本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
大家好。
Hello, everyone.
欢迎收听新一期的《The Circuit》。
Welcome to another episode of The Circuit.
我是Ben Bajarin。
I am Ben Bajarin.
各位程序，你们好。
Greetings, programs.
我是Jay Goldberg。
I'm Jay Goldberg.
嗯，我想任何长期收听我们播客的听众都注意到了，过去几年我们开始更多地讨论网络。
Well, we, as I think any long long term listener of this, of this podcast has noted, we we start talking a lot more about networking the last few years.
我们从第一天起就开玩笑打趣。
We have joked and jested since day one.
你们这些从一开始就在的人都知道，Ben从最初觉得网络很无聊、不想谈，到现在认为伟大的互联网络和收发器是让AI变得有趣的关键。
You've seen, all heard, who's been here from the beginning, Ben's evolution from networking stupid, I don't wanna talk about it, it's boring to now it's the great inter interconnected networking and transceivers is is the glue that makes AI interesting.
不过说到这个，光学相关的讨论最近有很多进展，我们之前聊光学聊得不多。
But in in that vein, a lot has been going on around this optical conversation, which we haven't spent a ton talking about optics.
我们俩都写过相关文章，而且我知道你我都各自做了不少独立研究。
We have both written about it, and I know you and I have both independently researched this quite a bit.
但我觉得现在是时候更深入地聊聊网络了。
But it's time to double click, I think, little bit more on networking.
我们会从这周Coherent和Lumentum的财报说起。
And we're gonna lead into that through both Coherent and Lumentum's earnings this week that happened.
两家公司的态度和反应显然不一样。
Different tones, obviously different reactions.
所以你先说说这两家吧，然后我们再深入探讨网络和光学领域的一些影响和观察。
So I'll let I'll let you start with the both of those and then we'll dig into some of the implications and observations around the networking and optical environment.
好的。
Sure.
Lumentum和Coherent这周都发布了财报。
Lumentum and Coherent both reported this week.
我手头没有确切的数字。
I don't have the exact numbers in front of me.
Lumentum全面超预期，还上调了指引。
Lumentum beat across the board, guided up.
Coherent的收入指引不错，但在利润率和其他方面有些吃力。
Coherent was a revenue guide was good, but struggled with margins and a few other things.
这周财报出来后，两家公司的股价表现差异很大。
And there is a a big divergence in their their stock prices this week following those results.
我觉得有意思的是，这两家都深度参与了AI和光模块的建设。
And I think I think it's interesting because both of them saw are are participating fully in in this AI build out, optical build out.
这个市场很大。
It's big.
我们聊过很多次了。
We've talked about it a lot.
所以虽然收入都不错，但情况并不完全一样。
So they have good revenue, but there is not all things are the equal.
他们实际卖的东西，细节里藏着很多门道，这从他们的利润率上就能看出来。
There's a lot of devil in the details of what they're actually selling, and that is conveyed through their margin profiles.
对。
Yeah.
我是说，我觉得，如果我说错了你纠正我。
I mean, I think the, correct me if I'm wrong.
我觉得大多数人都在关注他们的毛利率。
I think most people were pushing on kind of their gross margins.
你知道，两边电话会上都有一些有趣的讨论，关于两英寸晶圆，甚至追溯到两英寸，还有四英寸的。
You know, there was some interesting conversation on both of their calls about what was going on around both two inch wafers, you know, all the way back to two inch wafers, four inch as well.
我觉得，Lumentum说得挺清楚的。
I think, you know, Lumentum was pretty was pretty clear.
他们打算用那些两英寸的技术专利，也应用到NPO上，我觉得这让很多人挺高兴的。
They were gonna be able to use some of those IP from those two to also go into NPO, which I think made made people happy.
但就像行业里的陷阱链，你知道，转向六英寸晶圆，六英寸晶圆就是那种良率的问题。
But it's like the industry's trap chain you know, moving into six inch wafers and six inch wafers is sort of the yield.
对吧？
Right?
而且，当然，转向六英寸会有帮助，但你还是得从四英寸甚至两英寸的晶圆里尽可能榨出一切，对吧，根据Lumentum的说法。
And and and and sure, going to six inch is gonna help, but you still gotta, you know, squeeze everything you can out of your out of your four and even your two inch, right, from what Lumentum said.
对。
Right.
对。
Right.
所以半导体经济学101，晶圆的大小非常重要。
So semiconductor economics one zero one, the the size of your wafer matters a lot.
晶圆越大，能放的芯片就越多。
The more wafers the more chips the bigger the wafer, the more chips you can fit on it.
所以长期以来，数字逻辑一直在向300毫米晶圆过渡。
And so there's been a long history of moving towards 300 millimeter wafers for digital logic.
我们现在看到光学组件正在向六英寸晶圆过渡。
We're seeing this transition now to six inch wafers for optical components.
Lumentum 说，嘿。
And Lumentum is says, hey.
我们有四英寸的，而且做得非常好。
We've got four inch, and we're doing really well with that.
对。
Yep.
而 Coherent，我觉得他们可能觉得自己落后了，所以非常努力地推进六英寸晶圆，理论上这应该能带来好得多的毛利率，因为同样的晶圆，能切出更多的芯片，但实际情况并不是这样。
And Coherent, sort of, I guess, perceiving that they're behind, is pushing really hard into six inch wafers, which theoretically should offer much, much better gross margins because, again, for the same wafer, can get a lot more chips out of it, but that's not actually playing out.
关于具体原因，有很多争论。
And there's lots of lots of debate about why that is exactly.
我觉得要记住的是，六英寸的磷化铟晶圆是全新的东西。
And I think the the thing to keep in mind is that six inch indium phosphate wafers indium phosphate wafers are are brand new.
它的生态系统还不成熟。
It's it is not a mature ecosystem.
而且这些东西需要很多年才能发展起来。
It's and these things take years to develop.
所以你得更换工具，还有工具里的很多夹具，以及晶圆本身。实际上，六英寸磷化铟晶圆并没有好的供应商，只有一家，而且他们也是新手。
And so from you gotta change the the tools and a lot of the fixtures within the tools and the the wafers themselves, there's not actually a good supplier of there's only one supplier of indium phosphide six inch wafers, and that's and they're very new to it too.
所以看起来有些成长的烦恼。
So seems to be some growing pains there.
我的意思是，这不是什么大不了的事，这是正常的。
I mean, it's not it's not you know, this is normal stuff.
只是，如果这个技术再成熟几年，对Coherent来说会更好。
It's just, like, it would be better for Coherent if this were a few years more advanced.
完全同意。
100%.
而且我觉得，你知道，有两件事我一直在想。
And and I think, you know, what there's there's two things I I keep coming back to.
你知道，我在推特上和其他地方一直很直言不讳，而且未来几个月我们会有嘉宾来帮我们讨论光模块的时机、采用率和渗透率。
You know, I I have been very vocal on the Twitter sphere and others, and we will have guests in the upcoming months that can help us talk about the timing, adoption rate, attach of optical.
但是，我觉得有两件事正在发生，导致人们对所有这些光模块的渗透率过于乐观，当然，LPO除外，那是目前的标准。
But but there's two things that I think are going on, is leading to people to be a little bit too optimistic in the attach rate of all of this optical stuff, again, outside of LPO, which is sort of the standard today.
对吧？
Right?
那是个大插头。
It's a giant plug.
你把它插进去，它就在板子的边缘。
You go and stick it in, and it's at the edge of the board.
不是插进板子里。
It's not into the board.
对。
Right.
而这正是大多数人现在刚开始大规模采用的东西。
And and that's what most people are just starting to ramp, like, today.
比如，标准的光学方案就是LPO。
Like, your standard optical is is is LPO.
所以我们还没到NPO阶段呢。
So we're not even at NPO yet.
而NPO呢，就是下一代技术，把它放到电路板上，但还是没有直接集成到芯片里，而CPO的目标就是做到芯片里。
And NPO is, like, the next version that takes this onto the board, but it is still not, like, into the die, which is the goal of of of CPO.
再往后就是scale in，也就是放到基板上。
Then past that is is is scale in, which is onto the substrate.
对，没错。
Also so yeah.
所以我们得记住三个缩写，这就是……
So there's there's there's three acronyms we gotta And that's what
我想说的。
I wanted to do.
我们得把这个搞清楚。
We need to clear this up.
好的。
Okay.
所以有LPO，这是我们用了很久的技术。
So so there's LPO, which is what we've been using for a long time.
可插拔的。
Pluggable.
可插拔的。
Pluggable.
你把光模块插到服务器上。
You plug in an optical module into the server.
下一步就是插NPO。
Then the next step is you're gonna plug NPO.
你插的时候，不是直接插到服务器上，而是插到板上。
You're gonna plug the the instead of going to the server, it'll go onto the board.
插到板上。
Onto the board.
近封装光学。
Near packaged optics.
近封装光学。
Near packaged optics.
在板子上，里面还是有一些额外的组件。
On the board, it's still some extra components in there.
最终的目标，可以说是共封装光学（CPO），就是把光纤直接接到芯片上。
The holy grail, if you will, will be co packaged optics CPO, which is where you you attach the fiber strand directly to the chip.
这样做能带来各种节省和有趣、重要的好处。
And there's all kinds of savings and interesting things, important things about that.
但你可以想象，制造起来非常非常非常复杂。
But as you can imagine, it is very, very, very complicated to manufacture.
关于这一点，我想指出一件事，我为此开了很多会，在OCP峰会上这会成为热点。
The the one thing I wanna point out about this, and I have had numerous meetings on this, and this is gonna get hot at OCP Summit here.
显然，我们刚开完OCP台湾会议，我关注了很多相关的讨论。
Obviously, we just had OCP Taiwan, and I followed a whole host of notes that came out around that.
就像你说的，CPO的好处很明显，就是能降低每比特的能耗。
There there is the conversation to exactly like you said, which is, you know, the benefit to CPO is obviously you squeeze out more picajoules per bit.
对。
Yes.
而这就会带来经济效益的问题。
And that that leads to economics.
不过我觉得，目前关于到底能节省多少，以及相对于NPO带来的额外麻烦是否值得，还存在很大的争议。NPO本身也还处于起步阶段。
I do think, though, right now, it is highly debated how much of a savings that is and is it worth that additional headache over NPO, which again is is is in its infancy.
其实我们连CPO都还没真正实现，尽管大家都在谈论它，但规模还很小。
Like, we're not even really at CPO despite people talking about that at any kind of scale.
目前规模非常小。
It's very, very small.
我认为这种动态是需要解决的一个问题，因为实现CPO实在太难了。
And that dynamic, I think, is one of the things that'll have to get worked over because it's so hard to do to do CPO.
除了制造，还有很多其他问题，比如可靠性。
There's a whole lot of things, not just manufacturing, but reliability.
需要解决的事情太多了，所以我的观点是，如果节省的成本没有比工程投入多出显著的比例，那我觉得NPO可能会持续很长时间。
Like, so much stuff's gotta gotta get worked out that I guess my point is if the savings is not some significant percentage more versus the engineering effort, it's kind of why I think NPO kinda lasts for a long time.
而且我认为，关于CPO到底能节省多少成本，相对于需要投入的工程努力，这个争论还在继续。
And I think this is debated right about how much savings are you actually gonna get per the engineering effort that's gonna need to go into CPO is still a debate.
而且我觉得也值得指出，这绝对是连接领域的蛮荒西部。
And I also think it's just worth pointing out, this is absolutely the wild wild west of connectivity.
OCP的台湾分部刚发布了一份标准文件。
OCP's Taiwan just had a standards paper come out.
大概有300页那么长。
It's like 300 long.
Lightmatter帮忙牵头，还有其他一堆公司参与。
Light matter help late lead that along with a host of other names.
Lumentum不在名单上，我觉得Coherent也不在，显然，考虑到他们自己的规格，他们现在不一定想定标准。
Lumentum is not on that list, and I do not believe Coherent is either from obviously, they're not in a position to necessarily want a standard at this place given their their spec in.
但这就是蛮荒西部。
But it's the wild, wild west.
每个人都在做自己的实现，自己的定制光学设计，不管是用在哪儿，无论是谷歌的OCS，还是从LPO转向NPO。
Everybody is doing their own implementations, their own custom design for optical, where they're using optical, whether that's OCS with Google or or, you know, LPOs moving to NPOs.
现在就是这样。
Today.
这真是蛮荒西部，这让事情变得难得多，也难预测得多。
It's the wild wild west, which makes this much, much more difficult, much, much more unpredictable.
我们甚至不知道光纤电缆和其他一大堆协议会往哪个方向发展。
We don't even know which direction this takes in fiber cables, in a whole host of other protocols.
所以就是这样，对吧？
So there's that, right?
现在还是早期阶段。
Any early cycle.
所以人们总是过于乐观，觉得光通信的浪潮要来了，这些公司会变得非常庞大。
And so people just keep getting ahead of their skis of the optical trade is coming, these companies are gonna be gigantic.
但实际上，这是一个非常缓慢的过程。
It's like, this is a very slow roll.
另外我还注意到，大家都在讨论纵向扩展、横向扩展、跨区域扩展这些概念。
And and then the other part that I noticed is, you know, there's these conversations scale up, scale out, scale across.
有意思的是，很多人说纵向扩展的时候，其实指的是横向扩展。
And what's interesting is like a lot of people are using scale up when sometimes they mean scale out.
这也让大家很困惑，因为有些人说的scale up是指纵向扩展，也就是在机架内部，然后还有机架到机架。
And that is also confusing people because scale up in some people's nomenclature can mean vertically, so within the rack, and then also rack to rack.
所以你说的纵向扩展是从北到南，机架到机架，但也是从东到西，就是你旁边的机架，对吧，因为距离可能得是三米，如果你想更短的话，或者更长，抱歉，才能用光通信。
So you're scaling up is north to south, rack to rack, but also east to west, the rack right next to you, right, because it needs to be, say, three meters, right, if you wanna do or shorter, right, or or longer, sorry, to go to optical.
然后有些人就会说，你看你需要多少连接。
And so then some people are like, oh, you know, look at how many connections you're gonna need.
那肯定会直接推动CPO的发展。
Inter that's gonna just boom CPO.
其实不是。
It's like, No.
不是。
No.
机架之间，我是说，抱歉。
Inter rack is so I mean, sorry.
同一个机架内部，离大规模采用还远着呢。
Within the same rack is so far off from adoption.
机架到机架已经开始发生了，但他们以为你说的是纵向扩展。
Rack to rack is starting to happen, but but they're assuming when you say scale up.
对吧？
Right?
你懂我的意思吧？
You you know what I mean?
我完全把大家搞糊涂了，但我的重点是，在机架内部用光模块进行扩展，那还远着呢。
Like, I'm totally confusing everybody, but my point is my my point is scaling with optical within the rack is much farther off.
机架到机架是现在大家都在做的，这样才能在超过四米的情况下获得800G到1.6T及以上的吞吐量，而AEC大概就到四米为止了。
Rack to rack is what everybody is doing right now so that you can have the throughput you need at 800 g to to 1.6 and beyond at more than, call it, four meters, which is what AEC is gonna kinda tap out out.
我觉得这让大家困惑，因为其他厂商用“纵向扩展”指机架内部，用“横向扩展”指机架到机架。
That that, I think, is confusing people because other vendors use scale up to mean it with inside one rack and scale out to mean rack to rack.
所以术语混用让大家困惑，以为你会需要更多的光连接，但实际上可能不需要那么多，因为去隔壁机架不需要那么多连接。
And so we have this mixing of terminologies that's confusing people, leading them to assume that you're gonna have far more optical connections when you're probably not because you don't need as many to go to the rack right next to you.
总之，抱歉。
Anyway, I'm sorry.
抱歉，我把大家搞糊涂了。
I'm sorry for confusing everybody.
我只是觉得这个术语让大家很困惑。
I'm just this the nomenclature is throwing people off.
你这场仗有的打了。
You are going to be fighting this fight.
你这是在跟风车搏斗，得斗好几年呢。
You're gonna be tilting at this windmill for years.
这也不是你第一次提这个了。
I'm just this is not the first time you brought this up.
我知道。
I know.
甚至不是第二次了。
Or even the second.
这事儿永远不会有共识的。
No one's ever gonna agree on this.
我知道，但我认为搞清楚这一点很有必要。因为如果你打算按机架来假设光模块的部署位置，你就得知道它们到底是放在同一个机架内部、同一列机架内（目前这种情况并不常见），还是跨机架连接，甚至是跨Pod连接，距离可能长达100米。
I I I know, but I think it's worth knowing because if you're gonna make an assumption of attaching optical content per rack, you need to you need to know if it is within the rack, within the vertical of the rack, which is not happening, it is rack to rack, or it is pod to pod, so at distance, a 100 meters away.
在这两种情况下，所需的光模块数量都会少得多。
Very far less optical content in both of those two ladder scenarios.
所以，如果你要基于每个机架来假设光模块的配置，你就必须清楚实际连接的是哪些设备。
So you need to know what we are connecting if you're going to make an assumption of optical content attached per rack.
这点说得有道理。
That's a that's a fair point.
但我确实有点搞不懂，因为我们现在处在一个很魔幻的世界里，我一直觉得自己才是那个最懂光学的人。
But I I I'm confused by all this because we're in this bizarro world where I always think of myself as the ultimate optical bear.
我之前也聊过这个话题。
Like, I I've talked about this before.
我其实挺讨厌报道光学领域的，因为我做了这么多年，结果总是让人失望。
I hate covering optical because I've covered it for years, and it always disappoints.
不过说实话，我对光学基础设施的建设前景比你要乐观得多。
But, like, I'm more bullish on the optical build out than you are.
我其实觉得到2028年，我们肯定会开始看到NPO进入量产。
I I I actually I think in 2028, we will start to see certainly, certainly NPO and production.
那是肯定的。
For sure.
绝对没问题。
Absolutely.
而且CPO也会在一定程度上出现。
And and some some CPO to some degree.
对。
Right.
对。
Right.
我觉得是这样，但再说一次，是水平方向，而不是机架内的垂直方向。
I think that's But again, horizontally, not vertically with within the rack.
你可能需要NPO。
You you have to have NPO perhaps.
对。
Yes.
NPO可以在机架内开始应用，因为它必须可维护。
NPO can start to happen in the rack because it has to be serviceable.
一个500多GPU的机架，连接密度太高了，光纤用量巨大，对吧，你需要更多的交换机才能实现这一点。
The density of connections out of 500 and something plus GPU rack is so much optical, right, in term you need so much more switches in that to make that happen.
所以一旦实现，光学内容的需求量会非常庞大。
So the optical content, once that happens, is is enormous.
而且必须可维护，这正是NPO发挥作用的地方。
And it has to be serviceable, which is what NPO comes in.
我能看到NPO在那种机架中的应用，我的意思是，Lumentum在说，我们在做CPO，而且开始规模化。
I can see NPO in that rec my point is that you've got Lumentum out there going, we're doing CPO and we're starting to scale.
然后人们会说，哦，你说的CPO是指扩展，还是说他们说的扩展是指机架到机架。
And then people go, oh, by CPO, you mean is or they're saying scale up.
比如，Lumentum说的扩展是指机架到机架，而不是机架内部。
Like, Lumentum is saying scale up when they mean rack to rack, not within the rack.
因为这让大家很困惑。
Because and that's confusing people.
你会看到他们画这些图。
Like, you'll see them make these diagrams.
比如推特上那些机器人或者什么人，会说你看一个机架里需要多少光纤。
Like, these bots or whoever they are on Twitter is like, look at how much fiber you're gonna need in one vertical rack.
我就说，不对。
And I was like, no.
NPO，对。
NPO, yes.
不是CPO。
Not CPO.
在同一个机架内部，
In in in in within the same rack,
南北方向。
north to south.
但快了。
But it's coming.
快了。
It's coming.
快了。
It's coming.
快了。
It's coming.
时间上可以争论。
Can debate timing.
不是CPO。
Not CPO.
CPO也快了。
CPO is coming too.
不会缩减。
Not scale back.
不是亚太地区。
Not APAC.
你
You
没法维护那根光纤。
can't service that fiber.
它会坏掉的。
It's gonna die.
它会烧坏主板和GPU。
It'll it'll just kill boards and GPUs.
我觉得它还是会来的。
I I think it's I I still think it's coming.
现在问题是这样的。
It's gonna now here's the thing.
这是光学的。
It's optical.
这需要的时间会比我们任何人预期的都要长。
It's going to take longer than any of us expect.
是的。
Yeah.
对吧？
Right?
它总是需要光。
It's always gonna take light.
它总是会来得更晚。
It's always gonna come later.
所以要有心理准备。
So be warned on that.
但我认为这是有可能的。
But I I think I think it's possible.
对。
Yeah.
我的意思是，我不是说它不会出现。
I I mean, I'm not saying it doesn't show up.
我只是说它没法规模化。
I'm just saying it doesn't scale.
它不会变成那种大量的计算柜。
It's not going to be the vast number of attach of compute racks.
它就是没法规模化。
It just doesn't scale.
这里的制造瓶颈太深了。
The manufacturing bottlenecks here are too deep.
你只要去了解一下这东西制造起来有多难。
Like, all you have to do is dig into how hard this is to manufacture.
你很快就能估算出能做出多少连接。
You can very quickly come up with some decent estimates of how many connections can be made.
而GPU柜、加速器柜、CPU柜的制造能力会远远超过那个数量级。
And and GPU racks, accelerator racks, CPU racks will outstrip that by magnitudes of its ability to manufacture at at rack scale.
跟大家说一下，这不是只有我和本在争论。
Just so everyone knows, this is not just Ben and I having this debate.
我们请了好几位嘉宾，分别支持我们双方的观点。
We have several guests lined up to take to take both our sides in this.
好的。
Okay.
我们接着聊思科吧。
Let's move on to Cisco.
在网络趋势方面，我们刚才聊的是光通信。
In the networking trend, like, we've that's optical.
我们接着聊思科吧。
Let's move on to Cisco.
思科，我几年前就不太关注了，因为那时候它特别无聊，但现在它不再无聊了。
Cisco, which I I'd sort of stopped paying attention to years ago because it got very, very boring, is not boring anymore.
他们发布了季度财报。
They they reported their quarter.
我，我，我不知道，数字嘛，我猜是有点让人失望。
I I I don't you know, the numbers were, I guess, disappointing.
我肉眼看着还行，但市场在跌。
They looked okay to me at my eyeball, but, like, the street is down.
这周因为这个数字跌了几个百分点。
It's down a few percent this this week on those numbers.
但我读了他们的财报电话会议记录，觉得里面有些挺有意思的东西。
But I I read through their transcript, and I thought there's some really interesting things about it.
对吧？
Right?
首先，他们现在几乎完全用自己的芯片了。
I for one, they are they almost entirely use their own silicon now.
Silicon One 就是他们内部设计的芯片。
Silicon One is their internally designed silicon.
他们早在“定制ASIC”这个词出现之前就在做定制ASIC了。
They they they were they were doing custom ASICs before we even called it custom ASICs.
他们做自己的芯片已经二十年了，但现在几乎所有的产品线都跑在自己的芯片上，我觉得这挺有意思的。
They've been doing custom their own silicon for twenty years, but it's now pretty much all of their portfolio runs on their own silicon, which I thought was pretty interesting.
这先进多了。
It's much more advanced.
他们还反复强调说这些都是自己做的。
They also went out of their way to say repeatedly that they're doing it themselves.
他们以前用了很多年联发科的芯片，但现在所有能力都内部化了，这点在你看其他公司的定制业务时值得记住。
They'd used MediaTek for many years, but they're they now have all the capability internally, which is something to keep in mind when you look at other companies' custom business.
但我觉得，你知道，他们现在跟AI的关联度越来越高了。
But I I thought, you know, they're they're now getting a fairly healthy degree of AI attached.
而且我觉得，在AI之前的那十年里，他们一直被诟病的是在云方面没什么存在感。
And I think the the knock on them for the last, I don't know, pre AI, ten years leading up to that was that they really didn't have a cloud a presence in the cloud.
超大规模云服务商都想自己造白盒交换机。
Hyperscalers all wanted to build their own, white box switches.
所以思科在数据中心的影响力就慢慢缩小了。
And so Cisco was sort of shrinking, you know, shrinking relevance in the data center.
而AI似乎改变了这一点。
And AI seems to have changed that.
他们在各种交换机上都有很高的AI采用率，而且在超大规模数据中心方面也赢了不少。
They have a really good AI attach rate across, all kinds of switches, right, and, all kinds of hyperscaler wins.
他们谈了很多关于企业本地部署，用思科交换机搭建自己的AI集群。
They talked a lot about on prem enterprise building up their own switches using Cisco switches to build their on prem AI clusters.
而我觉得最有趣的是，他们谈了很多关于使用Mythos的情况。
And and the thing I I thought was most interesting, they talked a lot about their use of Mythos.
他们显然从Anthropic获得了许可，允许客户基本上对企业IT系统进行安全审计，通过Mythos进行，就是那种没有护栏的完整版Mythos，也就是完整的Anthropic，这暴露了各种安全漏洞。
They apparently have a license from Anthropic, and they they they, allow their customers to run basically security audits on enterprise IT systems to go through Mythos, like, full, you know, no guardrails kind of Mythos, like, the full Anthropic, and that is exposing all kinds of security vulnerabilities.
我觉得这对思科的安全业务来说是个挺聪明的销售策略，因为他们允许你做这种审计，可能很便宜甚至免费。
And I thought this is a pretty smart sales tactic for Cisco's security businesses because they they allow you to do this audit probably for cheap or for free.
当然，你肯定会发现各种漏洞，然后思科很乐意收费帮你修补。
And the of course, you're gonna find all kinds of vulnerabilities, and then Cisco would be happy to patch those for you for a fee.
而我觉得这里特别有意思的是，你想想思科的产品。
And I think where where this is pretty interesting, you think about, like, Cisco stuff.
思科的交换机到处都是。
There's Cisco switches everywhere.
有些已经存在很久了，而且你知道，它们都已经到了生命周期尽头，思科多年前就停止修补了。
Some have been around for a very long time, and, you know, they're they're all end of life, and Cisco stopped patching them years ago.
现在人们开始在核心交换机上发现漏洞，我觉得这是思科非常聪明的销售策略，因为他们现在可以为那些十年都没有许可证的硬件收费了。
And now people are finding vulnerabilities in sort of core switches, and I think it's a it's a it's a very smart sales tactic on Cisco's part because now they're able to charge for soft for hardware that they haven't, you know, haven't had on license in ten years.
但这确实是一个有趣的AI应用案例，Anthropic和思科在合作。
But it's it's an interesting, AI use case where where Anthropic and Cisco are working together.
挺聪明的。
It's pretty clever.
所以我想花点时间聊聊这个，因为我听到类似的用例现在到处冒出来，在一些CSO网络安全对话中，他们用Mythos来发现安全漏洞、做产品测试、寻找漏洞等等。
So I wanna talk about this a second because I have heard similar use cases now popping up all over the place in some CSO cybersecurity conversations that they're using Mythos for security vulnerabilities, product testing, trying to find vulnerabilities, etcetera.
实际上我在网络安全报告里有点预测到了这一点，我说过前沿实验室会非常适合与这些公司合作，开展他们的安全和网络安全工作。
And I actually somewhat predicted this in my cybersecurity report where I said that the Frontier Labs are gonna be very well positioned to work with these companies on their security and cybersecurity efforts.
我的问题是，我们没有任何证据，但我只是把这个当作一个理论提出来。
The question I have is, and we have no evidence of this, but I'm just throwing this out as as a theory.
但我得想象一下，Anthropic，以及某种程度上OpenAI，我知道，我们清楚他们向大型企业收取高额的许可证费用。
But I have to imagine Anthropic and to the degree that opening I know this, like, we know they charge these large enterprise licenses.
所以所有这些企业都在购买批量许可证，付给他们一大笔钱。
So all of these businesses are buying a bulk license and paying them a bunch of money.
但我得想象，他们可以，或者正在，或者应该为网络安全工作或某个模型收取更多费用。
But I have to imagine, like, that they could charge or they are charging or they should charge more for the cybersecurity efforts or a model a model.
比如说，他们开发了一个新模型。
Let's just say that they develop a new one.
Mythos很好，但他们又开发了另一个专门用于网络安全的模型。
So Mythos is great, but they develop another one just for cybersecurity.
我觉得你可以为此收取更多费用。
I I feel like you could charge more for that.
考虑到它为公司解决的问题，那在经济上应该是一个更高的许可证价格，但我们还不知道那方面的经济情况如何。
Like, that would be an economically higher license given the problem it's solving for the company, but we don't know what the economics of that look like yet.
对。
Yeah.
我们对Anthropic的经济状况和商业模式还不太了解。
We we don't have a good picture of Anthropic's economics and their business model.
你知道，他们申请了上市，所以希望很快我们就能看到那些详细的数据，了解更多。
You know, one of the you know, they filed to go public, and so hopefully soon we will get to look at all that juicy good data and learn about it more.
我强烈怀疑他们目前对Mythos的定价偏低，因为它刚推出，他们可能还不知道怎么收费。
My strong suspicion is that they are under monetizing Mythos at this point just because it's new and they probably don't know how to charge for it.
但你绝对能看到它的价值。
But you could absolutely see the value.
比如，思科的案例就是一个很好的例子，说明他们可以怎么做，因为你知道，Anthropic一直以来的故事就是他们非常专注于向企业销售AI，对吧，不像OpenAI那样做很多不同的事情，包括大力关注消费者市场。
Like, in in the Cisco case is a good example of how how they can do this because, you know, Anthropic's story has long been that they are very focused on selling AI to the enterprise, right, as opposed to OpenAI, which is doing a lot of different things, including a big focus on consumer.
我觉得人们心里有个印象，就是Anthropic比OpenAI做得更好。
And I think the sort of the heuristic that people have in their minds is that Anthropic is doing it better than OpenAI.
我不认为这是真的，但我觉得很多人都有这种偏见。
I I don't think that's true, but I think that's the bias a lot of people have.
对。
Yeah.
而且，正因为这种对企业的专注。
And and be because of that focus in the enterprise.
而且，说实话，OpenAI现在在企业业务方面也做得相当不错。
And in, you know, in fairness, OpenAI has a pretty good enterprise business now too.
但这两家公司，他们到底是怎么收费的呢？
But both of them, like, how are they charging for these things?
钱到底是从哪里来的？
Where is the money actually coming from?
是像我们这样成千上万的人每月花200美元订阅，还是他们从企业级API那里赚得更多？
Is it, you know, millions of us signing up for $200 a month, or are they actually generating far more money from their a p their enterprise grade API?
我对此非常好奇。
I'm I'm I'm super curious about that.
我们就是不知道。
We just don't know.
显然，关于这方面的数据并不多。
There's just not a lot of data out there about it for obvious reasons.
但是
But
但我觉得这说得通。
but I think it makes sense.
对吧？
Right?
从逻辑上讲，重点是如果你有一个模型，比如说它是Mythos级别或更高，然后你把它定位给企业，说，嘿，作为我们许可证的一部分，这个模型是用于你们的网络安全防御，以防那些能拿到开源模型就开始攻击、试图勒索你们业务的陌生人，同时也用于你们自己的软件或硬件，在这种情况下，芯片漏洞研究，那这个模型就比用来总结Teams会议的那个更有价值，对吧？
Logically, the point is if you have a model and let's say that it's Mythos class or beyond, and you're positioning it to enterprises as, hey, as a part of our license, this one is for your cybersecurity efforts to defend yourself against randos who can get an open model and start attacking and try to hold your business ransom and for your own software or hardware, in this case, chip vulnerability work, that's a more valuable model, right, than what you'd give for for summarizing a Teams meeting.
再给我做个PowerPoint。
Make me another PowerPoint.
对。
Yeah.
不。
No.
我的意思是，我们还没真正谈到这个，但现在确实有一个很大的争论，关于中国的模型、开源模型。
I mean, we we haven't really touched on this, but there's obviously been this big debate going on right now about Chinese models, open source models.
我觉得我们不需要深入讨论这个，因为很快就会变得非常复杂。
And I don't think we need to get into it all because it gets very complicated very quickly.
但是，再说一遍，现在很多圈子里普遍认为，中国的开源模型几乎和前沿实验室的模型一样好，甚至包括Mythos级别。
But, again, the the sort of the understanding in lots of circles right now is that Chinese open source models are almost as good as Frontier Labs up to including Mythos.
而且那些模型是免费的，我打个引号说，免费。
And those they are free, I say, with air quotes, free.
对。
Yeah.
所以，我是说，今天《金融时报》有个头条说Anthropic和OpenAI之间正在打价格战。
And and so, I mean, there's a headline in the Financial Times today is there's a price war going on between Anthropic and OpenAI.
我觉得我们不知道这一切会如何发展。
I I don't I don't think we know any how any of this is gonna play out.
中国的模型可能不会一直免费开放。
The the Chinese models may not stay open for free.
前沿实验室还在积极推进，但这一切都跟那个争论纠缠在一起。
The Frontier Labs are still moving forward pretty aggressively, but it it's it's it's all wrapped up in that debate.
我觉得这是其中一个很大的领域，我只能说，我们还不清楚这一切会怎么发展。
And I think this is this is one of these big areas where I just say, like, we don't know how this is all gonna play out yet.
它有很多不同的可能方向。
There's it can go a lot of different directions.
对吧？
Right?
对。
Yep.
好。
Okay.
所以挺有意思的，网络安全这些方面。
So interesting stuff, networking security, all those things.
我们换个话题，聊聊应用材料公司吧。
Let's talk to let's pivot to Applied Materials.
AMAT。
AMAT.
AMAT。
AMAT.
所以，我的意思是，那真的是非常看涨。
So, I mean, it was extremely bullish.
他们正在受益于我一直以来在思考的一些东西，我还在推特上发过，叫做LTA waterfall，我觉得这非常有意思，因为它和我们之前讨论的所有内容都有关联，对吧？
They're they're benefiting from something I've been I've been thinking about and I and I tweeted this called LTA water falling, which I think this is very interesting as it relate as as it relates to kind of all we've talked about, right?
这个行业的动态已经因为这些LTA而改变了，或者像SanDisk喜欢称之为新商业模式。
That dynamics of this industry has changed around these kind of LTAs or as SanDisk likes to call them new business models.
美光喜欢称之为SCA，随便什么。
Micron likes to call them SCAs, whatever.
这些其实就是客户来找他们，说，我想长期锁定产能。
These are just customers coming to them going, I'd like to secure capacity for a long period of time.
我要么直接告诉你我的可见度，要么在某些情况下，比如美光和内存以及活动中心，他们会预付定金来确保这一点，这样公司，比如美光，或者你最喜欢的CPU公司，或者其他什么，就可以去确保他们下游的供应链LTA，比如基板、晶圆、代工等等。
I'm either going to tell you exactly what my visibility is or in some cases with Micron and Memory and Event Centers, prepayments are happening to secure that so that the company Micron, maybe your favorite CPU company or whoever, can go and secure downstream their supply chain LTAs they need as well for substrates, for wafers, for foundry, etcetera.
所以你会看到这种，就像，一个非常大的公司A。
So you're getting this kind of, like, company a who's very big.
对吧？
Right?
他们作为关键供应商，拿到了长期协议。
They're the they're the key point supplier gets an LTA.
他们觉得，太好了。
They're like, cool.
然后他们就去跟自己的供应商签长期协议，因为要确保有足够的产能来兑现给客户的承诺，所以这个链条就一层层往下传。
I'm gonna go sign LTAs with my suppliers because I need to make sure I have the scale I need to meet the the commitment I just gave to one customer, and so you're getting this go down.
所以我觉得有意思的是，连晶圆设备这个层面，AMAT也第一次用了这种说法，说他们有八个季度的可见度。
So I thought it was interesting that even at the wafer equipment level, this is sort of the the first time we've heard this type of language around wafer equipment from AMAT, that they have eight quarters of visibility.
而且有意思的是，以前周期里他们只有两个季度的可见度。
And and what's interesting is that in prior cycles, they've only had two quarters of visibility.
那是客户能给他们的最好的情况了。
It was the best their customers gave them.
现在他们有八个季度了。
Now they have eight.
对我来说，这很能说明问题。
To me, this is very telling.
他们的业务在增长。
Their business is growing.
收入也在上升。
Revenues were up.
我觉得唯一的问题——我对此有个想法，还没告诉过你，所以你会第一次听到——就是他们在给出业绩指引时非常犹豫。
I think the one knock, which I have a theory on that I haven't told you yet, so we're gonna you're gonna hear it all fresh for the first time, is they were very hesitant to guide.
上次他们说，我们预计增长会超过30%。
Again, their last time they said is we think we're gonna grow more than 30%.
现在他们只是说，会比上次的数字有显著增长，但他们还不想给出明年的全年指引。
They're just saying now meaningfully more than our last number, but they didn't want to give away a full year guide for next year yet.
他们十月份有个投资者日，但我有个理论，为什么他们现在不想这么做，他们可能会在投资者日上给出，这个我留到你说完对应用材料的看法之后再说。
Now they have an October Investor Day coming, but I have a theory as to why they did not want to do that right now, and they probably will at their Investor Day, which I'll save after you give give your take on Applied.
好的。
Alright.
那我先退一步说。
So let let me let me back up.
应用材料昨天发布了财报。
Applied Materials reported yesterday.
今天是8月14号，星期五。
This is Friday, August 14.
我觉得他们的数据非常强劲。
And I I think they delivered very strong numbers.
他们的营收远超预期，比预期高了一点。
They they, revenues way above revenue was a little bit above expectations.
本季度和下一季度的盈利预期都大幅超出预期。
Earnings for both the current and the guided quarter were significantly above expectations.
所以他们表现不错，而且他们在有效提价，或者卖出了更好、更贵的系统。
So they're doing well, and they're raising prices effectively or selling better, more expensive systems.
数据真的很不错。
Really good numbers.
展开剩余字幕（还有 395 条）
对吧？
Right?
客观来说，从绝对值上看，数字非常强劲。
Objectively, in absolute terms, very strong numbers.
这是他们营收增长史上最快的环比增长。
It was the fastest sequential growth they've ever seen in revenue growth.
对吧？
Right?
是的。
Yeah.
历史性的。
Historic.
对吧？
Right?
对吧？
Right?
这里面有个周期。
There's a cycle.
大家都知道，而他们正好从中受益。
We all know about it, and they're benefiting from it.
但这还不够。
But it wasn't enough.
永远都不够，Jay。
It's never enough, Jay.
永远都不够。
It was It never enough.
我听得头发都快抓掉了，听电话会议里的那些问题。
I was pulling my hair out listening listening to the questions of the call.
全都是一个调调：季度不错，但我们想要更多。
They were all some variation of, yeah, good quarter, but, we want more.
我就觉得，这真的很能说明问题。
I'm just like it was it was very telling.
对吧？
Right?
有个人问了他一个问题，我就不点名是哪个分析师了。
There was there was one question somebody asked him, won't name the analyst.
他说，你的客户都在谈五年期的长期协议，但你只提三年期的。
He said, your customers are all talking about five year long term agreements, but you're only talking about three year long term agreements.
我当时就想，什么？
I'm just like, what?
十二个月前，一年期的长期协议都闻所未闻。
Like, twelve months ago, it was unheard of to have a one year long term agreement.
现在居然在争论三年还是五年。
Now we're debating three years versus five years.
所以，是的，市场想要更多。
And and so so, yeah, the the the street wanted more.
今天股价跌了5%。
Stock is down 5% today.
显然，市场的预期比华尔街公布的共识还要高。
Expectations were apparently higher than street printed consensus.
事情就是这样。
So it goes.
你的看法是什么？
What's your theory on
好的。
Okay.
嗯，有两件事。
Well, two two things.
我会告诉你我的看法，但咱们先这样。
I'll give you my theory, but let's say let's do this.
他们电话会议中有两点我也觉得很有意思，这也将引出我的看法。
The the two parts from their call that I also think were interesting was and this is gonna lead to my theory.
第一点是，他们说客户正在想各种办法腾出洁净室空间来安装更多设备。
So one was they said customers are finding creative ways to free up clean room space for more equipment.
所以那只是重新整理一下，比如把机器挪个地方，或者淘汰一些旧设备。
So that's just reorganizing, I don't know, moving a machine somewhere, decommissioning something somewhat.
他们就像是在想办法说，我有这个洁净室空间。
They're, like, figuring out how to say, I've got this clean room space.
怎么才能再多放几台设备进去呢？
How can I get a couple more, you know, a couple more tools in here in order to do that?
所以就有这么一点。
So there was that.
另外他们还提到，或者说暗示了，客户在加快一些建设进度，想早点把工厂建好。
The other one they said was customers are, or they alluded to, pulling in some construction timelines to get their fabs out sooner.
不错。
Cool.
抱歉。
Sorry.
所以就是这两点。
So you got those two things.
还有，每平方英尺的晶圆产量，对吧？
Also, wafer per square footage, right?
所以他们有机器，有一整条机器线，能让你每平方英尺的晶圆产量更高。
So they have machines, they have a whole line of machines that gives you better wafer output per square footage.
这对其他一些公司来说也开始变得重要了。
That's starting to come into play for some other Cool.
好的。
Okay.
我觉得他们不想现在给出这个数字，因为英特尔。
Think that they did not want to give this number yet because of Intel.
我觉得英特尔的融资刚完成。
I think Intel's capital raise is just happened.
对吧？
Right?
而且他们还没告诉英特尔未来两年需要多少台机器，因为他们得先拿到钱，然后得赶紧下单，这会在他们十月份的活动之前发生。
And they have not told them how many machines they want yet over the next two years because they first needed to get the money, then they need to go very quickly, place those orders, which is going to happen before their October day.
而且我觉得英特尔也在做我刚才说的事情，就是会用一部分这笔资本支出来加速数据中心或者洁净室的建设。
And I think Intel is also doing what I just said, which is gonna use some of this CapEx to accelerate data center or clean room construction.
所以我觉得他们一直在等。
And so I think they were waiting.
这就是那个变量。
That's the variable.
英特尔就是你刚才提到的那个变量。
Intel is the variable you just raised.
你会把这笔资本支出花掉的。
You're gonna spend this CapEx capacity.
你你你你需要这笔刚到的钱，才能搞清楚你的晶圆厂能多快上线，还有你需要多少台机器，而英特尔还没告诉他们。
You you you needed this money that just happened in order to figure out how much faster your your fabs can come online and two, how much your how much machines you're gonna need, and Intel has not told them yet.
所以我的理论就是，英特尔是那个变量。
So that's my theory is that Intel is the variable.
英特尔百分之百会在十月底之前完成所有这些，因为他们必须这么做。
Intel will 100% do all of this before the end of October because they have to.
然后AMAT就会知道需要多少英特尔的产品。
And then AMAT will know how much Intel it's gonna need.
我觉得确实是这样。
I I think that's that's true.
我确实在电话会议中感觉到，那个没明说的客人就是英特尔。
I definitely my my spidey sense during their call was there was, like, the unspoken the unspoken guest on the call was Intel.
就像是我们会绕着谈但不会直接点名的那件事。
Like, the the the one thing we're gonna talk around and not actually name.
我觉得你说得对。
I I think you're I think you're right.
我的意思是，另一个更简单的解释是，他们只是不想给出全年指引，因为公司不喜欢提前太远做预测。
I mean, in in their I mean, another simpler alternative is that they just didn't wanna guide for the year because companies don't like to guide too too far in advance.
但是，是的，我觉得英特尔目前是造成不确定性的一个主要原因。
But, yeah, I I think Intel is a is a is a prime prime cause of uncertainty at this point.
所以，好吧。
So okay.
我同意这个说法。
I buy that.
好的。
Okay.
不错。
Good.
嗯，我很高兴你没听到那个。
Well, I'm glad you don't hear it.
总之，我觉得这很合理。
Anyway, I I think it just makes sense.
我觉得这很符合逻辑。
I think it's logical.
但我这么说，是因为感觉即使是我之后读到的笔记，那似乎也是唯一的负面点。
But but but I say all that to say it felt like even the notes I read afterwards, like, that seemed to be the only real negative.
就像，你会说，嘿。
Like, yes, you were like, hey.
他们期望更高。
They expected more.
我同意。
I agree.
但他们觉得自己的指引已经很保守了。
But they wanted they felt like they were being conservative in their guide.
我的观点是，他们确实保守。
And my point and and they were.
而我的观点是，那是因为英特尔。
And my point is that's because of Intel.
到了他们十月份的那天，这一切都会尘埃落定。
And at their October day, this will all be settled.
对。
Yeah.
而且需要明确的是，我们刚才说他们的指引保守，但他们仍然在指引创纪录的业绩。
And just to be clear, when we were talking about conservative on their guidance, they're still guiding to record everything.
只是还不够创纪录。
Just not record enough.
就是……
It's just
说实话，也许你可以简单跟我解释一下这个。
To be honest with you, maybe you could just briefly explain this to me.
也许我们之前聊过这个，但就像我发的那条推文。
Maybe So we've talked about this, but, like so I tweeted out.
我当时就说，你看。
I was like, look.
我觉得现在大家普遍的看法是，到2030年，晶圆设备市场会达到3000亿或者更高。
I think broad consensus right now is moving wafer equipment toward 300,000,000,000 or north of 300,000,000,000 by 2030.
是的，有些事情可能会让这个时间提前。
Yes, there are some things that could move that forward.
但然后有人回复我的推文说，买方那边认为2028年就能到3000亿。
But then somebody tweeted me back and they were like, the buy side's at 300,000,000,000 for 2028.
我就想，为什么买方总是比卖方更激进呢？
And I was like, why is the buy side kind of consistently more aggressive than the sell side?
我搞不懂这一点。
I'm I don't understand that.
所以
So
有几件事吧。
so a couple things.
第一，我觉得这反映了整个华尔街的一个普遍现象，不管是我们之前聊的光学，还是半导体设备，都存在一种倾向，就是市场总是预期事情会比实际发生得更早，因为我们都是按九十天周期来思考的。
One is I think this is a symptom of just Wall Street in general where and this this applies as much to optical, what we were talking about before, as it does to semi cap, is there is a a tendency, a bias on the street to expect things to happen sooner than they actually do because we're all we're all on ninety day cycles.
我们都只想着九十天，然后觉得，哦，那还有六个季度呢。
We're all thinking ninety days, and we're like, oh, that's six quarters away.
那太久了。
That's a long time.
对吧？
Right?
记住，大多数人做模型都不会超过两年。
Remember, most most people don't have models that go beyond two years.
所以，华尔街总是很着急。
So every there is Wall Street's in a hurry.
对吧？
Right?
这就是那种倾向。
That's that's the bias.
至于为什么买方和卖方会有分歧，我觉得是因为卖方是公开的，更容易被要求对那个数字负责。
As to why there's a divergence between the the buy side and the sell side, I think that's because, the sell side is in public and will be more likely held to account for that number.
对吧？
Right?
如果我现在给出一个一年后的预测数字，一年后肯定会有人来问我。
If I if I put out a number now for a prediction for a year from now, somebody's gonna come and ask me a year.
他们会说，嘿。
They said, hey.
你一年前是这么说的。
You said this a year ago.
对吧？
Right?
而且我说有人，很可能就是，比如，我在CNBC上，有人问我相关问题。
And when I say somebody, it could easily be, like, you know, I'm on CNBC, and somebody asked me a question about that.
有记者问我，比如，嘿。
A reporter asked me, like, hey.
你之前不是预测过这个吗？
Didn't you predict this?
我理解。
Like, I get that.
我经常被问到这类问题。
I get those questions a lot.
你去年预测过什么。
You predicted something last year.
到底有没有应验呢？
Did it come did it come true or not?
而且大多数情况下，买方不需要公开预测。
And you're for the most part, buy side doesn't have to doesn't make public predictions.
对吧？
Right?
而且所谓买方共识这个说法本身也挺模糊的，因为根本没有什么公开的记录。
And and just the idea that there is a a buy side consensus is a little a little hazy too because it's not published anywhere.
就是几个人在房间里，或者几个在酒吧里，随便凑个数而已。
It's just, like, 10 guys in a room, five guys in a bar coming up with a number.
所以我觉得这就是分歧所在。
So I think that's that's the disconnect there.
对吧？
Right?
但是，好吧。
But but okay.
好吧。
Fine.
但是，我就是不太明白为什么这会这么重要，甚至能影响市场或者让它下跌。
But, like, I I guess I just don't understand why that I don't understand why that makes such a big deal and and can actually impact a market or or take it off.
比如，他们的预期没达到，所以股价就跌了。
Like, that their expectations weren't met, and so it's down.
嗯，因为买方是手里有钱的人，所以他们的预期会直接影响他们买股票或卖股票的决定。
Well, because the buy side are the people with money, and so what they are expecting matters to their purchases their stock purchase decisions or stock sale decisions.
对吧？
Right?
而且，你知道，他们有自己的预期，这些预期很重要，如果这些预期没达到，他们就会改变投资决策。
And, you know, that's they they have expectations, and they matter in the degree that if those back expectations aren't met, then they're gonna change their investment decisions.
明白了。
Got it.
好的。
Okay.
好的。
All right.
好吧，反正就是这样。
Well, anyway, that's that.
晶圆设备那边有好消息。
Good stuff with wafer equipment.
就像我说的，他们十月份有投资者日，还有Epic要发布。
They've got, like I said, October Investor Day, they've got Epic coming up as a launch.
我想最后强调的一点是，即使昨天桑迪斯克发布了消息，大家还是在说，在这个供应链的深处，他们所有的客户互动都变了。
Just, you know, the point the point I wanna end on on this is like, even even even with Sandisk's day yesterday, everybody just keeps talking about how in the deep of the weeds of this supply chain, all of their customer engagements have changed.
我觉得这不是人们以为的那种空话，这真的是供应链结构上的变化，你需要重新考虑怎么稳固你的供应链。
I just This is not lip service to the degree that people think is like, this is a structurally changing dynamic of how you need to shore up your supply chain.
全球最好的供应链都在签长期协议，这样他们就能保证需求。
And the best supply chains in the world are making these longer term agreements so that they have the demand.
这比周期性的东西要结构性得多，你懂我的意思吧。
Like, that's that's so much more structural than, do you know what I mean, than cyclical.
不是说没有影响，我只是觉得大家得明白客户关系的变化有多重要，这很关键。
Not to say that it doesn't, I just think people have to grasp how important the customer dynamic changes and that's significant.
而且我觉得很多分析都忽略了这一点。
And I just think that that gets missed in a lot of this analysis.
我还是觉得，这里存在一个时间上的错位。
I I again, I think there is a disconnect in timing.
长期对不同的人来说意味着不同的东西，我也不想对长期协议完全持怀疑态度。
Long term means different things to different people, and I I I don't wanna be totally skeptical about long term agreements.
我觉得这次和过去的周期确实有本质上的不同，但这些都还没经过验证。
I think there is a a qualitative difference between this and past cycles, but we haven't tested any of those.
对吧？
Right?
我们得看看三四年后这到底会是什么样子。
We'll we'll see what this looks like in three years, four years.
眼下这当然很好，而且会和过去的周期不一样。
In the meantime, it's great, and it's gonna be different than the last past cycles.
我非常同意这一点，但我们不会结束周期性。
I agree with that a lot, but it it's not gonna we're we're not ending cyclicality.
对吧？
Right?
也许我们只是在改变曲线。
We're we're changing the the curve maybe.
是的。
Yeah.
结构性底部。
Structural floors.
新的结构性底部。
New structural floors.
好的。
Okay.
行。
Alright.
咱们来说说NVIDIA和Neo Clouds吧。
Let's let's move to NVIDIA and Neo Clouds.
我还没跟你聊过这个，所以我很想听听你的看法。
I I haven't talked to you about this, so I'll be very curious on your take.
我看了CNBC那个圆桌讨论，就是黄仁勋和那些CEO一起的。
I watched the CNBC roundtable with Jensen and all these CEOs.
这个挺有争议的。
This is very con this was controversial.
NVIDIA和其他人做了很多努力，想让它不那么有争议。
A lot was done by NVIDIA and others to kinda make it less controversial.
但我觉得这里面有点意思。
But I think there's something interesting here.
我有点不同意黄仁勋的一个观点，所以想聊聊这个。
I sort of disagree with one of Jensen's points, so I wanna talk about that.
但我先想听听你的第一反应。
But I want your first reaction.
对。
Right.
所以这周初的大新闻，现在还在占据新闻头条的，就是NVIDIA和六家大型金融机构要一起搞一个平台——又是那种带引号的“平台”——来提供高达5000亿美元的数据中心融资。
So the the big news at the start of the week, really still dominating news cycles, is NVIDIA and six large financial institutions are putting together a platform, air quotes again, a platform to provide up to $500,000,000,000 of data center financing.
这个数字可不小。
And that's a big number.
NVIDIA似乎表示他们会承担其中25%的份额，这挺重要的。
NVIDIA seems to be saying that they're gonna effectively back 25% of that, which is significant.
但除此之外，我们知道的不多。
But beyond that, we don't know a lot.
他们没有透露太多细节，主要是因为协议还没最终敲定。
They have not revealed many details, not least because they haven't actually finalized the agreements.
很多事情都还可能变。
And lots lots can still change.
对吧？
Right?
这对我来说挺有意思的。
It is it is interesting to me.
我不太明白为什么英伟达现在就这么做。
I don't I don't quite understand why NVIDIA did it right now.
比如说，为什么不等到所有事情都签好字，再等上几周呢？
Like, why not wait a couple weeks till everything's signed?
因为，说实话，英伟达经常在事情还没完全准备好之前就发新闻稿，有时候那些事情根本就没成。
Because, like, let's face it, NVIDIA will press release before things are totally ready, and sometimes they they never get there.
对吧？
Right?
去年OpenAI就是这样。
That's what happened with OpenAI last year.
所以这笔交易最终可能会变得很不一样。
So the deal could end up being very different.
但咱们就先按字面意思来理解吧。
But let's just take it at face value.
英伟达正在推出5000亿美元的融资，用来建设基于英伟达的数据中心。
NVIDIA is launching $500,000,000,000 of financing to to build NVIDIA based data centers.
这个数字很大，让很多人感到震惊。
And it's a big number and has people freaked out.
好的。
Okay.
但你对黄仁勋说的那些观点怎么看？他说他试图解决这种循环依赖的问题，指出我们并不是只在互相卖东西。
But what did you take of Jensen's, like, points about I'm trying to address the circularity by pointing out that, you know, we aren't just all selling to each other.
我们主要是卖给主权国家、各个国家和企业，这也是我一直强调的观点。
We're selling largely to, you know, sovereigns, to nations, to enterprises, which is the point I keep making.
当人们说，天哪，他们只是在互相转圈圈。
Like, when people are like, oh gosh, they're just all circling to each other.
但实际上，这些基础设施提供商、实验室和模型公司正在准备向全球超过3亿家企业销售，不包括中国，我没把中国算进去，还有他们的所有员工，就像医生群体的先锋一样。
It's like, yes, but this crew of infrastructure providers, labs, model companies is gearing up to sell to 300,000,000 plus enterprises worldwide ex China, I'm not including China in that number, and all of their employees as the tip of the spear of doctors.
这就像是B2B2B2B2B的模式。
Like, it's b to b to b to b to b.
就是一连串的B2B，很长很长的那种。
Like, long long sets of b to b's.
所以那是一个非常庞大的客户群，他们会付钱给OpenAI、Anthropic这些公司，让他们在英伟达的基础设施上运行计算，然后这些钱再流到英伟达，等等等等。
And so that's a very big customer base who's going to pay the the OpenAI's of the world, the Anthropic world to run compute on the infrastructure that passes to NVIDIA, etcetera, etcetera, etcetera.
所以我经常觉得这一点被讨论到了。
So I I often feel that gets discussed.
黄仁勋就是顺带提了一下，没有深入展开，但他确实提到了这个观点。
Jensen just, you know, in passing does that, but he didn't deeply dive on that, but he sort of made that point.
所以，总之，我觉得他试图去回应这个问题，但我也在想，你知道，部分原因可能就是他确实希望英伟达的GPU在全球范围内有更多样的来源，这样就能形成一个巨大的英伟达算力池，来吸收未来所有企业的需求。
So, anyway, I feel like he tried to do that, but I also wonder if like, you know, again, part of this is just he does want that diversity of NVIDIA GPUs out there in the world from as many sources as possible so that there's just a giant pool of NVIDIA compute that can be there to absorb the coming demand of all of the enterprise.
所以我对这件事有两种看法，或者说这件事有两面性。
So I'm I'm of two minds about this, or there's two sides to this.
一方面，这在一定程度上就是厂商融资。
One is, to some degree, this is just vendor financing.
你买车的时候，你是在向丰田或通用汽车借钱，就是卖你车的那家公司。
You go buy a car, and you are borrowing money from Toyota or GM, whoever's selling you that car.
对吧？
Alright?
这很简单明了。
That's that's pretty straightforward.
而且NVIDIA也可以说，他们可能会从中获得收入分成，而且他们在投资这个热门的新领域。
And NVIDIA can also argue that they're gonna have probably have revenue share in this, and they're investing in this great hot new category.
好的。
Okay.
但另一方面，就像我一直说的，帮助客户购买东西、刺激需求和真正创造需求之间，界限非常模糊。
The the flip side of this is it's like I keep saying, there is a very fine line between enabling your customer to buy something, enabling demand, and then actually creating that demand where it doesn't exist.
我说的是对计算系统的需求。
And I'm talking about demand for compute systems.
至于那些B2B的最终用途是什么，最终需求在哪里，那是另一个完全不同的争论。
Whole other debate about what the the end of all those b twos are, where the end demand is there.
那是另一个话题。
That's a whole different topic.
我的意思是，如果英伟达没有介入为这些交易提供融资，这些交易很可能根本不会达成。
I'm just saying if in if NVIDIA did not step in to provide financing for these deals, the deals presumably would not get done.
对吧？
Right?
所以，你要么得相信一种可能性，也就是英伟达所说的，市场、资本市场没有理解这个机会，给它错误定价了。
And so there, you either have to believe that a one one possibility is that and this is what NVIDIA is arguing, is that the market, the capital markets, do not understand this opportunity, they're mispricing it.
所以英伟达要介入，弥合这个差距，修正定价，直到其他人都明白这有多天才。
And so NVIDIA is gonna step in and bridge that gap to fix the pricing until everyone else figures out how how genius it is.
另一种可能就是，这里面有很多泡沫在酝酿。
The other alternative is that there is, like, a lot of bubbling going on.
对吧？
Right?
其实黄仁勋在推特上发的关于这个的内容比新闻稿里还多一些。
Jensen actually posted on Twitter a little bit of more about this than was in the press release.
他实际上问了一个问题，这是不是循环融资？
And one of the one of the things he actually asked that question, is this circular financing?
然后他说了一大堆话，但根本没有回答那个问题。
And then he has a whole bunch of words, none of which are actually answering that question.
所以我觉得这里确实存在一定程度的循环融资。
And so I I think there is a degree of circularity here.
你从英伟达借钱，然后去买英伟达的产品。
You borrow money from NVIDIA to buy NVIDIA products.
这有多糟糕呢？
How how bad is that?
你知道，这个问题有两面性。
We we we can you know, there's there's two sides to that.
但我觉得关键是规模太大了，550亿美元可不是小数目，这让人们有点紧张。
But I think this this just the sheer size of it, I mean, $0.5500000000000 dollars is a lot of money, has people a little bit on edge.
对。
Yeah.
所以我要说几点。
So I'll say I'll say a few things.
我完全同意这些基础设施能赚钱，资本回报率也很吸引人。
I I I am completely am on board that that this infrastructure makes money, that there is an attractive return on capital.
我这么说，不只是针对英伟达。
And I'm saying this broadly, just for NVIDIA.
我也在说定制ASIC芯片。
I'm saying this for custom ASIC.
所以超大规模数据中心和其他公司，我理解那些新创公司对AI的敏感性，因为他们的业务就是AI，但再说一次，我认为这些资产在整个生命周期内是会赚钱的。
So the hyperscalers and others, I get the sensitivity of the neo clowns because their business is just AI, but but again, I think over the lifespan of of those assets, they will make money.
我还没完全想明白的地方，我会解释为什么，就是黄仁勋说的，算力是一种可投资的资产。
I I guess where I'm not totally solved, and I would say, I'll explain why, is the idea that compute is an investable asset, as Jensen says.
因为我觉得这意味着算力永远都不够好。
Because I think that necessitates that they never they're never good enough.
算力永远不够好，因为我觉得这意味着，如果你要持续投资算力，而算力是可投资资产，那它就必须保持对下一代产品的吸引力，才能放进你的数据中心。
That compute is never good enough because I feel like it means that if you're gonna keep investing in compute and compute is an investable asset, it has to maintain next generation's attractiveness to put into your data center.
而不是说，我可以一次性建好所有算力，然后它就放在那儿折旧，够用了，我就不再需要明年下一代的产品了。
Not that I can build it all, and then it just sits there and depreciates, and it's good enough, and I don't need next generation next year's products anymore.
也许我错了，但我是这么想的，就是我得不断往里投钱。
Maybe I'm wrong, but that's how I'm thinking about it, that it kind of has to I have to keep investing in it.
它得足够好，能推动总拥有成本，让我愿意继续投资。
It has to be good enough to push the TCO that I keep investing in it.
我的意思是，我不会投了之后就不再投了。
I don't invest in it and stop investing in it is my point.
我觉得这是英伟达商业模式核心的一个根本性难题，就是他们得不断推出新系统。
I I think this is a fundamental dilemma at the heart of NVIDIA's business model now, which is they need to keep coming out with new systems.
他们得不断提高价格，尽可能卖高价，让数字继续增长。
They need to keep raising prices on that, selling them for much as they they can to keep numbers moving.
但与此同时，他们也在很大程度上消耗客户的资产负债表。
But at the same time, they are taxing the balance sheet of their customers to a high degree.
这听起来很悲观，但我认为这里有一个合理的战略逻辑，那就是英伟达希望新云厂商存在，因为他们宁愿卖给200个客户，也不愿卖给五个拥有所有权力的客户。
And that sounds that sounds very bearish, but I think there is a plausible strategic rationale here, which is NVIDIA wants NVIDIA wants the neo clouds to exist because they they would rather sell to 200 customers than to five customers who have all the power.
这是一种非常非常不同的动态。
It's a very, very different dynamic.
所以，我我我我明白，我这边的问题是，所有这些讨论里有很多泡沫泡沫的说法，这也很正常。
And so I I I I get I mean, part of my issue here is that there's a lot of, like, bubble bubble talk in all of this discussion, rightfully so.
比如，很多Jensen的推文都挺泡沫化的。
Like, a lot of a lot of Jensen's twit tweet was kinda bubble speak.
所以我觉得，这确实让人有点紧张。
And so that I think that rightfully makes people a little nervous.
但这里有一个战略上的理由，就是说，过去一个月变得很清楚的是，AI竞赛不仅是技术之争，也是资本之争。
But there is a strategic rationale here, which is mean, what's become clear over the last month is the the race for AI is as much about capital as it's about technology.
大家都负债累累，把自由现金流烧成负数，来支撑这种建设。
Have everybody going into debt, burning the cash flow free cash flow into negative territory to fund this build out.
而且我觉得回报并不明确。
And I don't think the payout is clear.
我不是说不可能有回报。
I'm not saying it can't exist.
我只是觉得现在还不清楚。
I just don't think it's clear.
你得在模型里加入一些相当大胆的假设，才能真正让这些投资回本。
You have to put some pretty heroic assumptions into your models to really get these to pay off.
话虽如此，这周CoreWeave和Nebius这两家领先的新云服务商都发布了财报。
That that being said, we did have CoreWeave and Nebius, the two leading Neo Clouds report this week.
两家数据都不错。
Both had good numbers.
两家股价反应都非常强劲。
Both saw saw very strong positive stock reaction.
我觉得这里既有短期逻辑，也有长期逻辑。
And I think there's there is a a short term case here and a long term case.
短期来看，如果你现在能买到英伟达的GPU并部署起来，不到两年就能回本。
The short term is if you can buy NVIDIA if you can buy GPUs today and get them deployed, you can pay those systems back in under two years.
对吧？
Right?
英伟达在我们那次电话会议上也这么说了。
And NVIDIA said this on our call.
以前他们大概要三年才能回本。
It used to take them about three years to get payback.
现在两年就能回本了。
Now they're at two year paybacks.
两家公司都谈了很多，说他们现在以比以前高得多的价格续租Hopper和Ampere系统。
Both companies talked a lot about how they're renewing hopper and Ampere systems for much higher rates than the previously.
价格在上涨。
Prices are going up.
所以我不反对。
And so I don't disagree.
短期内，如果你有算力，你就能卖出去。
In the near term, if you have compute, you can sell it.
对吧？
Right?
但我不太清楚的是，这两家公司特别依赖少数几个客户，比如超大规模云服务商和前沿实验室，而这些客户现在都在建自己的数据中心。
But less clear to me is the extent to which those two in particular are very dependent on a small number of customers, the hyperscalers and the Frontier Labs, who are all now building their own data centers.
所以这就有点模糊了，因为这里面涉及太多层次。
And so it gets a little little murky because we have so many layers in this.
黄仁勋喜欢谈层次。
Jensen likes to talk about layers.
这里面确实有很多层次。
There's a lot of layers in this.
我觉得现在显然有很多热情。
And I I think there's clearly lots of enthusiasm right now.
但如果你说的是多年期的视野和未来可预见的大量资本支出，那我能理解为什么人们开始有点动摇了。
But if you're talking about multiyear horizons and trillions of dollars of CapEx for the foreseeable future, it it is I understand why people are getting a little wobbly on this.
对。
Yeah.
对。
Yeah.
我的意思是，想想看，我觉得这又回到了回报率的问题上，而回报率又有很多变量，显然包括GPU的利用率、你能收取的价格，以及所有这些在长期内高于你在建设和电力上的硬成本的部分。
I mean I mean, think, again, I think it comes back to the amount of the return, which again has a number of variables, which is obviously utilization of the GPU, price that you can charge, and all of those being normalized above your hard costs in construction and power over the length of time.
对吧？
Right?
你的固定一次性成本，相对于那些资产在GPU使用寿命内摊销的部分。
Your your fixed one time cost versus what's amortized over the length of your GPU value for those particular assets.
你明白吗？
You know?
而且再说一次，是的。
And and again, it's yeah.
我的意思是，我们有一个千兆瓦经济模型。
I mean, we have a gigawattonomics model.
你知道，我们之前谈过。
You know, we've talked about before.
我可以把这些假设都套进去。
I can run all these assumptions down to it.
然后，再把你验证CoreWeave所需的所有假设都带回来。
And, again, come back with all the assumptions you need to validate CoreWeave.
对吧？
Right?
他们每小时需要收多少钱？
How much do they need to val to charge per hour?
在什么样的利用率下？
At what utilization?
比如说，我们可以反推出来。
Like, we could we could back into that.
对吧？
Right?
我们甚至能算出它能赚多少。
We could even show how much it makes.
其实，我在推特上发过这个，假设十年寿命，不过微软和其他公司说他们会把寿命延长到二十年或二十五年。
In fact, I tweeted this how much that on a ten year assume a ten year life, which, again, Microsoft and others are saying they're gonna extend this to twenty or twenty five years.
我记不清了。
I can't remember.
但假设寿命是十年，并且按平均经济情况来算。
But assume a ten years life and average economics.
那么每小时6.5美元，利用率75%，我们算出来的资本回报率是超过200%的，对吧，在那个情景下。
So $6.50 an hour, 75 percent utilization, and we came back with well over just over 200% return on that capital, right, in that one scenario.
但你说得对。
But you're right.
问题是，这些投资是会停止，还是必须持续投入？
The issue is, do those investments stop, or do those investments have to keep continuing?
而且我觉得，除非你又说我理解错了，但黄仁勋说算力是可投资资产，我认为这必须假设它会持续增长。
And I guess and unless again, as you tell me I'm interpreting this wrong, but Jensen saying that compute is an investable asset, I think fee I think has to assume that it keeps going.
不是投完就停，然后折旧。
It doesn't invest and stop and depreciate.
对。
Yeah.
苹果农夫说苹果能治百病。
Apple farmer says apples cure all illness.
对吧？
Right?
你看，我跟你说。
I I I look.
我不是说这全是泡沫，也不是说这事永远不会成。
I I I'm not saying that this is all bubbly and this is never gonna happen.
我的意思是，没人知道。
What I'm saying is nobody knows.
没人知道五年后这回事会变成什么样。
Nobody knows where this is gonna be in five years from now.
而现在的问题是，我们开始签长期债务协议，一签就是五年、十年，我觉得这真的很难预测。
And the the problem we're having now is we're starting to sign long term debt agreements, extend out five and ten years, and I think it's very, very hard to forecast.
当然，今天我可以建一个模型，显示这些系统在两年内就能回本，回报非常可观。
Absolutely, today, I can build a model that shows payback on these systems is very strong in two years.
对吧？
Right?
但我们要达到的规模，尤其是随着价格上涨，比如记得詹森说过每吉瓦的价格在翻倍。
But the scale we're going to especially with prices increasing, like, remember Jensen saying prices per gigawatt are doubling.
你知道，在签这些长期协议的时候。
You know, at at you're signing these long, long term agreements.
我觉得你懂的。
I think you no.
任何人都不应该表现出高度自信，认为这肯定能卖得出去。
Nobody should have anybody who expresses a high degree of confidence that this is, you know, gonna be sell saleable.
五年后我们还在卖安培系统。
We're still gonna be selling Ampere systems five years from now.
也许吧，但没人知道。
Maybe, but nobody knows that.
所以我觉得，你知道吗？
And so I think there's you know?
而且，我们面对的是债券投资者。
And, again, we're we're dealing with bond investors.
债券投资者不关心人工智能。
Bond investors don't care about AI.
他们关心的是什么时候能拿回钱。
They care about when they're gonna get paid back.
到目前为止，他们愿意赌一把，因为短期数据太好了。
And so far, they're willing to take that bet because the near term numbers are so good.
你两年内就能把所有的钱还清。
You get to pay everything back in two years.
但我认为随着数字越来越大，这会变得更加困难。
But I think with the numbers getting as big as they are, that gets more challenging.
对。
Yeah.
我的意思是，我和一些从事GPU即服务公司的人聊过回报率，他们发现他们的融资在三年内就能还清，而且这种情况相当稳定。
I mean, like, I've talked with people who work with some of these GPU as a service companies on rates of return, and they've been seeing their as aided financing be paid back in three years pretty consistently.
所以，这确实在发生。
So again, that's happening.
我觉得我们一直在绕的那个点，就是你说某个东西是可投资资产的时候，我明白你的意思，关于农场和苹果，但是明年的苹果，我不会……天哪，我要把这个类比彻底搞砸了。
I think the the point that we're dancing around those again, like when you say something's an investable asset, and I get your point, right, about farming and and and apples, but next year's or next year's apples, I don't I don't take my god, I'm gonna completely bork this analogy.
我不会把我苹果园的一部分换成能让我苹果产量翻倍的新树。
I don't I don't I don't replace portions of my apple farm with new trees that double performance of my apples.
我才刚买下这个。
I just I just bought this.
但你明白我的意思吗？
But do you see what I'm saying?
就是说，你得继续……反正你已经在里面了。
Like, you have to keep you're you're in anyway.
对。
Yeah.
没错。
That's right.
我知道，当你建立这些模型的时候，那会成为一个重要因素。
I I and I just know when you when you build out these models, that becomes a major factor.
对吧？
Right?
到某个时候，你得达到稳定状态，让它们产生现金流。你看，亚马逊就是这么做的。
At some point, you need to get to steady state where they're generating And cash look, they're like, Amazon did that.
AWS也是这么做的。
AWS did that.
他们亏了好多年钱，现在却非常赚钱。
They went through years and years of losing money, and now it's hugely profitable.
但他们的升级周期要慢得多。
But it was on a much slower upgrade cycle.
而且相对于经济规模，他们的开支要低得多。
And and the the the expense was much lower relative to the economy.
对。
Yeah.
我的意思是，但与此同时，我觉得亚马逊——我特别提一下——不会否认，在他们建设AWS并持续运营AWS的过程中，他们的基础设施以及持续投入这些基础设施，并不是对资金的有效利用。
I mean but but at the same time, like, I I don't think Amazon, I'm gonna mention this specifically, wouldn't disagree that as they were building AWS and they continue to sustain and run AWS, that their infrastructure and continuing to invest in that infrastructure was not a valuable use of their money.
他们把这些基础设施变现，用来运营亚马逊网站，托管别人的邮箱和SaaS服务器之类的。
Like, they monetize that to, you know, run amazon.com and host people's email and SaaS servers and whatnot.
而且那个业务是可持续的，他们也从中赚到了钱。
And that that business was was sustainable, and and they made money on it.
所以AI也是类似的。
So AI is similar.
我另外想说的是，我觉得英伟达喜欢这个交易，喜欢这个想法，因为它把客户群扩大到了新云服务商之外。
The the other point I'll make is I think I think he like I think NVIDIA likes this deal this idea because it it it it widens the pool beyond the Neo Clouds.
而且我确实觉得，新云服务商想跟超大规模云服务商竞争，这种心态在某种程度上伤害了他们与那些超大规模云服务商之间的潜在长期关系。
And I do think there is some sensitivity around the Neo Clouds wanting to compete with the hyperscalers that is hurting some of their potential relationships long term with those hyperscalers.
而那些超大规模云服务商和其他实验室都在那儿说，我们需要更多的算力。
And the hyperscalers and other labs are sitting there saying, We need more compute.
所以如果算力就在那儿，这就是为什么我觉得像Crusoe、Lambda以及其他一些公司，作为超大规模云服务商或前沿实验室的算力合作伙伴很有吸引力。我认为他们可以在利润率上获得相当不错的经济效益，对于那些想把自己基础设施的溢出需求转移到这些即将建成的、纯粹是AI工厂的新数据中心的人来说。
So if it was there, which is why I think the likes of Crusoe and Lambda and some of these others are are attractive as partners for set hyperscalers or Frontier Labs for compute, that I think that they they can have pretty attractive economics for their margins, for people who want to do kind of spillover from their own infrastructure into some of these these new data centers that are gonna get built that are really just pure AI factories.
对吧？
Right?
不是想成为那种新云，也不是想成为超大规模云服务商。
Not trying to be a like a Neo Cloud is, like, not trying to be a hyperscaler.
所以我觉得这里也有一个动态因素在起作用。
So I think there's a dynamic here at play as well.
对。
Yeah.
我觉得，这就是我开始真正质疑黄仁勋说法的原因。
I I and I think I mean, this is this is where I start to really push back on Jensen's language.
因为他不是说算力是一种可投资的资产。
Because he's not saying that compute is an investable asset.
他说算力是一整类全新的投资。
He's saying compute is a whole new class of investments.
你会有股票、债券、货币、房地产，还有算力。
You'll have stocks, bonds, currency, real estate, and compute.
但我觉得这不是真的，因为你刚才提到的那些原因。
And I I just I I I I don't think that's true because of all the reasons you just mentioned.
对吧？
Right?
土地不会过时。
Land doesn't get obsolete.
对吧？
Right?
你不需要保留，你也不能造出新的土地。
You you don't need to keep you can't make new land.
就是说，它根本就不是同一类东西。
Like, it's just it's not it's not the same category of things.
这是债务。
This is debt.
这是一种债务形式。
It's a form of debt.
对吧？
Right?
可能还掺了一点股权。
Maybe a little equity mixed in.
谁知道呢？
Who knows?
但我认为这里确实存在一个矛盾：英伟达希望每个人每年都升级、购买新系统，但要做到这一点越来越难了。
But I I think there is a real tension here where gents NVIDIA wants everyone to keep upgrading and buying new systems every year, and it's getting hard to do that.
成本越来越高。
It's getting very expensive.
对。
Yeah.
我知道。
I know.
我是说，我认为在可预见的未来，这种情况会持续下去，因为每一代新产品的总拥有成本，以及最新一代的每瓦特收入，都会表现不错。
I mean, I think this happens for the foreseeable future because the TCO of each new generation and then revenue per watt with the latest generation is gonna be good.
但我的观点是，在十年的时间跨度里，情况还会是这样吗？
My point though is over the course of ten years, is that true?
到2035年，明年的计算能力比去年提升30%到40%。
That in 2035, next year's compute is 203040% better than last year's.
不过，也许我想错了，也许不该这么想，也许你就是打算每六年更换20%的基础设施。
Again again, may may I guess, maybe I'm thinking about this wrong, and that's not the right way to think about it, and you're just gonna do rip and replace and replace 20% of your infrastructure every six years.
那也行。
Fine.
但我的直觉是，要支持你那种说法，某种程度上意味着设备几乎不会过时。
But but but but my hunch is that to to make the claim what you're saying, to some degree necessitates very little obsolescence.
我觉得太远的事情很难说，因为两种可能性都存在。
I think in I I think it's it's too hard to say too far out because I I could see it either way.
对吧？
Right?
我不是在唱多或唱空。
I'm not, again, I'm not playing bull bear here.
我只是说，如果AI继续以现在的速度进步，对人们越来越有用，而且真的开始，比如，研发药物、治愈癌症，那么十年后我们可能仍然面临算力不足的情况。
I'm just saying if AI keeps improving at the rate it's improving and becomes more and more useful to people and it actually starts, you know, coming up with drug cure cancer cures, we could very well be in a place where we still don't have enough compute ten years from now.
是的。
Yep.
但我的直觉是，这不会是一个平稳的过程。
But my my intuition is that it will not be an even move.
它不会永远一直往右上走，而且我们很可能在未来的某个时候，至少暂时，算力会过剩。
It won't keep going up into the right forever, And we could very well be in a position where at some point in the future where we we have too much compute, at least for the moment.
而且我觉得没人真的知道。
And I just don't think anybody knows.
这是实话。
That's the honest answer.
没人知道。
Nobody nobody knows.
对吧？
Right?
对。
Right.
说实话，我觉得正是这一点让大家心里都没底。
And and I think it's that bit that weirds everybody out, to be honest with you.
我完全理解。
And totally get it.
我明白那种紧张感。
I get the tension.
就像我说的，我觉得两种可能性都存在。
Like I said, I could see it both ways.
就像我之前指出的，如果这种工作负载是图形和视觉艺术类的，那需求只会一直增长。
Like I pointed out before, if this workload is like graphics and visual arts, it just keeps going.
每年你都需要更多，因为你在追求逼真度。
Every year you need more because you're chasing fidelity.
如果你在追求通用人工智能，而我们又没真正达到，或者总是差那么一点基础设施，那我也能理解。
And if you're chasing AGI and we don't really get there or it's always the next bit of infrastructure away, I I get it.
所以你说得对。
So you're right.
但我觉得，这种敏感之处就在于，当你根本不知道的时候，很难把握那个愿景。
But I think that's the sensitivity to it is it's it's hard to grasp the vision when you just don't know.
从逻辑上讲，两边都有道理。
Like, logically, you could see both sides.
所以，时间会证明一切。
So time time will tell.
对。
Right.
时间会告诉我们答案。
Time will tell.
但你知道，我的意思是，在可预见的未来，我们还是会受限于算力。
But, you know, I mean, for the foreseeable future, we're compute constrained.
所以，这至少是我们共同的基础。
So that is at least our common denominator.
好了，各位。
So alright, everybody.
感谢收听，我们下周再聊。
Thanks for listening, and, we will talk to you next week.
谢谢大家。
Thank you, everybody.
记得点赞、订阅，告诉你的朋友们。
Click like, subscribe, tell your friends.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。