AI效率正在重新定价计算市场 | 史蒂夫·侯

关于前瞻性指引的任何表述，都不构成对任何投资或产品的买卖建议。

Nothing said on Forward Guidance is a recommendation to buy or sell any investments or products.

好了，各位。

Alright, everybody.

欢迎回到《前瞻性指引》的又一期节目。

Welcome back to another episode of Forward Guidance.

今天和我一起的，是节目的老朋友侯Steve，他几个月前刚来过，那时你还在彭博，现在你已经去了硅谷数据公司，担任研究主管。

And joining me today is repeat guest of the show, Steve Hou, who just joined me a couple months ago right at the the tail end of when you were at Bloomberg, but now you're at Silicon Data, head of research.

你现在是人工智能领域最重要的一些图表和指数背后的操盘手。

You are the man behind some of the most important charts and indices in the world of AI right now.

今天要聊的内容很多，不过真的很高兴你以新身份、新公司回归。

There's a lot to get into, but, yeah, really excited to have you back now under your new role at your new company.

恭喜你入职新公司，也很高兴你回来，Steve。

So congrats on on starting there, and, yeah, great to have you back, Steve.

是的。

Yeah.

谢谢您，费利克斯。

Thank you, Felix.

总是能见到您真好，也谢谢您再次邀请我回来。

Great to see you always, and thank you for having me back on.

是的。

Yeah.

太好了。

Awesome.

对于还不了解Silicon Data以及你们业务的人，我很想听听你为什么选择加入他们，你们具体做什么，还有在经历了这一切之后，你对当前AI领域的整体看法是怎样的。

For those that don't know about Silicon Data and what you guys do, would just love to hear a bit about, like, why you made the shift to joining them, what you guys do, and, yeah, just how you think about the the state of AI right now after.

我大约两个月前离开了彭博，差不多也是那时候加入了Silicon Data，时间点正好。

So, yeah, I left Bloomberg about two months ago, joined Silicon Data almost two months ago, back on the day.

Silicon Data是一家致力于为物理AI算力市场引入数据服务的公司。简单来说，我们想把期货和衍生品带到AI算力领域，帮助人们对冲当前这个规模高达数千亿甚至万亿美元、且快速演变的AI算力市场所伴随的核心风险。现在这些风险大多以股权或固定收益形式存在，但往往没有被高效定价。我们认为，许多与AI算力和GPU收入相关的风险，其实可以用传统金融工具来管理——这些工具在成熟资本市场中早已被广泛理解。对数据中心运营商、算力提供商，或者需要采购算力的企业来说，交易算力期货是一种非常自然的对冲方式，不仅能降低风险，还可能让你更有底气去扩大算力采购规模，避免像一些AI实验室那样，要么买得太少，要么买得太猛。

You know, Silicon Data is a company that tries to bring, well, data to physical AI compute market, and, you know, you think about I think the easiest thing to think about is we are trying to bring futures contracts, derivatives to the AI physical compute and allow people to hedge the essential risks that are involved with this now enormous and evolving AI compute market that's hundreds of billions, if not trillions in size, and there's a lot of risks that's being held in equity form and in a fixed income form, and our view is sometimes probably not perfectly efficiently, and there's a lot of, I think, risks that's now involved with AI compute and GPU income that can be handled with traditional financial instruments that is well understood by historical financial markets or capital markets, so to the extent that on the natural hedging side of the data center providers, compute providers, or the companies that are looking to buy compute, run compute futures contracts is a very natural way to hedge out their risk and maybe actually, in fact, help you be a bit bolder in terms of how much compute you actually acquire, so you don't find yourself I think, overexposed or maybe not having enough, which seems to have been the case with some of the AI labs.

我加入这家公司，是因为我觉得自己在彭博工作了六年，一直在做公开市场的基准分析和系统性指数。

I joined the company because I felt like Kai Haire I have been working on public markets, benchmarking, systematic indices at Bloomberg for six years.

这似乎是进入人工智能领域的一个很自然的方式，用的是类似的技能，只是应用在AI上。

This feels like a natural way to become exposed to the AI sector, similar set of skill sets, but apply to AI.

对，太棒了。

Yeah, awesome.

对，那我们直接切入正题吧？

Yeah, so why don't we get into it?

我们特别想听听，为什么对AI算力建设来说，对冲和期货合约是必需的？

We'd love to just hear a bit about, like who, what is, what is the, why is hedging and, you know, futures contracts a necessity for the AI compute build out?

那理想的目标客户是谁？他们想对冲什么风险？为什么？

Like who are, who are the ideal target customers and what are they trying to hedge and why?

是那些超大规模云服务商吗？

Like, is it the hyperscalers?

是前沿模型的开发方吗？

Is it the frontier models?

还是那些使用这些模型的下游公司？他们想对冲自己的推理需求波动？

Is it like, you know, just downstream companies that are utilizing these models and, you know, are trying to hedge out their their inference demands?

我是说，这到底是什么意思。

Like, what does yeah.

这个市场格局是什么样的？

What does that landscape look like?

我认为，这种需求应该来自市场上的所有主要参与者，就像原油市场或者历史上农产品市场那样，期货合约最初就是在那些地方发展起来的。

So, I mean, the demand for it should really come from all major participants in this market, in the same way you would expect in the crude market or even agriculture historically, wherever futures contracts first evolved.

随着这个市场最终变得更加分散，对吧？

As this market eventually become more, I think, fragmented, right?

目前，AI算力市场主要由少数几家大公司主导，卖家数量很少，市场非常集中。

Right now, the AI compute market is very much dominated by a couple of big players, a couple of big sellers, you know, very, very granular, right?

两大实验室——OpenAI和Anthropic——占据了AI算力需求的大约一半，而另一方面，超大规模云服务商则提供了绝大部分算力并负责建设这些基础设施。

So you've got the two major labs, OpenAI and Throtpic, that account for ostensibly half of the AI compute demand, and then on the other hand, the hyperscalers that are providing most of the compute and building most of the compute.

但我们现在看到，随着强大的开源模型和企业级AI应用的兴起，如果认真想想企业最终会如何采用AI，他们并不会简单地选一个赢家，来作为所有人都依赖的平台。

As we have now seen with the rise of very powerful open models and enterprise AI use Actually, if you think about how enterprises will ultimately adopt AI, they're going to actually not just basically pick a winner in this debate of who is going to emerge as the sort of platform on which everyone's going to build on top of.

我觉得这个问题其实已经很清楚了，对吧？

I think that's already been settled, right?

没人能彻底胜出，企业不会放心，就连那些实验室自己也觉得不太行。

Nobody's going to win it outright, Companies don't feel comfortable, and the labs themselves, I think, is just not.

所以真正会积累大量价值的，会是那个协调层。

So what's going to happen is that orchestration layer is going to accrue a lot of the value.

企业会通过让模型更具可替代性来保持自主权，一些低价值的任务可能会智能地交给开源模型或更便宜的模型，而高价值的任务则会使用前沿模型。

Companies are going to retain their sovereignty by basically making models more substitutable, and some low value tasks maybe are going to be routed smartly towards the open models, cheaper models, and then the high value tasks maybe will be using frontier intelligence.

在这方面，我们预计整个模式会进一步转变，尤其是推理需求会变得越来越重要，大部分AI算力需求将来自推理，而算力的投入也将来自众多企业，它们都在寻找算力资源。

To that extent, we expect there to be generally a transition away from the current paradigm, even more so, and with inference becoming an ever bigger deal that most of the AI compute demand is going to come from inference, and the funding of a compute is going to come from many, many enterprises companies that will be looking for compute.

随着模型不断演进，架构也在变化，你会发现更多模型不再只能跑在超大规模云上，而是可以部署在更靠近本地的小型集群或小型云平台上。

As the models evolve, the architecture of the models evolve, you're going to see that more models can be run on not just the hyperscalers, but maybe sort of more closely located smaller clusters, smaller clouds.

所以你能看到这种碎片化趋势。

So you can see this fragmentation.

因此，从卖方角度看，市场上正涌现出一批新的云服务商。

So the market, on the one hand, you have sellers of compute, these new clouds, cloud providers coming under the market.

它们希望获得更稳定的收入，而它们的投资方也希望看到更可靠的回报。

They want to have more certainty, and their lenders, their backers want to have more certainty over their revenue, right?

那你怎么会更不确定呢？

And how do you actually have less certainty?

一个很自然的方式就是，就像这个市场里任何大宗商品一样，用期货合约锁定未来的收入。

One natural way is to actually, just as you have with any commodity in this market, using futures contracts to lock in future revenue.

另一方面，是购买算力的客户，以及他们不同的合作方式。

On the other hand, you have the buyers of compute and the different ways in which you can work out.

有些公司可能不直接买算力，而是买代币，但总有人会实际去采购这些算力。

Some companies, maybe they don't directly buy compute, maybe they just buy tokens, but somebody will actually be sort of buying that compute.

我觉得我们就是这么看的。

I think that's how we sort of see it.

最终，市场正朝着这个方向发展，当然，如果你现在去问云服务商，他们肯定会强烈反对。

Eventually, the market evolving in that direction now, of course, along the way, if you talk to a cloud provider today, there's going be a lot of pushback.

没人愿意把自己的产品叫做大宗商品。

Nobody wants their product to be referred to as a commodity.

但我认为，无论如何，替代性总是会存在的，对吧？

And but I think the way I see it is that there's always going to be substitutability, right?

如果你说你的产品独一无二，那你怎么从别人手里抢客户呢？

If you tell me that your product is so unique, how do you win customers from someone else?

这就是说，在算力方面肯定存在替代性，只是这种替代不会完美无缺，所以你得想办法把客户从竞争对手那里拉过来。

That's how do you win clients away from someone that, someone that, so there is going to be substitute ability when it comes to compute, but it's not going to be perfect.

这就是我们对这个市场的看法。

And that's how we, you know, see this market.

太棒了。

Awesome.

好的。

Okay.

所以我想深入探讨一下你们构建的几个图表和指数，先从关于编排和替代性这一点开始聊。

So I wanna dig into a few of the different charts and indices that you guys have have built out and wanna start with just on that point about orchestration and and substitution.

想聊聊你们推出的代币指数。

Wanna talk about the the token Yes.

就是你们开发的那个指数。

Index that you guys built out.

这个指标不知道为啥火了，有它的道理，但也有些误解。

This one has, you know, gone gone sort of viral for maybe good reasons, also bad reasons.

网上对它的讨论特别多。

There's been a lot of takes on it.

确实如此。

Indeed.

其中有不少观点其实根本没搞清楚状况。

I would make the argument some some quite uninformed takes.

我想请你从最基础的原理出发，给我讲讲你们是怎么做的。

So I just wanna, like, let you explain through the methodology just from first principles.

我不会带任何偏见，纯粹想听听你的解释。

Like, I'm not gonna throw any sort of bias here.

这是你们自己做的令牌指数，对吧？

Just this is this is your token index.

能说说这个指数追踪的是什么，以及我们能从这张图里看出什么吗？

Tell me about the methodology of what it tracks and and what we can take away from this chart.

是的

Yeah.

对

Yeah.

确实如此

Indeed.

这个指数在我刚加入公司的时候就摆在那儿了，好几个月都没人怎么关注。

This this index was sitting on the shelves, you know, when I joined the company and, you know, for quite a couple months and, you know, and I think it was not getting a lot of attention.

我注意到，有些人看到它，其实误解了它的含义，以为它是价格指数或者总交易量指数。

Think I just noticed that people were sort of whoever noticed that was actually not interpreted incorrectly to have thought it was either a price index or total volume index.

我觉得它在终端上显示的名称也不太妥当，有点名不副实，叫什么‘代币支出指数’，但实际上根本不是那么回事。

I think the name of the index on terminal displayed has also got a little bit unfortunate, just a bit of a misnomer, because it's called the token expenditure index when it's actually neither.

它真正的本质是一个按支出加权的价格指数。

What it really is, is an expenditure weighted price index.

你可以把它看作是人工智能领域的PCE指数。

You can think of it as almost like the PCE for AI, right?

这意味着，如果你在追踪某种东西的通胀，要么像CPI那样固定一篮子商品，要么像PCU那样允许在相似替代品之间切换，因为用户是理性的，总会权衡质量和价格，而AI领域更是如此，尤其是当你能通过开放路由或推理平台自由选择不同模型的时候。

So what that means is that, you know, to the extent you're on track inflation of something, you can either fix the basket, which is what CPI does, or PCU allows substitution, right, across similar substitutes because users are rational and they always make the quality price trade off, and AI even more so, especially when you have sort of these open routing or inference platforms that allow you to choose different models, right?

有些模型能力超强，但价格贵得离谱，对吧？

You know, some models are really, really powerful, but they're crazy expensive, right?

就是那些行业里的顶级模型，对吧？

You know, your fables of the world, right?

它们一天就能花掉你一半的token预算，但另一方面，中国的一些开源模型却便宜得多，只是能力可能弱一点。

They burn half of your token budget, you know, in a day or something, but then you've got the Chinese open weights models that are, you know, super, super cheap, but they are maybe less capable.

我在六月初的时候，在公司官方的X账号上发了一篇帖子，说明了这个指数正确的解读方式应该是怎样的。

So I wrote a post on x on our corporate x account, towards the beginning of June saying that, no, this is the way you should be interpreting this index.

我们收集了市面上所有模型的价格，包括各种开源模型，同时也从几个公开的路由平台获取了输入输出的使用量数据。

We basically collect all the prices of all the models, render blast models out there, and also we collect the usage volume from a couple, you know, a few sort of these public routing platforms with the inputoutput volume.

你知道，每个AI模型都会对输入token和输出token分别收费，所以我们得为每个模型计算一个综合价格。

As you know, every ALM token model, there's a price charge on input token, how much tokens you put in, and there's also a price charge on output token, so we have to create a blended price for every single model.

我们根据每个模型的输入价格、输入量和输出量，算出它的综合单价，再按使用量加权汇总，形成指数，这样就能真实反映各个模型的实际使用情况。

We take the input price, input volume, output by highest output volume, create a blended price for each model, and aggregate them up to index level based on usage so that this thing is going to track how much each model is being used.

这个模型的价格动态一般是这样的：模型上线后，代币价格会有波动，但幅度不大，对吧？

To the extent that the general price dynamics of this model is such that once the model is launched, the token price moves but not that much, right?

所以大部分价格变动其实来自使用模式、用户行为和消费习惯，对吧？

So most of the movement comes from essentially usage pattern, usage behavior, consumer behavior, right?

我们所覆盖的数据并不是整个市场，对吧？

So our coverage of data is not a full market, right?

如果你把整个AI推理市场想象成一个正方形，

If you imagine the full market, the AI inference market as being like a square, right?

那我们这个数据只是其中一小块，不是垂直切片。

This is not a vertical slice of it.

它也不是一个有代表性的样本，对吧？

It's not a representative sample, right?

因为我们拿不到像OpenAI或者大厂内部的使用数据，更别提那些超大规模云服务商了。

Because we don't have data to We're not privileged to data of, say, an open air and drop it internal data usage or the hyperscalers, for the matter.

我们确实和几家在接触，但目前数据有限，只能截取了其中一角，这一角明显偏向那些已经使用这些推理平台的独立开发者和中小型企业。

We are talking to a couple of them, but at the moment we are slight, so we sliced off a corner, and that corner tilts admittedly towards probably like independent developers or small, medium enterprises that are already on to some of these inference platforms, right?

他们对价格的敏感度，可能比你想象的还要高，比如那些大企业，现在全靠云服务商撑着，员工根本不管预算，直到季度末、半年末，CFO才突然问：等等，这怎么回事？

And they are going to be necessarily the ones that are more price sensitive than you would imagine, like a large enterprise that's currently living over the hyperscaler that maybe the employees are not even responsible for the they just go crazy until, like, end of the quarter, end of the half year, and then CFO say, wait a second, like, you know?

所以这会是一个先行指标。我六月初写过一篇帖子，说这事儿好像有点停滞了，使用量在调整，可能要先歇一歇，才会再往上冲。

So this is going to be a leading indicator, and I wrote a post at the beginning of June saying, like, This thing seems to be plateauing a little bit, and the use correction situation is such, and this thing could be taking a break before it makes a leg up.

也许我们会看到一个超强模型一出来，大家就全冲过去了；但也有可能，大家会慢慢理性起来，开始权衡成本和效果，重新在质量和价格之间做取舍。

Maybe we see like a super powerful model and swapping just runs away with it, or there's a possibility that this thing actually mean reverts a bit as people become more rational about the cost of these things and sort of substitute between quality and cost.

事实上，这确实发生了，而且碰巧赶上了股市的转折点——部分原因是，现在AI资本支出的融资模式本身就有问题。

Indeed, that seems to have been what happened, and they also unfortunately coincided with the turn of the stock market because I think partly rationally, because the current paradigm of how the AI CapEx trade has been funded, right?

所有东西都按二阶导数来定价。

And everything's being priced on second derivative.

就算增长只是慢一点，只要市场觉得增速在放缓，估值就会立刻崩盘，你懂的。

So even if things keep going up just like a slower pace, things are going to sell off just based on the valuation service, you know, that, right?

但根本问题在于，如果市场普遍觉得AI支出正在理性化，而且前沿模型的利润可能被那些强大、便宜得多的开源模型，或者像Grok、MuseSpark这类非开源但性价比极高的模型冲击，那当前AI资本支出的融资逻辑就可能被动摇。

But the fundamental is such that if there is a perceived slowdown on rationalization, and there's a perception that the frontier models could see their margins challenged with very powerful, much, much cheaper open waste models or even just sort of not open, but capable, but cheaper models coming from Grok and MuseSpark and whatever, that this could actually challenge the current paradigm of how the AI CapEx is financed.

我想，这正是市场最近下跌的部分原因。

And I think this is part of the reason why we've seen markets selling off.

如果你看一下最新的图表，比如AI受益者指数或者SMH之类的，应该能明显看到那个转折点。

If you open over the latest chart with the AI sort of a beneficiaries index or the SMH or whatever, you probably see like that turning point.

但我们并不认为这是个长期看跌的趋势。

But we don't necessarily see this as a secular bearish index.

我觉得这中间其实没什么矛盾，也不是那种涨了就看空、跌了就看多的简单逻辑。

I don't think there is actually somehow a contradiction, and this is not one of those things where it goes up, it's bear bullish, goes down, it's very bullish either.

这其实是有条件的，对吧？

It's actually a little bit conditional, right?

就像你不会说通胀高就一定好，通胀低就一定好，对吧？

Just like you will never say, Oh, higher inflation is always good, Lower inflation is always good, right?

总有一天，我们会过渡到我之前说的那种新模式，对吧？

At some point, we're going to transition towards that paradigm I was talking about earlier, right?

那时候，AI的资金来源和需求会变得更广泛，模型的调度也会更智能，即使推理市场的规模急剧扩大，它在整体AI需求和算力中的占比也会大幅提升。

Where the funding of AI and the demand of AI becomes a lot more broad based, and, you know, you can have a lot more sort of smart routing of models, even as the inference market explodes in size, become a much, much bigger of the overall pie of AI demand and AI compute.

所以，从一个阶段过渡到另一个阶段，本来就会有点颠簸，对吧？

So this transition from one phase to another was always gonna be a bit rocky, right?

而且还有可能面临一种‘空档期’，一方面，当前模式下的资本支出和投资回报率并没有明显体现出来。

And even potentially runs in the risk of a bit of an air pocket where, you know, sort of on the one hand, the CapEx and ROI for the current paradigm doesn't show up quite enough.

另一方面，新的阶段——比如大型企业——要找到合适的流程并上手，也需要不少时间。

On the other hand, the new phase, like maybe the biggest enterprises takes them a while to find that workflow and pick up.

所以增长的步伐可能跟不上。

So the pace may not catch up.

你会发现，最快采纳并获得最高回报的往往是员工最少、流程最简单的小微企业，但就金额而言，它们在整个AI需求和算力市场里占比很小。

Like you will see the fastest adoption and most ROI show up in the smallest enterprise with the fewest employees and the least sort of frictional workflows, but those are going to be, by dollar amount, by dollar weighted, gonna be small.

对吧？

Right?

所以我们看不到整体生产率提升，部分原因也在这里。

So part of the reason why we don't see productivity increase in aggregate partly is also because of that.

对吧？

Right?

总之，我这番话有点啰嗦，都忘了你最初问的是什么了，但希望还能说清楚点。

So, anyway, that's kind of a long winded answer to I don't even remember your answer question was, but hopefully, it still made some sense.

是的。

Yeah.

不是。

No.

这非常有帮助。

That was super helpful.

我觉得里面确实有些。

I think there's some Yeah.

一些非常重要的细微差别。

Some some really important nuances in there.

不过，我说的这个观点是之前提过的，就是你看这个图表，三月初，差不多就是Mythos发布的时候，也正是那时，‘令牌最大化’这个概念开始在美国企业界流行起来。

But, yeah, I mean, I'll throw the proposition at you that was that was put forth, which is that, okay, if you look at the chart, you know, early March, which is around where when Mythos started to come out, it was also around the same time that this this idea of of of token maxing took hold of of US corporate America.

然后突然之间，大家都觉得，好吧。

And and suddenly, it's just like, okay.

就是令牌最大化，能用多少就用多少，尽量突破极限，之后这个图表就上去了。

You know, just token maxing, use as much as possible, use use the frontiers, and then that chart went higher.

然后我觉得，有些人从中得出的结论是——正如你提到的，这个指数可能有点用词不当，但大家把它理解为：企业对代币的真实需求确实上升了。

And then I think what some, you know, folks took away from, and I think to your point, you mentioned how, you know, partly it was just this this index, maybe it was a bit of a misnomer, but people took it as, okay, this is just actual demand for tokens from corporations.

接着，就在那些新闻标题开始出现的时候，情况发生了转变，比如优步一个月就花光了全年代币预算，大家一下子慌了。

And then you see the turn came forth right around the time where these news headlines started to come out of like, you know, Uber went through their entire year budget of token spend in like a month and everybody freaked out.

但你的意思是，这更多是替代效应，而不是纯粹的代币需求增加，对吧？

Then, but what you're saying is this more so about substitution than outright token demand, right?

你的意思是，企业正从‘代币最大化’转向‘代币效率化’。

Like what you're saying is that this is about going from Token Max into token token efficiency.

并不是说六月之后对代币的需求下降了。

It's not saying that demand for tokens has gone lower since June.

我们只是更喜欢用它。

We just love Yeah.

你能再详细说说吗？

Can you can you expand on that further?

当然可以。

Yeah, absolutely.

我觉得，所谓的‘Token Maxing’这个想法，本来就是为了让人们能自由尝试，对吧？

Like, I think, you know, this idea of Token Maxing was always the idea you want to actually let people experiment, right?

因为在复杂的流程和大型企业里，你根本没法提前知道具体会用在哪儿，而且直到今年，随着智能体技术的发展，模型才真正变得够聪明、够好用。

Because in complex workflows, large enterprises, you don't really know what the use case is gonna be, and also the models just were not didn't really become smart enough or good enough until this year with agentic and everything.

所以你需要Token Maxing，但这种模式本质上总是伴随着高昂的token成本。

So you need token maxing, but token maxing is fundamentally always with expensive tokens.

token太贵了，而且还受限，所以最终你必须让token变得更便宜。

Tokens are so expensive and also throttled, so you need eventually to get to a state where tokens are cheaper.

而智能调度模型这个思路，是不是一直都是正确的方向？

And this idea of smartly routing models, that's always going to be the right?

你看我以前的公司，彭博社发过一篇很不错的论文，讲他们怎么让模型跟数据互动，准确调用信息，但你根本负担不起那种高损耗的做法。

If you look at like my old employer, you know, Bloomberg put out this very nice paper on MCP of, you know, how they let the models interact with the data, you know, and correctly call it accurately with, you know, financially that you don't really kind of really afford, you know, high loss nation.

你看看Palantir那边的情况，还有他们怎么通过本体论让AI在企业环境中真正派上用场，Databricks也是同样的路子，对吧？

You see what's been happening at Palantir, right, or this idea of ontology of actually making the AI actually useful at the enterprise context and Databricks, same thing, right?

越来越多像开放路由这样的做法，早就不只是那种‘长期免费路由’的业余实验了，以前是开发者想试试不同模型，但现在，这种模式其实已经成了行业标准。

And more and more of this, you know, thing like open routing has turned from, you know, sort of open router will give you this sort of routing for a longer, as long as time, and that seems to be more of like a hobbyist, you know, developers trying to try different models, but I think that paradigm has actually become now the norm, right?

以后这种情况会越来越普遍，对吧？

Increasingly, should expect that, right?

说到可持续性和投资回报，我其实不太想用这个词，太老生常谈了，但说白了，关键在于怎么让企业真正用起来。

And this is kind of like to the extent, I hate to use the term now because it's so cliche, but like sort of the paradox, max, Basically you in the fuller context of it, ultimately how this thing can be truly sustainable and actually get an ROI is to actually get enterprise adoption, right?

成千上万家企业。

Millions of companies.

只有当AI不会把公司资源耗干的时候，才能实现真正的长期应用，不能整个公司都依赖一个模型，那根本行不通。

And the only way that you can actually get sustainable adoption of AI is when you don't suck dry the underlying company, and we just have like a single model that the entire company is built on top of, and that was not going to happen.

现在更不可能了，对吧？

It certainly is not going happen now, right?

相反，模型层必须能灵活替换——比如，如果只是数据清洗、选分析指标、做简单总结这类任务，完全可以用便宜得多的非前沿模型来处理，那就该用那些。

Instead, basically, the model layer has got to be substitutable where you can smartly substitute between, if I have a simple task, say just data cleaning or just sort of choosing which analytics or similar summarization or whatever, things that can be easily handled with much cheaper, less than frontier models, they should go to those things, right?

如果是需要更强推理能力的高价值任务，那就该交给更强大的模型。

And if I have a higher value to ask, some things that requires more powerful reasoning, they should all go to those, right?

能智能地在不同模型间切换，同时准确评估质量和成本，这事迟早得做，但现在做得还不够，难点在于怎么判断一个请求到底属于哪一类。

And the ability to route between them and correctly assess quality and cost is something that will have to happen, and it's not really been happening enough and is a challenging thing because just to figure out whether or not a certain ask is which type, right?

这并不总是显而易见的，对吧？

It's not always obvious, right?

有时候一个问题看起来简单得不得了，但实际上背后藏着很深的学问，只是提问的人自己也没想清楚要问什么。

Sometimes a question can seem benignly innocuously simple, but in reality, it's actually a pretty complex question that someone is just asked poorly because they don't really know what they are looking for.

我觉得这会成为新的挑战，也正因如此，你才会看到越来越多这类路由机制被提出来，比如Cell的方案，或者Ramp也在搞类似的东西，等等。

And I think that is going to be the new challenge, that's part of the reason why you are seeing more and more of these sort of routing things being activated, sort of the cells of the world or ramp is coming up with one and so on and so forth.

我觉得，这以后就会成为常态了。

So this is going to be the new norm, I think.

这真是太有帮助了。

That's super helpful.

好的，Steve。

All right, Steve.

你是个经济学家。

So you're an economist.

我想问你一个问题，这跟我们刚才聊的指数有点关系——在我看来，这个框架的关键在于，当这张图开始飙升的时候，Anthropic的ARR也在猛增，是不是意味着前沿模型这时候能独享大部分利润？

I want to ask you about this, but it feels like one of the big questions right now that sort of leads into what we just talked about with this index is this idea of, you know, feels to me like this framework is either the frontier models get to enjoy the margin like right around this time where this chart was surging was also the time where the ARR of Anthropic was surging.

而且你知道吗，加文·贝克提到过，有一个月Anthropic首次实现了盈利。

And, you know, Gavin Baker mentioned that there was even a month where Anthropic was like, was profitable for the first time.

这当然很不错。

So that's all great.

但现在我们看到趋势转向了令牌效率。

But now we're seeing the substitution towards token efficiency.

我们开始看到一些开源模型变得强大得多。

We're starting to see these open way models that are looking a lot more powerful.

我们正逐渐进入这样一个世界：前沿模型负责顶层协调，

We're starting to move to this world of, you know, perhaps frontier model, top line orchestration.

而大量的下游执行则由这些更高效、更便宜的开源模型来完成。

But a lot of the downstream execution is is these more efficient, cheaper, you know, open way models.

这感觉对终端用户来说真的很有利。

Like it feels like that is really good for for the end consumer.

好处最终都落到了用户身上。

Like, the surplus goes towards them.

所以感觉现在存在一种拉锯：前沿模型在追求利润率的同时，还得靠盈利来支撑一切，但好处又该更多地流向消费者。

So it feels like there's, this fight for margin between the the the frontier models in that AR and then being profitable and, everything that is almost dependent on that occurring versus the benefit going towards the consumer.

我很好奇，你是怎么看待这两者之间的平衡的？

So I'm just curious, like, what's your framework for how to how that balances out?

这两者能双赢吗？你觉得呢？

Like, can the both sides win, or or what do you think?

是的。

Yeah.

这就像你作为经济学家问我一样。

I mean, this is exactly like you asked me as an economist.

对吧？

Right?

经济学家通常会想，局部均衡和一般均衡这些事。

You know, economists like to think about, you know, partial equilibrium equilibrium and general equilibrium.

对吧？

Right?

那什么是局部均衡呢？

So what is partial equilibrium?

局部均衡就是当价格出现一个简单的波动时，对吧？

Partial equilibrium is when you just have like a simple shock to price, right?

价格减去成本再乘以销量，就是利润，对吧？

Price times minus cost times quantity gives you profit, right?

成本呢，就是计算和运营那些开销，如果价格被竞争对手压低了，你的销量不会变，但利润就会缩水。

So cost is kind of whatever, you know, cost of compute and operating, and if price is challenged by a competitor, you know, your quantity doesn't change, you know, you're going to see a shrinkage of the overall profit.

这会让你担心投资回报率之类的问题。

And that makes you worried about, you know, sort of ROI and so on and so forth.

但销量不是固定的，对吧？

But queue is not static, right?

这正是夏芒悖论的核心——需求是有弹性的，对吧？

You know, that's the whole point of Xiamang's paradox is there is elasticity, right?

如果整个市场变大了，比如现在企业采用率还很低，对吧？

And if the overall market has become much bigger, you know, like right now the adoption, enterprise adoptions are business small, right?

大多数人用AI，还只是停留在比普通搜索高级一点的层面，企业那边也还在摸索。

Most people are not using AI in the consumer context beyond like sort of glorified search, and enterprises are still trying to figure it out.

就算他们刚开始搞明白，心里还是担心成本问题。

And they are nervous about the costs, even as they just figure it out.

但如果你告诉我，前沿模型的利润空间可能会被压缩一点，但需求会暴涨得多得多，那整体还是划算的，对吧？

Now, if you tell me that, okay, the margin of maybe some of the frontier model is going to get challenged a little bit, but the queue is going to grow so much, much more, the overall thing still works out, right?

我常举个例子，这跟制药行业有点像，虽然不完全一样。

You know, the analogy I like to give a little bit, sometimes it's a bit like, this is not totally unlike like the drug market, right?

大家总在问，要是别人一推出仿制药，你还能靠什么创新呢？

You know, so people always ask like, who is going to innovate on pharmaceuticals if you just get generic drugs that sort of copy you and whatever.

确实会有一些利润被侵蚀的动态，对吧？

So there's going be some dynamic of erosion there, right?

有竞争对手在，你肯定拿不到垄断性的高额利润。

You know, you're probably not going to earn like full monopolist profits, you know, if you have competitors.

但同时，我也觉得这事儿其实没真发生过。

At the same time, I also don't think it's been true, right?

不管你看看软件、制药，还是别的领域，只要存在竞争，有模仿者之类的，不代表前沿领域就不会发展，也不代表领先者就赚不到钱。

Whether you look at software or look at pharmaceutical or whatever, just because there is going to be competition and there's going to be some degree of, you know, copycats or what have you, it doesn't mean that the frontier is not going to grow, and it doesn't mean that the leaders of the frontier will not be profitable.

不过呢，我觉得这确实会引发一个问题：最终的价值到底会落在谁手里？

That being said, I do think this is going to challenge the question, where the ultimate bulk of the value, you know, is going to accrue to, right?

我觉得说价值可能会从前沿AI实验室的双头垄断模式，更多地流向计算层或用户层，这个观点并不算太有争议，具体流向哪一边，还得看计算层最终怎么发展。

And, you know, I don't think it's going to be super controversial to think that at least direction on the margin, this means that some of the value that was perceived to have been certainly accrued to this duopolous model of frontier intelligence labs, probably gonna more accrue to either the compute layer or the user layer, depending on how the compute layer sort of settles out, right?

有道理。

Makes sense.

明白了。

Alright.

这正好可以转到聊聊新云和GPU租赁指数的话题。

That's a good segue into talking about neo clouds and GPU rental index.

这是你和团队推出的另一个指数。

So this is another index from from you and your team.

对。

Mhmm.

能给我们讲讲该怎么理解这个指标吗？它对当前的格局意味着什么？

Walk us through how to think about this and and, yeah, what does this what does it say for the current landscape?

是的。

Yeah.

我们实际上在追踪并构建这些租赁指数。

So, basically, we track, you know, and create these, you know, rental indices.

这有点像一个基准。

It's almost like a benchmark.

你可以想象一下，比如我想建一个纽约市无家具单间公寓的指数，就得收集全市不同地段、不同条件的租赁合同数据，对吧？

You think of something an an analogous like, you know, if I wanna create an unfurnished single bedroom apartment index in New York City, you know, you get ton you know, you observe rental contracts from across the city, a different term, in different shape and form, right?

有的带家具，有的不带，房间大小也不一样，诸如此类。

Some come furnished, some come unfurnished with a little bit of room and so on and so forth.

我们把这些数据全部汇总起来，用机器学习做标准化处理，把各种差异都对齐，最终为每种芯片生成一个统一的指数。

We aggregate them all, use machine learning to essentially create apples to apples comparison, right, and create a singular index for each chip out of all the data we look at.

这里这条蓝线是H100芯片的指数，而这条是A100，就是五年前就发布了的那款老芯片。

And here we have the blue line, that's H100 chip hopper, right, and then you've got A100, which is the ancient chip that's been announced for like five years or something, A100.

还有V200，就是这条橙线，波动更大，而且更新，部署得也少，还有H200。

And then you've got the V200, which is the orange line, is more volatile and also newer, less deployed, and also the H200.

我们在这方面的历史数据其实更久，但不知为什么，只保留了最近几周的品牌数据。

We have a longer history about here, only for some reason, only managed the brand of the most recent few weeks.

所以我的看法是，那些刚接触AI市场的人，往往只看H100这个指数，其实就是按需定价，对吧？

So the way I look at it is that, you know, amateur who follow the AI market, they just look at the H100, you know, sort of index, which is really actually the on demand price, right?

但这并不是完整的远期曲线。

This is not the entire forward curve, right?

它并不能告诉你一年后一小时GPU的使用成本会是多少。

This is not telling you like what, how much it will cost a GPU hour out, a year out or whatever.

他们一看这条蓝线涨了，就说需求上来了；跌了，就以为需求崩了。

They look at this blue line, if it goes up, say, oh, demand is going up, it's going down, like, oh, demand must be collapsing.

但事实上，我总是习惯先看横向对比，等有时间再进一步分析不同期限的结构变化。

But in reality, I always like to both look in the cross section, and then later, when we get time to also look at across the term structure.

因为在横向对比中，推理需求和训练需求之间的变化，尤其是推理需求快速增长，其实能明显看出工作负载的转移——不同实验室、不同团队对算力的需求类型不同，这些变化都能看出来。

Because in the cross section, what happens is that with between inference demand and training, especially with inference growing so much, you can actually see the shifting of workflows, and depending on different labs and sort of who's needing, seeing, demanding, what kind of thing, you can actually see shifts.

比如，今年年初以来，H100的价格一直稳步上涨，最近虽然有小幅回落，但你发现紫色线、黄色线和橙色线反而继续大幅上涨，对吧？

So for example, when H100 has been steadily going up right since the beginning of this year, and it has had some, you know, sort of fluctuations recently came down, when it came down, what you see is that the purple line and the yellow line or the orange line continue to go up a lot, right?

与此同时，如果你看A100，它的价格其实一直很稳定，对吧？

So meanwhile, if you look at the A100, you know, that's actually very steadily sort of holding, right?

根本就没往下掉。

It's not even coming down at all.

这种情况是完全说得通的。

You can have a scenario which, you know, this is consistent.

我不是说这一定就是真相，但你可以设想一种情况：推理需求暴涨，增长得特别快，对吧？

I'm not saying this is sure, but you can have a scenario where inference demand is just through the roof, growing like crazy, right?

这甚至推高了五年前的A100芯片的租赁价格，而且需求还很强劲。

And that's driving up rental rate for even A100 chip, like, you know, from five years ago, and strong, right?

同时，训练任务的工作流可能正从H100转向更新的芯片，对吧？

Meanwhile, you can have some of the workflow for training shifting away from H100 towards the newer chips, right?

等这些新芯片陆续上线、供应跟上，价格就会有些波动。

And as those chips come online and depend availability, so those things can fluctuate a little bit.

所以至少从按需指数来看，我是这么理解的，

So that's how I read at least the on demand indices for,

是的。

yeah.

太好了。

Awesome.

我觉得大家之所以这么关注这个，是因为现在世界上最大的公司之一就是英伟达。

I feel like people look really closely at this because of course, one one of the largest companies in the world right now is NVIDIA.

所以显卡的需求和纳斯达克指数以及整个股市的表现密切相关。

So obviously demand for the GPUs is very correlated with the performance of NASDAQ and the total stock market.

因此，如果市场对显卡在基础设施建设中的需求有任何担忧，都可能在整个系统中引发不小的震荡。

So obviously, you know, if there's any sort of concern for demand for GPUs in the build out, that could have some pretty significant shock waves throughout the system.

那么，我们能不能从目前这个行业里的现象中，看出一些端倪，来判断显卡需求到底处于什么水平，是已经见顶了，还是增长势头正在变化？

So is there anything that we can take away from what we're seeing right now in this industry to, you know, interpolate where GPU demand is at and whether like the build out is peaking out or whether the growth rates are changing there.

你那边看到什么情况？

What, yeah, what do you see there?

我们还可以从远期曲线中得到更清晰的信号，我觉得接下来我们可能会谈到这个。在这张图上，我最关注的其实是绿色线，而不是蓝色线。

So we can get an even cleaner read from the forward curve, which I think we hope we can come to next, I think on this chart, I actually, the favorite thing I look at on that, this chart is actually the green line and not so much the blue line.

你看看这个价格有多强，对吧？

Like, just look at how strong that price is, right?

年初的时候大家还老说，像迈克尔就一直讲，芯片这种东西最多就用三年，是快速折旧的资产，诸如此类的话。

Like people, there was all this talk about early in the year, like Michael very like, oh, chips are like only good for three, three years, fast depreciating asset, blah, blah, blah.

但事实上，就连A100芯片的租赁价格不光没跌，反而还在涨，一直稳在高位，对吧？

In reality, even the A100 chip rental rate is going up and has stayed up, has not come down at all, right?

这说明推理需求有多强劲。

That just tells you just how robust inference demand is.

再回到我们之前说的，关于不同工作负载的分配问题，其实有很多低垂的果实——那些简单、容易实现的推理任务，需求量非常大。

And going back to what we saying earlier about routing different workflows and whatever, there's just a lot of, I think, low hanging fruit, easier asked tasks, inference demand.

随着整体需求增长，大家对AI越来越熟悉，很多这类任务都自然而然地流向了那些主力芯片。

As this overall demand, you know, grows, people become familiar with AI, a lot of those things are being routed towards the workhorse chips, right?

所以，如果未来最广泛部署的Hopper芯片，虽然不再是最强的训练芯片，却成了新的主力机型，我一点都不会觉得意外。

So I wouldn't be surprised if all the most widely deployed hopper chips, which will soon become no longer the most powerful chip for training and will actually become a new workhorse, right?

所以这里也会出现类似的模式。

So you will see sort of a similar pattern to follow here.

这其实是个不错的信号，但咱们也可以看看远期曲线，对吧？

So that's actually a good signal, but then we can also look at the forward curve, right?

是的，没错。

Yeah, yeah.

你给我们讲讲这个怎么理解吧。

Walk us through there and how to understand and yeah.

简单说，远期曲线，我的理解是这样的，对吧？

So basically, you know, ForwardCourse, the way I think it was the following, right?

如果你在运行一块GPU，一些常见的用法能持续相当长的时间。

So if you are running a GPU, you know, some of these common tricks can run for a fairly long time, right?

你可以用一个月，三个月，甚至两三年都没问题。

You know, you can run it for a month, you can run it for three months, you can run it for as long as two, three years.

一般来说，任何商品曲线都应该是向下倾斜的，也就是所谓的贴水，你知道的，对吧？

And, typically, with any commodities curve, you expect a downward sloping, like, you know, backwardation or, you know, that's the jargon, right?

曲线向下倾斜，是因为预期供应会陆续增加。

So downward sloping curve, like, you because you expect supply to come online.

举个例子，如果你在纽约租一间公寓住满十二个月，每天的租金通常会比住三天酒店便宜。

In this case, like another example you can imagine is like, if you run an apartment in New York or anywhere for twelve months, your per day rental rate is probably going to be lower than if you run to the hotel room for three days, right?

为什么呢？

Why?

因为你避开了房东频繁换租客的麻烦，所以能拿到折扣。

Because you get a discount for avoiding the trouble for landlord, the turnover tenancy, right?

你一次性锁定长期租约，就不用老是去找下一个房客了。

You sort of lock in that thing, and it's like, doesn't have to just keep looking for the next renter.

确实，我们看到从去年11月22日开始，也就是生成式AI刚兴起、Opus模型上线的时候，曲线明显呈深度贴水、向下倾斜的状态。

And indeed, that's what we saw, like the blue line here from November twenty second last year, just when the agentic AI was sort of kicking off, the Opus model becoming online, the curve was very backwardated, downward sloping.

从那以后，直到最近几周，我们看到很多曲线都发生了变化，我来给你看看趋势。

And since then, like between November to like, this is a lot of curves, want to show you sort of how things have been involved in the last few weeks.

挺有意思的，对吧？

It's interesting, right?

这里的橙色线是三月底的情况，也就是3月31日，整个曲线实际上已经整体上移了，对吧？

The orange line here is the at the end of March, right, March 31, The entire curve has actually moved upward, right?

换句话说，每个合约期限的每GPU小时价格都上涨了。

In other words, every maturity, every length of the contract per GPU hour price has gone up.

不仅如此，长期端的曲线还隐约出现了轻微的正向市场结构，对吧？

Not only that, the curve has also become like on the long end, almost seemingly a little bit in contango, right?

也就是说，不再像以前那样呈向下倾斜了。

Like, sort of no longer downward sloping.

这意味着云服务商、租户这些云平台的‘房东’们现在觉得没必要再给长期合约打折了，宁愿让短期合约自动续期，好抓住机会再次提价。

What that means is that cloud providers, renters, the landlords of clouds are feeling comfortable not giving these long term contract discounts and just letting short term contracts roll over so that they have an opportunity to raise price again.

事实上，我们也从和云服务商的非正式交流中听到类似说法，他们确实在积极寻找各种机会涨价。

And indeed, you know, that's what we've been hearing from anecdotal evidence with anecdotal conversation we have had with cloud providers, that they really are very much aggressively looking for opportunities to raise prices.

当然，还是有人想要确定性，所以这才有了这个市场，对吧？

Now some still want that certainty, you know, so this is why there's a market, right?

之前我们聊过对冲，你知道的，不同人有不同的看法，对吧？

Earlier, we were talking about hedging, like, you know, people have different views, right?

如果你手头有东西，比如种点庄稼，或者产点石油，觉得价格只会一路涨，那根本没必要对冲，对吧？

If you have a thing, maybe you grow some crops, maybe you, you know, grow some oil, if you feel like prices can just go straight up, you don't hedge at all, right?

有意思的是，从三月到六月，你看这条紫线，那是六月二十五号的数据，当时市场确实有点恐慌，算力供应紧张，前端的H100价格稍微跌了一点，大家有点慌，但你再看一年期的价位，那些重要的计算任务合约，一年期的价格基本没动。

So what's interesting here is that, you know, between March and June, you know, if you look at the purple line, that's like June 25, like, you know, there was a bit of a scare in the market, you know, air compute capacity, you know, like basically on the front end, H100 came down a little bit and people freak out a little bit, but then you look at the one year mark, there's sort of a lot of the meaningful compute workflows that actually contracts that, like the one year term, that's pretty much unchanged.

更有趣的是，接下来几周，也就是上个月，我们看到一堆新闻，比如Meta在出租算力，各种模型上线，大家开始从前沿模型转向其他替代方案——这些消息都在暗示AI算力的供给可能过剩了，对吧？

And what's even more interesting is that in the ensuing weeks right, over the last month, you we have all this news about, you know, Meta leasing out compute and all these, you know, models and you're substituting away from the frontier models, like basically a bunch of news that will be indicative of AI, you know, supply maybe potentially being in excess, right?

因为市场估值太高了，大家总想找点理由来恐慌。

People, because the market valuation is so high, people are looking for every reason to freak out.

但从基本面来看，至少在算力方面，秋季的一年期租赁价格其实是一直在稳步上涨的，对吧？

What we see in the fundamentals, at least in terms of compute, is that rental prices at the one year mark for the fall, you know, fall rate is actually monotonically going up, right?

有两条线几乎重叠在一起，因为七月二号到七月十四号之间变化不大。

You know, there are two lines that overlap each other because there's not much change, like between second of July and July 14.

最近我们看昨天——七月二十二号，也就是二十号那天，绿色的线又往上蹿了一大截，对吧？

And then most recently, when we look at yesterday, you know, July twenty second, twentieth, the green line went up even more, right?

每次看到这种情况，都有好几个供应商同时在提价。

Every single time we saw multiple providers actually raising prices.

换句话说，这可不是偶然现象，至少从计算能力的基本面来看，需求强劲、供给紧张的迹象非常明显，价格自然得调整。

In other words, it's not just like a one off thing, you know, at least in terms of compute fundamentals, right, there is actually a pretty sort of strong indication of firming demand and supply, you know, being being being, you know, in shortage, therefore the price has to adjust.

你明白吧？

You know?

是的。

Mhmm.

太好了。

Awesome.

关于其他GPU曲线，你还有什么要补充的吗？比如怎么区分它们？

Anything you wanna add on the other GPU curves here in terms of distinguishing?

没有。

No.

我的意思是，情况差不多。

I mean, I think pretty much similar.

你看右边的图，A100也是长期稳步上涨，情况如出一辙。

You know, like, I think you see the similar picture on the right, you know, sort of just the a 100 and and is, you know, sort of secularly going up.

对吧？

Right?

最近几周可能没那么猛了，但整体还是挺坚挺的。

And, you know, maybe not quite as aggressively, you know, more recent weeks, but generally holding up pretty strongly.

B200的情况也总体在上升，不过有点波动。

And then the B200, you know, is also generally moving up, but then there's a bit of a fluctuation.

B200的情况是，从芯片整体动态来看，它还在持续部署中。

You know, so the thing with B200 and although I think just think about the chip dynamics generally is that B200 is still being deployed.

大部分数据中心还在陆续上线这些芯片。

Most of the data centers are still just bringing them online.

B200的供货量还没达到H100的水平，所以会有些波动，但总体来看，趋势方向是一致的。

Availability is not quite as high as the H100, so there's going be some sort of fluctuations, but generally speaking, I think the picture that emerges qualitatively is very much in agreement.

完全同意。

Totally.

我们之前主要聊的是GPU，但现在越来越多的AI基础设施建设，其实也体现在DRAM和内存方面。

Obviously we've been talking mostly about GPUs, but increasingly a lot of where the the change in in in, you know, the build out for for all this AI is also on the the DRAM side of things and the memory aspect.

我很好奇，你是怎么看这一块的？市场上这边的情况跟GPU这边相比怎么样？

I'm curious, how do you think about that and and and what's going on in that side of the market, compared to the

GPU吗？

GPUs?

显而易见，内存是GPU运行的基础，现在这些模型对内存的需求简直大得惊人。

Clearly, memory is input, right, into GPUs, and all these models are super memory hungry.

一般来说，对话越长、上下文越多，所需内存也就跟着增加。

Generally speaking, with longer context, longer conversations, the context grows with memory.

这也是为什么内存和存储需求暴涨——需求远远赶不上供应，价格自然一路飙升，这其实一点都不意外。

That's part of reason that's clearly the reason why, you know, memory as well as storage, you generate tons and tons of data that all just demand is not catching up with supply is not catching up with demand, And it's not surprising that price is shooting up the way it is.

但我们也知道，短缺和高价从来都是推动创新的催化剂。

But we also know that, you know, shortage and, you know, high prices are always the model of innovation.

所以毫不意外，我们现在看到很多算法上的突破，比如中国最近推出的QEMI这些模型，都在提升内存效率，让内存需求不再跟上下文长度成线性增长，这其实正是社区模型带来的那种KDA式创新的一部分。

So unsurprisingly, we're seeing algorithmic innovation, no less from the recent Chinese models like QEMI and so on that are actually making improvements to memory efficiency, and maybe the memory demand doesn't have to quite grow linearly with you know, context length, which just seems to be, I think part of the, there's what is sort of KDA, you know, innovation, you know, that came with the community model.

我觉得这没什么好惊讶的。

I think it's not surprise.

在我看来，社区这次的突破，感觉跟DeepSeek那次特别像。

To me, the community, you know, moment felt very similar to the deep seek moment.

我觉得，类似这样的效率提升还会持续出现。

Like we're going keep on having, I think efficiency improvements like that.

说实话，要是两年后内存厂商的毛利率还维持在85%左右，那我真会很惊讶。

I mean, clearly it's not going to be like, I will be very, very shocked if let's say in two years, the gross margin of the memory makers are still running at like 85%, you know, or whatever.

这其实就是一种压力下的挤压效应，对吧？

It's just like one of those things is a sort of a squeeze, right?

我觉得等熬过这段，还是会回到P乘以Q那个问题上。

So I think, you know, we get past it, you know, and, you know, I think there will be Again, it goes back to the question of P times Q, right?

我觉得价格可能会降一点，但需求量增长得足够快，大家依然能赚得不错。

Sort of, I think P is going come down some, the q product grows enough that, you know, everyone still, I think, do very well.

是的。

Yeah.

对。

Yeah.

关于Qiwi这件事，还挺有意思的。

It's it's interesting that the Qiwi thing yeah.

当我们讨论这些时，显然涉及需求层面，也就是模型的效率问题。

Like, there's obviously when we talk about these these there's the demand aspect in terms of just efficiency of the models.

如果效率提升了，比如KV缓存这些方面优化了，就不需要那么多内存了。

Like if, yeah, like if you don't need as much, if it's more efficient in terms of like KV cash and that whole thing, you won't need quite as much memory.

另一方面，还有供应端的问题，比如会不会有更多产能释放。

Then there's also the supply aspect of of whether we start to see more build outs.

听说中国生产的内存可能开始进入市场，这类消息也越来越多。

You know, there start to be talk of availability of of memory from China coming onto the market and that sort of thing.

但听起来，不管怎样，贾万悖论和这些观点依然成立。

But it sounds like, yeah, regardless of that, you know, Javon's paradox and those ideas still hold true.

不管这些细微变化如何，确实，当内存股已经涨了这么多，任何一点风吹草动都会让人觉得特别敏感、紧张。

And regardless of these, you know, marginal change, obviously, it's it can feel especially volatile and sensitive when you have these, you know, memory equities that have just ran like they have.

哪怕只是轻微的变动，加上市场里那么多杠杆仓位，短期来看感觉特别剧烈。但你的意思是，如果拉长视角看，这些其实都是边缘上的小调整。

Like, any sort of marginal shift and just with the amount of, like, leverage positioning, it feels very intense in the short But what you're saying is that regardless of that, like, you zoom out a little bit, you know, these are these are pretty small changes on the margin.

是的，说实话，这些内存厂商的股价到底怎么样，谁也说不准，对吧？

Yeah, I mean, I couldn't tell you, you know, in terms of sort of the stock price of these memory makers, like nobody knows, right?

现在所有价格都基于第二甚至第三层的预期了，但在我看来，笔记本电脑对内存的需求根本不会减弱。

Everything's being priced on second or third derivatives at this point, but it doesn't seem to me at all that there's going to be any laptop on the demand for memory.

我完全预料到可能会有意外的上涨，因为每次出现新突破，人们似乎都更愿意用这些模型做更多高效的事。

Fully expect to be surprised at the upside, because every single time we have had a new innovation that seems to only increase essentially people's willingness to try to use these models in more productive ways.

每当你这么做，就会产生更长的上下文，也会生成更多需要存储的数据。

Every time you do that, you generate longer context, you generate more data to be stored.

这不只是内存的问题，还涉及数据的存储。

By the it's not just memory, it's also storage of data.

我们还没谈到多模态，比如语音交互。我觉得过去几个月里，大家其实忽略了ShareGPT Live推出的那个非常出色的对话式AI，你可以和它自然对话，还能互相打断。

And we haven't even sort of gotten into the multimodal stuff with the voice, you know, like, I think, like for example, over the last couple of months, one thing that people really, I think, overlooked was that, know, ShareGPT Live, right, came up with this very, very good conversational sort of AI, right, where you're able to talk to it and you can interrupt each other.

现在AI甚至能主动打断你了，对吧？

AI actually is able to interrupt you now, right?

你有了一个特别强大的方式来和AI互动，比如让它调用代理帮你完成各种任务。

You have this really powerful way of engaging with the AI where you can ask AI to go and use agents to do things on your behalf.

你可以对着屏幕上的内容，直接用语音来操作。

You can interact with what you're looking at on the screen and just use voice to do it.

这意味着数据量会大得多，对吧？

And that also means it's a ton more data, right?

我们连视频的边儿都还没摸到呢，对吧？

So every time you have We haven't even touched, scratched the surface of video, right?

说白了，内存归根结底就是个大宗商品。

Like, you know, I just think like, you know, every commodity, like memory is ultimately a commodity.

我根本不在乎。

Like, don't care.

这确实是个超级周期，但大宗商品的超级周期也可能变得非常疯狂。

It is a super cycle, but super cycle of a commodity can, you know, also go pretty crazy.

所以，确实如此。

So, yeah.

这就是我看待这个问题的方式。

So that's sort of how I think about it.