们基本上是在说，东西已经在那里了，你们得来找我们提出下架请求。但我们基本上对此是默许的。我们甚至有点在推动你们说，你们创作媒体，但人们真的想通过把自己代入你们的版权材料中来加深联系。

This is Varun Shetty, the head of media partnerships in OpenAI. He does say we'll work with rights holders to block characters from Sora at their request and respond to takedown requests. I think this is nuts and a big deal because they are basically saying, like, it's out there and you're gonna have to come to us to bring takedown requests. But we're basically okay with this. And we're even kinda pushing you to say, you know, you create media, but people really wanna deepen their connections by putting themselves in your copyright material.

就像，我不知道这在OpenAI的整体故事中意味着什么，这是他们对版权和知识产权的态度。所以对我来说，我听到很多关于人们以及他们真正会上传到OpenAI的数据的敏感性问题。我觉得这提醒了我们，不管你有什么个人感受，你告诉ChatGPT治疗师的内容，有一天可能会自动发布成Sora的帖子。

Like, I don't know that to this in the overall OpenAI story, like, this is their approach to copyright and kind of intellectual property. So to me, you I I hear a lot of sensitivity around, like, people and what data are they really gonna upload to OpenAI. And I think this was such a reminder that, I don't know, basically, whatever your personal feelings you're telling your chat GPT therapist might get auto published into a Sora post, one day.

没错。而且Varun当然来自Meta。如果你曾经不得不处理Meta的版权侵权问题，那个过程糟透了，对版权持有人来说几乎是一种侮辱。它太繁琐了，不会激励你去处理，等到Meta下架时，东西已经传播到无关紧要的地步了。所以，我当然不感到惊讶。

Right. And Varun, of course, comes from Meta. And if you ever had to deal with Meta copyright infringement, the process, it sucks, and it's almost an insult to copyright holders to have to go through something like that. It's it does not incentivize it's so arduous, it does not incentivize you to work on it, and by the time Meta will take something down, the thing has already already spread to the point where it doesn't really make a difference. So, of course, it doesn't surprise me.

好吧。那么，这是否意味着OpenAI在这方面采取了正确的做法，令人遗憾地看涨？

Alright. Is is that sadly bullish OpenAI then in terms of this is the right approach?

我，也许这对业务有好处，但从道德上讲，我认为这最多是可疑的。好了，我们还剩五分钟。我们快速谈谈Tim Cook的潜在继任者。这是来自彭博社的消息，苹果将硬件主管John Turnis置于接班聚光灯下。

I I maybe it's good for the business, but, ethically, I think it's it's dubious at best. Alright. We have five minutes left. Let's talk quickly about the, the potential successor for Tim Cook. So it's from Bloomberg, Apple puts hardware chief John Turnis in the succession spotlight.

当Tim Cook最终卸任CEO时，他很可能以某种身份继续参与，或许担任董事会主席，这将使他走上Jeff Bezos、Larry Ellison、Bill Gates和Reed Hastings等其他科技领袖走过的道路。最大的问题是谁会在正式的CEO交接中日常运营苹果，硬件工程主管John Turnis仍然是主要竞争者。Gurman说苹果可能需要更多的技术专家，而不是销售或运营人员。尽管iPhone 17等产品显然深受客户欢迎，但公司在突破新技术类别方面一直挣扎。你怎么看？

When Tim Cook eventually steps down as CEO, it's likely he would remain involved in some capacity, perhaps as board chairman, that would put him on a path taken by other tech leaders including Jeff Bezos, Larry Ellison, Bill Gates, and Reed Hastings. The big questions who would run Apple on a day to day basis in terms of a formal tran CEO transition, hardware engineering chief John Turnis remains the leading contender. And Gurman says Apple probably needs more of a techno technologist than a sales or an operations person. The company has struggled to break into new technology categories even though products like the iPhone 17 are clearly resonating with customers. What do think about this?

我认为这是正确的、正确的、正确的举措。你的看法是什么？

I think it's the right the right the right move. What what's your perspective?

我认为需要一些新鲜血液。抱歉，蒂姆。我觉得这是，我不知道。我还是觉得我不知道。我在这里要做个决定。

I think some new blood is required. Sorry, Tim. I think it's, I I don't know. I still feel I don't know. I'm gonna make a call here.

他们应该收购Snapchat，让埃文·斯皮格尔当CEO。我说了，我说了。我希望公司重新有产品愿景。我们保留运营人员，蒂姆。从股东角度来看是有效的。

They should buy Snapchat and make Evan Spiegel CEO. I said it and I said it. I want some product vision at the company again. Let's keep keep the operational guys, Tim. It worked from a shareholder perspective.

但从真正的产品角度来看并不奏效。而且，他们只能从我这里榨取这么多29.99美元的Apple One订阅费，然后这必须转向其他地方。所以

It did not work from a true product per perspective. And the day, they can only squeeze so many twenty nine ninety nine Apple One subscriptions out of me before before this just has to go somewhere else. So

你知道谁会喜欢这样吗？埃文·斯皮格尔会喜欢的。我听过他身边的很多人说他自认为是史蒂夫·乔布斯。所以埃文肯定会全力支持。苹果会这样做吗？

And you know who would love that? Evan Spiegel would love it. I've heard multiple people, around him that he thinks he's Steve Jobs. So certainly, Evan would be all about that. Would Apple I do just

给你了，埃文。我刚给你了。

gave it to you, Evan. I just gave it to you.

我怀疑苹果不会这样做。不会。好吧。你认为蒂姆·库克应该在这个位置上待多久？我倾向于认为你应该可能应该尽早而不是晚些时候下台。

I doubt Apple would do it. No. Alright. When how long do you think Tim Cook should stay in the seat for? I tend to think that you should probably you should probably, step down sooner rather than later.

是的。我我觉得实际上，这可能是最顺利的那种情况，没有人会因此责怪他。这不是被解雇或被迫离开。只是时候继续前进了。这是公司发展的下一个阶段。

Yeah. I I think it's actually, like, it could be the most kind of smooth like, no one's gonna hold it against him. It's not a it's not like he was fired or pushed out. It's just time to move on. It's the next phase of the company.

很明显，他们必须弄清楚下一步该怎么做。我喜欢蒂姆，但他不是那个能为公司规划未来的人。我们已经看到了，这几年他们并没有这么做，而且其他公司都在推动创新和新产品开发，而苹果在这方面几乎毫无作为。

Like, it's clear that they have to figure out what's next. And I love Tim. He's not the guy to to figure out what's next for the company. We've seen it. We've seen this for a few years now that they're not going to and again, like, companies are just driving ahead with innovation and new product development that, like, Apple hasn't done anything with it.

Internus自2001年7月起已在苹果工作了24年。我实际上更喜欢Spiegel的想法，因为我坚信苹果确实需要一位非史蒂夫时代的CEO，因为公司内部很多事情之所以那样做，仅仅是因为史蒂夫·乔布斯当年就是这么做的，比如部门隔离、保密文化，显然这为他们带来了很大成功。但最终你必须说，好吧，让我们尝试点新东西。

Internus has been at Apple for twenty four years since July 2001. And I almost I I actually like the Spiegel idea even more because I'm of the belief that Apple really needs a not Steve era CEO because so many things inside that company are done there just because that's the way Steve Jobs did them, the silos, the secrecy, and obviously it served them very well. But eventually you have to be like, alright, let's try something new.

是的。不。我觉得，好吧。你在这里听到了。这是Kantrowitz和Roy的提议。

Yeah. No. I think, okay. You've heard it here. This is the proposal from Kantrowitz and Roy.

这虽然可能性不大，但已经提出来了。

It's it's a long shot here, but it's out there.

更疯狂的事情都发生过。Ranjan，非常感谢你的到来。一如既往地很高兴有你参与。

Crazier things have happened. Ranjan, thank you so much for coming on. Really great having you as always.

好的。下周见。

Alright. See you next week.

下周见。各位，FirstMark Capital的董事总经理里克·海茨曼将于周三做客节目，继续我们关于人工智能经济学的讨论，然后兰詹和我将在下周五与大家再次相聚。非常感谢大家的收听，我们下次在《大科技播客》中再见。

See you next week. Folks, Rick Heitzman, the, managing director of FirstMark Capital is gonna be on the show on Wednesday to continue our conversation about AI's economics, and then Ranjan and I will be back with you next Friday. Thank you so much for listening, and we'll see you next time on big technology podcast.