S&P hits all-time highs, AI trade does all the work

欢迎大家回到投资专家播客。

Welcome back everybody to investing experts podcast.

今天我又请到了阿姆丽塔·罗伊和乌塔姆·戴，他们是我们Seeking...的一对分析师夫妻。

I am here once again with Amrita Roy and Uttam Dey, a married pair of analysts we have at Seeking...

我们很幸运能在Seeking Alpha拥有他们。

We are fortunate enough to have at Seeking Alpha.

之前他们来过我们播客，但确实隔了太久了。

We have been fortunate enough to have on the podcast before, but it's been too long to be sure.

乌塔姆和阿姆丽塔，欢迎回到节目。

Uttam and Amrita, welcome back to the show.

欢迎回到播客。

Welcome back to the podcast.

很高兴你们能来。

It's great to have you.

如果你们想开始的话，你们自己决定谁先来吧。

If you wanna just start, and I will let you decide who goes first.

如果你想先打个基础，说说你怎么看待当前这个市场，现在是2026年10月12号，你怎么看科技板块，以及你重点关注哪些领域，我很想从这里开始。

If you wanna just lay the groundwork, how you're looking at this market, it's 10/12/2026, how you're thinking about tech and the areas that you're focused on, I'd love to start there.

好的。

Sure.

那我就先开始吧。

I'll get it started.

所以我会，我会，我会先提供一个

So I'll I'll I'll provide

一个

a

大致的概述，先来个宏观概述，然后我们可以慢慢深入探讨我们认为在未来几个月和几个季度会受益的关键科技领域。

broad overview, like a macro overview first, and then we can, like, slowly start start drilling into the key areas of tech that we believe will benefit in the coming months and quarters.

所以标普500指数上周终于创下了历史新高。

So the S and P five hundred finally managed to create an all time high last week.

我觉得它现在仍然在历史高点附近徘徊，尽管今天市场略微下跌或持平。

I think it is still sort of hovering around an all time high even though, you know, markets are slightly low to flat today.

但有趣的是，推动标普500创下历史新高的，全是人工智能和科技股。

But the interesting bit is that it is all AI, all tech that actually did the work for the S and P five hundred to create that all time high.

所以AI交易领涨，而其他板块，比如金融、工业、小盘股，基本上都没怎么参与。

So the AI trade led the way while other other sectors like financials, industrials, small cap, you name it, they pretty much did not participate at all.

奇怪的是，尽管AI交易把标普500推到了历史新高，但SMH，也就是Van X半导体ETF，这个可能是最常用、最能代表AI交易的基金之一，却没有创出新高。

And the strange bit in all of this is that even though the AI trade took the S and P 500 to all time highs, SMH, the Van X Semiconductor ETF, which is probably one of the most used and one of the best proxies for the AI trade, has not made a new all time high.

所以我们这里要面对几个矛盾。

So there are a couple of contradictions we are dealing with over here.

与此同时，在宏观经济方面，十年期国债收益率差不多处于本世纪高位，粗略来说是这样。

And meanwhile on the macroeconomic front we have the ten year yields that are at the century highs pretty much, a rough figure.

美元处于历史...

Dollar is at an all time...

美元目前相当强势。

Dollar is pretty strong at the moment.

油价并没有...

Oil is not...

油价就是不肯跌破每桶88美元，这好像在强化“高利率持续更久”的说法，也让流动性状况变得更糟。

Is refusing to move lower beneath the $88 per barrel level, which is like reinforcing the higher for longer narrative and also worsening liquidity conditions.

而流动性一旦变差，通常就意味着市场要回调了。

And when liquidity conditions worsen, that usually calls for a pullback.

但这次不一样，我们正处在中期选举的月份和季度，从历史来看，10月和11月在中选期间……

Except this time, we are dealing with a midterm election month, quarter, where we're looking at like history, October and November during mid...

中选期间往往是一年中最强劲的月份。

During a midterm tends to be the strongest months in a year.

所以现在市场上有好多矛盾在同时上演。

So you have a bunch of contradictions that are playing out in the market right now.

我们觉得现在的情况是，资金正在重新流回科技股。

So the way that we see this playing out is that currently what we are seeing is that money is rotating back into tech.

比如从6月开始……

Like since June...

从6月到……

From June till...

到目前为止，我们看到资金从科技股流出，转向再通胀板块，比如软件行业。

Up until now, we saw money rotate out of tech into, you know, your reflationary sectors and also, software, for instance.

现在，我认为从上个月末开始，资金终于又流回科技股了。

And now finally, I think starting late last month, money is moving back into tech.

是的，市场广度不佳，但资金仍在回流科技股。

Yes, breadth is poor, but still money is moving back into tech.

我看到网上流传着很多图表，人们把标普500的走势和互联网泡沫时期做对比。

I see a lot of charts going around in the internet where people are drawing parallels between the price action of the S and P five hundred versus the .com bubble.

我个人认为，至少目前，我们并不处于泡沫之中。

I just personally don't believe that we are in a bubble at all, right now at least.

因为估值相当稳定且偏低。

And that's because valuations are pretty stable, low.

盈利非常强劲。

Earnings are very strong.

目前，AI收入每7.2个月就翻一番。

When we look at AI revenue, it is doubling every seven point two months at the moment.

供应链非常紧张。

Supply chain is extremely constrained.

我是说，CPU的交货时间目前已经变得非常紧。

I mean, at CPU lead times, which have gotten quite tight at the moment.

AMD和英特尔都在提高CPU价格。

AMD, Intel are increasing their CPU prices.

AWS、Nebius，等等，他们都在提高GPU价格。

AWS, Nebius, you name it, they're increasing their GPU prices.

所以是的，这是一个需求极其强劲、供应还在努力追赶的世界。

So yeah, this is a world where demand is extremely strong and supply is still trying hard to catch up.

所以我不认为我们处于任何泡沫区域。

So I don't think we're in any kind of bubble territory.

话虽如此，考虑到我们现有的中期支撑或底部，我确实认为标普500可能还有一点上涨空间。

Having said that, given the sort of the midterm support or the floor that we have, I do think that, you know, there's probably still a marginal high left for for the S and P 500.

所以我认为我们可能还会继续走高，可能会到8000或8100点。

So I think we probably will still move higher from here, possibly 8,000, 8,100.

但之后，我们预计指数可能会有5%到15%的回调，具体取决于流动性状况如何发展。

But I am, we are expecting probably a five to 15% pullback in the index after that, depending on how the liquidity conditions play out.

我在Seeking Alpha上所有的宏观文章里，一直在讲我们是如何应对这种情况的……

And I have been talking a lot in through all my macro posts on Seeking Alpha as to how we are navigating this scenario in the...

我们计划在未来几个月如何应对这种情况，通过平衡投资组合中的现金、增加现金储备，并选择性地投资一些高确信度的股票。

How we plan to navigate this scenario in the coming months by balancing our cash in the portfolio and raising cash and selectively investing in certain high conviction names.

标普500指数可能还有1%到2%的上涨空间，随后根据流动性状况，可能会出现5%到15%的回调。

There is still probably a one to 2% left in the S and P 500 followed by a 5% to 15% pullback depending on liquidity conditions.

作为AI投资者，我们并不认为基本面在恶化。

As AI investors ourselves, we don't think that the underlying fundamentals are worsening by any means.

当然，我们确实需要有利的流动性条件。

We do need favorable liquidity conditions for sure.

因此，在那次回调中，我们会非常看涨，并将其视为一个绝佳的长期布局机会。

So we will be very bullish in that pullback and view that as a tremendous long term setup.

所以，我想这就是我们目前对科技股的看法，以及未来大约一个季度的展望。

So that's sort of, I think, where we see tech currently and playing out in the coming maybe quarter or so.

在这个范围内，是的，我们……

And within that, yeah, let's let's...

我们可以深入聊聊这些领域。

We can we can get into sort of the areas.

我知道现在智能体AI非常热门，上个月Meta Muse刚发布。

I know agentic AI is huge right now with with Meta Muse launching last month.

然后还有OpenAI的Dots……

And then OpenAI Dots...

OpenAI发布了Dots，这是他们的个人AI智能体，然后谷歌也推出了Gemini……

OpenAI releasing Dots, which is their personal AI agent, and then Google launching Google Gemini...

或者说Gemini智能体，就在上周。

Or Gemini agent last last week only.

智能体AI领域现在竞争非常激烈，个人和……

The agentic AI space is getting really heated up, and the lines between personal and...

消费级AI与商业AI或商业智能体之间的界限正在变得模糊。

Consumer and business AI or business agent is blurring.

像OpenAI和Gemini或者谷歌这样的公司，是从企业角度切入的，而Meta则是从另一个角度切入的。

You have companies like OpenAI and Gemini or Google who are approaching it from the enterprise angle, whereas Meta is approaching it from...

你知道，Meta通过让Muse保持广泛免费，尽可能让更多人用上，来推动普及，主攻消费者市场。

Know, by keeping Meta Muse broadly free and getting it in the hands of as many people as possible to like drive adoption and sort of attack the consumer space.

所以，我觉得会有很多受益者出现，尤其是在超大规模数据中心领域、网络安全领域，甚至整个AI基础设施领域。

And as a result, there are lots of beneficiaries that I think will come about, especially within the hyperscaler complex, within cybersecurity, and even in the whole AI infrastructure space.

Amrita刚才引用了一个很好的概述，讲的是宏观层面正在发生的事情。

Amrita just citing did a pretty good overview about, you know, what's been going on at a at a macro level standpoint.

我想补充的一点是，当我想深入AI领域时，我们在宏观层面看到的那些噪音，可能正在引发一些担忧，这些担忧被传导到比如债务或者特殊目的载体债务上。

The only thing that I would add when I'm trying to get vertical within, you know, just the AI space is that a lot of that noise that we see, that we are seeing in the macro level is kind of probably triggering some kind of fears that is, you know, that's just getting funneled down into, let's say, something like debt or something like special purpose vehicle debt.

这是一个方面。

That's one area.

另一个方面是，嘿，等等，也许整个存储领域正在变得通胀，对数据中心来说。

The other area is, hey, wait, maybe the whole memory space is is getting it's it's getting inflationary for for data centers.

这也在很大程度上主导了叙事，并让叙事左右摇摆。

And that's kind of crowding a lot of the narrative as well as swinging it from side to side.

这让某些投资者真的很难办，挑战很大。

That's making it, you know, really difficult and challenging for certain investors.

所以我觉得，除了阿姆丽塔刚才讲的大方向之外，我想补充的就只有这一点。

So I guess, I think that's the only other additional piece that I would just layer on top of big picture that Amrita gave.

我之前有几个点想提一下。

There's a few points that I wanted to pick at before.

我很想听听你对各个子板块的看法，哪些你比较看好，哪些不太看好。

I'm happy to hear your thoughts on the various subsectors that you are more excited and or less excited about.

但关注我们播客的朋友都知道，我每天早上都会和史蒂夫·克雷斯聊，他一直在讲中期选举的季节性效应，以及它对市场的影响。而且通常来说，中期选举之后，历史上市场往往会大涨。

But as followers of this podcast know, I've been talking every morning to Steve Cress, and he's been talking a lot about this midterm election seasonality and how it's affecting things and that subsequent to the midterm elections, typically, historically, the market has exploded thereafter.

首先，我想问问你，作为至少是个宏观观察者，你怎么看我们现在所处的这个季节性节点？这跟你说到的科技股回归又有什么关系？

A, I'm wondering as a, you know, a macro observer at the very least, what you would say about the seasonality point that we're in, how that pertains to also what it seems to be, as you said, a return to tech.

然后如果可以的话，如果不……

And then if I may, if it doesn't...

我想知道，关于科技股的泡沫讨论，对你有没有什么参考价值？

I'm wondering if the bubble conversation about tech, does that serve you at all?

这对你的洞察或分析有帮助吗？

Does that serve your insight or analysis at all?

你觉得讨论这个是不是浪费时间？

Do you feel like it's a waste of time to have that conversation?

你觉得我们可能快要进入泡沫了吗？

Do you feel like we were maybe about to be in the bubble?

比如，你的……

Like, what's your...

我很好奇你对整个这种说法，以及围绕这个话题的讨论有什么看法。

I'm curious your thoughts about that whole narrative and and kind of the discussion around that topic.

因为这个话题确实被炒得很热，但我不确定这些热度有多少价值。

Because it sure gets a lot of play, and I don't know how worthy all that play is.

当然。

Sure.

这个问题问得真长啊。

That's a long that's a long question.

请享用。

Enjoy.

这个问题问得好。

It's a great question.

就是……

Was just like...

我的意思是，你知道，如果一两周内没听到任何关于AI泡沫的讨论，我们就会想，好吧。

I I mean, you know, if if if you don't hear any conversation around AI bubble for, like, a week or two tops, we'll be like, alright.

下一个头条什么时候来？

When is the next headline coming?

你知道，这可能会吸引很多点击。

Like, you know, it it it probably gets a lot of clicks.

每个人都在讨论AI泡沫。

It's Everybody's involved in the AI bubble conversation.

你叔叔、你爷爷都在讨论。

Your uncle, your grandfather is involved.

现在基本上人人都知道AI泡沫这回事了。

Like, everybody knows about an AI bubble now at this point.

但从基本面来看，确实是这样。

But from a fundamental standpoint, yes.

我是说，我确实觉得我们最终会走到那一步。

I mean, I do think we will eventually get there.

我们现在到了吗？

Are we there yet?

还远着呢。

Not at all.

就像我说的，我们才刚开始。

We are just getting started, like I said.

你看那些超大规模企业的资本支出，直到去年年初，全都是靠……

Like when you look at like hyperscaler CapEx, like up until early last year, all of it was used...

全都是靠经营现金流来支撑的。

All of it was funded using operating cash flows.

现在，超大规模数据中心的资本支出中，越来越多的一部分是靠债务来融资的。

Today, are seeing an increasing portion of hyperscaler CapEx being funded using debt.

甲骨文是受影响最严重的公司之一，因为它的信用评级被下调，信用违约互换利差也在飙升，诸如此类。

Oracle is one of the companies that is getting hit the worst because of its, you know, of its credit rating getting downgraded and its CDS spreads just like, you know, ballooning and things like that.

而且债务与资本支出的比率总体上也在上升。

And debt to CapEx ratios have been going up generally as well.

但再看看底层的指标，比如云收入在加速增长，云收入中的人工智能收入也在加速增长。

But then you look at the underlying, you know, the metrics underneath it, you know, things like, you know, your cloud revenue is accelerating, AI revenue within cloud revenue is accelerating.

积压订单在爆炸式增长。

Backlog is exploding.

最重要的是，我觉得投资者已经……

And most importantly, like, I think investors are up...

投资者常常忘记这一点，或者说，因为短期主义有时在我们心中根深蒂固，是的，超大规模数据中心现在在增长，但最终会开始正常化。

Investors often forget this or sort of like, because, you know, the short termism is like so ingrained in us sometimes is that, yeah, the hyperscaler is growing today, but it's eventually going to start to normalize.

然后它开始正常化，你今天那两万亿美元的积压订单，包括甲骨文的，也会开始转化为实际收入。

And then it starts to normalize and your $2,000,000,000,000 backlog that is there today, including Oracle's will also start to convert into real revenue.

然后就是……

And then it's...

而且不仅仅是转化为收入。

And and it and it is not just that it's converting into revenue.

它还是以更高的租金率转化为收入，因为GPU的价格实际上在上涨。

It is also converting into revenue at higher rental rates because your GPU prices are actually going up.

所以你把资本支出的正常化，与积压订单以更高的租金率转化为收入，再加上你将采购的额外AI工作负载量，结合起来。

So you you match your normalizing of Capex with, your backlog converting into revenue at higher rental rates plus additional volume of AI workloads that you will be procuring.

而且你的运营杠杆实际上会在一年内，大概十八个月左右，转为正数。

And your operating leverage will actually turn positive within a year, within probably eighteen months or so.

所以我认为自由现金流的拐点肯定会到来，投资者可能心里有数。

So I think a free cash flow inflection is definitely coming, which investors probably know at the back of their head.

但话说回来，我觉得大家还是有点担心，比如，好吧，但大部分承诺还是来自OpenAI和Anthropic，我同意这确实是个潜在风险点。

But again, like I think there is just a fear around like, okay, but most the commitments are still like OpenAI and Anthropic, which I agree is an area of potential risk.

不过，从基本面来看，我们并没有看到超大规模云服务商削减资本支出。

But yeah, like from a fundamental standpoint, we are not seeing hyperscalers pull back Capex.

实际上，上个季度，除了微软之外，其他三家——Meta、亚马逊和谷歌——都提高了他们的资本支出。

In fact, last quarter, except from Microsoft, all three of them, Meta, Amazon, and Google, they actually raised their Capex.

所以，是的，我确实认为泡沫的讨论会反复出现。

So, yeah, so I think, yeah, I definitely think that the bubble conversation will come in and out.

这也经常给我们提供很好的买入机会，尤其是当你清除噪音、着眼于长期前景的时候。

It also provides us sometimes very good buying opportunities, like especially when you like sort of clear out the noise and look at the long term picture.

比如，上个月，当AI恐慌发生，因为那个关于放缓的讨论，还有那个什么诉讼……

Like for instance, last month when the AI panic happened over the slowdown thing with all the cyber suit...

还有那个……

With all the...

还有Dario Amodi关于放缓前沿的论文。

With with Dario Amodi's essay on pacing the frontier.

然后大家都说，天哪，AI现在要减速了。

And everybody was like, oh my god, like AI is going to now slow down.

第二天，AI股票就出现了相当大的回调，我们得到了一个很好的买入机会，比如。

And the next day we got like a pretty massive correction in AI stocks and we got a great op buying opportunity, for instance.

所以，是的，这其实就是在管理围绕嘈杂叙事的波动性，同时保持对长期前景的关注，而我们认为这个长期前景目前依然完好无损。

So so yeah, it's it's it's it's it's managing volatility around noisy narratives while staying sort of focused on the long term picture, which don't think, which we think is fully intact at this point.

我喜欢嘈杂叙事这个词。

I like noisy narratives.

我喜欢这个说法。

I like that term.

非常贴切。

Very apt.

乌塔姆，关于阿姆丽塔的观点，你有什么要补充的吗？

Uttam, anything to add to that, to Amrita's point there?

是的，实际上我一直在看……

Yeah, actually I was looking at...

所以我觉得很多关于泡沫的讨论其实都源于资本支出，以及超大规模企业试图发行创纪录规模的债务。

So I think a lot of the bubble conversation actually was even coming from Capex and then how hyperscalers are trying to issue record volumes of debt.

所以现在有一种令人困惑的说法，就是你需要更多的资本支出来推动人工智能基础设施的繁荣。

And so there's this confusing narrative going out there that you're gonna need more Capex to fuel the AI infrastructure boom.

我可以这么说，我觉得2023年是个大年，那时候基础AI刚刚开始渗透到各个企业，尤其是大型企业。

And I can say this much that I think 2023 was one big year where you saw just this foundational AI just you know getting in and starting to percolate through all of the enterprises, especially the large ones.

然后2025年又是一个大年，至少对我来说，我会回头看看，可以说，我觉得那一年大家都开始关注AI到底怎么赚钱。

And then 2025 is like one another year at least in my, you know, that I will look back to and I can say like, I think that's the year where everyone started looking at how probably AI can actually monetize.

你可以看到，从那时起，内存需求开始上升，价格也开始上涨，还有光通信，整个转向光通信和网络，在2025年非常火爆。

And you can see like, you know, memory started to go up from there and then the prices started to go up from there and optical, you know, this whole switch to optical and networking, you know, was was huge in 2025.

这一切都是因为AI数据中心需要快速高效。

And that's that's all because like AI data data centers, they they needed to be quick and fast.

我觉得人们，尤其是在...

And I think people, especially in the...

在超大规模数据中心领域，他们能预见到会有一波影响浪潮即将到来。

In in in hyperscalers, they they they could foresee that there is gonna be some kind of a influence boom, that is gonna that is gonna come.

我觉得2025年就是我们看到这一切发生的大年。

And I think 2025 was one big year that we saw this happening.

所以这就是为什么当你看到资本开支增长率，也就是2023年到2025年的复合年增长率，大约在67%、68%左右。

So which is why when when you see the Capex growth rate, just the CAGR growth rates between 2023 and 2025, it's about 67%, 68 CAGR rates.

现在再看明年超大规模数据中心的资本支出，大概是一万三千亿到一万四千亿美元。

And and now when you look at like where hyperscaler Capex is gonna go into next year, that's about 1.3, 1,400,000,000,000.

所以从2025年到2027年，实际上年复合增长率还是保持在70%左右。

So between 2025 and 2027, that's that's still actually just about 70% CAGR growth rates.

但从2023年到2025年，有一个...

But from '23 to '25, you have a...

大量的资本支出被提前投入了。

It's a lot of Capex that's that's front loaded.

所以先是快速增长，然后逐渐趋于平稳。

So that's fast growth and then normalizing.

而从2025年到2027年，就是趋于平稳然后开始放缓。

And now from '25 to '27, you have that normalizing and then that's gonna slow down.

在整个资本支出增长曲线中，像微软和Meta这样的公司，也提前投入了大量的资本支出。

In that, you know, in that entire curve of of of CapEx growth, you have people like Microsoft and and people like Meta also that that had front loaded a lot of their CapEx.

所以这些公司现在正试图放缓他们的资本支出。

And so these guys are trying to slow their CapEx down.

他们可能会看到投资回报率的大幅提升。

They're probably gonna see, you know, some big booms in their in their ROI.

有一个指标叫ROI。

There's this metric called ROI.

看，这实际上说明了……

See, that actually illustrates how the the...

怎么说呢……

How are...

这些大型超大规模企业投入的资本和所有投资，其ROI到底怎么样。

How the ROI on Capex and all the investments that these big hyperscalers are making.

我觉得随着时间推移，我们会开始看到ROIC这个指标有显著改善，尤其是对超大规模企业来说。

I think we're gonna start to see some some major improvements in that ROIC metric, especially for hyperscalers as we move forward.

所以我……

So I I...

我对未来一到两年相当乐观，因为我们会看到推理真正转向大家期待已久的实际变现，这也是过去两三年一直在讨论的。

I'm I'm quite optimistic about where we're gonna go, you know, over the next one to two years as as we see inference switching gears into the real monetization that everyone really wanted to see and and it was talked about over the past two or three years.

我觉得我们还没到泡沫的地步。

I don't think like we're in this bubble territory yet.

现在的季节性因素可能会成为顺风，尽管流动性状况在十年期国债等方面有所恶化。

The seasonality right now is probably going to act as a tailwind despite the liquidity conditions worsening ten years and everything.

我认为这可能会成为标普500下一波上涨的顺风，虽然涨幅不会太大，但可能也就是8000、8200点封顶。

I think it's probably going to act as a tailwind for the next leg of the S and P 500, which I don't think is going to be much, but probably, you know, 8000, 8200 tops.

我觉得一旦中期选举的影响消退，我们就会遇到一些颠簸，但就像我说的，这对AI交易整体来说是好事。

I think once midterms fade, are in for some rough, you know, some rocky road, which again, like I said, is good for the

AI交易整体。

AI trade overall.

你有不同的看法吗？

Have a A different opinion?

稍微有点不同，这要追溯到2018年。

Slightly different And that's going back to 2018 Okay.

当时中期选举在11月举行，然后美联储主席杰罗姆·鲍威尔突然让市场意外，接着在12月突然加息，导致了12月的闪崩。

When the midterm elections happened in November and then at the time the Fed chair Jerome Powell had to suddenly surprise markets and then suddenly you know, increase rates in December and that caused this flash crash in December.

我觉得我们可能得做好准备，对。

I think we should maybe be just prepared for this Yeah.

另一种风险情景是，我们可能会突然看到，嗯，你知道，沃尔什主席试图不透露太多信息，就让市场自己去决定自己的方向和结论，关于利率前景的。

Alternate risk scenario where we could suddenly see a a you know a a you know chairman Walsh who is trying not to give as much information and and just let the market dictate their own forces and their own conclusions on the the interest rate outlook.

我觉得这算是我们大家都应该关注的一个风险情景……

I think that's that's kind of like one risk scenario that I feel like we should all be just looking in in...

向前看，或者往后看，看看会不会……

Looking forward in the rear or or looking forward ahead and see if that...

沃尔什主席有可能被逼到墙角，被迫加息，而且幅度远超市场预期。

That's something that could be possible that Chairman Wash could be pressured and cornered to to raise rates much more than than than what markets were expecting.

我觉得这是我们唯一需要考虑的另一种情景。

I think that's the only other scenario that we should factor

还有一件事，我之前和史蒂夫·克雷斯聊过……

Another thing I was talking about with Steve Cress actually was...

我想听听你对今天报道的两笔交易有什么看法，有报道说亚马逊在洽谈收购笛卡尔，英伟达在洽谈收购反思AI。

And I'm curious your opinion of the two deals that were reportedly in in talks today with Amazon reportedly in talks to acquire Descartes and NVIDIA reportedly in talks to acquire Reflection AI.

你对并购这块有什么看法？你觉得整合会怎么进行，或者你对现在这个周期所处的阶段感到兴奋吗？

Any thoughts on the m and a section and how you see consolidation happening happening and or if you're excited about it, where where we're at in that cycle?

最简单的回答，直截了当地说，并购……

The the easy answer, the straight answer is m and a...

如果并购在发生，那就说明AI业务的前景依然强劲，这是一个信号。

If if m and a is happening, that's one sign of of of AI business that that the outlook for AI is still strong.

如果并购在发生，那就说明背后有基础动力，每个人都在努力保持竞争力，都想收购竞争对手或者收购别人来获得一些优势，从而在某个领域里过关斩将，变得更有竞争力。

If if m and a is happening, you know that there is the underlying, you know, everyone's trying to be competitive, everyone's trying to buy out the competitor or trying to buy out someone else to gain some leverage to to, you know, jump through the hoops and be more competitive in in a certain space.

比如……

For...

就拿亚马逊收购，我记得是收购Descartes来说吧。

Let's let's take for example Amazon's acquisition of, I think was deck...

Descartes。

Tech Art.

亚马逊为什么想这么做，意图其实很明显。

It's quite clear like the intentions why Amazon was trying to do that.

这有点像是……

It's it's kind of...

至少有两方面吧……

There's there's, you know, at least two...

我读了那份新闻稿，我觉得我能……

I was reading through, you know, the the the press release and I think I can...

我能想出至少一两个他们这么做的原因。

I could come up with at least one or two reasons why they wanted to do that.

一个原因是他们有自己的芯片，就是他们的Trainium芯片，他们可能也想效仿谷歌的TPU模式，直接卖芯片。

One is they have their own chips, their own trainium chips that they're they're also trying to probably, you know, follow a Google TPU model where they'll they'll try to like just sell the chips directly.

另外别忘了，亚马逊是……

Also remember Amazon has...

亚马逊是唯一一家自己造芯片的超大规模云服务商，同时还跟英伟达有合作，用的是NVLink融合协议，那是英伟达用来构建AI数据中心扩展网络的专有技术。

Is is one of the only hyperscalers that makes their own chips and has a partnership with NVIDIA via the NVLink fusion protocol that is that is NVIDIA's IP for building the scale up networks in in AI data centers.

所以我觉得，通过收购Descartes AI，他们可以利用这家初创公司的芯片优化软件，用来精准地优化工作负载的分配，把它拆成最优的块。

And so I think by by buying Descartes AI, they could leverage the chip optimization software from this from this from this startup, which could then be used to, you know, to to to accurately and and and to optimize how workloads get split up into the best pieces.

然后一部分工作负载可能会被送到英伟达的GPU上，另一部分则可能被送到亚马逊的Trainium芯片上。

And then part of it could be sent to maybe NVIDIA's GPUs, another part could be sent to Amazon's Tranium chips.

所以他们是想尽量给亚马逊的客户提供最好的价值。

And so they're they're trying to like, you know, give the best value out to whoever is is Amazon's customer.

我觉得这大概就是他们这么做的动机吧。

I I guess that's kind of the motive for doing that.

你可以看到，他们试图通过这一切来提供价值，以保持竞争力。

And you can see like they're trying to, you know, provide value in all of this to stay competitive.

一方面，通过销售这些价值，他们可能甚至能提升AWS的毛利率上限，这对亚马逊来说是个需要持续展示的重要指标。

A, by by selling the value, they can probably, you know, even, you know, push the the limits of their own gross margins for AWS, which is kind of like a big deal that Amazon needs to keep showing.

所以我觉得这对亚马逊、对AWS、对AI来说都是一个巨大的加分项。

So I think this is just like, I think it's it's a great plus for for Amazon and for AWS and for AI.

至于英伟达的交易，我没太关注，但我觉得这大概也是为了让英伟达在这个市场上保持竞争力。

With the Nvidia deal, didn't catch much of that deal though but I think like it's kind of also so that Nvidia can just stay competitive in this whole marketplace as well.

现在关于开放权重模型和开源模型的讨论真的很多。

There's so much of talk going on about open weight models, open source models now.

所以我觉得英伟达其实也是在努力保持竞争力。

So I think NVIDIA's trying to really also stay competitive around this.

你怎么看各个子板块？

What would you say about the sub sectors?

从软件和超大规模数据中心的角度，你怎么看？

What would you say in terms of software hyperscalers?

哪些公司值得看好？

Who is there to like?

哪些公司不太值得看好？

Who is there not to like?

为什么选择个股或特定ETF？

Why stock specific ETFs?

很乐意听听你的想法。

Happy to hear your thoughts.

我可以先从超大规模数据中心和软件说起，然后基础设施部分你来接着讲吧。

I can start again with the hyperscalers and software, And then I think you can take over the infrastructure bit.

就像我在宏观部分第一个问题里说的，代理式...

So like I said in the first question in the macro section that agentic...

关于代理式AI，我觉得我们现在真的还处于非常非常早期的阶段，第一批真正的个人AI代理已经交到用户手里了。

Around agentic AI, I think we are just in the very, very early stages of Agentic AI right now with the first actual personal AI agent muse that is in the hands of people.

这也标志着一个关键的转变，从被动的聊天机器人转向自主代理，它们可以在很少或没有人类监督的情况下完成任务。

It also marks a pivotal shift away from this passive chatbots into autonomous agents who can work, who complete tasks without or with minimal human supervision.

所以这是一个范式的转变。

So it's a paradigm shift.

这是整个领域的一个根本性变化。

It's a fundamental shift in the whole sort of space.

说到整个超大规模云服务商，我想先从Meta开始。

I think when it comes to hyperscalers as a whole, I think I'll just start with Meta.

我知道我在之前的播客里也聊过Meta。

I know I spoke about Meta in the previous podcast as well.

当时Meta表现平平，但我仍然把它列为我最看好的股票之一。

When Meta was like going nowhere and I was still like pushing Meta as one of my highest conviction pick.

我觉得，它是我投资组合里第三大的持仓。

It is I think, my third largest position in my portfolio.

Meta 终于……

Meta is finally...

之前 Meta 一直没做好的地方在于，公司投入了大量资本开支，但非广告业务的收入却拿不出手。

Mean, was the thing that was not working for Meta up until now was that the company was spending a lot of Capex and there was not enough non ads business revenue to show for it.

所以投资者想看到的是独立的收入来源，也就是非广告收入。

So like investors wanted to see like evidence of standalone revenue, which is not ads revenue.

而 Meta 直到上个季度都拿不出来。

And Meta up until last quarter just couldn't show it.

可它又花了那么多资本开支，而且……

And yet it is spending so much Capex and yet...

同时它的利润率也受到压力，自由现金流也受压，所有这些。

And at the same time its margins are coming under pressure, free cash flow, all depressed, all of that.

我觉得 Meta 终于掌控了这个局面。

And I think Meta finally got control of that narrative.

而且，不只是听投资者怎么说，显然在那之前它就已经在研发了。

And I mean, listen to not just listen to investors, obviously it was working on news even before that.

随着MetaMuse的推出，它基本上已经证明了，确实有一个重要的业务线正在崛起，这个业务线不是广告，而且可以通过订阅，尤其是基于交易的收入来实现商业化。

With the launch of MetaMuse basically, it now has demonstrated that it can that there is a significant business line that is coming up, which is not ads and which can be monetized using subscriptions and most importantly, transaction based revenue.

而且我觉得目前消费级人工智能的机会非常大。

And I think the opportunity in consumer AI currently is huge.

比如说，Meta不只是把Meta Muse作为一个可下载的应用，还把它集成到了Meta AI眼镜里。

I mean, for instance, like Meta is not just sticking to Meta Muse as something downloadable, but it's also integrating it in their Meta AI glasses.

上周好像有个广告，人们用手机横着看网球赛，抱歉，是美网公开赛。

There's this commercial, I think, that came out last week of people using their phones to go sideways to watch a Wimbledon match, sorry, a US Open match.

然后，戴Meta眼镜的人只需要看一眼。

And, you know, the person wearing a Meta Glass would just, like, look.

所以你不需要

So you don't

你不需要

have to you don't have

不需要拥有那个设备。

to own the device.

就像你

Like, you

不需要，你不需要

don't have to you don't have

通过手机屏幕去看现实。

to live the reality by looking through a phone.

你其实可以直接戴上它。

You can actually just wear it.

我觉得这真的很酷。

And I thought it was very cool.

所以，是的，我认为MetaMuse真的能让他们的消费者业务突飞猛进，同时也能带动Reality Labs，那个部门之前一直有点……

So I think, yeah, so I think MetaMuse could actually skyrocket their consumer side of the business, as well as the Reality Labs, which also had been sort of...

它一直在进步，但和他们的广告业务相比，收入还是微不足道的。

It had been making progress, but it's tiny revenues compared to their overall ads business.

同时，我觉得Meta就是没有坚持做消费者市场。

And at the same time, I think Meta just didn't stick to the consumer lane.

他们还推出了企业平台，Meta企业平台，面向企业推广Muse。

They also rolled out their enterprise platform, Meta enterprise platform, where they're rolling out Muse for businesses.

我觉得OpenAI和谷歌在大企业方面有优势，而Meta在小企业方面有很强的定位。

And I think where like OpenAI and Google have like an upper hand with large enterprises, Meta has a pretty strong positioning with small businesses.

因为，你知道，大概有2亿企业已经在使用Meta在Instagram和WhatsApp上的服务了。

Because, you know, there's like probably 200,000,000 businesses who are already using Meta's services on Instagram and WhatsApp and things like that.

所以我觉得Meta有很多新的收入潜力，而且很快就会显现出来。

So I think there's a lot of new revenue potential for Meta that is coming up and it's going to come up very quickly.

投资者最初觉得，好吧，有多余的计算能力。

Investors were originally like, okay, there's excess compute capacity.

为什么不直接租出去呢？

Why not just rent it out?

但Meta说，等等，我们有更有战略意义的事情在做，而且会非常庞大。

But Meta was like, hold on, We have more strategic things we are working on and it's going to be huge.

我觉得我们开始看到这一点了。

And I think we are starting to see that.

之后股价显然飙升了。

The stock obviously skyrocketed after that.

不过还没创下历史新高，但我认为未来几个月、几个季度或一年内，Meta 有明确的路径成为千元股，这是我的看法。

Not yet made a new all time high though, but I think there's a clear path for Meta in the coming months, quarters, or a year to be a thousand dollar plus stock in my opinion.

所以，是的，我认为 Meta 是代理式 AI 的直接受益者之一。

So, yeah, so I think Meta is one of the immediate beneficiaries of agentic in my opinion.

我还认为，随着代理式 AI 的发展，实际上让我退一步说。

I also think that with agentic AI, actually I'll take a step back.

上个月很有意思，因为你有 Meta Muse 和……

Like last month was interesting because you had Meta Muse and...

它引领了代理式 AI。

Which spearheaded agentic AI.

与此同时，所有那些 OpenAI 的研究人员都离开了公司……

And at the same time, you had all these OpenAI researchers leaving the companies...

离开OpenAI，而且现在有很多关于AI安全的担忧。

Leaving OpenAI and there's so much just AI safety concerns going...

因为担心超级智能会毁掉人类生活之类的。

Because of fears of superintelligence wrecking human lives and whatnot.

所以问题是，有了代理式AI，网络安全只会变得更加重要。

So the thing is that with agentic AI, cybersecurity is only going to become so much more important.

你可以从网络安全股票看出来，比如CrowdStrike、Cloudflare，尽管它们的估值很高，但股价一直在飙升。

And you can see that with cybersecurity stocks like CrowdStrike, Cloudflare that have just been skyrocketing despite their premium valuations, for instance.

特别是Cloudflare很有意思，因为我知道在我的……

Particularly Cloudflare is interesting because I I know in my...

在上一期播客里，还在讨论SaaS对Calibs的影响，但我记得我当时就说，不，网络安全肯定会赢，Palantir也会。

In the last podcast, like, there was still SaaS for Calibs going on, but I remember I was like, no, cybersecurity will definitely be a winner and so will Palantir.

但说到网络安全，显然我是CrowdStrike的坚定看多者。

But coming to cybersecurity, obviously I'm I'm I'm a huge CrowdStrike bull.

我的投资组合里有CrowdStrike，但比如Cloudflare，我想，我是说，我写过关于它的文章。

I own CrowdStrike in my portfolio, but with Cloudflare for instance, I guess, I mean, I wrote about it.

我可能因为估值太高，太早放弃了那个机会。

I probably left that opportunity too early because of valuation.

但我觉得Cloudflare在代理式AI领域的独特优势在于，当不只是大型企业，而是消费者和小企业都在大量使用AI推理时，你需要能够在靠近终端用户的地方本地运行推理。

But I think where Cloudflare is uniquely winning in agentic AI is because when you have When it's not just large enterprises, but like consumers and small businesses using so much AI inference, you need to be able to run that inference locally, close to the end user.

这正是Cloudflare的CDN服务非常有用之处，再加上他们的网络安全产品。

And that is where Cloudflare's CDN services come very useful, along with their cybersecurity offering as well.

所以对我来说，Meta肯定还是一个大赢家，还有整个网络安全领域。

So yeah, so I think for me, Meta definitely continues to be a big winner along with the whole cyber secure...

比如网络安全行业，但特别是像CrowdStrike这样完全集成的平台。

Like the cybersecurity landscape, but specifically like the fully integrated platforms like CrowdStrike.

我也喜欢Palo Alto，但我确实持有……

I do like Palo Alto as well, but I do own...

但我实际上持有的是CrowdStrike。

But I actually own CrowdStrike.

我真的很喜欢……

And I I I I really like...

我觉得Cloudflare是我太早放弃的一个机会。

Think Cloudflare is something I give up too quickly.

现在我在等回调，想再买回来，因为长期前景看起来很好。

And now I'm looking for maybe a pullback to get back in because I think the long term picture looks great.

你说你太早退出Cloudflare了。

You said that you pulled the plug too early on Cloudflare.

关于这个，你有什么经验教训可以跟投资者分享的？

What would you share with investors in terms of lessons learned there?

我觉得有几件事吧。

I think there are a couple of things.

首先肯定是风险管理，这个占了很大一部分。

There's obviously your risk management, which plays a big part.

比如管理投资组合的时候，你只能持有一定数量的股票。

Like when you're managing a portfolio, only have a certain number of stocks.

而且当你不是百分之百确定的时候，资金又有限，你还想跑赢指数，那就更难了。

And when you're not a 100% confident, given your limited capital, you're trying to generate superior returns compared to the index.

所以，我觉得当时我们在资金分配上做了个决定，就是退出Cloudflare。

So, I think there was a capital allocation decision at that time that we made to get out of Cloudflare.

然后当时想把更多的资金投到CrowdStrike那边。

And I think allocate more towards CrowdStrike at that time.

还有就是关于技术面，你知道，基本面和技术面的权衡。

There is also the bit around like tech, you know, fundamentals versus technicals.

我的意思是，我们主要是基本面投资者，但确实也会看图表，用来确定买入点和卖出点之类的。

I mean, we are, you know, primarily fundamental investors, but we do look at charts in order to like sort of determine buy points and exits and things like that.

仓位管理真的很重要。

Allocation sizing is really important.

所以，对，我觉得这就是那种情况，基本面投资比短期技术图表管理更有效，我想。

So yeah, I think it's one of those cases where being a fundamental investor beats short term technical chart management, I guess.

而且，就像我说的，我觉得Cloudflare的关键就是长期看好。

And yeah, like I said, I think with Cloudflare, was just about just believing in the long term.

说实话，我觉得当时有足够的证据支持这一点。

Like, I mean, and I think there was plenty of evidence to be very honest.

技术面是一方面。

The technical bit is one.

第二方面是关于仓位大小的配置。

And then the second bit is about allocation sizing.

对于像Cloudflare这样的公司，虽然目前看起来不错，但当时股价波动对我们来说挺大的，再加上估值溢价，所以风险确实存在。

Think we, for companies like Cloudflare, which a, could be risky, yes, currently the picture looks great, at that time, like the price volatility in the stock was quite a lot for us, plus of its premium valuation.

所以关键在于仓位大小的配置。

So it's about allocation sizing.

所以先从小仓位开始。

So keep it at starter.

我觉得我们当时加仓太快，可能是导致后来感到不安的原因之一。

I think we had, you know, increased our position too quickly was probably one of the reasons why we just got uncomfortable at one point.

所以如果你不是完全有信心，那就先从小仓位开始，然后逐步加仓。

So if you're not fully confident, then just start at basic, like a starter allocation and scale it up.

我想补充一点，从AI基础设施的角度来看，我认为这将进入一个加速发展的新阶段。

The other piece that I would add is from an AI infrastructure standpoint, which I think is going to switch gears into a different kind of an acceleration mode.

我们三个四月份那次聊天，我记得聊的是主题投资。

This conversation that you know, the three of us had in April, we were talking about thematic investing I think.

当时我们提到了瓶颈交易，那会儿还是个新话题，刚冒出来，现在瓶颈交易真是到处都在爆发。

And at the time we had spoken about bottleneck trades, which was kind of just a new tone that was turning around the corner and we see bottleneck trades just you know, explode everywhere.

有CPU瓶颈、光网络瓶颈、电力瓶颈，还有内存瓶颈，很明显像美光、SK海力士那些公司的情况。

There's CPU bottlenecks, there's optical networking bottlenecks, there's power bottlenecks, memory bottlenecks obviously with Micron and whatever's going on with SK Hynix and you know, the like.

所以这只是我们能展开聊的一个话题。

And so that's just one conversation that we could break this out into.

就凭四月份聊到的那些瓶颈，我们当时提到了三只股票。

There were three stocks just based on these bottleneck rates that we had talked about in April.

其中两只就是Marvell，从四月到现在涨了大概120%到130%。

Two out of them were Marvel which is up I think a 120, a 130 since April.

它现在还是我们投资组合里很重要的一部分。

It still is a big part of our portfolio.

对，是吧？

Yes, it's Right?

我觉得这已经是我们前五大的持仓了。

It's quite I think it's a top five position for us.

虽然随着行情变化我们调整过仓位，但我觉得Marvell表现还是很棒的。

Resizing the position as we've kept going, but Marvel is is I think great.

我始终认为Marvell接下来会进入一个新的加速增长阶段。

I I continue to reiterate that Marvel's I think gonna switch into this new mode of of acceleration.

不管他们去年十月收购那个叫Celestial AI的有趣初创公司是为了什么战略转型……

Whatever they did last year in October when they acquired this this really interesting startup called Celestial AI to pivot...

那个……

That...

这帮助马维尔转型成了光通信领域的大佬。

That's helped pivot Marvel into this giant like, know, in in optical networking.

我觉得他们做得很好。

I think they're doing great.

他们上周刚开了投资者日，所以我们有一些关于他们前景的最新数据。

They just had their investor day last week, so we have some fresh data points in terms of their outlook.

所以他们有……

And so they have...

Marvel 已经公开表示，他们打算……

Marvel's come out and said that they're going to...

他们修订了 2027 财年的短期展望，我觉得那对应的是 2026 日历年。

They've revised their their short term outlook for FY '27, I I think which is CY twenty six.

他们把那个数字调高到了 120 亿美元。

They've revised that to 12,000,000,000.

这大约是 47% 的增长。

That's about 47% growth.

记住，这是过去六到八个月内他们第三次上调全年展望，这真的非常乐观。

Remember, that's a third upgrade in their outlook for the whole year in the past six or six or eight months, which is really bullish.

他们预计今年营收增长 47%。

And they expect 47% of revenue growth this year.

明年这个增速会加快到大约 67%，所以加速之所以发生，是因为市场对 Marvel 的光学互连需求巨大，这些需求来自他们自己的知识产权，也来自他们通过收购 Celestial AI 所获得的技术，也就是那笔并购。

That accelerates to about I think 67% next year and so that the acceleration is just happening because there's so much of demand for Marvel's optical interconnect and that's coming from, you know, from their own IP, that's coming from whatever they've acquired through the Celestial AI takeover, the M and A.