加文·贝克：SpaceX 可能是有史以来最伟大的公司

欢迎你，加文·贝克，来到我们的节目。

Gavin Baker, welcome to the show.

你最近怎么样？

How are you doing?

还不错。

Doing

挺好的，兄弟。

great, man.

你们那边怎么样？

How are you guys?

好得不得了。

Fantastically.

对吧？

Right?

SpaceX上市了。

SpaceX IPO.

一切都进行得完美无缺，简直天衣无缝，连一点戏剧性都没有。

It seems like everything went just flawlessly to a tee, like, to the point where it wasn't even dramatic.

给这位银行家鼓个掌吧。

Round of applause for the banker.

那位银行家。

The banker.

股价几乎精准地暴涨了2020%。

It's, like, 2020% pop almost to a tee.

是的。

Yeah.

我的意思是，如果股价先是涨了70%，然后跌了30%，再反弹50%，那才叫有戏剧性。

What what I mean by that is, like, is, like, it would have been more dramatic if it had, like, popped up 70% and then went down 30 and then went up 50, and it was drama.

但这次是从头到尾都完美执行，一点波折都没有。

But it was just, like, perfect execution from start to finish.

你是这么看的吗？

Was that how you interpreted it?

我觉得高盛和摩根士丹利做得非常好。

I thought Goldman Sachs and Morgan Stanley did a very good job.

对。

Yeah.

你觉得市场接下来最关注SpaceX的什么？

What do you think the market is looking for SpaceX next?

大家是不是都在盯着他们的第一份财报？

Is it all eyes on the first earnings report?

那接下来该看什么呢？

Like, where do we go from here?

因为IPO之前，SpaceX的整个故事都是关于十年后、五年后、两年后的规划。

Because there's so much of the SpaceX story pre IPO that was a decade out, five years out, two years out.

但市场会不会像对待其他公司一样，开始逐季地评估它的表现？

But is the market going to be processing quarter by quarter plays like like many other companies?

我觉得市场始终都会看重季度业绩。

Well, I think the market is always you know, quarters matter.

是的。

Yeah.

你知道吗，在X平台上那些回复的人，就像什么话题都能侃上几句的专家。

You know, the runners on x, you know, it's like they're these reply people on x for any topic or experts.

对。

Yeah.

但有个人曾经跟我说，一场马拉松其实就是二十六个一英里跑组成的。

But somebody once told me a marathon is twenty six one mile runs.

对。

Yeah.

结果所有的跑者都反驳说，根本不是这么回事。

And so, you know, and then all the runners were like, no, that's not true at all.

没错。

Right.

很明显，跑一英里和跑完整个马拉松完全是两码事。

You know, obviously, running miles different than running a marathon.

但你知道，每个季度都很重要。

But, you know, every quarter matters.

不过话说回来，如果你看看过去五六年间市场对特斯拉的态度，再想想亚马逊在发展AWS和搭建零售配送体系那会儿的情况，你知道，他们在这些投入周期里的表现。

But at the same time, I do think if you look at how, the market has has looked to Tesla over the last five, six years, you know, Amazon during kind of the days of, you know, when they're building out AWS and then, you know, kind of their their retail distribution infrastructure when they go through these investment cycles.

我觉得公开市场对长期投入的包容度，以及他们的投资视野，其实比风险圈里很多人以为的要宽得多。

I do think the public market has a much greater tolerance for investment and a much longer time horizon than a lot of people in the venture ecosystem give it credit for.

对了，这是坐在后面的Foxy。

By the way, this is Foxy in the back.

他是我们Tradies这边的一位分析师。

He's our he's he's one of the analysts here at the tradies.

我刚听了他在Brad Gerstner播客里的发言。

I was just listening to him on the Brad Gerstner podcast.

我觉得他当时也插过话。

I think he chimed in on that.

对吧？

Right?

是你把他请来的？

You brought him in?

好了，就这样。

There we go.

是的。

Yes.

是他做的。

He did.

他负责了 SpaceX 的分析，我觉得你们可能会对这个感兴趣。

But he did the work on SpaceX, so I thought you guys might wanna Fantastic.

咱们聊聊这个吧。

Talk about that.

但我认为，事情没你刚才说的那么夸张。

But I do think I'm not sure the story is as far out as you made it seem to be.

对。

Mhmm.

我觉得未来一年有两个关键因素会特别重要。

I there are two variables that are gonna matter a lot over the next year.

对。

Mhmm.

第一个是他们能多快地部署地面算力。

The first is just how quickly can they bring on terrestrial compute.

是的。

Yeah.

看起来他们每吉瓦的变现速度确实比其他人都快。

It does seem like they monetize gigawatts at a higher rate than anyone else.

没错。

Yeah.

根据詹森的说法，他们在建设数据中心的速度上确实快过其他人。

And we know from Jensen that they bring on data centers faster Yeah.

这是詹森提到的。

Than anyone else per Jensen's words.

对。

Yeah.

整个生态系统中的每个人都非常有动力把风电送到能用起来的人手里，因为只要GPU运转起来，大家都能赚更多钱。

And everyone in the ecosystem is really incented to get wind power in the hands of people who can energize it, because everybody starts making more money when the GPUs are energized.

所以，你知道他们当时每吉瓦的收入是多少吗？福克斯，记得那个数字吗？

So, you know, if they can what what are what what was the altimeter figure for how what they were monetizing gigawatts at, Foxy?

你记得吗？

Do you remember

哪个合约？

for which deal?

我觉得大概是每吉瓦两到三亿美元吧，至少是普通云服务商价格的两倍以上，这跟规模、客户质量这些因素有关，但确实很可观。

Well, I think it was like two to three wasn't it like at least two times higher than like average neo cloud pricing and that's partly due to scale and customer quality and things like that, but it was meaningful.

他们跟谷歌的这笔交易，每吉瓦收入达到了500亿美元。

So they're they're doing 50,000,000,000 a gigawatt on Google deal.

天啊。

Wow.

所以他们能多快增加吉瓦容量真的很重要。

So how quickly they can add gigawatts really matters.

对。

Yeah.

对吧？

Right?

如果你能在明年把两三个甚至四个吉瓦的电力用起来，

If you can energize two, three, four gigawatts

没错。

Yep.

那可是相当可观的收入了。

In the next year, that's a lot of revenue.

当然，价格可能会涨，也可能跌，但这一点至关重要。

Now, you know, the prices may go up, they may go down, but that's really gonna matter.

所以短期内，大家的目光都集中在Colossus三号、四号、五号、六号、七号、八号、九号、十号这些项目上？

So all eyes on Colossus three, four, five, six, seven, eight, nine, ten, etcetera, at least in the short term?

是的。

Yeah.

宏观层面越来越难，难上加难。

Our macro hard, macro harder, macro Hardest.

就是这个最难的宏观层面。

To that Macro hardest.

是的。

Yeah.

我觉得光标是个关键变量。

And then I think cursor is another big variable.

是的。

Yeah.

我们之前和布拉德聊过，Composer 2.5经过三周的强化学习和监督微调之后，已经在Colossus 2上表现得明显更优了。

You know, we we talked about it with Brad about how Composer 2.5 after three weeks of, you know, RL and supervised fine tuning Colossus two is kind of Pareto dominant.

那要是把它用在更大、更强的基座模型上，又会怎样？

So what's gonna happen when it's applied to a bigger, better base model?

我觉得 Cursor 已经渗透到一半的财富500强企业里了，确实挺有意思。

I do think Cursor being in, you know, half the Fortune 500 Mhmm.

这挺有意思的。

Is interesting.

所以我觉得这两件事——说实话，我对去火星、小行星采矿真的特别期待。

So I think those two things, you know well, I mean, I'm super excited for going to Mars and asteroid mining and Yeah.

还有在月球上建城市、搞电磁弹射器之类的，这些事简直太酷了。

And, you know, a city and mass drivers on the moon, like, that stuff is awesome.

我觉得在我有生之年，这些事有很大可能实现。

And I think there's a, you know, decent chance it happens in my lifetime.

但我还是觉得，眼前这些更实际、更紧迫的驱动力更重要。

I just think there is there's much more tangible near term drivers here.

是的。

Yeah.

你怎么看云业务和他们自己的应用与模型业务之间的矛盾？

How do you think about the tension of the cloud business versus their own application and and model business?

很多人对Anthropic最初那笔交易感到意外，主要是因为了解迈克尔·特鲁尔的人知道，他一定会非常兴奋

I think a lot of people were surprised when the initial Anthropic deal happened purely because, knowing Michael Truell, he would have been very excited

能接触到更多的算力

to see compute.

没错

Yeah.

确实如此

Yeah.

能拿到你几年前的资源

Get access to your couple of ago.

一起分享吗？

Share?

这儿有个叫戴夫·奇尔德的人

Dave child around here.

而且感觉就像SpaceX，随时可能迅速增加大量算力

And it and it feels like, you can imagine SpaceX, you know, bringing on a lot more compute very quickly.

但与此同时，每个实验室都得面对这样的难题：到底该把多少算力分配给训练，多少给推理。

But at the same time, every lab has had to go through this tension of like how much how much compute do we allocate to training versus inference.

是的。

Yeah.

换个角度看，这说明SpaceX的团队对自己快速部署吉瓦级算力的能力很有信心。

A way to interpret that is that maybe the team at SpaceX is pretty confident in their ability to bring gigawatts online.

是的。

Yeah.

这说得通。

That makes sense.

而且他们订购了20%的Rubin芯片，这可是个重磅芯片。

And the altimeter figure that they've ordered 20% of the Rubens, and the Rubin is an epic chip.

就像Blackwell，当初想把它用起来可太难了。

Like, Blackwell, it was really hard to get it online.

但Rubin就相对容易多了，可以直接替换上手。

Ruben is kind of a more drop in replacement.

这芯片真的非常出色。

It's a really, really good chip.

在未来六到九个月内，Grok LPUs 就会集成进去。

You'll have the, you know, the Grok LPUs integrated at some point in the next six, nine months.

所以换个角度看，他们对自己快速建成数据中心很有信心。

So one way to interpret that is they're pretty confident that they can bring data centers online pretty quickly.

不过我记不太清所有细节了，但我记得 Anthropic 的协议里好像有个条款，如果他们……

But I also I I don't remember all all of the details, but I do think there there was an out in the anthropic agreement that if they Yeah.

还有谷歌的那个合作。

And the and the Google and the Google deal.

是的。

Yeah.

可能是短期的，我的意思是……

It could be short term that's what

我就是想说。

I just meant.

确实存在一种张力，是的。

It's like it is there is like a tension Yeah.

比如，有时候你可能急需算力，但收入还没完全到位。

There where if you, you know, there's a moment where maybe you want the compute, but you don't necessarily have all of the revenue yet.

所以无论如何，我非常有信心。

And so you could see anyways, I'm I'm very confident

你

that you

能搞定。

can figure it out.

对。

Yeah.

说到埃隆的云服务，有没有可能 SpaceX 最终会搭建起更多类似 AWS 的基础设施？

On on on the topic of like Elon web services, is there a world where SpaceX winds up building out more of what looks like AWS?

你看，在 OpenAI 和 Anthropic 之间就存在这种张力：因为 ITAR 规定，或者为了融入生态，又或者是企业客户本身就倾向于使用 AWS，所以必须得在 AWS 上运行。

Like, we've seen between OpenAI and Anthropic, there's this tension of, like, you gotta be on AWS because ITAR or, you know, the ecosystem or enterprises just wanna be on AWS.

他们甚至都不会去用 Azure 或者 Google Cloud。

They won't even go over to Azure or Google Cloud for whatever.

你可以想象一下，光做一个 token 工厂够不够用？

And you could imagine, like, it might is it enough to just be a token factory?

还是说你得有数据库、内存存储，还有云存储、冷存储，以及 AWS 提供的那些所有东西？

Or do you need to have a database and an in memory storage and you know, cloud storage and cold storage and all the things that AWS provides.

我觉得，在未来五到十年里，光做 token 工厂就完全够了。

My thought is that being a token factory is more than enough for the next five to ten years, man.

我们是的。

We are yeah.

我们需要更多的代币工厂。

We need more token factories.

对。

Yeah.

不过我们走着瞧吧。

But we'll we'll see.

对。

Mhmm.

如果他们真需要那些东西，那当然。

And if they need that stuff, like Yeah.

他们会自己搞出来的。

They will build it.

没错。

Yeah.

你对其他超大规模云服务商，比如其他大科技公司，怎么看？

How are you thinking about the other hyperscalers, the other big tech companies?

我们刚才还在琢磨Meta现在的策略。

We we were sort of puzzling over Meta's strategy at this point.

他们有海量的资源、现金流和数据中心。

Tons of resources, cash flow, data centers.

他们在这一块也有经验。

They have experience there.

也许他们没SpaceX那么快，但实力不容小觑。

Maybe they're not as fast as SpaceX, but there's a lot there.

但转去做令牌工厂完全是另一条路，可需求依然很大。

But it's a completely different motion to go into token factory, and yet there's still a lot of demand.

那你觉得这些会怎么发展呢？

So how do you think all of that plays out?

我觉得通往令牌工厂这条路有好几种方式。

Well, I think there's many ways to go to the token factory business.

如果你是Meta的话，

You could if you're if you're meta

是的。

Yeah.

你完全可以对Fireworks这样的公司说：我给你十万张显卡，我们按收入分成，看看结果如何。

You could just say to someone like Fireworks, hey, Here are a 100,000 GPUs.

我们会拿一部分收入，然后静观其变。

We'll we'll take a revenue split and we'll see what happens.

对。

Yep.

我越来越把EV除以净PPNE当作一个估值指标来看待。

I am increasingly looking at EV to net PPNE as a valuation metric.

毕竟，地球上部署的这些算力设备，我觉得它们的价值会上升。

Just with the idea being installed atoms on Earth, I think are gonna appreciate in value.

这有点像一种光环交易，我觉得可能是高盛提出来的——资产价值高，但过时风险低。

And this is a little bit that halo trade that I think maybe Goldman Sachs came up with, you know, high asset value, low obsolescence.

是的。

Yeah.

但Meta的EV除以净PPNE倍数现在处于一个挺有意思的位置。

But Meta, you know, Meta's EV to net PP and E multiple is in an interesting place.

这说明市场对它们变现自身资产的能力非常不信任。

It just suggests that the market has an immense amount of skepticism Yeah.

对它们把自家资产变现的能力持高度怀疑态度。

About their ability to monetize their own asset base.

而且我觉得，考虑到管理层目前只抛出了‘个人超级智能’这个概念，这种质疑是有道理的。

And and I think that's I think that's warranted given that all we've really got out of management is this concept of personal super intelligence.

确实。

Mhmm.

情况要是不同的话，比如 Meta AI 能排在应用商店榜首，大家聊天时都会说‘对，我在用 Meta AI’，那就不一样了。

And it would be different if Meta AI was like at the top of the app store and everyone you were talking to was like, yeah, know, I'm using Meta AI.

或者 Instagram 上真出现了一个杀手级功能。

Or there even there was like a killer feature within Instagram.

对。

Yeah.

就像，哦，现在人们都在Instagram上购物了。

It's like, oh, people are shopping in Instagram now.

而且我觉得没人会怀疑，一旦扎克做出一个能带来十亿美元收入的人工智能产品，他就能把它做到数百亿美元。

And I think no one no one doubts that like once Zach has an AI product with a billion dollars of revenue, he can take it to to tens of billions.

确实。

Mhmm.

但这一点还不清楚。

But it's unclear.

我觉得，进入企业市场后，这个API业务到底会变成什么样，还真是个问题。

I feel like, you know, going into enterprise, like, you know, what what is this API business gonna gonna look like?

我们知道他们确实能用GPU来优化广告产品，这没问题。

We know they can use the GPUs just on making the ads product Yeah.

归根结底，他们做得更好，但还有很多问题。

Better at the end of the day and but there's a lot

很多疑问。

of questions.

听我说，我觉得有几点值得注意。

Look, I think a few things.

如果我看一下各个玩家的举动，感觉大家都觉得我们正进入最后阶段。

I do think if I look at the way different players are behaving, it feels like everyone thinks we're entering the endgame.

人人都喜欢用下棋来打比方。

Everybody, you know, loves to use these chess analogies.

看起来，这场终局可能比我们想象的来得更快。

And it looks like the endgame may be here sooner than we think.

我觉得这就是OpenAI要下调Codex价格的原因，毕竟代码相关的使用量太值钱了。

I think that's why OpenAI is cutting Codex pricing simply because coding, you know, coding tokens are so valuable.

如果你手里的代码令牌不够，可能就进不了大家都在抢的这个正向循环。

And if you don't have enough of them, you may not be able to get into this RSI loop that that everyone is, you know

大家都在拼命争取的。

Pushing for.

如果你是扎克伯格，这种感觉一定特别强烈。

And I think if you're Zuckerberg, feel that acutely.

我一直在想，Muse是什么来着？

I thought Muse, what is it?

Muse Spark。

Muse Muse spark.

对我来说，这真是个意外的惊喜，让我突然想起《塔拉迪加之夜》里那句啥来着？

It was to me, it was an upside surprise and it made me feel like, you know, what's the what's what's what's from Talladega Nights or what is it?

所以你是说还有可能？

So you're saying there's a chance?

当然有希望了。

Hell, yeah.

对啊。

Like, yeah.

你是说真的还有机会？

So you're telling me there's a chance?

再说一遍塔拉迪加之夜里的那句名言：如果你不是第一，那就是最后一名。

I mean, the other quote from Talladega Nights is if you're not your fur if you're not first, you're last.

嗯，没错。

Uh-huh.

而且你知道的，

And and, you know

很多人都是这样

That's how a lot of

人们从《塔拉迪加之夜》里能找出太多经典台词了。

people There's so many good quotes from Talladega Nights.

全是挡风玻璃的事。

He It's all the windshield.

没错。

Yeah.

百分之百。

A 100%.

但我觉得，马克·扎克伯格这么多年来证明自己是个出色的创业者，他愿意调整方向，这挺有意思的。现在Facebook，也就是Meta，跟投资者说，根本不可能把GPU拿出去赚钱。

But I think if I'm pretty confident Mark Zuckerberg has proven to be a good entrepreneur over time, and he's been willing to pivot, you know, and it's just it's so funny, you know, right now, Facebook is, you know, Meta's telling investors there's, like, no chance that we're gonna, you know, you know, monetize GPUs externally.

就在他们宣布大规模裁员前十天左右，我记得那应该是2022年底。

Well, like, ten days before they announced all those layoffs, which I think was at the end of twenty two.

要是不记得具体时间，那可能就是裁员 announcement 之前不久。

If not, I don't remember when they started to announce the layoffs.

就在布拉德之后不久。

It was short shortly after Brad.

也许是在布拉德给马克写那封信，说你是国王之后的两个月吧。

Maybe at two months after Brad wrote that letter to Mark where he said, you're the king.

你想做什么都行，但我还是想礼貌地建议你这么做。

You can do whatever you want, but this is what I might respectfully encourage you to do.

你知道吗，他们几天之内就从‘我们还要继续投资’变成了‘必须严格控制运营支出’。

You know, they went from saying we're we're not gonna do that, we're gonna keep investing to, you know, super focused on OpEx in a couple of days, man.

所以啊，人们去见Meta的时候，听到他们说‘我们不会把GPU拿出去赚钱’。

So it's like people they meet with Meta and they hear, oh, you know, we're not we're not gonna monetize our GPUs externally.

这个嘛，我也不确定。

Well, I don't know.

过几个小时再看看吧。

Check back in a few hours.

是的。

Yeah.

对。

Yeah.

不对。

No.

说到底，每个人都有自己的底线。

It's like everyone it it comes down to like everyone has a price.

你看看四个月前埃隆和Anthropos的关系，再看看后来机会展开的过程，其实挺自然的。

You look at you look at Elon and Anthropix relationship four months ago, and then you look at kind of kind of as the opportunity unfolded Yeah.

他们选择合作是明智的。

It made sense for them to partner.

更好了。

Better.

我想问一下关于

I wanted to ask how

关于Meta这件事，我觉得有一点很重要。

On the meta thing, I think there is one point that's important.

对。

Mhmm.

这些大型上市公司最终推动行为的动机是什么？

What ultimately drives behavior for all these big cap companies?

说到人工智能，我觉得这对每家公司来说都关乎生死存亡。

Like, getting to AI, I think that does feel existential for all of them.

但股价也同样重要。

But the stock price also really matters.

没错。

Sure.

如果股价下跌，你之前给员工发的期权就贬值了，工程师们自然会不满，没有足够的激励，你很难达成目标。别再说什么十倍工程师了，现在都流行说一万倍工程师了。

Because if you don't if the stock goes down, you know, you gave people a grant at this price, stock goes down, you know, your engineers are unhappy and it's tougher to get to where you wanna go if you don't have, you know, these you know, now forget the 10 x engineer people are talking about 10,000 x engineer.

如果你留不住这些一万倍的工程师，根本不可能走到那一步。

You know, you're not gonna get there if you don't keep those 10,000 x engineers.

是的。

Yeah.

所以我觉得，最终哪种方式更好，我们还得走着瞧。

So just I think, like, we'll see we'll see what's better.

是的。

Yeah.

对。

Yep.

这次IPO有没有改变你对零售投资者在资本市场中作用的整体看法？

How did this IPO update your your general kind of framework for the role that retail is playing in capital markets?

上周看到有人说，零售投资者认购才两倍超额，真的挺搞笑的。

A lot of it was very funny last week to see people like say, like, oh, the the IPO is only two x oversubscribed by retail.

是的。

Yeah.

他们把其他所有十亿美元级别的IPO都抛在后面了。

They're dumping all other billion

而那些IPO的认购倍数可是十倍啊。

dollar IPOs were were 10 x oversubscribed.

这中间差了一个数量级。但我觉得，未来十二个月，甚至更久，零售投资者依然会扮演非常重要的角色，尤其是现在那些大型云服务商都已经现金流为负了。

It's like that we're that we're talking about, a very order of magnitude But difference it still feels like retail is going to play a very, very important role over the next twelve months, maybe more so than ever given that the the hyperscalers are are now running, you know, cash flow negative.

中东地区正在发生冲突。

You have the conflict in The Middle East.

对吧？

Right?

那里有很多重建工作需要开展。

There's rebuilding efforts that needs to needs to happen.

所以我觉得，现在正是散户大显身手的时候，可能比以往任何时候都更关键。

And so I feel like retail this is this is kind of retail's moment now maybe more than ever.

是的。

Yeah.

嗯，有几点我想说。

Well, so a few things.

第一，有些人提到散户时带着贬义，但我觉得，蠢不蠢得看实际表现。

One, people say retail in a pejorative way and I would just say stupid is as stupid does.

我们一直在跟踪那些散户偏爱的股票，说实话，它们涨得可真不少，涨得相当猛。

And like, you know, we track you know, there's all these indices of retail favorites And, man, they're up a lot, and they're up a lot

是的。

Yeah.

25岁，而且2024年涨得更多。

25, and they're up a lot in '24.

对。

Yeah.

所以说，傻人有傻福嘛。

You know, so kinda stupid is as stupid does.

我觉得，零售投资者的表现很可能超越了绝大多数专业基金经理，不管是私募、风投，还是公募，反正都是。

So I'm not, you know like, think retail is probably outperforming the overwhelming majority of professional money managers, whether they're, you know, private equity managers, venture, you know, public equity managers, whatever.

所以我觉得，说‘零售’根本不是贬义。

So, like, retail, I don't think it's pejorative.

是的。

Yeah.

说到SpaceX，我其实觉得，你可能接收的信息太多了，我不确定。

I do actually think in the case of SpaceX, I'm not sure, you know, you process so much information.

我不太清楚这说法是从哪儿来的。

I'm not sure where this comes from.

但我觉得，IPO之后，不知什么原因，部分散户买的股票可能被转手了，因为据我了解，当时机构买入的更多。

But I actually think after the IPO, you know, for whatever reason, it some of that retail stock may have may have gotten allocated to people who are flipping it because my understanding is there was more institutional buying.

我不明白为什么有人把IPO之后和散户买入混在一起说。

I don't know where this after the IPO, plus retail buying.

我知道散户对动量特别敏感，所以今天的情况可能改变了这一点。

Now I'm sure, you know, retail is very sensitive to momentum, so maybe what's happening today changes that.

但我认为，SpaceX上市这件事里，另一个被严重忽视的点是，有一万名员工里有十个人在IPO时买了股票。

But I think the other really underappreciated thing about the SpaceX IPO is that 10 over 10,000 SpaceX employees bought on the IPO.

大家总在分析锁定期，算着有多少股票会流入市场，肯定有那些三层结构的SPV之类的，可能确实有。

And that's just, you know, everybody does all these lockup analyses, you know, how much stock is gonna come come to market and, you know, for sure, there's all these triple layer SPVs or whatever, you know, maybe there are.

确实有。

There are.

有人猜测有。

People speculate there are.

我确信其中一部分会慢慢平复。

And I'm sure some of that will get unwound.

对。

Mhmm.

但现实是，SpaceX的员工持有大量这些股票。

But the reality is is that SpaceX employees own a lot of this.

如果你是员工，或者你是通过SPV投资、甚至直接在股东名册上的投资者，过去十年里，你每六个月就有一次卖出的机会。

And if you're an employee or even if you're an investor who has an s an SPV or you're directly on cap table, you've had a chance to sell every six months for the last ten years.

如果你真想卖，早就卖了，比如去年下半年那次融资的时候。

If you wanted to sell, you could have sold and, you know, whatever, when they did that offering in in the second half of last year.

是的。

Yeah.

所以我觉得这里的供需关系会非常有意思。

So I think the supply demand dynamics here here are gonna be very interesting.

至于散户整体而言，老实说，过去三年半里，他们在市场上的影响力比以往任何时候都大，甚至超过2021年那会儿——不过顺便说一句，这跟那时完全不是一回事。

But with retail in in general, yeah, man, they're they've been a more powerful force in the market over the last three and a half years than at any other time in my career, including the year February, which I think this is nothing like, by the way.

对啊。

Mhmm.

但没错，他们确实重要，蠢事就是蠢事。

But yeah, they they matter, and stupid is as stupid does.

我想更深入探讨一下这个观点：价值会聚集到那些处于代币路径上的公司。

I'm interested to go a click deeper on the idea of, like, value will accrue to companies that are in the token path.

我觉得这句话现在特别有力量，但我想进一步展开，弄清楚‘代币路径’的边界在哪里，因为我们之前聊过Cloudflare的马修·普林斯。

I think that phrase is extremely powerful right now, but I wanna sort of expand it and understand the fringes of the token path because we talked to Matthew Prince at Cloudflare.

股票都创了新高。

Stocks at all time highs.

它在边缘节点上分发代币。

It's delivering tokens at the edge.

显然，有大量AI代理在与不同网站交互，这些都需要分布式部署。

And obviously, there's an immense amount of AI agents that are interfacing with different websites that need to be distributed.

所以这里面蕴藏着巨大的价值。

So there's a lot of value there.

但如果Cloudflare最终在边缘把令牌包发给你，那它算不算在令牌路径上？

But is Cloudflare in the token path if they're the one finally sending you the packet of tokens at the edge?

还是说这不算？

Or does that not count?

你怎么看这个问题

How does that how

你觉得呢

do you

你怎么想？

think about that?

每个CDN都是这样

Every CDN Yeah.

都有这种业务

Has has this business.

你知道的，Akamai已经签了

You know, Akamai, they signed

是的。

Yeah.

那是什么？

What was it?

和Anthropic签了18亿美元的合同。

A $1,800,000,000 with Anthropic.

如果你拥有这么多边缘节点，

And if you have all these points of presence

是的。

Yeah.

就能为高价值用户低延迟地分发令牌。

And you can deliver really low latency tokens to high value users.

我觉得对所有这些CDN来说，有趣的是，

I think what's interesting for for cloud for all these CDNs

是的。

Yeah.

他们因为低延迟而获得了巨大的溢价。

Is they're they're getting a massive premium for that latency.

对。

Yeah.

你知道吗，从Cerebras身上学到的一点是，人们真的愿意为速度买单。

You know, it turns out, you know, one of the lessons from from Cerebras is people are willing to pay for speed.

是的。

Yeah.

你看这些公司定价多高，但我觉得，从全球消耗的token总量来看，这些CDN服务所处理的还不到1%，甚至可能连0.1%都不到。

And you see that in the prices these guys are commanding, but I do think, I mean, my gut is in terms of tokens consumed on planet Earth, these CDNs are delivering less than 1% of them and I mean, maybe less than 10 basis points on them.

大部分token运算都在内部完成，你只拿到最终结果。

Most of the tokens are happening internally, and then you just get the result.

对。

Yeah.

我会说这些公司真的处在token传输路径上吗？

Would I say that these guys are in the token path?

我的意思是，他们确实有办法进入令牌处理的流程。

I mean, I would say they have a path towards being in the token path.

对。

Yeah.

他们目前的部分业务其实已经涉及其中了。

Then they have part of their they have part of their business that's in it today.

是的。

Yeah.

他们正努力，像其他人一样，把更多业务往这个方向迁移。

And they're trying and like everyone, they're trying to move it more and more there.

对。

Yeah.

马修之前提到过，把英伟达的服务器部署在边缘节点，专门用来运行语音模型，因为这类服务对延迟要求极高。

I I mean, Matthew was talking about co locating NVIDIA servers on the edge, specifically for the delivery of voice models because you want those to be very low latency.

至于代码生成模型，你或许可以让它在弗吉尼亚那边跑上二十分钟甚至一小时，再把结果传回来。

Maybe you want your coding model to go off and cook in Virginia for twenty minutes or an hour and then come back to you with the final result.

但如果你真的要跟模型对话，那肯定越快越好。

But if you're trying to actually talk to a model, you probably want it as fast as possible.

那相反的情况呢？

What about on the opposite end?

之前不是有个初创公司，NVIDIA刚投资了他们，专门做……

Wasn't there that startup that NVIDIA just invested in that, like, will

哦，对。

Oh, yeah.

装在你家外面。

On your house.

他们会在你家外墙装一台四卡GPU服务器。

Money that put a GPU you know, four a four GPU server on the outside of your house.

我的意思是，确实如此。

I mean, yeah.

GPU那玩意儿

GPUs That'd

到处都在普及。

are going everywhere.

迟早会进你的卧室。

Going in your bed eventually.

这其实是个完整的循环。

I mean, it's full circle too.

你得先暖和一下

You do warm

在里面。

in there.

贾森以前不是说过类似的事吗？就是加密货币热潮那会儿。

Wasn't Jensen talking about stuff like that back during, like, the original Yeah.

那时候人人都在说要挖矿。

Crypto cycle being, like, everyone's gonna be, you know, mining?

是的。

Yeah.

特斯拉的Powerwall其实就是一种用来储能的设备，算是这种思路的一个版本。

And I mean, the the Tesla Powerwall is, like, a version of that for storing energy.

它不存储计算能力，但能在电网边缘本地储存能量，这背后能解锁很多强大的应用。

It doesn't store compute, but it stores energy locally on the edge off the grid, and there's a lot of powerful things that get unlocked by that.

那土地呢？

What about land?

土地会不会在某个时候，成为这条链条最末端的代币化对象？

Is land in the token path at some point on the extreme other end?

因为大家显然在不断往更上游的方向探索。

Because people are obviously searching for go going further and further upstream.

有人甚至找到了一家马桶公司，它们生产的某种材料会进入半导体供应链。

They found a a toilet company that makes a particular material that goes into semiconductor supply chain.

人们现在就像在搜寻最后那一种关键原料。

People are hunting for, like, the final the final ingredient.

也许那根本就是沙子，最终能变成硅。

Maybe it's just sand that turns into silicon.