第35集 - 技术尽职调查、业绩预测、基准测试、供应链、投资论点（咨询）| Abhilash Jain, Jordan Nanos

大家好，

Hello,

各位好。

everyone.

欢迎回到SemiAnalysis。

Welcome back to SemiAnalysis.

我是Jordan。

My name is Jordan.

今天和我一起的是Abhilash。

I'm here with Abhilash.

这周我们打算做点不一样的。

This week, we are gonna do something a little bit different.

我们不聊文章。

We're not talking about an article.

而是给大家简单介绍一下SemiAnalysis的咨询业务。

Instead, we're giving everybody a little bit of an explanation on the SemiAnalysis consulting business.

Abhilash负责我们的咨询业务。

Abhilash leads our consulting business.

他有一个很棒的团队。

He's got a great team.

很多人对我们的咨询工作有疑问，比如我们做什么、服务哪些客户、做哪些项目，以及我们学到了什么。

And a lot of people have some questions about like what we do in consulting, what types of clients we serve, what are the projects that we do, and things that we've learned.

这涵盖了从投资者或考虑并购活动的公司的尽职调查，到那些想超越核心产品、定制我们研究产品（如加速器模型、数据中心模型等）的投资者，以及为众多客户和行业提供战略指导、产品路线图反馈和市场推广活动。

This covers everything from like DDs for investors or companies that are looking at M and A activity to investors who wanna go beyond the core products and, you know, customize something about one of our research products, the Accelerator Model, Data Center Model, things like that, as well as strategy guidance, you know, feedback on product roadmaps, go to market activities for lots of clients and industries.

那么，Abhilash，欢迎来到节目。

So, Abhilash, welcome to the show.

今天很高兴你能来。

Excited to have you today.

谢谢你，Jordan。

Thank you, Jordan.

很高兴终于能在著名的SemiAnalysis Weekly节目中亮相。

Excited to finally make my debut to the renowned SemiAnalysis Weekly.

阿比拉什和他的团队在这个播客里经常发表很多观点，但这是我们第一次直接听他本人说。

Abhilash and his team have lots of opinions that get expressed on this podcast, but for the first time, we're gonna hear it from him directly.

所以我很期待。

So I'm excited for that.

哦，当然。

Oh, absolutely.

好的。

Okay.

给我们简单介绍一下吧。

Give us a quick overview.

SemiAnalysis的咨询业务是做什么的？

What's SemiAnalysis consulting business?

我们具体做什么？

What do we do?

我觉得……

I think...

顺便说一下，SemiAnalysis 从Dylan创办开始就一直在做咨询。

So so, by the way, SemiAnalysis has been doing consulting ever since, I think, Dylan started SemiAnalysis.

但我觉得是去年，我们才组建了一个专门的团队。

But it was only last year, I would say, when we set up a dedicated team.

原因是SemiAnalysis做的事情很多。

And the reason is SemiAnalysis does a lot of things.

我们发布通讯。

We we publish our newsletter.

我们有机构模型。

We have institutional models.

我觉得我们现在有差不多13到14个不同的机构模型。

I think we have close to 13 to 14 different institutional models right now.

但其中很多……

But a lot of this...

我们还有开放项目……

And we have open projects...

比如Infinex和PlasmaMax这样的开放研究项目。

Open research projects like Infinex and PlasmaMax.

所以有很多事情在发生，但其中很多其实都是现成的。

So there's a lot happening, but a lot of this stuff is actually very off the shelf.

所以如果任何行业参与者或投资者想更深入地与SemiAnalysis合作，我们内部其实没有一个运营模式来服务这些需求。

So if any industry participant or any investor wants to engage more deeply with SemiAnalysis, we didn't really have an operating model internally to be able to service those requests.

这就是SemiAnalysis咨询业务的起源。

So that's the genesis of SemiAnalysis Consulting.

我觉得我们做的事情就是与各种不同的客户进行非常专注的合作。

I think what we do is we work with various different clients on a very dedicated basis.

我们做的所有工作都是高度定制的。

All the work that we do is very custom.

我们通常会配备一个专门的团队，时间从三到四周到一年不等。

We typically staff a dedicated team for anywhere between three to four weeks to all the way from a year.

我很乐意多聊聊这个，但基本上这就是我们做的事情。

And I'm happy to talk more about it, but that's essentially what we do.

我们做的所有工作都是定制化的。

All the work that we do is custom.

太棒了。

Awesome.

定制化工作。

Custom work.

也许可以先聊聊我们服务的客户类型。

Maybe start by talking about the types of clients that we serve.

我之前大概介绍了一下我们做过的不同类型的项目。

I gave a little higher level overview of, like, the different types of projects that we've done.

对，那你带我们详细说说吧。

But, yeah, take us through that.

在我看来，我们正在见证一种叫做……的东西。

In my opinion, I think what we're witnessing is something called...

我们总是谈论硬件、软件核心设计。

Like, we we always talk about hardware, software core design.

我觉得在AI基础设施这个领域，还有另一件事也在发生，我称之为技术与金融的协同设计。

I think in the world of AI infra, there's something else that's also happening, which I call tech finance co design, I would say.

所以，我们需要让技术更懂金融。

So, we need to teach tech grows more finance.

我们也需要让金融更懂技术，这基本上就是我们在做的事情。

We need to teach more finance grows more tech, and that's essentially what we do.

我觉得我们服务的主要有两类客户群体。

I think there are two sort of client groups that we serve.

第一类是金融机构投资者，主要是大型私募股权基金。

The first one is the financial institutional investors, and these are large private equity funds.

还有大型对冲基金，他们都是非常精明的投资者，对投资AI基础设施非常感兴趣。

These are large hedge funds who are very savvy investors, and they are very excited about investing in AI infra.

现在，当他们准备向芯片公司、新云服务商或数据中心投入十亿、二十亿美元时，他们希望SemiAnalysis介入，帮他们理清交易的各个方面。

And now, when they're to drop $1,000,000,000, $2,000,000,000 in a chip company, in a neocloud, in a data center, they want SemiAnalysis to come in and essentially help them figure out various aspects of the deal.

所以我们为他们做各种尽职调查。

So we do various sorts of diligences for them.

我们还为对冲基金做很多投资主题研究。

We also do a lot of thesis development for hedge funds.

很多对冲基金都在寻找下一波浪潮。

A lot of hedge funds are looking for the next wave.

你知道，大家之前对内存很兴奋，对CPU也很兴奋。

You know, people got excited about memory, people got excited about CPU.

所以他们现在在想，好吧，接下来是什么？

So they're trying to think about, okay, what's next?

我们还和很多对冲基金合作，把各种不同的数据源和市场叙事结合起来，找出下一波浪潮是什么。

And we also worked with a lot of hedge funds in combining various different data sources and different market narrative to figure out what's the next wave.

另一方面，我们也服务很多行业客户，这些客户主要是超大规模云服务商、AI实验室、芯片公司和半导体公司。

On the other hand, we also serve a lot of industry clients, and these are, I would say, largely the hyperscalers, the AI labs, the chip companies, semis companies.

在这里，话题的范围可以非常非常广。

And here, the range of topics can can be really, really large.

从帮助半导体公司制定产品和工程路线图，到……

It could range from helping a semi company figuring out their product and engineering road maps.

所以一个常见的问题是，如果大语言模型在朝着某个方向发展，那么三四年、五年后我们应该考虑什么样的产品。

So a common question is, if, you know, LLMs are evolving in a certain way, what should be the products that we should be thinking about three years out, four years out, and five years out.

也可能是帮助某个超大规模云服务商在印度偏远地区选址建设数据中心。

It could also be helping certain hyperscaler figure out data center sites in remote remote locations in India, for example.

也可能是为那些产品很强但不知道怎么连接客户、理清客户旅程的公司制定市场推广策略。

It could also be doing go to market strategy for various different companies who have a strong product but don't know how to connect and figure out their customer journeys.

所以选项很多，但基本上这就是我们SemiAnalysis咨询服务的客户类型。

So it's a range of different options, but that's essentially the type of clients that we serve as part of SemiAnalysis Consulting.

有道理。

Makes sense.

顺便替Abhilash打个广告，如果你感兴趣的话。

And quick advertisement for Abhilash on his behalf, if you're interested.

如果你在听，并且对科技金融联合设计感兴趣，可以考虑申请加入SemiAnalysis的咨询团队。

If you're listening and you're interested to tech finance co design, you should consider applying to work on consulting at SemiAnalysis.

这份工作很有意思。

It's a lot of fun.

我们正在大量招人。

We are hiring a lot.

所以，厚着脸皮打个广告，如果你对AGI充满热情，对半导体感兴趣，又好奇想学习，请联系我们。

So, shameless plug, you're AGI filled, excited about semis, curious to learn, please hit us up.

我们急需人手。

We are desperately in need of people.

那么，给我讲讲这些定制化的例子吧。

So, take me through some of these customization examples.

我参与这些工作的地方在于，我们把我和其他团队成员做的研究打包成更容易理解的形式，并根据每个人的具体兴趣进行定制。

Like the place where I get involved in some of this work is that we're taking research that me and other guys on the team are doing and trying to package it up in a more digestible way as well as customize it for somebody's specific interest.

那么，给我讲讲这个过程是怎样的，我们如何利用现有的数据源来实际执行这些项目。

So take me through what that process looks like where we lean on our existing data sources to actually go and execute on these projects.

我觉得因为SemiAnalysis...

I think because SemiAnalysis...

在SemiAnalysis，我们有很多不同的事情在发生。

At SemiAnalysis, we have so many different things that are happening.

通常，如果你能把一些数据产品以一种更易懂的方式结合起来，我们就能回答那些单个数据产品无法回答的问题。

Usually, if you are able to just combine some of these data products in a more comprehensible manner, we are able to answer some of the questions that each of these data products are not able to answer by itself.

所以我觉得最近有个例子，一个对冲基金客户非常想知道，假设芯片安装基数一定，全球总的token供应量是多少。

So I think one of the examples was very recently, there was a hedge fund client that was very interested in understanding what's the total token supply in the world, assuming the chip installed base.

你想想看，这其实是个挺简单的等式。

And if you think about it, right, it's a fairly simple equation.

现在全球总的安装基数是多少？

What is the total installed base in the world today?

然后呢……

And what's...

未来五年预计会怎么增长？

How how is it expected to grow in the next five years?

其中有多少算力用于训练，多少用于推理？

How much of that capacity is allocated to training versus inference?

如果你有了安装基数，等式的第二部分就是每块GPU的吞吐量是多少。

If you have the installed base, the the second part of the equation is what is the throughput per GPU?

多少……

How much...

每块GPU在一天或一年内能生成多少token？

How many tokens can each GPU generate in a certain day, in a given year?

最后，这些token的价格是多少，按每百万token多少美元算？

And finally, what's the pricing in terms of dollar per million tokens for these tokens?

这些不同的……

Each of these diff...

三……

Three...

这个公式的三个部分各自用不同的方法解答，但最复杂的是估算每块GPU的吞吐量，因为我们必须结合很多已经公开的推理数据，但这些数据涉及尚未发布的芯片和尚未发布的模型。

Each of these three different parts of the equation got answered in a different manner, but the most complex was estimating the throughput per GPU because we had to combine a lot of our inference x data that's already public, but on chips that are not out yet and on models that are not out yet.

所以我们不得不构建一个推理模拟器来估算，比如，如果NVIDIA推出具有某些特性的Feynman芯片，那对吞吐量意味着什么？

So we had to build an inference simulator where we had to estimate, okay, if NVIDIA comes up with Feynman with certain set of characteristics, what does it mean for the throughput?

如果内存规格被降级，就像已经发生的那样，对吞吐量有什么影响？

If memory gets despec ed, like it has happened, what's the impact on throughput?

如果OpenAI训练一个一万亿、一万五千亿、两万亿参数规模的模型，这对吞吐量意味着什么？

If OpenAI trains a 10,000,000,000,000, 15,000,000,000,000, $20,000,000,000,000 model, like, parameter model, what does it mean for throughput?

其实很多这些数据并不是现成的，我们得自己专门去构建。

Now, a lot of this is not readily available and something that we had to build out custom.

我们还得跟SemiAnalysis的各种推理工程师合作，才能做到这一点。

And we had to work with, you know, various inference engineers at SemiAnalysis to be able to do that.

但这就是一个需要定制工作的项目例子。

But that's an example of a project that would be custom work.

好的。

Okay.

带我看看那些假设吧。

Take me through some of those assumptions there.

比如，在推理模拟器里，我知道……

Like, in the inference simulator, I know...

嗯，也许先解释一下它是怎么交付给客户的，因为我觉得很多人听到咨询就会想到幻灯片。

Well, maybe explain how it gets delivered to the clients because I think a lot of people hear consulting and they think slide deck.

我们确实做了很多幻灯片，但对于像这样复杂的问题，我们的趋势实际上是转向实时仪表盘，让人们可以随时间调整他们的假设，而我们需要维护这些。

And we make plenty of slide decks, but for complicated questions like this, the trend for us has been to actually move towards live dashboards where people can edit their assumptions over time, and we have to maintain this.

所以这有点像我们交付给他们的一个软件产品。

So it's kind of like a software product that we're delivering to them.

没错。

Exactly.

所以这个产品里有很多排列组合，我们不可能把它放在一个静态的Excel里。

So so this was like like there were so many permutations and combinations in this product that it was not feasible for us to put this in a static Excel.

所以我们不得不构建一个动态仪表盘，这基本上就是交付给客户的东西。

So, we had to build a dynamic dashboard, and that was essentially what was delivered to the clients.

这是一个他们可以使用的仪表盘，可以用非常直观的方式摆弄不同的假设组合，并且看到每一个数字都在变化。

This is a dashboard that they could use, play around with different sets of assumptions in a very visual manner, and see literally every number changing.

所以，这里的关键假设，我想，首先是你分配多少算力给推理，多少给训练？

So, the key assumptions here were, I think, to start with, how much how much of your compute are you allocating towards inference versus training?

什么是...

What are...

我们知道现有的芯片规格，但你们想象中，AMD、TPU、NVIDIA正在开发的下一代芯片规格是什么样的？

We know the existing chip specifications, but, like, what are the chip specifications that you're imagining for the next set of silicon that are being developed by AMD, TPUs, NVIDIA?

我们从两三集前的某一集里知道，这些规格变化很快。

We know from one of the episodes, two or three episodes ago, that those specs can change pretty quickly.

Ruben Ultra刚刚从1TB的HBM降级了，对。

Ruben Ultra just got despec from a terabyte of HBM down to Yeah.

56、128或者192。

56 or one twenty eight or one ninety two.

所以……

So...

你的……

What's your...

比如，另一组假设可以是关于你的输入输出比率的。

Like, another set of assumption could be, like, around what's your input output ratio?

你们在用agent x harnesses吗？

Are you using agent x harnesses?

你是用固定的序列长度、上下文长度，还有数据类型这些吗？

Are you using, like, fixed sequence lens, context lengths, like, sort of data types?

你的缓存命中率是多少？

What's your cache hit ratio?

比如说，怎么……

Like, how do...

如果缓存命中率从80%提高到85%再到90%，吞吐量会怎么变化？

How does throughput change if you're hitting, like, 80% cache versus 85 versus 90%?

所以我觉得这些假设就是让问题变得有意思的地方。

So I think those are some of the assumptions that make it interesting.

完全说得通。

Makes perfect sense.

现在，给我讲一个不同类别的例子，比如策略方面的。

Now, tell me more about an example that goes for a different like category, which is like strategy.

这可能是关于产品路线图的，也可能是关于自研、购买还是合作这样的商业决策。

You know, this could be on product roadmap, it could be on, like, a business decision about build versus buy versus partner.

你现在在帮人们解决哪些领域的问题？

What are what are some of the domains that you're helping people with right now?

我觉得策略这个词挺宽泛的。

I think strategy is a very broad term, I guess.

它可能指的是产品策略。

And it could it could mean, like, product strategy.

也可能是市场推广策略。

It could mean, like, go to market strategy.

不过我可以给你举两个例子。

But I can give you an example of both.

就像我之前说的，我们在内部建了一个推理模拟器。

So as I said, like, one of the things that we have built internally is an inference simulator.

有一家存储公司就用这个模拟器，回放各种模型和芯片的推理轨迹，来规划他们下一步要开发哪些存储产品。

And and, like, one of the, memory storage companies is essentially using that simulator to replay eugenic traces for a combination of different models as well as chips to figure out the next set of memory products that they should be developing.

他们想搞清楚，在跑生产环境的推理任务时，内存带宽和容量之间的权衡到底怎么影响性能。

They're answering the questions like, how does memory bandwidth versus capacity trade offs play out in serving production inference workloads.

这就是我们正在做的一个产品策略方面的例子，而且我们和团队的合作非常紧密。

So that's one example of a product strategy thing that we're doing, and we're working with the teams on a very dedicated basis.

另一个例子是，我们在帮一家半导体公司确定他们理想的客户群。

The other example could be, we're helping you know, one of the Semi's companies figure out their ideal set of customers.

他们产品是有的，但不知道卖给谁。

So they have a product, but they don't know who to sell to.

比如，可能随时都有大概一百个客户会买他们的产品。

Like, there are probably like 100 customers that could buy their product at any given point in time.

但这是个销售周期很长的产品。

But it's a long lead time product.

要认证一个客户至少需要三到四个月。

It takes at least three to four months to qualify a customer.

所以，实际上不可能去和一百个不同的客户做产品认证。

So, it's not realistically possible to go and qualify your product with a 100 different customers.

所以我们帮他们找出最应该优先考虑的十到十五个客户，以及怎么去开展销售对话，客户会关心哪些关键技术点，还有你该怎么定位自己，才能满足这些客户想要的技术规格。

So, we are helping them figure out who are the top 10 or 15 customers that they should be prioritizing, what's the right way for them to approach sales conversations, what are the key technical topics that they, as customers, would be interested in, And how should you position yourself to be able to deliver on those technical specifications that these customers want?

说起来容易。

It seems easy to say.

但我觉得执行起来非常难，因为我们需要亲自去跟二三十甚至四十个不同的客户聊，听取他们的看法，然后汇总出一个市场观点。

It's very difficult, I think, to execute because we need to go out and literally speak with 20, 30, 40 different customers, get their perspectives, aggregate a market view.

而这是很多公司做不到的，因为SemiAnalysis的人脉非常广。

And this is something that a lot of different companies are not able to do because SemiAnalysis is so very well connected.

我们其实不通过专家机构来安排这些专家访谈或者别的什么。

We don't really use an expert agency to carry out these expert calls or whatever you call them.

因为我们的网络很广，我们直接利用自己的关系就能搞清楚那些对整个生态系统都很重要的问题。

Because we have such a wide network, we are able to just use our relationships to figure out important questions that would help the wider ecosystem.

这也会反过来影响我们的研究。

And this also informs our research.

我的意思是，也许你可以解释一下，以前上过这个播客的人，比如几周前讲模块化数据中心的Eric，他们做这些研究工作时，是怎么为SemiAnalysis的产品做贡献的。

I mean, maybe you can explain a little bit about how people who have been on this podcast in the past, like say, Eric, a few weeks ago working on modular data centers, are, you know, when they go and do this research effort, contributes to SemiAnalysis products.

而且在很多方面，我们卖的一些机构产品，比如数据产品，最初都是从一个咨询项目开始的，或者至少是有人问我们相关问题，才启发了我们开始做这些产品。

And in many ways, some of the institutional products that we sell, the data products, originally started as a consulting project, or at least the inspiration to start working on that actually started because somebody's asking us about this.

他们显然觉得这很重要，所以我们就去做吧。

They clearly think it's important, therefore, let's go do it.

因为我觉得我们和很多客户都是定制化合作，而且对某个话题会挖得很深，所以很多时候我们能发现一个……

Because I think because we work on a very custom basis with a lot of these clients and we go very deep on a topic, I think oftentimes we're able to identify that there is a...

比如说，如果某个问题或需求对一个客户很重要，而我们又能找到一个标准化的解决方案，那我们就能把它推广到更多客户，这就变成了一个产品。

Like, if this question or problem is relevant for one client and we have been able to figure out a standardized solution for that, I think we're able to scale that up to a broader set of clients and that becomes a product.

我觉得我们的数据中心模型最开始就是给一个客户做的咨询项目。

I think our data center model started as a consulting project for one of the clients.

也许ClusterMax最开始也是咨询项目。

Maybe ClusterMax started as a consulting project.

我不太清楚。

I don't know.

这个你得告诉我。

You have to tell me that.

不。

No.

不是，ClusterMax不是，但AI云TCO模型绝对是始于咨询项目的，对。

No, ClusterMax, but the AI Cloud TCO model definitely started with consulting projects, which was Yeah.

你知道，我们继续用它来做分析，比如，GPU集群的投入成本是多少，那些基于芯片选择、运营支出以及你说的推理端点利用率等因素来提供token服务的公司，盈利能力如何。

You know, we continue to use to build analysis on, like, how, you know, what are the input costs for GPU clusters, how profitable are given companies that are serving tokens based on the choice of chips, based on the amount of OpEx, based on the, as you're saying, utilization ratio on a inference endpoint or something like that.

我们真正深入去做的方式是，因为那些你信任的人，你的客户，他们在认真对待一些问题并提出疑问，你受到启发去做这项工作，然后，咱们去找答案吧。

Way we actually dig in and do that is like, you get inspired to do the work because people that you trust, that your clients are taking some problems seriously and asking questions and, you know, let's go get answers.

这就变成了一个有用的产品。

Becomes a useful product.

所以，就这样吧。

So, anyway.

没错。

Exactly.

而且我觉得另一个例子……

And I think the other example...

最新的例子就是我们正在构建的推理模拟器。

The latest example of that is the inference simulator that we're building.

我觉得我们最开始是为了一个客户做这个的。

I think we started building that for one client.

甚至不只是一个客户。

And not even for one client.

我觉得它最初是SemiAnalysis内部一个挺酷的科研项目，然后我们把它提供给了第一个客户。

I think it started as a cool, like, science project within SemiAnalysis, and then we offered it to one client.

他们觉得很有价值。

They found value in it.

然后其他几个客户也觉得有价值，现在它正在变成一个咨询……

And then a few other clients found value in it, and now it's becoming a consult...

有点像独立的产品了。

Like, sort of a product by itself.

所以……

So...

确实如此。

Absolutely.

完全说得通。

Makes perfect sense.

那尽职调查呢？

So how about DDs?

这对投资者以及一些对并购活动感兴趣的公司来说，显然是个大事。

That's obviously a big thing for the investors as well as some companies that are interested in m and a activity.

我觉得，你知道，丹提到有七万亿美元的债务被发行来资助AI基础设施。

I think, you know, Dan talks about $7,000,000,000,000 of debt being issued to fund AI infra.

所以有很多投资者，我觉得，正在进入这个领域。

So there's a lot of, I think, investors that are coming into the game.

还有很多新的玩家也在进入这个领域。

There are a lot of newer players that are also coming into the game.

所以尽职调查项目基本上就是投资者来找我们，问我们，嘿，你们能不能帮我们……

And so what a DD project essentially looks like is an investor coming to us and asking us, hey, can you either help us...

我们拿一个新云举个例子吧。

Let's let's take a neocloud for an example.

因为这是我们经常做的事情。

Because that's what we do a lot of the times.

如果一家私募基金想向一家新云公司投入十亿美元，他们主要关注两件事。

If if a private equity fund wants to put in a billion dollars in a neocloud, essentially, they're looking for two things.

他们需要技术分析，也需要商业分析。

They're looking for a technical analysis, and they're looking for a commercial analysis.

在技术方面，他们关心的是新云公司向客户提供的服务等级协议（SLA）是什么。

On the technical side, I think what they're interested in is what are the SLAs that a custom...

以及这家新云公司能否兑现这些SLA承诺。

That the neocloud is offering to the customer, and will this neocloud be able to deliver on those SLA guarantees?

如果做不到，会有一系列更广泛的影响，我们稍后可以讨论，但对任何一方都不好。

Because if they don't, there are, like, wider set of implications that we can talk about later, but it's not good for any of the parties.

他们还希望了解OEM的支持、保修结构和合同条款，因为这里面有很多细节问题。

They're also interested in understanding what is the OEM support and warranty structure and contracting looks like, because there's a lot of devil in the details there.

如果你没有和OEM正确交易，可能会面临各种风险。

And if you're not transacting properly with the OEMs, you could expose yourselves to a variety of risks.

他们还对了解集群性能感兴趣。

They are also interested in understanding the cluster performance.

所以我们能用一种非常受ClusterMax启发的方法来测试他们的集群，并准确指出他们擅长什么，以及他们非常不擅长什么。

So we are able to test their clusters with a very ClusterMax inspired methodology and be able to tell what exactly are the things they're good at and what exactly are the things they're really, really bad at.

以及他们在一到八十天内改进这些方面的计划是什么。

And what's the one and eighty day plan for them to be able to improve on those things.

我觉得在商业方面，因为这是一个非常需要财务智慧的决定，他们需要了解这些账单的资本支出和运营支出，以及它们与市场标准相比如何。

I think on the commercial side, because it's such a financially savvy decision, they need to understand what's the CapEx and the OpEx for these bills and how does it compare really to the market standard.

因为作为我们总拥有成本模型的一部分，我们能够实时基准测试成本，以及通过其他各种方式，比如性能。

Because as part of our TCO model, we are able to benchmark in real time costs, as well as, you know, like through various other means, like performance.

对。

Yep.

我们能够为这些客户实时基准测试这些指标。

We are able to, benchmark these things in real time for these clients.

所以这就是我们典型的尽职调查项目的样子。

So that's what, like, our typical DD projects looks like.

而且这取决于你是在尽调一家新云公司、芯片公司还是模型公司，你做过……

And it it varies depending if you're diligencing a neocloud versus a chip company versus a model company you have done...

或者数据中心、电力公司，我们真的什么都做过。

Or a data center or a or a power LAN, and we have literally done everything.

自从你来了之后，这一年左右我们做了很多，我的意思是，我们就是在不断积累经验，组建团队。

We've done a lot in just a year or so since you've been here and I mean, we're just like stacking up experience across it, building the team.

看到它随着人成长真的很酷，真的，随着越来越多行业里的人开始认真对待这些东西，因为我觉得对很多这类项目进行技术尽调是至关重要的。

It's pretty cool to see it kinda grow as people, really, as as more and more people in the industry take this stuff seriously because I I think the need for technical due diligence on a lot of these projects is of the utmost importance.

所有这些都是新的。

It's all of it's new.

并不是说某些投资者能理解或者有能力理解，你知道，一个六个月或三个月前的新模型。

It's not like certain investors can understand or have the capabilities to understand, you know, a type of model that's six months old or three months old.

我的意思是，对每个人来说这都是新的。

I mean, it's new for everybody.

总之。

Anyway.

没错。

Exactly.

对，这对每个人来说都是新事物。

It it it's new for everybody.

而且我觉得，现在的情况是，购买方不再仅限于顶尖实验室，新的实验室也开始活跃在市场交易中。

And also, think what's happening is that, in my opinion, I think the off taker base is diversifying beyond the top labs to also like now the new labs who are actively transacting in the market.

新云运营商的群体也在多样化。

The neocloud operator base is also diversifying.

你想啊，像Skillers是元老，然后是加密货币矿工，还有像Core Web这样的专业玩家。

If you think about it, right, like, skillers are the OGs, and then you have the crypto miners, and then specialist players like, you know, Core Web and others.

但现在你看到很多房地产开发商，他们……

But then now you see a lot of real estate developers who have...

他们没有任何云运营经验，也开始进入这个领域。

Who do not have any experience in operating cloud entering the scene.

你看到有钱人的家族办公室现在也在尝试建新的云。

You see family offices of rich people trying to build a new cloud now.

所以SLA就变得更加重要了，因为如果你以前没有运营这类东西的经验，而且SLA又对你不利的话，那后果会非常严重。

And so SLAs become even more important because if you don't have the experience of operating such things before, and if SLAs are not in your favor, like, the implications would be really massive.

我们来聊聊这两种类型吧。

Let's talk about the two types of them.

显然，这个我挺熟悉的。

Obviously, is something I'm quite familiar with.

有一种SLA是和交付时间相关的。

There's one type of, you know, SLA that's related to the delivery timelines.

其实SLA说白了就是，如果出了什么问题，合同可以被终止。

Really, an SLA is just like, if something happens, a contract can be terminated.

这大概就是服务级别协议背后的法律条文的意思。

That's roughly the the legalese behind a service level agreement.

而且，合同终止权可能是个更宽泛的定义，我们这里说的其实就是这个，而不是SLA本身。

And it doesn't, you know, contract termination rights is maybe a more broad definition, and that's really what we're talking about here as opposed to SLAs.

你可能能终止合同的第一个原因就是交付延迟。

The first reason why you might be able to terminate a contract is because of late delivery.

集群晚了好几个月才上线，所以合同就取消了。

Cluster shows up months late, so therefore cancel the contract.

但最常见的情况是停机。

But the most common one is downtime.

如果有人装了GPU，他们期望有99.9%、99%、90%的可用性，如果低于90%，也就是说一个月里10天有1天，或者30天里有3天，你没法用你付了钱的GPU，通常你会得到一些补偿，或者你有权终止合同。

If somebody's got GPUs installed, they expect to have three nines, two nines, one nine of availability and if you're under 90%, you know, one out of 10 days in the month, three out of 30, you don't have access to the GPUs you're paying for, generally, you're getting some credits in response or you have the right to terminate the contract.

那你能不能跟我讲讲，贷款方、新云服务商，甚至买家，为什么需要某种保险来应对合同终止的这两种可能性？

So maybe talk to me about the motivation for the lenders or for the neocloud itself or even for the buyer to have some sort of insurance against either of these possibilities of a of a contract termination?

他们想要下行风险保护。

They want downside risk protection.

没错。

Exactly.

我觉得，乔丹，你刚才真是一针见血。

I I think, so so, Jordan, you hit the nail right on head right there.

所以首先，为什么SLA很重要，然后我再讲为什么保险很重要。

So so one, why SLAs are important, and then I'll talk about why insurance is important.

你说得对。

I think you said it right.

首先，作为一家新型云服务商，如果没有强有力的服务等级协议……

The first thing is, as a neocloud, without strong l s...

没有SLA，你就拿不到融资。

SLAs, you won't be funded.

就是这样，没得商量。

Like, period.

你连起步都做不到。

Like, you won't be able to start your journey.

所有投资者和贷款方最关心的就是他们的风险敞口有多大。

Like, all the investors, all the lenders, the most important thing that they care about is what is their risk exposure.

如果他们每个月或每个季度能收到利息，那他们就要考虑，如果集群宕机了，这笔钱还能不能按时到账。

If they're getting interest payments every month, every quarter, what is the chance of that payment not hitting their bank account because their cluster is down?

所以这是最重要的一点。

So that's the number one thing.

我觉得第二点是，SLA其实是为困难时期准备的。

I think the second thing is, I would say SLAs are really for the bad times.

你知道，我们在各种场合都见过这种情况。

You know, we have seen this in various settings.

如果一切顺利，没人会提SLA。

If things are going well, nobody talks about the SLAs.

如果事情不顺利，大家就会提SLA，这能让各方有机会无责退出合同。

If things like, you know, don't go well, people talk about the SLAs, and it gives sort of the parties to be able to get out of the contract spot free.

第三点是，我们之前聊过，至少对于新云来说，他们通常会签十五到二十年的数据中心租约。

And the third thing is, you know, we have talked about it, at least for neoclouds, they typically sign a fifteen, twenty year data center lease.

他们通常签五年的GPU合同。

They typically sign the GPU contracts for five years.

所以如果你不承诺或者不履行SLA，你就会有声誉风险，这意味着你拿不到第二笔交易，也拿不到第三份租约。

And so if you're not committing or if you're not sort of delivering on your SLAs, you carry a reputation risk, which means that you will not be able to get your second deal and you'll not be able to get your third lease.

如果你做不到这些，那你怎么支付二十年的数据中心租金呢？

And if you're not able to do that, then how do you pay for twenty years of data center leases?

所以SLA很重要。

So SLAs are important.

我觉得对贷款方来说，这之所以重要，是因为他们在寻求下行保护。

I think the reason why it's important for lenders, as I said, is they're looking for downside protection.

而保护下行风险的一个方法就是购买SLA保险。

And one way to protect your downside is by having an SLA insurance.

我们现在知道有几家公司正在为新型云提供SLA正常运行时间的保险。

We know a few companies now who are insuring SLA uptime for the neo cloud.

所以如果你……

So if you...

作为新型云，如果你支付一定基点的保费，你就能获得一定的保护，比如如果你需要支付一定金额的罚款，保险公司会帮你处理。

As a neo cloud, if you pay a certain basis points of premium, you know, you get a certain protection that if you have to pay an x amount of penalty, the insurance company will take care of it.

在我看来，获得这种保险并不容易，因为这些保险公司在承保这些新型云时也非常精明。

It's not easy, in my opinion, to get this insurance because these insurance companies are also pretty savvy in terms of how they underwrite these neoclouds.

所以，表现不好不能成为借口，你不能继续表现不佳，否则你将无法……

So it's not an excuse to have bad performance that you can continue to have bad performance, but then you will not...

然后你通过保险来弥补这一点。

And then you'll kind of compensate for it by having the insurance.

如果你表现不好，你甚至拿不到保险。

If you have bad performance, you won't even get insurance.

但如果你表现好，我觉得保护下行风险、让合同更具投资价值的一个办法，就是考虑选择这些保险。

But if you have good performance, I think a way to protect your downside and make maybe your contract more investable is to probably think about electing for some of these insurance.

我们刚才一直在说SLA，也聊了不少相关的东西。

Now, we've thrown around the term SLA and and talked about a few different things.

显然，一个集群、数据中心或站点有很多不同的部分都可以附带SLA。

Obviously, there's many different components of a given cluster or data center or site that can have an SLA attached.

所以，这可以适用于GPU或节点本身。

So, this could apply to the GPUs or the nodes themselves.

也可以适用于机架。

It could apply to the racks.

还可以适用于集群。

It could apply to the clusters.

它也可以适用于站点本身。

It could apply to the site itself.

它还可以适用于那里的电力和冷却系统。

And it could also apply to the power systems and cooling systems in there.

而且当有这么多方参与其中，这么多合同在流转时，实际上有很多理由让你希望对某个你不太放心的单个部分进行风险评估，相比之下，你对合同的其他部分更有信心，可能是因为你之前和那个供应商合作过，或者因为你更了解这项技术。

And when there are so many parties at play doing this and there are so many contracts trading hands, there's actually a lot of reasons why you might want to have a risk assessment done for just even one individual part that you're less comfortable with when compared to, you know, some other part of the contract that you have more confidence in because maybe you've worked with that vendor before or because you understand the technology better.

跟我聊聊风险评估的概念和范围吧，因为，嗯，我知道……

Talk to me just about the concept of a risk assessment and the scope because, well, I know the...

我这是在引导证人发言。

I'm bleeding the witness here.

但范围可以非常窄、非常快，也可以非常大、非常广。

But the scope can be pretty tight, pretty quick, or it can be pretty big, pretty broad.

我觉得，正如我所说，风险评估可能相当关键，尤其是在贷款方……

I think, as I said, like, a risk assessment could be, like, fairly critical, especially in cases where lenders are...

或者投资者要求的情况下，这是他们为某个特定集群提供资金的一个必要条件。

Or investors are demanding for it, and that's a necessary condition for them to fund a particular sort of a cluster.

我觉得可以很简单，比如对新建云数据中心最重要的合同做一次审查。

I think it could be as lightweight as doing a contract review of the of the most important, like, sort of contracts for a neo cloud.

这些合同包括承购合同、OEM合同和数据中心合同。

And these are the off taker contract, the OEM contract, the DC contract.

因为你向客户承诺了某些东西，然后你依赖数据中心合作伙伴和OEM来提供服务。

Because you're promising something to the customer, and then you're relying on the DC partner, and you're relying on the OEM to be able to service that.

还有其他一些附属合同，我们也可以讨论，但现在不必深入。

There are other side contracts as well that we can get into, but we don't have to right now.

但这是其中一种。

But that's one.

第二种是稍微更深入一点的风险评估，可以称之为数据中心就绪度评估，也就是了解数据中心何时能通过L3、L4、L5测试，以及何时能让你进行各种其他测试。

I think the second thing is a more like, a slightly higher touched review of risk assessment, you can call it, which is data center readiness, which is understanding the milestones around when will the data center clear like L3, L4, L5 testing, When will the data center be made available to you to do, like, a bunch of different other testings?

机架什么时候能到数据中心？

When will the racks arrive on the data center?

什么时候能安装电力和冷却系统？

When can you install, like, sort of power and cooling?

那所有这些能不能作为一个整体系统一次性测试呢？

And, like, when all of these things can be tested as a system all at once?

我们也可以聊聊其他方面。

And we can talk about, like, other aspects here as well.

第三件事就是做裸机审查，乔丹，我觉得这方面你可是专家，毕竟你一直在做各种机房访问测试。

The third thing is doing a bare metal review, which, Jordan, I think you are the master of with all the classroom access testing you have been doing.

所以我想这部分就交给你来，然后我们再回来讨论最后一项。

So I think I'd like to let you do it, and then we can come back for the for the final one.

简单来说，就是我们能访问这些系统。

I mean, the summary is basically that we get access to the systems.

我们登录进去。

We log in.

然后进行测试。

We test them.

我们会模拟故障之类的，然后测试确保监控系统能设置好，让大家能发现故障发生，以及多快能恢复。

We do things like simulate failures, and we test to make sure that people can have the monitoring system set up so that they can identify that the failures occurred and how quickly they can recover from it.

集群出故障的方式有很多很多种，你的监控系统必须覆盖全面，因为我们见过各种奇葩的故障方式。

There's many, many different ways in which things can fail in a cluster and your monitoring systems have to have full coverage because we see things fail in spectacular ways.

即使在我们短短几周的测试里，那还是个完全人为控制的环境，大家都全力以赴去搭建，就为了做集群微测试。

Even in just our few weeks of testing where it's like a fully contrived environment, people have all hands on to set it up just through cluster mics testing.

然后我们还听到这些服务商的所有客户讲他们体验到的服务质量有多差。

And then we hear the stories from all the customers of these providers about how for the quality of services that they're experiencing.

而且根本原因可能真的五花八门。

And the root cause can be, like, really, like, sort of varied.

我觉得这个评估能帮你搞清楚性能差的真正原因到底是什么。

And I think this assessment helps you understand what exactly is the root cause of bad performance.

比如，是因为电力配置有问题吗？

Like, is it because problem in sort of, like, the way your electricity is set up?

是因为冷却系统设置的问题吗？

Is it because of the way your cooling is set up?

还是机架本身有挑战？

Is it because of the challenges in the racks itself?

是因为软件方面的问题吗？

Is it because of the challenges in software?

而且最终，其实只需要一个……

And ultimately, it's like only one...

就算你的集群表现不佳只有一个原因，你仍然得承担SLA的罚款。

Like, even if there is only one reason that, like, your cluster is not performing well, it still does not mean that you don't have to pay the SLA penalty.

所以，只要有一个环节出了问题，你就得把所有的SLA积分退还给客户。

So, like, only one thing has to go bad for you to, like, be liable for paying all the SLA credits back to the customer.

这就是为什么我们要回到合同审查阶段，也就是第一步。

I mean, that's why we go back to the contract review stage, which is the first one.

人们在制定这些合同时需要很多建议，因为正如你所说，合同是为了应对糟糕时期，而不是好时期。

People need lots of advice on how to set up these contracts because as you said, contracts are for when there's bad times, not when there's good times.

好的。

Okay.

让我问一个相关的问题，这样我们可以继续聊下去。

Let me ask a kind of related question, but it can keep us going here.

当我们谈到提供战略建议这类事情时，很多人正在进入这个领域，考虑是自建还是购买，特别是针对新云，因为你知道，交付周期那么长，找到数据中心站点太难了，我们也做过一些这样的咨询工作，帮助人们做出自建、购买还是合作的决策。

When we talk about giving strategy advice and stuff, a lot of people are getting into this space where they're considering build versus buy specifically with the neo clouds because, you know, there's such long lead times, it's so hard to find a data center site, and we've done a little bit of that consulting work where we've helped people make that build versus buy versus partner decision.

而且我们不仅仅在新云决策上做过，还做过其他方面的，比如如果你是芯片公司，供应链管理之类的。

And we've done it in more than just the neocloud decision for other stuff, let's say supply chain management if you're a chip company or something.

所以也许你可以谈谈那些例子。

So maybe you can talk a little bit about those examples as well.

我觉得因为我们确实看到很多限制，关于获取托管容量来进行大型GPU集群的站点建设。

I think because we are definitely seeing a lot of constraints around getting colo capacity to do sort of large site build outs for GPU clusters.

我们也看到芯片方面有些紧张。

We are also seeing some tightness, you know, around chips.

而且劳动力肯定有短缺。

And there are, like, definitely shortages around labor.

所以原则上，即使很多公司最终想自建集群，通常也不是获得近期容量的最可行途径。

So in principle, even when, like, a lot of the companies want to eventually build their own clusters, is oftentimes not the most feasible path to get near term capacity.

所以最终，这归结为几个因素。

And so ultimately, it boils down to a few factors.

第一个是上市时间。

And the first is time to market.

第二个是你愿意为获得算力支付溢价的程度。

Second is your willingness to pay a premium to get compute.

第三个是你自己团队的技术水平。

Third thing is how technically savvy your own team is.

你们有没有运营集群的经验？

And do you have any experience in running cluster operations?

最后，从融资等方面来看，你们在资本市场上是否有足够的信誉，能够筹集资金从零开始建立集群，或者成为某家公司的承购方。

And finally, in terms of financings and everything, do you have enough credibility in the capital markets to be able to raise funding to set up a cluster from scratch or to be able to be an off taker of a company.

所以最近，我们和几个客户合作，我们让他们……

And so recently, we have been working with a few clients where we have let...

我们得出了几个不同的结论。

We have we have arrived at a few different answers.

我记得有一个案例，我们最近建议一家公司直接去规划，因为他们想在接下来的六个月内就要100兆瓦的电力。

I think in one of the cases, we have recently advised a company to be able to just like go out and plan because they wanted 100 megawatts in literally in the next six months.