AI 基础设施能否持续指数级增长？

AI 要发展到多大规模，才能证明数万亿美元的基础设施支出是合理的？接下来，我们将与 Semi Analysis 的权威人士探讨这个问题。欢迎来到 Big Technology Podcast，一档冷静而细致地讨论科技世界及其之外的节目。AI 基础设施的数字似乎越来越大，看不到尽头。所以，我们将讨论整个建设过程的物理规律，以及它能否持续自我维持，也许不是毁灭世界，而是毁灭经济。因为我们有一些非常有趣的嘉宾来和我们讨论这个问题。今天，我们请到了 Semi Analysis 的 CEO 兼创始人 Dylan Patel，以及 Semi Analysis 的技术团队成员 Jordan Anos，来聊聊基础设施建设的方方面面。Dylan、Jordan，欢迎来到节目。

How big does AI have to get to justify the trillions in infrastructure spending? Let's talk about it with the leading authorities from Semi Analysis right after this. Welcome to Big Technology Podcast, a show for coolheaded and nuanced conversation of the tech world and beyond. The AI infrastructure numbers just seem to be getting bigger with no end in sight. So, we're going to talk about what the physics of this entire buildout looks like and whether it can continue to sustain itself without not maybe not destroying the world, but destroying the economy. Uh, cuz we have some very interesting guests to speak with us about it. Today, we're joined today by Dylan Patel, the CEO and founder of Semi Analysis, and Jordan Anos, member of the technical staff at Semi Analysis to talk all about everything uh going on within the infrastructure buildout. Uh Dylan Jordan, welcome to the show.

谢谢。很高兴来到这里。

Thanks. Great to be here.

我们先来谈谈 AI 基础设施建设规模有多大。Dylan，现在 X 上流传着一张图表，将 AI 基础设施建设与过去的建设相比，比如铁路、高速公路。图表显示，如今 AI 建设占 GDP 的百分比约为 3.6% 到 6%，抱歉，中位数约为 3.6%，平均年份，而铁路为 2.2% 到 2%。基本上，评论认为这要大得多，对吧？即使按占 GDP 的份额计算，这也比铁路大 50%。评论认为，这是我们从未见过的规模，实际上是前所未有的。我想听听你的看法，这到底有多大，这种看待方式是否正确，以及如果我们建设这么大的东西意味着什么。

Let's just start by talking about the scale of what we're seeing uh in this AI infrastructure buildout if we can. Uh Dylan, there's like a chart going around X right now that compares uh the AI infrastructure buildout to previous buildouts in the past like railroads, um like highways. And the chart is showing that the percent of GDP for the AI buildout today is about 3.6% 6% or sorry the P the median is about 3.6% uh the average year compared to 2.2% 2% for the railroads. And basically the commentary is that this is far bigger, right? That would put it basically 50% bigger than railroads even in terms of its share of GDP. And the commentary is this is bigger than anything we've ever seen before on a scale that is, you know, effectively unprecedented. Um I'd love to hear your perspective on just how big this is, whether that's the right way to look at it. Um and then what it means if we're building out something this big.

是的，我的意思是，当那张图表流传时，我第一反应是，人们的数字太低了。实际上已经比那高得多了。嗯，

Yeah, I mean uh the first thing I when that chart was going around, I was like, well, people's numbers are too low. It's actually much higher than that already. Um,

真的吗？好吧。那分享真实数字吧。

Really? Okay. So, share the real numbers.

是的，是的。如果你看明年，美国的资本支出大约在 2 万亿美元左右，不仅包括数据中心和芯片，还包括人们投资的其他所有供应链。所以资本支出接近 2 万亿美元。然后再加上 GDP 接近 30 多万亿美元。所以你在 5% 或 6% 左右，而不是 3% 或 2%。所以我的想法是，哇，这些数字比实际要低，因为人们低估了美国、北美和全世界正在发生的资本支出的数量和规模。我认为部分原因是，对于铁路，你的资本支出在你部署的国家。对于光纤，你的资本支出在你部署的国家。但对于 AI 基础设施，你在美国的资本支出实际上很大一部分用于服务美国以外的用户，对吧？例如，欧洲的大部分 Anthropic 推理在美国，欧洲的大部分 OpenAI 推理也在美国。类似地，如果你看很多亚洲国家，所以美国的资本支出实际上要高得多，因为第一，所有训练都在这里进行，第二，你实际上从美国的数据中心服务其他国家。所以最终，在这个阶段，产能的经济影响比以往任何物理基础设施投资周期都要大得多。

Yeah. Yeah. So, so like I mean it's it's if you look at like next year um capex across the US will be on the order of $2 trillion uh include not just data centers and chips but also all the rest of the supply chain that people are investing in. Um so you're close to $2 trillion of capex. Um and then you stack on top of that uh GDP is like close to 30 30 and some change trillion. So you're at like you know five or 6% right not not 3% or 2%. Um so so sort of my thought process is actually like oh wow these numbers are lower than the actual because people are understating the amount and size of capex that's happening in the US and and and North America and the world. Um, and I think like part of this is because, you know, with railroads, your capex is in the country that you're deploying it in. With fiber, your capex is in the country you're deploying it in. But for AI infrastructure, your capex in America is actually a lot of it's being used to serve outside of America uses, right? So, if you look at, for example, Europe, most of Europe's inference for anthropic is in America and most of OpenAI's inference in Europe is in America. Um and and similarly if you look at a lot of Asian countries um and so there is a bit of this like this this you know capex in America is actually like much higher as a result um because one you're doing all the training here and then two you're actually serving other countries from data centers in America um and so ultimately you're sort of at this stage where capacity is much um the economic impact of it is much larger um than any other sort of physical infrastructure. investment cycle ever before.

好的，Jordan。你是《Clustermax 3.0》的作者。这是 Semi Analysis 对 GPU 云系统的排名。基本上把许多超大规模云厂商和新兴云厂商分 tier，谈论它们的性能，我们稍后会谈到。但你在报告中全面审视了这一点，我想请你评论一下。基本上，我们看的是超大规模云厂商季度年同比资本支出增长的趋势，在 2026 年第三季度，增长率是 116%，年同比增长，这很惊人。然后共识预期，也就是华尔街的预期，在第四季度下降到低于 100% 的年同比增长，然后明年第一季度在 70%，之后继续下降。我认为这是我们习惯的周期，就像，我们在更大的基数上，所以不可能再次超出预期，不可能再次翻倍。所以我很好奇，从你的研究角度看，你认为我们真的会看到增长减速的趋势，还是我们又陷入了这个陷阱，或者华尔街又陷入了这个陷阱，就像，你已经到了你能想象的最大数字，但你仍然需要想得更大。

All right, Jordan. And you are the author of Clustermax 3.0. This is semi analysis rankings of the GPU uh cloud systems. Uh basically putting a lot of the hyperscalers and the neoclouds uh in tiers talking a little bit about their performance and we're going to get to that in a bit. But I I you know you do look across the board in your report and this is something that I I'd like you to comment on. Uh basically we're looking at um where the uh the hyperscaler quarterly yearon-year capex growth is going and in the third quarter of 2026 uh the growth is at 116%. Year-over-year which is astonishing. um the consensus estimates so what Wall Street is expecting to come after that you know they they drop down in the first quarter well in the fourth quarter to below 100% year-over-year growth and then the first quarter next year uh in the 70% and it continues to drop from there and I think this is a cycle that we're used to right where it's like well you know we're we're on a bigger base so it can't possibly exceed expectations again and it can't possibly double again um so I'm curious to hear from your perspective in your research, uh, do you think that we're actually going to see that trend like the growth is going to decelerate or are, you know, are we again falling into this trap or is Wall Street again falling into this trap where it's like, all right, you've you're basically at the biggest numbers you could possibly imagine, but you still need to think bigger.

是的，我认为我们肯定会在明年年底和进入……时遇到限制。

Yeah, I think we're definitely going to hit limits towards the end of next year and into

等等，Jordan，在你认真回答之前，我得说，就像你问一个孩子能想到的最大数字，他们说 100，对吧，现在就像，不不不，我们到了万亿，宝贝。

Wait, Jordan, before you answer seriously, I got I gota I got, you know, it's like when you ask a kid what's the biggest number they can think of and they say like,00 right and and now and it's like no no no we're at a trillion babe.

是的。

Yeah.

是的。四个逗号俱乐部。嗯

Yeah. Four commas club. Um

所以等等，Dylan，在这个例子中，孩子是华尔街。并不是说人们是孩子。只是，嗯，请继续。抱歉。

so wait in in Dylan in in this example the kid is Wall Street. Not not making any any statements that people are children. Just uh yeah, go ahead. Sorry.

Dylan 可能有时习惯把他的员工和客户称为孩子，但除此之外，嗯

Dylan may have an a habit of referring to his employees and his customers as kids sometimes, but that aside, um

我想我用一个更刻薄的词。

I think I think I use a much more mean word.

好吧。是的，

Okay. Yeah,

我已经喜欢这个走向了。

I like where this is going already.

我们快用完大数字了。在建设方面，有很多东西要看。这可能是从工业设备到准备做 EPC 工作（如工程、采购、施工）的公司，到芯片，到冷却系统，到你知道的所有外部组件，基本上是一个 IT 资产。有很多工作正在扩大供应链，以容纳所有这些资本支出。其中很多正在起作用，所以我们将看到巨大的增长，但是的，你不能只是有三位数的增长并进一步加速二阶导数。你会用完东西，我认为现在最大的事情是双重的。

We're we're running out of big numbers. There are there are things to uh look at when it comes to the buildout. This could be everything from like industrial equipment to firms that are prepared to do EPC work like engineering, procurement, construction to chips to cooling systems to you know all the components that go outside of that and um an IT you know estate basically. Uh there's a lot of work being done to expand this supply chain in order to be able to accommodate for all of this capex to to go into it. Um, and it's a lot of it's working and so we're going to see massive growth, but yeah, you you can't just have tripledigit growth and accelerate the second derivative, you know, even further from where it is right now. Um, you run out of things and I think the biggest thing right now is uh twofold.

其一，对于最大的新型云厂商和超大规模企业来说，要在他们已经在建的基础上再建一个吉瓦的设施，组织和调配所需资源变得非常困难。而且他们筹集资金也变得相当困难。尽管很多这类投资已经有惊人的回报，你仍然需要找到资金投入这些项目，当我们谈论的是数万亿美元的数字时，这很难做到。

One is that it becomes really hard for the biggest NeoClouds and hyperscalers to just organize and marshal the resources required to build another gigawatt on top of what they're already trying to build, for example. And it becomes pretty hard for them to raise the capital. Even though there are incredible returns on a lot of these investments already, you still just need to find the capital to pour into these projects, and that's a difficult thing to do when we're talking about numbers in the trillions of dollars.

我想说的另一点是，从定义上讲，数字必须缩小。今年我们有 10 倍增长——Anthropic 从不到 100 亿的营收增长到超过 1000 亿。所以他们有 10 倍增长。从定义上讲，他们到明年年底不可能有 1 万亿美元的营收。所以是的，增长在减速。英伟达的营收增长仍然是同比 2 倍，但不如早前那么高。因为从定义上讲，在同样的基数上再增长 2 倍或 3 倍更难。所以其他公司也一样。所以确实在减速。

I would say the other thing is that by definition the numbers have to shrink. This year we have 10x growth—Anthropic going from sub-10 billion of revenue to north of 100 billion. So they had 10x growth. Definitionally, they're just not going to have a trillion dollars of revenue by the end of next year. So yes, there's some decelerating growth there. Nvidia's revenue growth is still 2x year over year, but it's not as much as it was at an earlier point. Just because it is definitionally harder to grow 2x off the same base, or 3x off the same base. So the same applies to other firms. So it is definitely decelerating.

好的。所以支出的增长在某个时候需要减速。但营收和利润的增长必须加速，超过——基本上必须跟上这个大型基础设施建设的步伐，并且可能加速到超过我们看到的资金投入水平。我的意思是，你会想象为了促进所有 AI 活动。关于这一点有更多的讨论在流传。Dylan，让我把这些数字抛给你。这是一份高盛的报告。抱歉，高盛，他们说 Meta——这是一条推文，对吧？Meta 在承担了无法逾越的债务后已经企业自杀，公司将难以偿还。还有 Bane。Bane 说超大规模企业到 2031 年需要 6 万亿美元的营收——不是曾经，是每年——才能盈亏平衡。还有哥伦比亚大学教授 Stinvan Newberg，他说美国到 2032 年每年需要在 AI 服务上花费大约 3.5 万亿美元——占 GDP 的 8.8%——才能证明今天大规模数据中心投资的合理性。你怎么看这些关于支出和营收需要达到什么水平才能收回投资的评估？他们有机会兑现这些投资吗？

Okay. So the growth on spending is going to need to decelerate at some point. But the growth on revenue and profit for that matter is going to have to accelerate beyond—it's basically going to have to trail this big infrastructure buildout and accelerate probably beyond where we've seen the dollars being spent. I mean, you would imagine to facilitate all the AI action. And there have been further conversations that have circulated about this. And Dylan, let me just throw these numbers out to you. So this is a Goldman Sachs report. And sorry, so Goldman Sachs, they say Meta—this is a tweet, right? Meta's committed corporate suicide after taking on insurmountable debt that the company will struggle to pay back. There's Bane. Bane says hyperscalers will need 6 trillion in revenue—not ever, this is per year—by 2031 to break even. And there's a Columbia professor, Stinvan Newberg, who said that America will need to spend roughly 3.5 trillion annually on AI services by 2032—8.8% of GDP—to justify today's massive data center investment. What do you think of these assessments of where the spending and where the revenue needs to be to pay off the investments, and do they have a chance at making good on those investments?

是的。我认为，考虑到这次建设的速度，每个人对营收需求的数字都相当低。例如，我认为到 2031 年营收会远高于那个数字。这里有几个点，对吧？一是人们在资本支出上花了多少，然后营收总是滞后于资本支出。这些基础设施投资在会计基础上是六年——至少是可折旧资产——尽管有相当多证据表明它们实际上会持续更久，但六年是——我们就继续用六年。所以如果我今天在 AI 数据中心和 GPU 上花一万亿美元，GPU 持续六年，数据中心持续十五年。但这就意味着从营收角度，要么我假设它是直线，那么六年里我需要 1.5 万亿美元的营收，所以那一万亿美元资本支出每年需要 2500 亿美元营收；要么它是有斜率的，第一年少得多，最后一年高得多，才能收回投资。至于对 Meta 的说法，这有点愚蠢，因为 Meta 随时可以停止购买新东西，他们的现金流可以轻松支付所有已签署的东西，对吧？他们不必明年建一千亿美元的数据中心。他们可以随时拔掉插头。显然有他们已承诺的合同，但也有很多他们尚未承诺的东西。所以就 Meta 的偿付能力而言，他们完全有偿付能力。他们完全没问题。然后问题是他们在此基础上能赚多少营收？至于 AI 对美国经济的影响，是的，如果 2031 年 AI 营收只有 3 万亿美元，那么我们会有大问题。但我相信仅 OpenAI 和 Anthropic 到明年 12 月就能达到 6-7 亿。我认为仅这两家公司到 2027 年底甚至能达到 1.5 万亿。所以那是 27 年底。那么回报就很清楚了，对吧？你知道他们从哪里获得。最终，人们建设的产能与他们认为业务会来的方向相匹配。所以如果我们处于建设阶段，你花多少基础设施与 AI 营收之间总会有不匹配，因为今天的 AI 营收实际上基于你前几年花的基础设施。而现在资本支出没有每年增长 10 倍，但推理营收对 AI 实验室来说刚刚同比增长了 10 倍。所以我们看到 Anthropic 现在在算力成本减去营收——或者说营收减去算力成本——训练和推理合计方面已经盈利，OpenAI 将在接下来几个季度，也许两三个季度内达到，至少基于我们在 tokconomics 模型中的预期。所以我认为那些公司已经盈亏平衡了。如果你看亚马逊，例如最大的 AI 基础设施建设者，他们所有的 AI 基础设施投资今天都是盈利的。现在，他们的客户可能会破产——那些客户是 OpenAI、Anthropic 等。但目前他们有偿付能力，毛利率非常强。事实上，AI 使他们的毛利率在最近一个季度上升了，来自 Bedrock 和 Bedrock 的 AI 营收。所以是的，AI 营收需要大幅增长。但我认为，能够自动化我们一直能够做的所有事情，以及能够建造世界一直能够建造的所有东西，最终将为消费者产生有意义的营收和盈余。

Yes. I mean, I think everyone's numbers for revenue requirements are quite low given the pace of where this buildout is going to go. I think revenue will be much higher than that by 2031, for example. I mean, there are a couple points here, right? One is how much people are spending on capex and then revenue, and revenue always trails the capex. These infrastructure investments are six years on an accounting basis—at least depreciable assets—although there's quite a bit of evidence that they're actually going to last longer than that, but six years is—let's just continue to use six years. So if I spend a trillion dollars today on AI data centers and GPUs, the GPUs are lasting six years, the data centers are lasting 15. But that means definitionally in terms of revenue—either I could assume it's a straight line, right? So then over the 6 years I need to call it 1.5 trillion of revenue, so $250 billion of revenue a year for that trillion dollars of capex, or it's going to be something sloped where the first year is much less and the last year is much higher to get the payback. As far as statements on Meta, it's just kind of silly because Meta at any point can just stop buying new stuff and their cash flow can easily pay for all the stuff that they've signed, right? They just don't have to build a hundred billion dollars of data centers next year. They could pull the plug whenever they feel like it. There are obviously contracts that they've committed to, but there's also a lot of stuff they haven't committed to yet. And so as far as the solvency of Meta, they're completely solvent. They're completely fine. And then the question is what revenue do they make on top of that? And as far as AI's impact on the US economy, yeah, I mean, if AI revenue is only $3 trillion in 2031, then yeah, we'll have big problems. But I'm of the belief that we can get to like 600-700 million across OpenAI and Anthropic alone by December of next year. I'm of the opinion that you can get to 1.5 trillion maybe even by end of 2027 with just those two companies. So that's end of '27. So then the payback is clear, right? You know where they're getting that from. Ultimately, the capacity that people are building out is matched to what they think business is coming to. And so there's always going to be a mismatch if we're in the building phase of how much infrastructure are you spending on versus what's the AI revenue, because that AI revenue for today is really based on the infrastructure you spent in prior years. And right now capex is not going up at 10x a year, but inference revenue just went up 10x year on year for the AI labs. And so we've seen Anthropic is now profitable in terms of compute cost minus revenue—or revenue minus compute cost—both training and inference combined, and OpenAI gets there within the next couple quarters, maybe two or three quarters, based on at least what we expect in the tokconomics model. So I think those companies are already breaking even. If you look at Amazon, for example, the biggest builder of AI infrastructure, all of their AI infrastructure investments are profitable today. Now, the customers for them could go insolvent—those customers being OpenAI, Anthropic, etc. But right now they are solvent on those, making very strong gross margins. In fact, AI has made their gross margin go up in the most recent quarter from Bedrock and AI revenue from Bedrock. So yes, AI revenue needs to grow a lot. But I would posit that being able to automate all the stuff that we've been able to and being able to build all the stuff that the world has been able to is going to generate meaningful revenue and surplus for the consumer ultimately.

是的，我认为我们需要在头脑中稍微扩展一下，以理解未来我们将如何衡量 AI 营收。

Yeah, I think we need to stretch it a little bit in our minds to understand what we are going to measure as AI revenue going into the future.

如果这些公司真的拥有世界上最好的模型，能够有效地完成所有知识工作，你可以想象它们会进入除编程之外的许多不同行业，并获得与它们在编程领域所见类似的投资回报。所以你可以看看机器人技术和自动驾驶汽车。你可以看看药物发现。你可以看看整个健康与生命科学领域。我们可以把这些行业都考虑进来，说如果它们想将产品推向市场，比如抗癌药物或自动驾驶汽车，你可以想象基于它们今天的投资，它们可能带来多少收入。如果它们拥有 AI，拥有能够出色地完成构建该产品所需的知识工作的系统，我认为它们是唯一能够在做这些事情的前沿真正竞争的公司。所以我认为人们缺乏创造力，无法理解除了今天的 Claude Code 订阅和按 token 定价模式之外，回报还可以如何被看待。

If these companies really do have the best models in the world that are capable of effectively all knowledge work, you can imagine them going into a whole bunch of different industries beyond coding and having returns on their investments similar to what they've seen in coding. So you can look at things like robotics and self-driving vehicles. You can look at drug discovery. You can look at just health and life sciences generally. And we can take this across all these industries and say if they want to bring a product to market like a cancer drug or like a self-driving vehicle, you can just imagine how much revenue they can potentially return based on the investments they have today. And if they have the AI, the system to be able to produce the knowledge work required to build that product incredibly well, I think they're the only ones that can really compete on the frontier of being able to do these things. And so I think people lack the creativity to understand how returns can be viewed beyond the like Claude Code subscription and per token pricing regime of today.

是的，有这张图表，研究我们支出占 GDP 的比例。这只是 2025 年的数据。医疗占 GDP 的 18%，是最高类别。食品是 9.1%。AI 服务是 8.8%。它在 2025 年是第三大类别。我想它会在 2026 年超过食品。所以，我很好奇这些钱从哪里来？比如，假设 OpenAI 和 Anthropic 明年合计达到 7000 亿美元收入。这些钱是来自这些公司的劳动力池吗？还是只是因为它们能卖更多而增加的？这些公司是这些公司的客户，这些钱从哪里来，能 funnel 到 AI 公司？

Yeah, there's this chart that looks at how much we're spending as a share of GDP. This is just 2025. So a share of GDP health is 18% that's the top category. Food is 9.1%. AI services is 8.8%. It's the third category in 2025. I imagine it's going to lap or it will pass food in 2026. So guys, I'm just curious where's this money coming from? Like, let's say OpenAI and Anthropic combined get to $700 billion in revenue next year. Is the money coming from these companies' labor pools? Is it just additive because they can sell more? Where is the money that these companies are the customers of these companies? Where's that money coming from that they can funnel it into the AI companies?

这是几个因素，对吧？首先，显然公司能够加速它们正在构建的东西，因此它们可以投资更多。研发支出中越来越大的份额将流向 AI。销售成本中越来越大的份额也将流向 AI。但也会有大量资金来自劳动力，对吧？劳动力，我不认为劳动力本身会萎缩。我认为其他经济领域增长如此之快，至少在短期内，劳动力会保持平稳。所以就像过去，我们在经济中经历了不同的周期，但一个长期存在的周期是，几十年来劳动力占 GDP 的份额一直在下降，自 70 年代以来一直缓慢而稳定。所以很多人说“哦，美国经济很糟糕”的许多纷争，其实是因为经济规模更大了，劳动力占经济的份额缩小了，这就是为什么实际上更多的劳动力价值正在流向资本。AI 所做的是，它将加速这一趋势，劳动力份额与资本份额的长期趋势，因为 AI token 基本上只是资本，以芯片和数据中心的形式存在，创造劳动力，所以劳动力份额将下降得更快。所以我希望，对社会来说，这是我的信念，也许这是一个非理性的信念，但我相信我们会让经济增长得如此之快，以至于以美元计算的劳动力不会萎缩。它甚至可能缓慢增长。只是其他经济领域增长得太快了。这就是它的来源，对吧？所以，所有这些提供各种服务的公司，它们的服务会变得更好。它们能够收取更多费用。它们能够卖出更多。而且它们能够用更少的人做到这一点。或者另一种看待方式是，它们将拥有相同数量的人，而它们的业务会变得更好。这就是，而且有很多行业这是可能的，对吧？显然编程是人们已经做了很多的一个领域。显然一般白领工作正在迅速自动化，但公司也能够比过去做更多的事情。但有很多行业，比如制药、交易和芯片设计等等，对吧？芯片设计就是一个很好的例子，对吧？芯片。从事芯片设计的美国研发工程师数量基本上 20 年来一直持平。然而其经济规模却爆炸式增长。现在，显然制造业不是，我们不算制造业。我们只算设计师。但设计师就像，你知道，世界上最有价值的两家公司，对吧？英伟达、博通，以及谷歌和所有其他制造自己芯片的公司等等，对吧？这些设计师在人数上并没有真正增长多少。作为一个行业，对我们来说一直持平，但积累的价值却爆炸了。为什么？因为有很多工具。现在 AI 辅助芯片设计做了什么？现在经济中所有的芯片都变得越来越好。你的手机会变得更好。你的 AirPods 会变得更好，里面的芯片。机器人的芯片会变得更好。显然 AI 的芯片会变得更好。各种东西的芯片。蓝牙和 GPS 以及所有其他东西。博世制造的用于气体泄漏等的传感器。这些都会变得更好，因为 AI 可以帮助你设计芯片。你做同样数量的人，收入增长，利润率增长，一切看起来都很棒，对吧？这就像有很多行业可以复制这一点。在我看来，

It's a few factors, right? First off is obviously companies are going to be able to accelerate what they're building and so they can invest more. Increasing share of R&D dollars will go towards AI. Increasing share of cost of goods sold dollars will go towards AI. But also you'll have a significant amount of money sort of coming from labor as well, right? Labor and I don't think labor shrinks per se. I think the rest of the economy grows so fast, at least in the short term, that labor sort of stays flattish. And so like in the past, we've had different cycles across the economy, but one long-standing one has been labor share of GDP has fallen over the decades, and it's been a steady slow since the 70s. And so a lot of the strife that people say, "Oh, America's economy sucks." And yet, oh look, the economy is bigger is actually well like labor share of economy is shrank and that's why in reality a lot more of the labor the value is getting accruing to capital. What AI does is it's going to supercharge that trend, long-term trend of labor share versus capital share because AI tokens are gener basically just capital in the form of chips and data centers that are sitting there and creating labor and so that labor share is going to fall even faster and so I would hope and for society that that's my belief and maybe it's like a irrational belief but I believe that we will just grow the economy so fast that labor in terms of dollars will not shrink. And it might even grow slowly. It's just that the rest of the economy grows so fast. And that's where that's sort of where it comes from, right? So, all these companies that do offer various services, their services are going to get better. They're going to be able to charge more. They're going to be able to sell more. And they're going to be able to do it with fewer people. Or another way to look at it is they're going to have the same number of people and their business just going to get better. And that's and there's a number of industries where this is possible, right? Obviously coding is an area where people have done it quite a bit. Obviously general white collar work is getting rapidly automated, but also companies are just able to do a shitload more than they were able to do in the past. But there's a lot of industries like pharma and trading and chip design and such, right? Like chip design is a great example, right? Chips. The number of American R&D engineers working in chip design has been basically flat for 20 years. And yet the economy for it has exploded. Now, obviously manufacturing isn't, we're not counting manufacturing. We're just counting the designers. But the designers are like, you know, two of the most valuable companies in the world, right? Nvidia, Broadcom, as well as like Google and all these other companies that are making their own chips and many others, right? These designers have not really grown that much in terms of number of people. It's been flat as an industry for us and yet the value that has accrued has exploded. And why is that? There have been a lot of tools. Now what does AI assisted chip design now do? Well, now all of the chips that are across the economy get better and better. Your phones are going to get better. Your AirPods are going to get better, the chips in them. Chips for robots are going to get better. Chips for obviously AI are going to get better. Chips for all sorts of things. Bluetooth and GPS and all these other things. Sensors for gas leaks or whatever that Bosch makes. These are all just going to get better because AI can help you design chips. You do the same number of people, revenue grows, margins grow, everything looks fantastic, right? And that's sort of like there's many industries where this can get replicated. In my view,

Dylan，你曾经说过，我想你过去说过类似“知识工作完蛋了”的话。所以你现在不再相信了吗？

Dylan, you've said that I think in the past you've said something to the extent of knowledge work is effed. So do you no longer believe that?

我认为大多数知识工作，好吧，让我们分开来看，我对 2035 年乐一通式愿景的信念与对 2028 年的信念。我认为在 20，我认为在未来 X 年，只要 AI 非常强大但并非在大多数方面都比人类更好，最具能动性的人将能够做最酷的事情，他们能够对世界产生更大的影响。现在，在 AI 比所有人都更好的情况下，那可能会急剧变化，对吧？那么知识工作就完蛋了，实际上重要的是人们能够玩游戏、享受乐趣，人们能够表演，人们能够做人类做的事情，这些是只有人类才能互相欣赏的，如果 AI 不杀死我们所有人的话，对吧？但这有点像，你知道

I think most knowledge work okay like let's separate like my belief in like Looney Tunes 2035 versus like you know 2028 I think in 20 I think for the next you know X number of years as long as AI is very powerful but it is not better than humans in most every way the most the most high agency people are going to be able to do the coolest stuff they're going to be able to make way bigger impacts on the world Now now in like the case where AI is just better than everyone that that could change drastically right now then then like knowledge work is screwed and actually like you know what matters is like people being able to like play games and have fun people being able to act people being able to like you know do the things that humans do that humans only appreciate in each other if the AI doesn't kill us all right but that's sort of like you know

我们今天不谈这个。

We won't go there today.

Jordan，这算是你的专长领域。我想先聊聊风险，比如什么可能会让这个故事脱轨。我们之前已经谈过，有些合同像 Meta 这样的公司可能不必履行。你显然在大量研究 Neocloud 的生意，而这些生意依赖于 Meta 这样的公司真正兑现它们对这些公司许下的建设承诺。这是不是一种被低估的风险？现在如果你去问这些 Neocloud，一切看起来都挺好，因为他们会告诉你，他们拥有世界上最好的客户，他们有 Meta，Meta 会兑现承诺，而且他们正基于此发债。但现实中的风险可能是，像 Meta 这样的公司可能就决定，对，我们不干了，而这些合同可能并不像 Neocloud 告诉我们的那样有约束力。

Jordan, this is kind of in your area of expertise. I want to get started talking a little bit about the risks, like what might derail this story. We'd already talked about the fact that there are some contracts that a company like Meta may not have to go through with. You're obviously looking a lot at the businesses of the Neoclouds, which are dependent on companies like Meta actually making good on the promises that they've put forth to build with these companies. Is this sort of an underappreciated risk? Everything looks fairly fine right now if you speak to one of these Neoclouds, because they'll tell you that they have the best customers in the world, they have the Metas, and Meta will make good on the promise, and they're raising debt based off of this. But in reality the risk might be that a company like Meta might just decide, yeah, we're not going to do that, and that the contracts might not be as binding as the Neoclouds tell us they are.

好。我想先澄清一件事，关于 Dylan 刚才说的终止合同，与 Meta 和 Neocloud 之间的情况，区别在于：一般来说，这些超大规模厂商或大型前沿实验室跟 Neocloud 签合同，并没有按月随意取消的权利，不能想拔插头就拔插头、合同就结束了。一般来说，他们签的是五到六年的期限，把全部容量都包下来，有固定的每小时费率和总合同价值，这对 Neocloud 供应商非常有吸引力，因为这些人已经出去拿地、拿电、建站，甚至在某些情况下在真正签合同之前就已经把芯片买好了。Dylan 所说的放缓、暂停或取消项目的可能性，更多是针对自建站点，也就是说 Meta 有庞大的园区，未来将作为容量的基础负载，他们可以选择买芯片，也可以选择让它空置多年，如果他们想放缓、不想全力推进的话。而且一般来说，建设成本虽然非常昂贵，但在一个新芯片项目的总成本中只占很小一部分。换句话说，混凝土、钢铁和墙壁很重要、很大，也确实是个实实在在的挑战，但从金额上看，HBM、光模块和 GPU 本身才是账单的大头。所以现在的情况看起来是，所有这些大公司都真的需要应对来自客户群的巨大增长需求——不管是 Meta、OpenAI 还是 Anthropic，它们的消费级应用都有需求，比如 ChatGPT、Claude Code、Codex，甚至 Meta 的 Muse 之类的。为了服务这些需求，他们需要 GPU，而服务这些需求是有利可图的，所以他们需要扩产，准备好把这东西铺到 Meta 各平台服务的数十亿月活用户，ChatGPT 接近十亿月活，Claude Code 之类则是数亿。

Okay. So I want to be clear about one thing Dylan was talking about there, in terms of termination versus what's going on with Meta and the Neoclouds, which is that generally speaking these hyperscalers or large frontier labs who are signing up with contracts for the Neoclouds do not have arbitrary cancellation rights on a monthly basis where they can just pull the plug and the contract's done. Generally speaking, they're signing five-year, six-year term lengths and they're off taking the full capacity, and they have a set hourly rate and total contract value that's very, very attractive to the Neocloud providers, because those guys have gone out and secured land and power and built a site and even in some cases purchased the chips respectively before they've actually gone and done that contract. What Dylan's referring to with the potential to slow down or pause or cancel projects, it's more with respect to the self-build sites, which is to say Meta has massive campuses that are going to serve as more of a base load of capacity going into the future, which they can either choose to buy chips for or just leave as an empty shell for years if they want to slow down and not go as hard as they can. And generally speaking, construction costs, while very expensive, are a small percentage of the total cost of a new chip project. In other words, the concrete and the steel and the walls is quite, you know, it's important and big and a very real challenge to get it done, but on a dollar basis, the HBM and the optics and the GPUs themselves are the bigger portion of the bill. And so what it's looking like right now is that all of these big companies are really needing to contend with the massive growth that they're expected to pursue by their demand from their customer base — either that's Meta or OpenAI or Anthropic, who have demand on their consumer-facing applications like ChatGPT or Claude Code or Codex or even Muse from Meta, things like that. In order to serve that demand they need GPUs, and it's profitable to serve the demand, so they need to ramp up and be prepared to roll this thing out to the billions of monthly active users on the Meta surfaces that they serve, and approaching a billion of monthly active users for ChatGPT and hundreds of millions for Claude Code and things like that.

那么，带着这个区分，项目取消或需求下滑之类的风险是什么？

So anyway, with that distinction in mind, what's the risk with project cancellation or a downturn or things like that?

我的意思是，要让我们预测供需曲线出现某种变化，你真的需要看到需求放缓。芯片能生产多少、数据中心能建多少、这些东西能多快上线，都有硬性约束。而截至目前，我们能做的所有建模基本上都显示需求超过供给。那么，现在需求是比供给多 10%？多 50%？还是无限大？我们没法回答这个问题，因为以现在能上线的 GPU 供给，我们追不上需求。所以当这个问题有答案的时候，如果有答案的话，如果我们发现知识工作的需求、经济对智能的需求存在上限，那我们就能弄清楚谁需要在什么时候、在哪里放缓，以及事情会怎么演变。但到目前为止，我想这些上限还没有被找到。

I mean, in order for us to forecast some sort of change in the demand-supply curve there, you'd really need to see demand slow. There are hard constraints on how many chips you can produce, how many data centers you can build, and how fast this stuff is going to come online. And as of right now, all of the modeling we can do basically shows that demand outstrips supply. And so, is demand 10% more than supply right now? Is it 50%? Is it infinite? We don't get to answer that question because we are not able to catch up to demand with the supply of GPUs that you can bring online right now. And so when that question gets answered, if it gets answered, if we find limits to the demand for knowledge work, to the demand for intelligence from the economy, then we'll be able to figure out who needs to slow at what time, where, and how that plays out. But so far, I guess those limits haven't been found.

好。在进入 Cluster Max 之前，我还想聊最后一个话题：一场即将到来的大 IPO，当然就是 Anthropic 的 IPO。Dylan，你发过一条很有意思的推文，我直接念一下。你说：“你不能在 2 万亿美元的估值上投资 Anthropic 的 IPO，因为它可能归零，也可能到 20 万亿，但到了 20 万亿，我们就全完了。”你能解释一下，并跟我们说说你是什么意思吗？

All right. Before we get into Cluster Max, one last topic I want to cover: big IPO on the way, which is, of course, the Anthropic IPO. Dylan, you had a very interesting tweet about it. I'm just going to read it. You said, "You can't invest in the Anthropic IPO at $2 trillion because it could be zero, or it could be 20 trillion, but at 20 trillion, we're all fucked." Can you explain that and tell us a little bit about what you mean by that?

先说清楚，这完全就是一条帖子。我当时就是——

To be clear, it's entirely a post. I was like—

你知道吗？我明白，但我觉得你说的有些道理。跟我们聊聊吧。

You know what? I get it, but I think there's some truth to what you're saying. So talk to us about it.

我总体上极度看好 Anthropic。我认为 Anthropic 会成为世界上非常成功的公司。Anthropic 甚至可能成为世界上第一家市值 10 万亿美元的公司。我就是这么看好 Anthropic。但与此同时，要成为 10 万亿或 20 万亿美元估值的公司，尤其是 20 万亿，首先，他们的收入倍数得大幅下降。他们以前是按 30 倍融资的，现在 IPO 时，就算他们有一千亿美元收入，可能也就 20 倍。如果他们要达到 10 万亿，我打赌他们得有大概一万亿美元的收入之类的疯狂数字，对吧？这在 28 或 29 年的某个时候可能发生。当然这取决于——股市也可能疯掉，让倍数也疯掉。但在他们值 20 万亿的世界里，我不觉得他们的收入——我觉得他们现在连 10 倍市销率都不到，大概也就 5 倍甚至更低——那样的话，如果他们收入是 20 万亿，那他们是 2、3、4 万亿？20 万亿估值，2、3、4 万亿收入，到那个时候，如果在一个 150 万亿美元的全球经济里，他们产生数万亿美元的收入，并且是迄今为止世界上最大的公司，那 AI 的能力得疯狂到什么程度。

I'm generally extremely bullish on Anthropic. I think Anthropic will be a very successful company in the world. Anthropic could even be the first company to be a 10 trillion valuation company in the world. Like I'm that bullish on Anthropic. But at the same time, for them to be a 10 or 20 trillion valuation company, especially 20, first of all, their multiples on revenue are going to have to collapse quite a bit. So they used to raise at 30x, and now the IPO is going to be, let's call it, they're at a hundred billion dollars of revenue and they're going to maybe be at 20x. If they get to 10 trillion, I bet you they would need to be at like a trillion dollars of revenue or something crazy, right? Which could happen at some point in '28 or '29. It depends, of course — the stock market can also go crazy and make multiples go crazy. But in the world where they're worth 20 trillion, I don't think their revenue — I don't think they're even at a 10x sales multiple now, they're probably at like a 5x or even less — in which case, if they're at $20 trillion of revenue, are they at like 2, 3, 4 trillion? $20 trillion valuation, 2, 3, 4 trillion revenue, and at that point, what is the state of AI if, on a $150 trillion global economy, they're generating multiple trillion dollars of revenue and are by far the largest company in the world? AI's capabilities are freaking insane.

在这种情况下，所有关于 AI 会杀死我们所有人，或者 AI 太好以至于每个人都失业的担忧，因为收益分配将严重倾向于持有 Anthropic 股票的人或其供应链上的公司，而且分配不均，那么人们会感到痛苦并推翻政府，因为所有增长都积累在极少数人手中。所以这些事情可能会撕裂社会结构，因此如果你相信 Anthropic 会达到 20 万亿美元，整个社会可能会……

In which case, all of this worry about whether it's AI killing us all or AI being so good that everyone loses their jobs, and because the distribution of that earnings is going to be heavily leveraged to the people who own Anthropic stock or the constituent supply chains of it and not distributed well, then people are going to have distress and overthrow governments because all of the growth is accumulating to a very small number of people. So these things could tear the fabric of society apart, and therefore if you believe Anthropic will get to $20 trillion, society as a whole could be...

Jordan，你有类似的担忧吗？

Jordan, do you have similar worries?

我觉得是的，我认为要达到那个目标还有很多事情要做。实际上我不觉得自己一定像 Dylan 那样担心。

I think yeah, I think there's a lot to do to get to that point. I actually don't think I'm necessarily as worried as Dylan.

顺便说一句，我是乐观的。我只是描绘了悲观的图景。我来谈谈乐观的。

I'm optimistic, by the way. I just painted the pessimistic picture. I'll go optimistic.

是的。我的意思是，乐观的美好观点就是我之前描述的那样：如果他们达到 20 万亿美元的收入，这不是靠对使用 Claude Code 作为产品的用户按 token 计费来实现的，而且我不认为你会看到 Claude Code 在质量上比现在好 20 倍并推动收入。我认为他们会进入其他业务领域，而且我会相当信任 Anthropic 将，如我之前所说，一种抗癌药物推向市场之类的。我们最近看到 GLP-1 类药物发生了类似的事情，对经济中的许多人来说意义重大，因此像礼来这样的公司从那些带来帮助人们的不可思议药物的惊人研究中获得了巨大回报。所以我认为，对于理论上 Anthropic 达到 20 万亿美元估值的未来，我会更乐观。但我确实认为沿途会有很多颠簸。也就是说，回到你之前的问题，他们怎么可能服务所有这些需求并训练这些模型，并让智能扩散到经济的各个部分，以便人们能够实现收益。我认为做到这一点的方法是建设算力，而这非常依赖于建设和安装 GPU 并可靠运行等细节。对我来说，在这条从 2 万亿美元 IPO 到公开市场 20 万亿美元估值的道路上，世界的中心是 NeoClouds。

Yeah. I mean, the rosy optimistic view is kind of what I described earlier, which is if they're getting to $20 trillion of revenue, this is not done on a per-token metered basis for people using Claude Code as a product, and I don't think you go and see Claude Code go 20x better from where it is right now in terms of quality and that drives the revenue. I think that they get into other areas of business, and I think I would trust Anthropic quite a bit to bring, like I said earlier, a cancer drug to market or something like that. And we've seen similar stuff happen recently with GLP-1s just being such a meaningful thing for so many people in the economy, and as a result, companies like Lilly have had massive returns on that incredible research that led to those incredible drugs that help people. So I think I would be more optimistic for the future where theoretically Anthropic's at a $20 trillion valuation. But I do think that there are going to be a lot of bumps along the road along the way. Which is to say, going back to your question previously, how can they possibly serve all of this demand and train these models and have intelligence diffused to all of these different parts of the economy so that people can realize the benefits. And I think the way you do that is to build compute, and it so depends on the nitty-gritty details of construction and installing GPUs and running them reliably and all of that. To me, the center of the world is the NeoClouds on that road from a $2 trillion IPO to a $20 trillion valuation in the public markets.

是的，Anthropic 是一个有趣的商业案例，对吧？他们把所有这些钱花在基础设施上。即使在 S-1 中，你也看到伴随收入而来的巨大亏损。尽管，Dylan，正如你提到的，看起来他们所做的是稍微扭转了局面，现在他们赚的钱比花在算力上的钱多。稍微谈谈，因为我认为这里的信念是像 Anthropic 这样的公司是低利润业务，或者至少比软件低得多。但实际上似乎并非如此。你们已经计算过了。

Yeah, Anthropic is kind of an interesting business case, right? They're spending all this money on infrastructure. Even in the S-1, you saw big losses accompanying their revenue. Although, Dylan, as you mentioned, it looks like what they've done is they've turned the tide a bit and they're making more money right now than they're spending on compute. Talk a little bit because I think the belief here is that a company like Anthropic is a low-margin business or at least much less than software. But that doesn't actually seem to be the case. And you guys have run the numbers.

我想说，如果你看看 Salesforce，它可能是最伟大的 SaaS 公司之一，我的意思是，至少持续成长为最大的 SaaS 公司之一，他们的毛利率大约是 75%。而 Anthropic 在推理上的毛利率是 75%，甚至略高一些，尤其是在前沿模型上。所以这就像是，Salesforce 花了相当长的时间才达到那里，对吧？他们 IPO 时并没有 75% 的毛利率。他们花了相当多的时间。实际上，他们当时是，但后来在 2000 年代中后期，他们还没有完全达到。Anthropic 达到那种强劲毛利率水平的时间要短得多。如果你看看许多其他 SaaS 公司，他们的毛利率甚至没有那么高。然后另一件事是他们的客户获取成本高得离谱，对吧？如果你看一家 SaaS 公司，就像是，好吧，我的销货成本很低，但我的客户获取成本可能高达 30%。而 Anthropic 现在几乎没有客户获取成本。显然，他们有巨大的研发支出，但他们的研发支出增长慢于收入。所以现在他们已经实现盈利，没有理由这个趋势不会继续，对吧？他们继续变得更盈利。所以至少在营业利润率方面。然后可扩展性是另一件事，对吧？许多 SaaS 公司的美妙之处在于，即使你的客户获取成本很高，只要你的利润率足够高来支付初始成本，最终你不需要重新获取那个客户。现在你有了这个超级盈利的业务。但对于 AI 实验室来说，他们的业务似乎有些粘性。是的，有所有这些信息说，哦，Meta 在削减支出或微软在削减支出之类的，但最终收入在图表上只是一条直线上升，几乎每个客户的收入都在增长。而且转换成本，即使你知道一些其他中国模型出现或出现更好的模型，也不会真正改变现有客户的支出份额。你最终会得到一个相当粘性的客户，可能以某种速度增长收入，没有客户获取成本，利润率很高，所以可以说，AI 实验室是比软件更好的业务，如果你能成功……

I would say if you look at Salesforce, which is probably one of the greatest SaaS companies out there, I mean, I think at least consistently grown to become one of the biggest SaaS companies, their gross margins are like 75%. And Anthropic on their inference is doing 75% gross margins, a little bit higher even, especially on the frontier models. So it's sort of like, well, and it took Salesforce quite some time to get there, right? They were not at 75% when they IPOed. They took quite a bit of time. Actually, they were, but then in the mid to late 2000s, they were like not quite there. It took Anthropic quite a bit less time to get to that strong level of gross margin. And if you look at many other SaaS companies, they're not even that high. And then the other thing is their customer acquisition cost is insane, right? If you look at a SaaS company, it's like, fine, my COGS is really low, but then my customer acquisition cost can be as high as 30%. And Anthropic has near no customer acquisition cost right now. Obviously, they have a humongous R&D spend, but their R&D spend has grown slower than their revenue. And so now they've reached profitability and there's no reason why that trend doesn't continue, right? And they continue to grow more profitable. So at least operating margin-wise. And then scalability is the other thing, right? The beautiful thing about a lot of SaaS companies even if your customer acquisition cost is so high, as long as your margins are high enough to pay for it initially, eventually you don't have to reacquire that customer. Now you've got this super profitable business. But with the AI labs, their business seems somewhat sticky. Yes, there's all this messaging of like, oh, Meta's cutting spend or Microsoft cutting spend or what have you, but ultimately the revenue has only been a straight line up on the chart and almost every customer's only grown on revenue for them. And switching cost even when some other Chinese model comes out or comes out with a better model doesn't really shift the spend share that much for existing customers. You end up with this pretty sticky customer potentially who is growing revenue at some pace and there's no customer acquisition cost and the margins are high, so it's arguably a better business than software is, an AI lab, if you can successfully...

但风险有多大，因为出现了这种“控制前沿”的运动？

But how much is the risk the fact that there's this pacing the frontier movement coming out?

我认为那只是信号。我认为这完全是……

I think that's just signaling. I think it's completely...

谈谈那个。你不认为那会……那根本不是真的。

Talk about that. You don't think that's going to... that's true at all.

我认为如果你相信我们可能会杀死自己，那么控制前沿是一种负责任的做法。不过我不一定认为他们会过多地控制前沿。我认为鉴于他们所处的竞争，他们能做的控制很少。人们已经对此做了滑稽模仿，但 Dario 不能慢下来，因为他不信任 Sam，而 Sam 不会慢下来，Sam 也不能慢太多，因为 Dario 也一样。显然，当他们发生这些大规模安全事件时，他们会稍微慢一点。而且他们俩都不能慢下来，因为中国落后不太远。所以这非常困难。

I think pacing the frontier is a responsible thing to do if you believe that we could kill ourselves. I don't necessarily think that they will pace the frontier too much though. I think there's very little pacing they can do given the race that they're in. And people have made skits about this, but Dario can't slow down because he doesn't trust Sam, and Sam's not gonna slow down, and Sam can't slow down much because same with Dario. Obviously, they're going to slow down a little bit when they have these massive security incidents. And neither of them can slow down because China's not too far behind. So it's very difficult.

这真让人放心。好吧。

That's reassuring. All right.

嗯，我们稍作休息，回来后我们会聊一聊 AI 云的状态，当然还有 Cluster Max，然后继续问一些关于技术和业务的问题。所以，休息一下，马上回来。

Um, look, we're going to take a quick break and then when we come back, we're going to talk a little bit about, um, the state of the AI clouds, uh, cluster max, of course, uh, and then we'll continue on with some more questions about the technology and the business. So, we'll do that when we're back right after this.

我们回到 Big Technology 播客，嘉宾是来自 Semi Analysis 的 Dylan Patel 和 Jordan Anos。Semi Analysis 想必大家都很熟悉，它是 AI 基础设施领域的圣经。这两位是最棒的。非常感谢你们抽出时间，我强烈推荐大家至少去看看博客。你也可以在 subscanalysis.com 上获取 Semi Analysis，它是排名第一的技术通讯，名副其实。恭喜你们。

And we're back here on Big Technology podcast with Dylan Patel and Jordan Anos from Semi analysis. Semi analysis, I'm sure you're familiar. It's the Bible for anything AI infrastructure. Uh, these guys are the best. very grateful to have your time both of you and um you know definitely recommend you at least uh check out the blog. You can also get semi analysis on subscanalysis.com. It's the number one technology newsletter uh and welld deserved. So congratulations guys.

在通讯里你可以找到 Cluster Max 3.0。即使你不是技术背景，我也和 Jordan 聊过，读起来很有趣。它深入探讨了建设现状。副标题告诉你，它已经以正确的方式启动，详细描述了 GPU 云提供商的可靠性、性能、支持、定价和安全性。里面内容很丰富。

In the newsletter you can uh find the clustermax 3.0. Uh it's it's actually even if you're not technical I was talking with Jordan about this. It's a fun read. It goes into the state of the buildout. Um, and the subhead tells you, you know, it's already kicked off in the right way in the gory detail, the reliability, performance, support, and pricing and security of the GPU cloud providers. There's a lot in there.

嗯，Jordan 和 Dylan，我们可以深入很多领域，但让我们从 Nvidia 的角色开始。报告虽然大部分时间花在技术方面，但确实提到了 Nvidia 在保持这些集群提供 AI 基础设施方面扮演着重要角色，可能面向一些前沿实验室，但实际上是全方位的，让整个事情运转起来。正如你们所写，Nvidia 一直在通过收入下限、房东担保和租赁来支持他们的许多业务，并计划将这些转移给第三方，帮助 Neo Labs 和 AI 实验室在没有超大规模企业参与的情况下获得投资级融资。Jordan，这个问题给你。你知道，有一种批评，类似于 Edotron 对这个行业的批评，认为 Nvidia 基本上是在为所有这些 Neocloud 的需求提供资金，如果没有这种支持，整个事情基本上会消失，让 Nvidia 看起来小一些，让这些 Neocloud 变得不可行。在深入研究了这个问题之后，你怎么看？

Um, there's a lot of areas we can go in this uh Jordan and Dylan, but let's begin with the role of Nvidia here. Um the report you know though it spends most of the time on the technical side of things does go into the fact that uh Nvidia is playing a ma major role in keeping these clustries providing uh infrastructure AI infrastructure maybe to some of the frontier labs but really across the board uh and making this whole thing work and Nvidia has been backstopping uh a lot of their um their businesses with uh as you write um revenue floors landlord guarantees and leases that it plans to transfer to third parties helping the Neo Labs and AI labs in obtaining investment grade financing without a hyperscaler being involved. Um Jordan to you on this one. Uh you know there's the the criticism sort of like the Edotron criticism of this uh of this industry is that Nvidia is basically funding uh the demand for all these Neoclouds and without that uh backs stop um you know this whole thing would basically evaporate and make Nvidia look a bit smaller and make these neoclouds um you know non nonviable. What's your thought on that after going so deep into this?

嗯，是的，我的意思是，在 Cluster Max 和我们最近另一篇名为《Nvidia 的支持宇宙》的文章中，我们深入探讨了支持是什么以及它们如何帮助 Neocloud 筹集部署 GPU 所需的资金，而客户目前对这些 GPU 的租赁需求非常强劲。我们目前追踪到 Nvidia 正在支持的价值 5880 亿美元的表外支持，这是我们在本播客中列出的又一个令人难以置信的数字。Nvidia 的收入正接近 4000 亿，即将突破 5000 亿。所以我们谈论的是非常大的数字。作为参考，这大约相当于美国 M2 货币供应量的 2.5%。显然，在这个我们称之为“宇宙”的领域中有巨额资金在流动。我只能说，这对 Nvidia 来说似乎非常务实。他们真的希望客户多样化，不想依赖超大规模企业来获得全部收入。他们也不希望客户依赖从三四家超大规模企业那里获得芯片。因此，他们有充分的动力建立一个健康的市场，提供多样化的选择和大量选项供所有客户合作。为了实现这一点，人们面临的问题不仅仅是筹集购买 GPU 所需的资金。首先，他们需要保证，如果他们筹集资金，债权人将看到回报。这需要某种投资级的承购方，Nvidia 在许多情况下可以充当这个角色。但到目前为止，运作方式是 Nvidia 作为承购方签约，然后当真正需要为自己内部研究工作购买这些 GPU 时，会有其他人以更高的价格更想要它们。于是他们卖给那些人，而不是用于内部研究，然后继续下一个。对于那些说 Nvidia 这样做是出于某种邪恶目的或维持泡沫的人，我建议你看看 Nvidia 研究的质量。他们在 Hugging Face 上发布的模型比任何人都多。他们在 GPU 支持的软件套件中拥有令人难以置信的 CI，并且训练像 Neotron 这样的模型。我们不是这些模型或软件的最大粉丝，我们认为应该有一个多样化的生态系统，但认为 Nvidia 不使用自己的 GPU 进行服务于公司目的的研究本身就是错误的。然后我们看到的动态是，这些计算量实际上并不是 Nvidia 收入的 meaningful 部分，只占 Nvidia 总收入的很小一部分，用于他们自己的内部研究工作负载。所以我们看到的是，他们只是在支持 Neocloud 的建设，以满足客户目前极其强劲且难以满足的需求。也许最后要说的是，不仅仅是芯片。我们追踪了他们支持的广泛领域。这包括建设成本、内存供应商、建设晶圆厂的人、建设工业设备的人。不仅仅是“我需要帮助为我的 GPU 筹集资金”，而是“我需要帮助为供应链中的所有其他事情筹集资金”，而这正是特定项目的关键路径。

Uh yeah, well I mean both in Clustermax and in another article we did recently called Nvidia's backs stop universe, we go into a little bit of the dynamics of what backs stops are and how they help the Neoclouds raise the money they need to deploy the GPUs which then customers are renting pretty strongly at this point. And uh the number that we have is that currently we've got tracking of $588 billion of offbalance sheet backs stops that Nvidia is supporting right now, which is one of yet another unbelievable number in the list that we've been going through on this podcast so far. Um, I mean, but Nvidia's revenue is approaching 400 about to cross 500 billion. So we're talking about really big numbers there. For context, that's like 2.5% of the M2 money supply in the US. So obviously there's massive massive amounts of money going on uh in this um universe as we refer to it and uh I mean the only thing I can say about this is that it just seems really pragmatic from Nvidia. They really uh want a wide diversity of customers. They want don't want to be dependent on the hyperscalers for all of their revenue. Um they also don't want their customers to be dependent on getting their chips from a select set of three or four different hyperscalers. And so they have every incentive to have a a nice healthy market with a diversity of options and uh lots of selection for all of the customers to go uh work with. And in order to deliver on that, people have more problems than just uh raising the money required to buy the GPUs. Well, first of all, they need the guarantee that if they raise the money, those creditors are going to see returns. And that requires you know some sort of investment grade offtaker which Nvidia can serve in these cases in many in many of these cases. But the way it's worked so far is that Nvidia signs up as that offtaker and then when it comes time to actually you know take those GPUs for themselves for their own internal research work. Somebody else comes along and wants them more at a higher price. And so then they sell them to their to those guys. They don't take them for their internal research and they roll on to the next one. And for everybody who says that Nvidia, you know, is doing this for some nefarious purpose or to to keep some bubble going, I'd encourage you to look at the quality of Nvidia's research. They publish more models on Hugging Face than anybody else. They have incredible CI across the suite of software that they they support on their GPUs and they train models like Neotron. We're not the biggest fan of these models or this software. you know we think there should be a diverse ecosystem but the idea that Nvidia does not use their own GPUs for research that serves the purposes of their company is you know itself wrong and then the dynamic that we've seen is that this amount of compute is actually not a meaningful portion of Nvidia's revenue it's it's a very small percentage of Nvidia's total revenue that is you know going back for their own internal research workloads so what we're seeing is just that they are supporting Neocloud's build out the capacity for the demand from the customers which is so incredibly strong right now and hard to serve. Um, and then maybe the the last thing to say about this is that it's not just the chips. Like we we we track a whole wide range of things that they're backstopping. This is everything from like, you know, the construction costs, the memory suppliers, the people building fabs, the people building industrial equipment. I mean, it's not just like I need help raising money for my GPUs. is I need help raising money for all of the other things in the supply chain which is just the long pole in the tent for that specific project.

是的。让我们谈谈安全问题，因为这份报告的第一张图片是 Ilia Sutskever 在 Hugging Face 事件后发布的一条推文的引用。Ilia 说 Neocloud 的网络安全有限。下次智能体成功失控时，他们会试图接管一个 Neocloud 来运行更多副本。这很糟糕。因此，Neocloud 应该大大加强网络安全，每个拥有强大网络模型的公司都应该帮助做到这一点。这是 Ilia 的一条罕见推文。Dylan，你怎么看？这听起来像是如果他自己说出来，记住他自己也在构建这些东西。我想说超级智能，我们可能有问题了。

Yep. Let's talk about security because um you know the first image in this report uh is a quote tweet of an Ilia subsever post which was made right after uh the hugging face incident. Uh Ilia says neoclouds have limited cyber security. Next time agents successfully go rogue, uh they'll try taking over a Neocloud to run more copies. This is bad. Thus, Neocloud should greatly strengthen their cyber security and every company with strong cyber models uh should help with that. This is a rare tweet uh from Ilia. Dylan, what do you make of this? Uh is this sounds to me like if he's out saying this, right, remember he's building this stuff himself. Uh I'd say super intelligence that we might have a problem here.

是的。所以我们发表了一篇题为《大多数 Neocloud 在安全方面很糟糕》的文章。

Yeah. So, we we posted an article titled most Neoclouds suck at security.

顺便说一句，那个插图太棒了。我们会把它放在屏幕上，给那些正在观看的人看，这家伙被所有人捅刀子，而且它就像符合 SOC2 标准一样。

By the way, the art for that is incredible. We'll put it on the screen for those who are watching this guy getting stabbed by everybody and it's like it's SOC2 compliant.

是的，没错。

Yeah, exactly.

所以我们发布了那份报告，然后大约八小时后，Ilia 发了他的推文。因为他们非常使用各种云。他们显然超级神秘，非常关心他们正在做的事情的保密性，但他们也是超级智能的忠实信徒。我猜是安全超级智能，SSI。他们是第一个——你知道，特朗普实际上抄袭了 Ilia。Ilia 是第一个提出另一个的，对吧？第一个提出推理，第一个提出 LLM 缩放，现在第一个提出超级智能品牌。但 Ilia 在推特上发布了这个，它呼应了我们写的内容，对吧？就像，看，这些云大多数在安全方面很糟糕。你可以让开源模型直接黑进 Neoclouds 或访问其他 Neoclouds 内的东西。GLM 5.3 尤其擅长这个。所以，开源中国模型可以直接黑进现有的几十个 Neoclouds，你可以去看看其他人在做什么。所以我们在其他云上也见过类似的事情。有一个云，我们甚至没有打算，但只是意外地最终看到了那个集群上的所有其他客户，看到他们存储中有什么东西，看到他们在 Slurm 中有什么东西，等等。而且还有像情报机构，对吧？就像在这个上。所以我们立即向他们报告了。我们说，哟——

So we posted that report and then maybe eight hours later or whatever, Ilia posted his tweet. Because they are very much using various clouds. They're obviously super secretive and care a lot about the secrecy of what they're working on, but also they are big believers in superintelligence. I guess safe superintelligence, SSI. They were the first—you know, Trump actually copied Ilia. Ilia was first with another one, right? First to reasoning, first to LLM scaling, and now first to superintelligence branding. But Ilia sort of tweeted that out, and it sort of echoed what we wrote, right? Which is like, look, most of these clouds are terrible at security. You can get open-source models to just hack Neoclouds or get access to stuff within other Neoclouds. GLM 5.3 especially is really good at this. So, open source Chinese models can just hack dozens of the Neoclouds that exist out there and you can go look at what other people are doing. And so we've seen stuff like this at other clouds. There was a cloud that we didn't even intend to, but we just accidentally ended up seeing all of the other clients on that cluster, seeing what stuff they had in storage, see what stuff they had in Slurm, etc. And there were like intelligence agencies, right? Like on this. So we immediately reported it to them. We're like, yo—

哪些情报机构？

Which intelligence agencies?

我们不能说。我认为那不是合适的说法。国家像联邦安全机构政府——

We cannot say. I don't think that's a valid thing to say. National like federal well security agencies governments—

他们不是美国人？

They were not American?

他们不是美国人，但我们说这是为一个 GDP 排名前 10 的国家，不是像某个小国。

They were not American but we said it was for a country that was in a has a top 10 GDP it's not like some tiny country.

我在这里要疯了，那居然可能。

I'm losing my mind here that that was possible.

是的，就像如果我们有恶意，顺便说一句，可能还有十几个其他客户，也许其中一个是恶意的，我们本可以就像看到他们在训练什么，对吧，他们在为自己训练模型和运行模型，大概是因为这是 GPU 基础设施。所以你会用一堆 Blackwell GPU 做什么别的？

Yeah and it's like if we were malicious and by the way there were probably a dozen other customers and maybe one of them was malicious we could have just like seen what they were training right and they were training models and running models for their themselves presumably because this was a GPU infrastructure. So what else would you do with a bunch of Blackwell GPUs?

是的，我们经常提到主权 AI 这个词，它多次与新云的概念纠缠在一起，人们甚至在融资时也经常使用这个词，因为他们将算力视为基础设施。我认为这完全正确。这就像电信建设一样，部署和参与其中的主权国家将这种基础设施视为国家想要控制和拥有主权的资产，他们相信，就像你在节目开始时所说的，他们想成为净代币进口国或出口国，而这由他们建设的基础设施驱动，这意味着内部装有 GPU 的数据中心，所以我认为联邦政府研究和情报机构在一些这样的集群上与他人共享基础设施并不奇怪。你希望他们有一个气隙数据中心，但如果整个国家只有一个设施拥有 GB300、NVL72，我的意思是，这是一个卖方市场。

Yeah, we throw around this word sovereign AI and it gets entangled with the concept of a neocloud many times and people even when they're raising money, they use this word a lot because they view compute as infrastructure. And I think that's exactly right. This is just like the telecommunications buildout where this infrastructure is viewed by the sovereign nations that are deploying it and are involved in this as an asset that the country would like to control and have sovereignty over and they believe that similar to what you're saying at the beginning of the show they want to be a net token importer or exporter and that is driven by the infrastructure they build which means data centers with GPUs inside of it and so I think it's actually no surprise that federal government research and intelligence agencies are on some of these clusters sharing infrastructure with others. You would like them to have an air gap data center, but if there's exactly one facility in the entire country with GB300, NVL72, I mean, it's a sellers market.

天哪。好吧。今晚我睡不着了。所以让我们——你知道，在这份报告中，你还将许多超大规模云厂商和新云分层，顶级是 CoreWeave 和 Nebius。我更感兴趣的是第二层，即 Oracle 和 Google Cloud，主要是 Jordan，因为你知道，在我为今天的对话准备的文档中，我有很多从报告中提取的主题和句子以及问题，然后对于 Oracle，我只写了 Oracle 和一堆问号，因为我对那家公司就是这种感觉。我不知道该怎么看待他们。显然，他们的债务被评级机构评得不好。他们正在做这个全押赌注。他们的评级相当高，而且在你的报告中，他们被归功于开创了一些技术。但我很想听听你对 Oracle 的解读，以及这对他们来说是否是一个明智的赌注，以及他们是否会荣耀崛起还是像很多人预期的那样灰飞烟灭。

Goodness. All right. Well, there goes my sleep tonight. So let's let you know you do in this report also rank a lot of these hyperscalers into tiers and the neoclouds into tiers and at the top tier is CoreWeave and Nebius. I'm more interested in the second tier and that is Oracle and Google Cloud mostly Jordan because you know I have like in the document that I put together for our conversation today I have like a lot of like topics and sentences that I've like pulled out of the report with questions and then with Oracle, I just have Oracle and then a bunch of question marks because that's kind of how I feel about that company. I don't know how to feel about them there. Obviously, their debt has been rated not well by the rating agencies. They're making this all in bet. They rate pretty high and they actually in your report have been credited with pioneering some techniques. But I would love to get your readout on Oracle and you know whether this has been a smart bet for them and whether they're going to rise in glory or sort of fall in flames like a lot of people anticipate.

我认为 ClusterMAX 的关键在于它不是股票推荐，对吧？它与嘿，这个新云明年要建一个吉瓦或 200 兆瓦无关，这与股票更相关，而不是你知道他们运营集群的能力有多好，对吧？与嘿，他们说他们明年要交付的这个数据中心容量被延迟了 3 个月无关，那不是 ClusterMAX 衡量的。ClusterMAX 明确衡量的是，如果你租用 GPU，这个集群对某人使用来说有多好。它有多可靠？有多安全？存储怎么样？网络怎么样？生命周期怎么样？Kubernetes 上的 Slurm 怎么样？所有这些事情怎么样？在这种情况下，我们只是试图更客观。显然，制作一个评级金字塔就像，你知道，这是你必须要做的事情。如果你能只做一个长报告，没有人关心是否有排名，那就好了。我的意思是，那会发生。所以，你必须要做排名。并设定某些界限，你总是可以就为什么以及什么在哪个层级进行无休止的争论。但就像我们认为 Oracle 在服务质量方面是顶级新云之一，这是一个客观事实。另一方面，这并不意味着嘿，在 SemiAnalysis 四月份，我们向客户发布，他们的新墨西哥数据中心将延迟，因为他们不会得到管道。这对 OpenAI 来说是一个大问题，因为 OpenAI 与他们在新墨西哥签订了一个巨大的数据中心合同。但如果这些管道不建，这个数据中心就无法上线。

I think the key thing about ClusterMAX is it is not stock recommendations, right? It is not anything to do with hey this neocloud is going to build a gigawatt next year or 200 megawatts which is far more relevant for the stock than you know how good are they at operating a cluster right has nothing to do with hey this data center that they said they're going to deliver with this much capacity next year got delayed by 3 months that's not what ClusterMAX measures. ClusterMAX is explicitly measuring sort of how good is the cluster for someone to use if you rent the GPUs. How reliable is it? How secure is it? How's the storage? How's the network? How's the life cycle? How's the Slurm on Kubernetes? How are all these things? In which case, it's just sort of we try to be more objective. Obviously, making a rating pyramid is like, you know, it is something you kind of have to do. It would be nice if you could just make a long report and no one cared if there were no rankings. I mean, that's what would happen. So, you kind of have to do the rankings. And set certain boundaries and you can always litigate the hell out of why and what is in what tier. But like it is an objective fact that we think Oracle is one of the top Neoclouds in terms of quality of service. Now that doesn't mean on the other hand, hey at SemiAnalysis in April, we published to our clients that their New Mexico data center was going to be delayed because they weren't going to get the pipelines. And this was a big issue for OpenAI because OpenAI has contracted a huge data center with them in New Mexico. But if these pipelines don't get built, this data center can't go up.

然后后来我们在 5 月和 6 月又说了，最近 Oracle 宣布了强制措施，股价因此大跌。所以我们在区分两件事，对吧？Oracle 是一家很棒的云，管理他们的托管集群，是顶级 Neocloud 之一，但与此同时他们也可能有执行问题，比如这个新墨西哥州数据中心的监管问题，他们无法获得管道，或者资产负债表上的其他问题。所以这种区分——什么对股票有利或不利，与我们要测试的是什么。第一个 ClusterMAX 的目标是创造一些租用 GPU 的人可以使用的东西。现在事实证明有很多其他用途，人们试图强行塞进去，比如，嘿，我们要为这个 NeoCloud 写债务。如果他们从铜牌升级到银牌，他们的利率实际上会下降。或者，嘿，我要投资 ClusterMAX 顶部的股票。显然这些不是决策——这些不是 ClusterMAX 设计的目的。这是你在与这些公司打交道时应该使用的数据点。但就 Oracle 作为一家公司而言，有两件事，对吧？一是他们与 OpenAI 签署了大量 AI 数据中心协议。是去年吗？是的，是去年。天哪，才去年。感觉像很久以前。我想可能正好是一年前。但有趣的是，他们签署了这些巨额协议，其中嵌入了许多风险。一是他们基本上以固定利润率签署了这些协议。所以现在 GPU 价格对云来说上涨了很多，他们无法利用这种更高的利润率。所以他们有点被困在那里。第二，他们筹集了大量债务，然后市场开始担心公司能有多少债务，所以他们被吓到了，尽管 Oracle 完全有偿付能力。第三，他们有一些执行问题。我提到了监管问题。还有其他一些数据中心也有延迟等，但主要是这个新墨西哥州站点，由于管道问题可能延迟多年。所以这些事情可能对股票产生非常负面的影响，但我认为公司仍然没事。现在，他们不是 Nebius。Nebius 可以转身，他们拥有 ClusterMAX 中顶级云的最短平均合同长度。所以 Nebius 能够转身并真正提高定价并利用这一点，对吧？而 CoreWeave 和 Oracle 尽管也有良好的集群管理和可靠性、安全性等，但无法利用那么多，因为他们主要将自己锁定在长期合同中。

And then later we said it again in May and June, and then most recently Oracle announced a force measure on it and their stock tanked because of it. So we're disentangling two things, right? Oracle is an amazing cloud, managed for their managed cluster, one of the top Neoclouds, but they also at the same time may have execution issues, whether it's regulatory in the case of this New Mexico data center where they can't get the pipeline there, or other issues on their balance sheet. And so that disentangling—like what is good or bad for the stock versus what are we trying to test here. The goal in first ClusterMAX was to create something that people renting GPUs can use. And it turns out now there's a lot of other uses for it that people try to hamfist in, whether it's like, hey we're going to write debt for this NeoCloud. If they upgrade rankings from bronze to silver, their rate will actually go down. Or like, hey I'm going to invest in stocks that are at the top of ClusterMAX. So obviously these are not the decisions—these are not what ClusterMAX is designed for. It's a data point that you should use if you're doing anything with these companies. But as far as Oracle as a company, there are two things to go here, right? One is they signed a ton of AI data center deals with OpenAI. Was it last year? Yeah, it was last year. Holy crap, it was only last year. Feels like eons ago. I think it might have been exactly a year ago. But the interesting thing is they signed these massive deals, and there are a number of risks embedded with them. One is they signed them with a fixed margin essentially. So now that the price of GPUs has gone up a lot for clouds, they can't take advantage of this higher margin. So they're kind of hammed in there. Number two, they raised a bunch of debt and then the market started getting freaked out about how much debt companies can have, and so they spooked out even though Oracle is completely solvent. And then number three, they have some execution issues. I mentioned the regulatory one. There are some other data centers that have some delays and such there as well, but the main one being this New Mexico site that is delayed potentially multiple years because of this pipeline issue. And so these things can have very negative impacts on the stock, but I think the company is still fine. Now, they're not Nebius. Nebius gets to turn around and they have the shortest average contract length of the top clouds in ClusterMAX. And so Nebius has been able to turn around and really jack up pricing and take advantage of this, right? Whereas CoreWeave and Oracle, despite having also good cluster management and reliability and security and all this stuff, were not able to take as much advantage because they locked themselves into long-term contracts primarily.

哦，那太迷人了。好吧，Jordan，让我问你一个技术问题。正如 Dylan 所说，这是一个金字塔，有这些顶级组，比如我们谈到的 CoreWeave 和 Nebius、Oracle、Google Cloud 都在上面，但在不推荐部分有很多标志，有表现不佳和不可用的，我想不可用部分的标志比前四层加起来还多。你认为这说明什么，我的意思是，你认为这说明这项技术什么，有这么多在外面，只是没有达到你的标准？

Oh that is fascinating. All right Jordan let me ask you a technology question then. As Dylan put out there it is a pyramid and there's like these top tier groups like we talked about CoreWeave and Nebius Oracle Google Cloud is up there but in the not recommended section there's a lot of logos there's underperforming and unavailable I think there's more something like more logos in the unavailable than top four tiers combined What do you think that says about the fact that I mean, what do you think that says about this technology that there's so many that are out there that just aren't meeting the bar for you?

看，作为试图代表买家利益的人，我认为这让我们感到难过。很难过看到几乎一年半，你知道，自 ClusterMAX 第一版发布以来实际上接近两年了，自从我们在 Semi Analysis 创造了 neocloud 这个词以来，很多人就是没有在我们所写的事情上取得进展，而我们写的事情正是我们与这些 neocloud 的客户谈论的内容。所以我认为这只是卖方市场的动态，意味着这些公司的优先事项是如何尽可能多地筹集资金，如何尽可能快地建立数据中心，以及如何尽可能快地部署芯片，而不是致力于底层服务的真正质量。这就是你看到的一点动态，人们可以拥有这些伟大的企业，而不必太担心提供客户真正喜欢并推荐给其他人的优质服务。另一件事是，我认为有一种持续的动力，我们将随着时间的推移继续看到，那就是只能有这么多领导者，因为只有这么多真正创新的人，然后其他人复制他们。当你有一个市场，根据我们的跟踪，截至今天早上实际上有 323 家提供商，我们在我们的 AI 云 TCO 模型中跟踪，对于 Semi Analysis 的订阅者他们有这个完整列表，只有一定数量的人是相关的，比如足够大可以放在列表上，然后其中一定数量的人可以通过我们认为需要看到的标准来推荐他们。你知道，在某些方面，我们认为被推荐的标准相当低。比如在安全性和可靠性方面失败的基本事情，这两个是最大的，在我们看来非常简单，只是难过看到其他人没有足够关注或关心实际实施这些东西。

Look, as people who are trying to represent the interests of buyers, I think it just makes us sad. It's sad to see that like a year and a half almost, you know, since the first version of ClusterMAX was published actually approaching 2 years now, and since we coined the term neocloud at Semi Analysis, that a lot of people have just not made progress towards the things that we were writing about and the things we were writing about is just exactly what we talk to the customers of these neoclouds about. And so I think it's just like the dynamic of it being a sellers market means that the priority for these companies is how can they raise as much money as they can, how can they build data centers as fast as they can, and how can they deploy chips as fast as they can rather than working on the true quality of the underlying service. And that's a little bit of the dynamic that you see represented there where people can have these great businesses without actually worrying too much about delivering a great service that their customers really enjoy and recommend to other people. The other thing is that I think there's like this constant dynamic that we'll continue to see over time which is that there can only be so many leaders because there's only so many people who truly innovate and then everybody else copies them. And when you have a market that by our tracking is actually 323 providers as of this morning, which we track in our AI cloud TCO model for people who are subscribers to Semi Analysis they have this full list, there's only a certain amount of them that are even relevant, like large enough to put on the list and then a certain number of those that can pass the bar for what we believe is what we need to see to recommend them. And you know, in some ways, we think that the bar for being recommended is quite low. Like the basic things that fail with respect to security and reliability being the two top ones are just in our view dead simple and it's just sad to see that others aren't paying attention enough or caring enough to actually implement this stuff.

好的，在我们走之前两个快速问题。只是一个更广泛的问题。你知道 Google 显然他们宣布他们有这个 Gemini Argon 即将推出。Gemini for Argon。你的观点是，你知道有些人呼吁 Google 就基本上成为一个 NeoCloud，出租 Google Cloud 的 TPU 和计算。你认为他们会朝那个方向走吗，还是你认为他们仍然会将很大比例的计算用于模型训练和推理？

Okay, two quick ones before we go. Just more broad question. You know Google obviously they've announced they have this Gemini Argon coming out. Gemini for Argon. Is it your view that that you know there have been some people who have called upon Google to just be, you know, basically a NeoCloud and rent out the TPUs from Google Cloud and the computing. Do you see them going that way or do you think that they're still going to use like a large percentage of their compute for model training and inference?

我的意思是，从定义上讲，他们在过去 6 个月分配给模型训练和部署的计算比两三年前少。而且这个百分比分配随着时间的推移而下降。这是我们开玩笑的事情之一，也许甚至不是玩笑，而只是陈述 Thomas Curran 赢了而 Dennis 输了，对吧？现在我想是 Karai。Google 正在生产显著更多的计算。

I mean definitionally they have allocated less compute to model training and deployment in the last 6 months than they had two or three years ago. And that percentage allocation has fallen over time. This is sort of one of the things that we made a joke about, like maybe not even joke but just a statement that Thomas Curran won and Dennis lost right? Now I guess Karai. Google is producing significantly more compute.

逐年来看，投入 DeepMind 的算力增长并没有那么快。并不是说 DeepMind 曾经拥有谷歌三分之一的算力，现在仍然拥有三分之一。实际上这个比例在下降，这非常有意思，因为现在如果我只比较 DeepMind 与 Anthropic 和 OpenAI，到今年年底，DeepMind 在三者中算力最少。所以这确实是个大问题。

Year on year, the amount that's going to DeepMind is not rising as fast. It's not like DeepMind had a third of Google's compute and now it still has a third. It's actually falling as a percentage, which is really interesting because now if I look at just DeepMind versus Anthropic and OpenAI, DeepMind has the least compute of the three by the end of this year. So it's a really big deal.

那么谷歌，不管有没有新云，新云这个词到底是什么意思——卖其他模型的 token 算不算新云？从某种意义上说，新云业务就是卖 AI 模型的 token，对吧？所以也许那就算新云。

And so Google, whether or not there's a neocloud, what the word neocloud even means—does selling tokens of other models count as neocloud? In a sense, that's what the new cloud business is: selling AI model tokens, right? So maybe that's a neocloud.

黑石去搞新建设项目，建你的 GPU，直接部署给核心平台之外的客户——或者卖芯片，对吧？那算新云吗？

Blackstone and going and doing new construction projects to build your GPUs to deploy directly for customers outside of your core platform—or selling chips, right? Is that neocloud?

不算。我是说，他们向 Anthropic 卖了很多芯片。所以是物理上卖芯片，而不是仅仅出租。所以我会说谷歌的商业模式在演变，他们想从技术栈的每一层赚钱。但按照定义，用 TPU 这样大力帮助 Anthropic，用其他东西这样帮助他们，是在损害 Google DeepMind 的业务。

No. I mean, they're selling a lot of chips to Anthropic. So it's physically selling the chips rather than just renting them out. So I would say Google's business model is evolving and they want to make money at every layer of the stack. But by definition, helping Anthropic this much with TPUs and helping them this much with these other things is hurting Google DeepMind's business.

我觉得他们得找到自己的 niche，老兄。就拿 Gemini 来说，没人想要第三、第四或第五好的编程模型。所以如果他们想让 DeepMind 继续在 AI 研究前沿竞争，就得找到一个能推动 Gemini token 销售的 niche。因为现在，即使发布了 Gemini 4 Argon，在编程方面跟排名前二三的模型比，还是不够看。

I think they need to find a niche, man. Like for Gemini, I mean, nobody wants the third or fourth or fifth best coding model. So they need to find a niche that drives Gemini token sales if they want DeepMind to continue competing on the frontier of AI research. Because right now, even with this Gemini 4 Argon release, it's just not there on coding compared to the top two or three.

这里的风险是，明年当 OpenAI 和 Anthropic 都试图各自为其实验室增加 5 到 10 吉瓦的额外算力，而谷歌只增加几吉瓦时，会发生什么。现在他们的算力赤字——谷歌曾经在算力上遥遥领先，而他们有点落后或持平，主要是稍微落后一点。现在他们落后更多了，但算力还是那么多。那么当他们算力更少时会发生什么？他们能神奇地追上吗？有可能，对吧？他们有才华，有人才，他们需要重组并真正集中精力。但同样，越来越多最优秀的人不断被挖到 Anthropic、OpenAI，还有 Meta。

And the risk here is what happens next year when OpenAI and Anthropic both attempt to add 5 to 10 gigawatts of additional compute for their labs each, and then Google only adds a few gigawatts. Now their compute deficit—Google was way ahead in compute and they were kind of behind slash equivalent, mostly just a little bit behind. And now they've fallen further behind, but their compute is the same. So what happens when they have less compute? Do they somehow magically catch up? It's possible, right? They have the talent, they have the people, they need to reorganize and really focus up. But also, more and more of the best people just keep getting poached to Anthropic, to OpenAI, and it's Meta.

发现循环。

Discovery loop.

是的。

Yeah.

发现循环。

Discovery loop.

发现循环。没错。

Discovery loop. Right.

好的。走之前我想再问一个问题，可以吗？我想 Dylan 你过去说过，从创建的 AI 模型中获得大部分价值的人是最终用户，对吧？比如在底层使用云的人。当然，Anthropic 通过 API 提供那个模型，并进入各种用途。它将不得不做更多的许可，谁知道他们能否为所有这些做产品。我很好奇你是否认为，鉴于我们过去谈到的那些大数字，或者他们更长时间地把前沿模型留在家里，久得多，并用它们来构建自己的产品。就像我们知道他们有湿实验室，我们知道他们有 Claude Code 和 Co-work，然后他们不开放这些，因为他们想从他们购买和建设的基础设施中获得更多利润。

Okay. I want to ask one more before we go if that's okay. I think Dylan you've said in the past that the people who are getting most of the value out of the AI models that are created are the end users, right? The people that are using the clouds for instance at the base. And of course, Anthropic makes that model available via API and to go into all these various uses. It's going to have to do more licensing and who knows if they'll be able to do products for all this. I'm curious if you think if you see a scenario where given the big numbers that we talked about in the past or they keep their frontier models at home for longer, much longer, and use them to build their own products. It's like we know they have the wet lab, we know they have Claude Code and Co-work and then they don't make them available because they want to get more margin out of the infrastructure that they've bought and built.

是的。所以，我的意思是，我认为我们已经看到了这一点，对吧？Anthropic 在二月份就准备好了 Mythos。他们的模型安全卡指出，自 Mythos 以来他们还有两个其他模型已经发布，而公开的我们仍然没有完整的 Mythos。我们有一个被阉割的版本叫 Fable。而且 Fable 甚至花了很多很多个月才出来。他们有 Glass Wing，就像 Mythos 但只针对某些子行业，在某些领域被阉割了，不是所有领域。所以很明显 Anthropic 没有发布新模型。OpenAI 同样，Astra 实际上几个月前就完成了，它才出来，你知道，他们花了几个月才发布 Astra。还有新模型 Bell——他们甚至展示了一个博客，展示了它在数学上的表现以及如何解决所有这些东西，但他们没有发布它。所以很明显他们只是在内部使用，不对外。所以实验室已经沿着这条线移动了，就像“哦是的，他们和中国开源之间的差距仍然是 6 个月”，他们有点保持这一点，但实际上他们只把几个月的进展留在内部，因为一旦你把它外部化，人们就可以蒸馏它等等。所以我认为 Anthropic 可能会在未来几周为他们的 IPO 发布一个新模型。我认为 OpenAI 也可能不得不在未来几周发布一个新模型，因为竞争环境太激烈了。但与此同时，他们很多内容只留在内部。

Yeah. So, I mean I think we've already seen this, right? Anthropic had Mythos ready in February. Their model safety cards noted that they have two other models since Mythos that have released and publicly we still don't have full Mythos. We have a neutered version of it called Fable. And it took many many months before Fable even came out. They had the Glass Wing which was like Mythos but for only some subsectors and it was neutered in some areas, not all areas. And so it's clear that Anthropic has not released new models. OpenAI likewise, Astra was actually done a couple months ago and it only came out like, you know, it took a couple months before they released Astra. And the new model Bell—they even showed a blog where they showed the performance of it in math and how it solved all this stuff and they have not released it. So clearly they are just using it internally not externally. And so the labs have already moved along this line right where it's like oh yeah the gap between them and Chinese open source is still 6 months, they're kind of keeping that but in reality they've kept months of progress internal only because the moment you make it external people can distill it and all these other things. So I think Anthropic will likely release a new model in the coming weeks for their IPO. I think OpenAI will likely have to release a new model in the coming weeks as well just because it's such a competitive environment. But at the same time they are like keeping a lot of their stuff internal only.

是的，我们内部讨论了一些资本主义激励。如果激励真的会促使人们发布更多模型并让消费者使用，因为按 token 的 API 定价有惊人的毛利率，或者激励是将其保留在内部，以便你可以追求最终用户应用，这些应用比按 token 出售模型访问权更有价值。看看如何发展会很有趣。但在其中一种情景中，资本主义的做法将是为自己控制前沿节奏，对吧？

Yeah, we were discussing internally some of the capitalistic incentives. If the incentives are actually going to motivate people to release more models and get them in the hands of consumers because of the incredible gross margins on the per token API pricing or if the incentive is to keep it internally so that you can pursue the end user applications which are even more high value than selling the access to the model per token. It's going to be interesting to see how it plays out. But in one of those scenarios, the capitalist approach will be to pace the frontier for yourself, right?

那你直觉上觉得会走向哪里？你认为哪种情景会上演？

And which what's your gut on where it's going to go? Which one of those scenarios you think plays out?

我认为 Anthropic 公开表示过，他们通过限制自己来控制前沿节奏。而这是他们的博客文章在现实中体现的唯一方式。没有其他人在减速。他们只是在减慢自己。我认为那是真的。我认为这就是正在发生的事情。而且我认为，通过解决像癌症这样毫无疑问是好的、每个人都会同意是好事的问题来提升 AI 地位的激励，将是一种积极的激励。然后我认为如果你做那样的事情，也有很多钱可赚。所以我打赌我们会继续看到 OpenAI 和 Anthropic 都发布最小可行模型，以保持对中国和其他开源生态系统的领先。

I think Anthropic has publicly stated that they are pacing the frontier by keeping themselves restricted. And that's the only way that their blog posts have manifest in reality. Nobody else is slowing down. They're only slowing themselves down. I think that's true. I think that's what's happening. And I think the incentives to both improve the status of AI by solving things that are, you know, like cancer that are just unequivocally good that everybody can agree is a good thing would be, you know, a positive incentive. And then I think there's also plenty of money to be made if you do something like that. So my bet is that we continue to see the minimum viable models released by both OpenAI and Anthropic to keep a lead over Chinese and the rest of the open source ecosystem.

网站 semianalysis.com——就像 Dylan 说的，去订阅博客，像我一样订阅那份通讯。那里有丰富的洞见和分析，如果你还没订阅，我想你会喜欢的。而这场对话正是它的体现。那么，Dylan，很高兴再次见到你。Jordan，很高兴认识你。欢迎你们随时再来，也感谢你们今天抽时间。

The website semianalysis.com — sign up for the blog as Dylan says, sign up for the newsletter as I have. It's a wealth of insight and analysis, and I think you'll love it if you're not there already. And this conversation has just been the manifestation of that. So, Dylan, great to see you again. Jordan, really nice to meet you. You guys are welcome anytime, and we appreciate your time today.

谢谢你邀请我们，Alex。

Thanks for having us, Alex.

谢谢，Alex。

Thanks, Alex.

好，很棒。感谢大家的收听和观看，我们下次在 Big Technology Podcast 再见。

All right, good stuff. Thanks everybody for listening and watching, and we'll see you next time on Big Technology Podcast.