SpaceX 与蓝色起源的“繁荣” | 头脑风暴 第 134 集

这是我见过最大的火箭爆炸之一，恐怕也是过去一周里任何人见过的最大爆炸——蓝色起源在发射台上炸了。

One of the largest explosions of a rocket I have seen, maybe that anyone has seen this past week, Blue Origin exploding on the pad.

这对他们来说是个重大打击，对美国、对NASA想登月的计划也是个沉重打击。

This is a big setback for them, a big setback for The United States, for NASA trying to get to the moon.

丹尼尔，真高兴能当面见到你。

Daniel, great to see you in person.

是的，没错。

Yes, yeah.

我特意穿得正式一点来参加这次活动。

I dressed up for the occasion.

我穿了衬衫，对的。

Wore my shirt, yeah.

我也为这种场合特意打扮了一下。

What I also dress for the occasion.

从美国航天政策的角度来看，你怎么看这件事？SpaceX马上就要上市了，Starlink和蓝色起源正在做的事形成鲜明对比，你能帮我们分析一下吗？

What do you make of this, know, in the context of US space policy, SpaceX on the verge of going public, Starlink versus what Blue Origin is working on, can you contextualize it for us?

当然。

Sure.

我来简单说明一下发生了什么。

So maybe just to explain what happened.

蓝色起源的‘新格伦’火箭正在进行它的第四次飞行测试。

So Blue Origin's rocket called New Glenn was taking its fourth flight.

当时它在发射台上进行点火测试，结果意外爆炸了。

It was conducting a test fire on the stand, and unfortunately, it blew up.

最引人注目的是，爆炸发生在发射台上，几乎把整个发射设施都毁了。

The most notable thing is that it blew up on stand, which pretty much destroyed its whole launch complex for the most part.

当时大家都是这么想的。

And that was the thought at the time.

偏偏倒霉的是，就在两天前，蓝色起源刚拿到NASA一笔4.7亿美元的合同，要为2028年登月任务提供运载工具。

Why the timing was unfortunate is just two days before New Glenn or Blue Origin got like a forties $470,000,000 contract from NASA to deliver vehicles to the moon in in 2028.

所以当时很多人评论说，从图像上看，损坏程度很可能让下一次发射至少推迟一年，因为那是他们唯一在用的发射台。

So at the time, a lot of the commentary was saying that, you know, the damage that they could see from imagery looks like it'll probably push the nearest launch out by at least a year because it was their only operational launch pad.

不过幸运的是，今天首席执行官戴维出来表示，发射台的关键部件大多完好无损，他很有信心，今年年底前就能重新发射。

However, thankfully, today, the CEO, David, came out and said that it looks like a lot of the long lead items for the complex are actually still intact, and he's optimistically confident that they can launch towards the end of this year.

这就是事情的经过，接下来我简单说说可能的影响，咱们之后可以再深入讨论。

So that's that's what happened, and then maybe just some quick implications, and I'm sure we can go back and forth.

首先，这件事清楚地表明，垂直整合非常重要，因为很多卫星公司都高度依赖火箭发射服务，而这个市场本来就很紧张。

But firstly, it really just shows that vertical integration is important because there's a lot of satellite providers that were really relying on launch providers, which is already a constrained market.

亚马逊的‘柯伊伯计划’几周前就宣布了，说他们正在制造大量卫星，但根本找不到发射运载工具。

Amazon's Project LEO already came out a couple of weeks or months ago and said, we're building so many satellites, but we can't get launch vehicles.

所以他们之前对新格伦火箭的依赖程度非常高。

So it was pretty heavily reliant on New Glenn.

同时，它在NASA重返月球的计划中也扮演着关键角色。

It was also playing an important role in NASA's mission to get back to the moon.

现在得看这次延误会持续多久，按一年算的话，整个行业又得重新依赖SpaceX和星舰项目了。

So contingent to see how long that'll be delayed based on the twelve month, and then also just the industry falling back on SpaceX and again, and reliance on the Starship program.

所以，影响确实挺多的，但目前这些只是初步的看法。

So, yeah, a lot a lot of kind of implications, but they're kind of the first take so far.

是的。

Yeah.

我不确定这对SpaceX来说是不是真的有利。

I don't know if it's actually necessarily even good for SpaceX.

我的意思是，SpaceX的猎鹰重型火箭其实可以承接一部分原本要给新格伦火箭的发射任务。

As in, I think SpaceX, Falcon Heavy, can potentially take some of the volume that otherwise was going to go to New Glenn.

但SpaceX面前还有很多执行上的挑战，而且把资源集中在制造星链卫星、装进星舰火箭、实现可重复使用并发射上，回报潜力巨大。

But SpaceX has a lot of execution in front of it and huge return on capital potential in just building Starlink satellites, sticking them in the Starship rocket, getting that reusable, and launching it.

从资本效率的角度来看，航天业务通过生产发射载具来支撑发射技术的研发，我觉得这是个不错的思路。

And the space business is in a capital efficient way to produce launch vehicles, I think is a good way to think about it, to fund R and D for launch.

但他们并不想把资源和精力花在用猎鹰重型去填补新格伦延期留下的空缺，毕竟另一边的商业前景实在太好了。

But they don't necessarily want to have to devote resources and bandwidth to picking up some of the schedule that would slip from New Glenn in Falcon Heavy just because they have such an amazing set of commercial prospects on the other end.

所以我认为，这对SpaceX和整个行业来说都是个好消息。

So I do think it's really good news, I think, for SpaceX and for the industry as a whole.

如果蓝色起源能在一年内重新成功发射，那将远远超出我们之前的预期——毕竟之前发射台那场爆炸，简直像奥本海默电影里诺兰用IMAX摄像机拍出来的场面一样震撼。

If Blue Origin can launch again within a year, that would be very, very, very positive relative to what we were anticipating based upon the tremendous Oppenheimer level Christopher Nolan IMAX camera explosion that was on the launch pad.

有个问题想问一下Oi。

Question for Oi.

美国现在有多少个在用的发射场？SpaceX又有几个？

What and how many launch facilities are active in The US or how many does SpaceX have?

关于在用的发射场，我一时也说不上具体数字。

So in terms of active, I don't know off the top of my head.

SpaceX那边，我记得大概有四五个吧？你清楚吗？

SpaceX has, I know it has maybe four or five, You know off the top of your head?

嗯，确实如此。

Well, yeah.

这得看他们用的是哪种火箭。

It depends on what rocket they use.

而且他们

And they're

正在建设中，没错。

constructing Yes.

好几个。

A number Several.

更多是给星舰用的。

More for Starship.

是的。

Yeah.

没错。

This was Right.

只有蓝色起源公司。

Blue Origins only.

只有正在运营的。

Only operational.

他们实际上正在附近建一个，另外还在加利福尼亚的范登堡建一个。

They're constructing one actually close by to it and then another one in Vandenberg, California.

但即便如此，这些设施也做不到在十二个月内投入使用。

But again, they wouldn't be ready within twelve months.

他们大概会重建被毁掉的那些设施。

Think the thought is that they'll just rebuild what was destroyed.

根据我们的模型，SpaceX 在未来几年里需要投入大约八十亿到一百亿美元，用来建设发射台，以支持星舰计划。

In our modeling, SpaceX needs to build on the order of, like, 8 to $10,000,000,000 of launch pads over the next couple of years to support the Starship.

这些发射台的成本真的非常高。

And these launch pad launch pads are really expensive.

最明显的例子就是ULA的情况，他们当初把建造合同给了贝克特尔公司，预算才几亿美元，结果现在花了几百亿，连个发射塔都没建完。

There's no better evidence than what ULA was trying to do where Bechtel had this contract for, I think it was a couple $100,000,000, and now they've spent like a couple billion dollars and they haven't completed a launch tower for kind of the United Launch Alliance.

SpaceX 显然能做得更便宜，因为他们不是按成本加利润的合同来运作的。

SpaceX obviously is doing it cheaper because they're not on a cost plus contract.

这些钱都是他们自己掏的。

That's actually their own money they're spending.

SpaceX 的发射塔不只是用来发射火箭，还要能接住返回的火箭，把它放回发射台。

But the SpaceX launch towers are not only like launch the rocket, but also catch the rocket and put it back on the sand.

所以这些设施结构极其复杂，建造周期很长，而且得在好几个地方同时施工。

And so these are very, very hard, long construction projects built in multiple places.

所以埃隆以前说过，成功的发射，尤其是最近这次星舰，是他们在新发射台上的首次发射。

That's why Elon has said in the past, like, a successful launch, particularly, like, the most recent Starship was their first one on the new pad.

它可以在太空中直接炸掉。

It can blow up just blow up in space.

千万别在发射场炸了，那会拖慢几个月，甚至好几年。

Don't blow up on the complex because it just delays everything by months, if not years.

不会。

No.

我猜他们的发射台也不互通，比如

I assume they're also not interoperable as in like

对。

Yeah.

蓝色起源不可能说：‘埃隆，我们能用你们的发射台吗？’

Blue Origin can't say, hey, Elon, can we use your launch?

不行。

No.

不行。

No.

这完全取决于火箭的大小。

That's purely by rocket size.

新格伦火箭每次能发射大约四万五千公斤。

So the New Glenn can launch around 45,000 kilograms per launch.

最初的星舰能载十万公斤，最终目标是二十万公斤，所以火箭的运力不一样。

The initial Starship is a 100,000, eventually going to 200,000, so different rocket sizes.

对。

Yeah.

还有一点值得注意，第一，太好了，没人受伤。

And it is worth noting, one, it's great, no one was injured.

是的。

Yes.

第二，

Two,

你

you

你知道吗，Origin是唯一另一家成功回收轨道级助推器的公司，这真让人遗憾。

know, it's particularly sad, Origin is the only other company that's landed a orbital booster.

没错。

Yep.

对吧？

Right?

虽然比SpaceX落后了十年，但好歹还是第二名。

And that's ten years behind SpaceX, but second place nonetheless.

是的。

Yes.

现在又出了这么个挫折。

And now there's this setback.

你觉得这对所有相关人员和卫星公司会有什么影响？

What do you think is going to be the implication for all of the people and satellite companies now?

我觉得这影响更大，就像你提到的，本来就已经很受限了。

I think that's kind of the bigger impact where already, as you mentioned, it was constrained.

这些小公司财务上撑得住再烧一年甚至更久的钱吗？

Do these small companies have the balance sheet to handle another burn rate of, I don't know, another year plus?

是的。

Yeah.

至少在公开市场的发射服务商里，有个火箭实验室。

I I think for for launch providers at least, like in the public market, there's Rocket Lab.

他们现在的火箭叫电子号，每次只能运大约三百公斤。

Their current rocket is called Electron, but it's only 300 kilograms roughly per launch.

他们正在研发一种叫中子号的火箭，运力大概一万三千公斤。

They are developing Neutron, which is around 13,000 kilograms.

所以，还是远比新格伦号小得多，根本没法执行深空或登月任务。

So, again, still significantly smaller than New Glenn, so that wouldn't have the capability to go in out to deep space into the moon.

但很可能你会看到更多公司涌向公开市场，靠融资来不断学习和迭代。

But it's likely you'll see probably a lot more companies come to the public market to get the equity to learn and iterate.

Stock Space 是私营领域另一家挺有意思的公司。

Stock Space is another interesting company in the private market.

他们是唯一一家直接跳过部分可重复使用，直接迈向完全可重复使用的公司。

They're the only one kind of leapfrogging the partial reusability and going to full reusability.

但说到萨姆的观点，蓝色起源这次冲击这么大，是因为它是SpaceX之外，全球唯一成功回收助推器的公司。

But to to Sam's point, Blue why it was such a blow is because Blue Origin, it's the only company who's landed the booster outside of SpaceX anywhere in the world.

所以在这个市场如此受限的当下，这一点格外引人注目。

So that's why it was just so notable given how constrained the market is right now.

我们换个话题吧，布雷特，你还有别的惊人见解吗？

Shifting gears here, Brett, unless you have any other explosive comments.

没有，咱们换吧。

No, no, let's shift.

好的，SpaceX S1A 结束。

Okay, SpaceX S1A out.

英伟达，作为每年150亿美元合同的一部分，提供了多少块GPU？

Nvidia, number of GPUs as part of the $15,000,000,000 annual contract.

你算过了，网上也算过了，但伊隆给出了回应。

You've done the math, the internet's done the math, but you've got a response from Elon.

你的修正意见是什么？

What's A your correction.

对，继续说，布雷特。

Yeah, away Brett.

你能解释一下，这个GPU数量到底能说明多少背景信息吗？

But can you break down how much context does that number of GPUs provide?

这对SpaceX意味着什么？我觉得这不如它对整个行业意味着什么来得重要。

What does it mean for SpaceX, which I think is arguably less important than what it means for the industry?

如果你能给我们详细分析一下就好了。

And so if you could break that down for us.

当然可以。

Sure.

根据伊隆公布的S1信息，Anthropic这家开发Claude的公司，是从SpaceX旗下的xAI租用GPU的，而xAI拥有Colossus 1和Colossus 2两个数据中心。

So there's this in the S1 and announced by Elon, Anthropic, maker of cloud, was renting GPUs from SpaceX xAI, which has its Colossus 1 and Colossus 2 data centers.

S1文件显示，他们每年为此租赁支付150亿美元，但我们当时并不清楚这些数据中心到底租用了多少。

And the S1 revealed they were going to pay $15,000,000,000 a year annualized for that rental, but we didn't know how much of those data centers were being rented.

举个例子，如果我将来去市场上租GPU，可能每吉瓦要花150亿美元。

So for context, if I go out into the market and ask for GPUs sometime in the future for rental, I might pay $15,000,000,000 per gigawatt for those GPUs.

Anthropic这类公司愿意付这么高的价格，是因为他们把GPU转化成token后，每吉瓦能收300亿美元，甚至可能到400亿或500亿。

The reason the likes of Anthropic are willing to do that is if they take the GPU and they convert it into tokens, they can charge something on the order of $30,000,000,000 per gigawatt, maybe even 40, maybe even 50.

我们不清楚Anthropic和OpenAI背后的财务细节，但目前token非常值钱，大家愿意为数字智能花大价钱。

We don't know exactly what's going on financially behind the scenes for Anthropic and OpenAI, but tokens are very valuable right now and people are willing to pay a lot for them for digital intelligence.

这就是这些前沿模型公司的商业模式。

So that's the business model of these frontier model companies.

S1的修订版让我们估算出，Anthropic至少每吉瓦支付了250亿美元，实际可能接近300亿美元。

And the amendment to the S1 allows us to estimate that at a minimum, Anthropic was paying 25,000,000,000 per gigawatt, but probably more like $30,000,000,000 per gigawatt to rent the GPUs.

换句话说，这几乎是市场价的两倍——如果你想提前两三年锁定算力，可能根本拿不到。

So call it 2X the market rate if you wanted to go out a couple of years and try to secure compute, and you probably can't even get it a couple of years forward.

他们愿意付这么高价钱，是因为已经签下了大量Claude的客户，但客户抱怨Claude经常用不上，他们只能降低模型性能来应付需求。

So the reason they're willing to do this is because they've signed up all these customers for Claude, but the customers are upset that Claude isn't always available, and they're having to dumb down the models in order to serve the customer demand.

所以他们真的需要更多GPU来维护和服务已经签约的客户。

So they literally need more GPUs to retain and service the existing customers they've already signed up.

事实上，你可以说他们的收入增长受限的不是需求，而是手头有多少GPU。

In fact, you could argue that their revenue growth is constrained not by demand, but by the amount of GPUs they have.

另外要注意的是，这些数据中心的建设成本，SpaceX比行业平均水平还要低30%甚至更多。

So if you do The other thing to note about these data centers is SpaceX is somewhere around 30 and even more percent more capital efficient in building data centers than the industry.

因此，SpaceX不仅能以高于市场价的价格向Anthropic提供短期GPU资源。

And so SpaceX is not only able to charge a higher than market rate to give anthropic short term GPUs.

而且他们在搭建这些基础设施上的投入也更少。

They also spent less capital building the underlying infrastructure that they are monetizing.

如果这个合同能持续五年，仅凭这个数据中心，SpaceX就能净赚500亿到600亿美元。

So if you said, Hey, this contract's going to last for five years, it would actually produce net $50 to $60,000,000,000 in cash to SpaceX just on this data center that they have.

不过埃隆的回应是，这只是一个短期协议，确实如此。

Now, Elon's response was, this is a short term deal, which is true.

合同里写得很清楚，任何一方都可以随时退出。

Contractually, it's basically like either party has the ability to be like, I'm out.

我觉得他们只需要提前九十天通知，甚至更短时间，就可以说不干了。

And I think it's ninety days, maybe even less that they have to give notice of like, no, we're out.

从Anthropic的角度来看，他们只是需要短期的算力支持。

And so I think from Anthropics perspective, they had a short term need.

他们愿意支付高于市场价的费用。

They're willing to pay over the market.

只要能拿到更好的算力资源，他们就会转投别的合作方案。

If they can line up better compute capacity, they'll swap out of this into a different deal.

而对SpaceX来说，他们打算收购Cursor公司。

And from SpaceX's side, they have Cursor that they're going to presumably acquire.

他们拥有收购Cursor的选项。

They have an option to acquire them.

而且他们自己也需要大量算力来训练模型，以便在前沿模型的商业化竞争中站稳脚跟。

And they have a lot of compute training needs so that they can themselves compete on a monetization front with the frontier model providers.

所以他们可能会把算力收回来，用在自己的训练任务上。

And so they might want to call the compute back because they need it for training purposes.

所以，这两点说明双方都不太可能长期履行这份合同。

So those are the two reasons to believe that neither party is going to stay in the contract for a long period of time.

我觉得更稳妥的说法是，你建好服务器和数据中心，租出去一年，等于白得了一个数据中心。

I think it's safer to say you build the computer, you build the data center and you rent it out for a year and you get the data center for free.

也许不一定满一年，但大致的算账就是这么个逻辑。

Maybe it's not going be a full year, but that's the way the rough math works out.

这也说明，在当前全球计算资源严重短缺的环境下，谁能快速高效地部署算力基础设施，谁就占了大便宜。

And it's also a sign that it's really good to be a company that can very efficiently and quickly put compute infrastructure in place in a world that is net short of compute infrastructure.

另外一条消息是，Anthropic已经公开提交了上市申请。

And then the other news is that Anthropic has publicly, confidentially filed to go public.

我觉得他们是公开提交了保密文件，不对，等等。

I think it's publicly filed a conf no, yes.

刚才说‘保密’这个词用得不太对。

Think confidential was in the wrong space in that sense.

他们是公开宣布了保密提交上市申请。

Publicly announced a confidential Yes,

谢谢。

thank you.

对，我很好奇，从数据中心的角度来看，你觉得整个市场理解你刚才说的这些经济逻辑吗？

Right, I'm curious, just in terms of, do you think that the rest of the market understands the economics you just laid out from the data center side?

因为我们也看到， Alphabet 正在市场里试图通过股权融资八百亿美元，用于未来的 AI 基建。

Because we also had, I think Alphabet is in the market trying to raise $80,000,000,000 in equity for future AI build out.

但你要是看股价，这消息一出来，股票反而跌了。

But the stock is, if you look, it's selling off on the news.

所以我只是想知道，别人是不是真懂这些数据中心背后的经济模式和后端变现能力。

And so I'm just curious if you think others appreciate the economics and the back end monetization that these data centers provide.

说实话，我觉得他们并不懂。

I mean, no, I don't think that they do.

我觉得 Alphabet 的股东们可能有点不安，明明是个能持续产生大量现金流的业务，却还要卖股票来投更多钱。

I think Alphabet shareholder base is kind of maybe a little queasy stomach to say that this large cash flow generating business is selling off a piece of equity to invest more.

我以为他们现金已经很充裕了。

I thought we had plenty of cash flow.

所以，可能是股东本身还没做好心理准备面对这种情况。

So it could be the shareholder base itself is not prepared for this.

这也说明整个市场或许短期上认可这些经济模型，但担心这些资产三四年后的价值到底如何。

And that's a sign that the market as a whole, maybe they believe the economics on a short term basis, but they're worried about what are these assets worth three and four years from now.

目前来看，这些数据中心的实际变现能力远超当初评估时的预期。

Right now, the demonstrated monetization of these things is much higher than people were anticipating when they originally underwrote the data centers.

也就是说，当初的评估模型是假设这些芯片用个四五年就该退役了。

Meaning, the underwriting model was on like these chips last maybe four or five years, and then that's it.

但实际情况是，这些芯片在标称寿命结束后还能继续创造收益，因为可以用它们运行更简单的AI代理。

And the reality is it looks like the chips are going to continue to monetize long after their rated life because you can have dumber agents running on those chips.

你可以把AI模型的预填充阶段拆出来，让模型并行处理大量数据，再单独处理结果生成的部分。

You can actually parse out kind of pre fill where the AI model is interpreting a bunch of data in parallel and decode where it's actually generating the results for you.

而预填充这种任务可以用更便宜的芯片来完成，因为它能并行处理。

And you can run that pre fill on less expensive chips because you can do it in parallel.

所以，这些芯片的实际使用寿命比预期长得多，变现周期也因此被拉得更长。

And so there's lots of ways in which these chips are turning out to be useful for longer, and so then have a better monetization curve.

如果这个趋势持续下去——我们觉得它肯定会——那么投资数据中心并出租的资本回报率会非常高。

If that trend continues, which we think it will, then kind of the return on capital for investing in data centers and renting them out is very, very, very, very, very, very positive.

所以，我觉得市场还没真正明白这一点。

And so, I think the market doesn't really get it.

或者说，他们以为自己觉得现在和明天算力还不够，但担心三年后算力会过剩，大概就是他们心里的算法。

Or believe, I think they believe they believe we're short compute today and tomorrow, and they're worried we're going to be oversupplied of compute three years from now, is probably like the rough math.

也许我们没必要深究这个，但布雷特，我觉得至少该问问你。

And maybe we don't want to go down this rabbit hole, but I think worth at least asking you, Brett.

外面确实有质疑声，比如迈克尔·伯里就在大声提醒，有些公司用表外融资的方式来为这些基建项目筹资。

There are questions out there, and I think Michael Burry is out there beating the drum around off balance sheet financing that some of these companies are using to fund some of this build out.

你们对SpaceX和XAI的测算，有没有考虑这些融资工具的影响？

Do your numbers for SpaceX, XAI factor in any of those vehicles that are being used?

你怎么看这种结构？

And what do you make of that structure?

因为他直接说这是彻头彻尾的欺诈。

Because he's calling it an outright fraud.

我对这个还真说不准。

I don't know entirely what to make of it.

我想听听你的看法。

So I'm curious if you have thoughts.

我稍后可能让丹尼尔来补充，但SpaceX确实租了一些数据中心，不过根据会计准则，这些租赁支出是计入资本开支的。

I may defer to Daniel here in a second, but SpaceX did lease some of its data center, but our interpretation of the financial statements is those do appear in the CapEx numbers because of the way the lease is treated by GAAP accounting standards.

所以这些支出都体现在资产负债表上。

And so, they are fully on balance sheet.

为了评估SpaceX在数据中心方面的资本效率，我们从多个角度做了分析，主要看资本开支和固定资产。

And also to calculate the capital efficiency of SpaceX, particularly on the data center line, we've gone about it in a few We look at the CapEx line and we look at property, plant, and equipment.

我们用了几种方法交叉验证，确保没有误判他们的资本效率。

So we've gone about it in a few ways triangulating to make sure we're not misstating their capital efficiency.

结果来看，他们的数据是站得住脚的。

And it looks like they triangulate successfully.

丹尼尔，你怎么看？

Daniel, thoughts?

是的。

Yeah.

在S1文件里，这叫一个‘失败的出售并回租’交易，我觉得是Velour公司。

In the S1, it's a transaction called a failed sale and leaseback, where it's I think it's Velour.

是的。

Yeah.

是的。

Yeah.

理论上，虽然他们把资产卖了又租回来，但实际控制权其实并没有真正转移。

In theory, because they've sold it and they've leased it back, they haven't necessarily transferred control even though on paper they have.

所以从会计角度看，这笔资产仍然记在他们的账上。

So from an accounting perspective, they keep it on their books.

对。

Right.

对。

Right.

好的。

Okay.

所以你才能把他们的PPNE当作一个大致的参考，尽管从账面上看，他们已经不再拥有这些设备了。

So that's why you can kind of look to their PPNE as a rough gauge despite like on paper, the fact that they don't physically own the billing anymore.

对。

Yeah.

就是为了防止这种欺诈行为——明明还在继续使用，却把资产从账上抹掉。

To prevent all that fraud from like pulling it off your books even though in practice you're actually still using it.

所以不是短期的。

So not the short

第二部分。

part two.

对。

Yeah.

我觉得谷歌正在筹集股权资本，而且这些数据中心到目前为止主要靠股权融资，当然我不是说没有债务融资，但确实有相当大一部分资金是来自股权的。

I think that the fact that Google is raising equity capital and the fact that these data centers have thus far largely been funded by equity though is, there is debt financing of data centers, don't get me wrong, there's a big chunk of equity going into data centers.

此外，数据中心的建设速度受到两方面限制：一方面是芯片供应，以及台积电愿意多快提升产能；另一方面是电力供应，看人们能多快把设备接上电源。

Plus the rate at which data centers are being built is constrained on the one hand by chip supply and how quickly Taiwan Semi is willing to increase its foundry, and on the other hand by power supply and how quick people can line up and plug these things in to power.

这两者都是供应端的制约因素，目前正好在控制着整个周期的节奏。

Those are both kind of supply constraints that are keeping the cycle in check thus far.

但同时你得相信一件事：有大量表外融资在进行，而我们现在正处于周期尾声，三年后，这些东西会多到泛滥，我们根本不知道该怎么处理。

And then there's like a, you have to simultaneously believe, hey, there's all this off balance sheet financing going, and we're at the end of the cycle, as in three years from now, there's going be too much of this stuff, and we're not going to know what to do with it.

我觉得，从整体来看，这两种观点都不对。

And I think on a mixed basis, both of those are wrong.

我认为需求端的实际情况，远比人们想象的要深刻得多，大多数人根本没认真去衡量过。

I think the demand story, I think is actually much more profound than people understand, and most people haven't tried to dimension it.

他们只是觉得：怎么可能有这么大的需求？

They're just like, no way there's this much demand.

我看了加里·马库斯的观点，他说这些东西迟早会撞上天花板。

I read Gary Marcus and he says these things are going to hit a wall.

而在供应端，我确实认为还有其他因素在限制发展，目前融资风险还没那么大。

And on the supply side, I do think that there are other factors that are constraining things and there's not yet so much kind of like financing risk.

迟早会到那一步，虽然今天还不是。到时候会出现大量这类情况，短期内市场根本消化不了。

At some point, but not today, at some point, we will be in like a, there's a lot of this stuff happening, much more than can be stomached by the market in the near term.

接着你就会看到，甚至在需求端，企业也会开始嘀咕：哎呀，我们在 token 上花了不少钱。

And then you will see, and even on the demand side, you could see like businesses being like, Oh, we're spending a lot on tokens.

这些投入到底带来了多少生产力提升？

Are we getting enough productivity out of it?

他们甚至不需要削减开支。

We need to They don't even need to change the amount that they're spending.

他们只需要大声说出来，大家就会非常恐慌，市场也会出现一波回调之类的情况。

They just need to say that in a loud enough way and people will get really freaked out and there will be a bunch of market correction type stuff.

但我觉得，如果以两到三年的时间跨度来看，需求会持续旺盛，收入增长也会非常惊人。

But I think that measured over two and three years, this is going to be continued a lot of demand and amazing revenue generation.

好的。

All right.

谢谢你，丹尼尔。

Thank you, Daniel.

谢谢，布雷特。

Thank you, Brett.

谢谢，尼克。

Thank you, Nick.

非常感谢。

Thanks so much.

大家再见。

See you everyone.

是的。

Yeah.

就在这里。

Here it is.

大家轮流发言挺好的。

It's a good go around the room.

ARK投资管理有限责任公司是一家经美国证券交易委员会注册的投资顾问公司。

ARK Investment Management LLC is an SEC registered investment advisor.

ARK、WOLF Financial 和 Public 是彼此独立且无关联的实体，彼此之间不存在任何关于对方产品或服务的营销或销售关系。

ARK, WOLF Financial and Public are separate unaffiliated entities and do not have a relationship with respect to any company marketing or selling the products or services of the others.

ARK、WOLF Financial 和 Public 不对其他实体及其关联方、高管、员工或代理的行为、疏忽、损失或损害作出任何声明或保证，并明确拒绝承担任何责任。

ARK, WOLF Financial and Public make no representation or warranties regarding and expressly disclaim any responsibility or liability for any acts, omissions, losses, or damages of the other entities or their affiliates, officers, employees, or agents.

任何公司均不对因本播客或其内容而产生的间接、偶然、后果性或特殊损害承担责任。

No company shall be liable for any indirect, incidental, consequential, or special damages arising out of or in connection with this podcast or its contents.

本节目所提供的信息仅作参考之用，不可作为任何投资决策的依据，且可能在未通知的情况下随时更改。

The information provided in this show is for informational purposes only and should not be used as the basis for any investment decision and is subject to change without notice.

这不构成 ARK 明示或暗示的任何服务或产品提供，投资者应自行判断某项投资管理服务是否符合自身的投资需求。

It does not constitute either explicitly or implicitly any provision of services or products by ARK, and investors should determine for themselves whether a particular investment management service is suitable for their investment needs.

节目中关于公司或证券的所有陈述，均仅代表 ARK 或节目嘉宾的个人观点，并不构成 ARK 对任何公司或证券的推荐，也不建议买入、卖出或持有任何证券。

All statements made regarding companies or securities are strictly beliefs and points of view held by ARK and or show guests and are not endorsements by ARK of any company or security or recommendations by ARK to buy, sell, or hold any security.

历史表现不能作为未来表现的依据。

Historical results are not indications of future results.

本节目中的某些陈述可能涉及对未来预期和其他前瞻性表述，这些表述基于 ARK 当前的观点和假设，包含已知和未知的风险与不确定性，可能导致实际结果、表现或事件与陈述中所表达或暗示的内容存在重大差异。

Certain of the statements contained in this show may be statements of future expectations and other forward looking statements that are based on ARK's current views and assumptions and involve known and unknown risks and uncertainties that could cause actual results, performance, or events to differ materially from those expressed or implied in such statements.

ARK没有义务更新任何前瞻性信息。

ARK assumes no obligation to update any forward looking information.

ARK及其客户，以及相关人士，可能持有但不一定持有所讨论证券或发行方的经济利益。

ARK and its clients, as well as its related persons, may, but do not necessarily, have financial interest in securities or issuers that are discussed.

部分信息来源于ARK认为可靠的渠道。

Certain information was obtained from sources that ARK believes to be reliable.

但ARK不保证从任何第三方获取的信息准确或完整。

However, ARK does not guarantee the accuracy or completeness of any information obtained from any third party.