解读SpaceX首份财报

上周在节目中我们说过，8月4日是SpaceX必须证明自己的日子。

Last week on this show, we said August 4 was the day that SpaceX had to prove themselves.

第一份财报出来后，华尔街的分析师们基本上都持悲观态度。

The first earnings report, every analyst on Wall Street basically was bearish.

他们说埃隆·马斯克不可能做到，说这玩意儿就是个烧钱的无底洞。

They said that Elon Musk could not pull this off, that this thing is a money vacuum.

结果恰恰相反。

Turns out the opposite was the case.

SpaceX彻底惊艳了所有人。

SpaceX absolutely smashed it.

营收达到了78.1亿美元，同比增长92%，对于他们这个规模来说简直不可思议，而且星链的用户数翻了一番，达到了1200万。

The revenue is up to $7,810,000,000, which is 92% year on year, which is crazy for the size that they're at, and Starlink numbers doubled to 12,000,000 subscribers.

尽管如此，今天早上股价还是下跌了。

That being said, the stock is still down this morning.

大家担心的是明天将发生的大规模解禁，价值1000亿美元的SpaceX股票将上市流通。

People are worried about the major unlocks that are happening tomorrow where a $100,000,000,000 worth of SpaceX stocks goes live.

但我和Josh在这里要告诉你，情况可能并非如此，市场也可能是错的。

But Josh and I are here to tell you why that may not be the case and why the market could be wrong.

对。

Yeah.

嗯，我理解为什么大家会看跌。

Well, I understand why people are bearish.

比如，他昨天在财报电话会议上，我旁听了，他说得好像一切都很顺利。

Like, he was on the earnings call yesterday, which I sat in and listened to, and he's claiming he's like, alright.

我们要在月球上建带质量驱动器的机器人，还要把地球经济翻一千倍。

We're gonna build robots with mass drivers on the moon, and we're going to thousand x r economy on Earth because of it.

听起来确实很离谱。

And it sounds ridiculous.

我一边看这些数字，一边慢慢了解大公司是什么样的。

And I was looking through the numbers, as I was reading this, I was, like, slowly starting to inform myself on what large companies look like.

Ejaaz，其实我也想先问你一个问题。

And Ejaaz, I wanna start with a question too, actually.

你知道哪家公司年收入最高吗？

Do you know what company had the most annual revenue ever?

就是，哪家公司每年赚的钱最多？

Like, what what company is making the most money every year?

这真的很让人意外

This was really surprising

对我来说。

to me.

比如，现在？

Like, currently?

就是当下？

Like, right now?

对。

Yeah.

就是每年的总收入。

Like like, total revenue per year.

那谁赚得最多呢？

Like, who's making the most?

我觉得要么是英伟达，要么是亚马逊。

I would have to say it is either like an NVIDIA or an Amazon.

其实根本不是科技公司。

It's not even a tech company.

是沃尔玛。

It's it's Walmart.

沃尔玛。

Walmart.

沃尔玛。

Walmart.

营收最高达到6800亿美元，听起来太夸张了。

Topped out at $680,000,000,000 of revenue, which seems outrageous.

而且，这就涉及到定价的问题，比如沃尔玛并不是世界上最大的公司。

And, I mean, it gets into the pricing thing where, like, Walmart's not the biggest company in the world.

这是为什么呢？

Why is that?

嗯，因为他们的利润率低，市盈率也低。

Well, because their profit margins are low, their multiples are lower.

但是SpaceX和埃隆在电话会议上说，嘿。

But SpaceX and Elon went on this conference call, he said, hey.

我们实际算了一下，不是2031年，而是到2030年，他们预计收入能达到一万亿美元。

We actually ran the numbers, and instead of 2031, by the year 2030, they expect to have a trillion dollars in revenue.

这比当今全球最大的公司还要多40%或32%。

That is 40% or 32% more than the largest company on earth today.

我觉得我们得好好分析一下这意味着什么，因为，那可是巨大的价值，而且他们刚刚报告了5亿美元的净亏损，对吧。

I think we kind of have to unpack what that looks like because, I mean, that is a tremendous amount of value given the fact that they actually just reported a net loss of $500,000,000 Mhmm.

虽然有所改善，但仍然是净亏损。

Which is an improvement, but still a net loss.

如果你能把公司的财报分成三大块，首先是连接业务，比如星链和他们的所有卫星项目。

If you could kinda break down the company's earnings report into these three pillars, first, it is connectivity, which is like Starlink and all of their satellite situations.

然后他们还有AI业务，同比增长了250%。

Then they have AI, which was up 250% year over year.

然后还有太空业务，同比增长了30%。

And then they had Space, which was up 30% year over year.

而且，我觉得你提到的重点，Ejaaz，就是32%的同比增长，不对，抱歉。

And the I mean, the important thing that you noted, Ejaaz, is that 32% growth year over year is sorry.

92%的增长是非常大的，因为对于年化收入300亿美元的公司来说，典型的增长率大概只有10%。

92% growth is huge because for companies with $30,000,000,000 of annualized revenue, typical growth is, like, 10%.

所以很明显，他们

So clearly, they're

做对了某些事情。

doing something right.

我昨天一直在关注他的社交媒体，他说的一件事是，我们有望在今年年底达到1000亿美元的年化经常性收入。

I was keeping an eye on his social media yesterday, and, one of the things that he said is we are on track for a $100,000,000,000 ARR by the end of this year.

在2020年，我听到过类似预测的只有Anthropic。

Now 2020 the only other company that I've heard that projection being made for is Anthropic.

现在，当我看到SpaceX和Anthropic时，我觉得它们可能在同一个领域竞争，但我不认为SpaceX在AI方面已经完全赶上，但这就是大多数人忽略的地方。

Now, when I look at SpaceX and I look at Anthropic, I'm saying, okay, they may be competing in the same sort of field, but I don't think SpaceX is quite caught up on the AI side, but that's what most people miss.

SpaceX不仅仅是一家AI公司。

SpaceX isn't just an AI company.

他们是一家计算公司。

They are a computation company.

他们拥有最大的武器库，最大的NVIDIA GPU规模，我得说。

They have the largest arsenal, largest GPU footprint of NVIDIA GPUs, might I add.

而且他和Jensen签了一个重要的合作，我们稍后会谈到，但他是唯一一个把这些东西发射到轨道空间的人。

And he signed a major partnership with Jensen, which we'll get to in a second, but he's the only person that's launching these things into orbital space.

你知道，他拥有地球上最大的数据中心集群，而且他还在做Starlink连接的事情。

You know, he has the largest data center cluster on earth and he's also doing the Starlink connectivity thing.

所以这些都是不同的收入来源。

So all of these are different revenue drivers.

我喜欢这份报告的地方在于，他仅仅通过数字就证明了很多看空者是错的。

And what I like about this report is he's proved a lot of the bears wrong simply by the numbers.

咱们来具体看看这些数字吧，对吧？

Like, let's actually like walk through some of these numbers, right?

营收方面，预期是69.3亿美元，结果他多赚了将近10亿美元，远超预期。

So revenue, the estimate was $6,930,000,000 He smashed it by earning almost a billion dollars more than that.

对吧？

Right?

再看每股收益，大家原本以为会每股净亏损0.26美元。

And then when you look at earnings per share, people were like, okay, they're gonna be like at a net loss of $0.26 per share.

结果不是。

Nope.

实际是每股盈利0.09美元。

Came in at 0.09.

现在盈利了吗？

Is it profitable yet?

还没有，但已经比大家预期的好太多了。

No, but it is still much better than people expected.

然后说到净利润，就像你之前提到的，大家原本预期会亏损20亿美元。

Then the net income, as you mentioned earlier, people were expecting a $2,000,000,000 loss.

他把这个亏损缩减了三分之二，降到了5亿美元。

He shaved that down by three X to $500,000,000.

那么，这真的是他们能交出的最好的财报吗？

Now, again, is this the best earnings report that they could have come up with?

可能不是，但对于像SpaceX这样资本开支很重的公司来说，这些都需要时间。

Probably not, but these things take time for a CapEx heavy company such as SpaceX.

我对他们能取得这样的成绩感到非常佩服。

I'm really impressed with what they've been able to pull off.

而且我觉得，这也正好说明了埃隆的言论有多疯狂。

And I think like, it just kind of like shows that Elon has pretty crazy statements.

我记得他大约十二个月前提出太空GPU和星链机器人的愿景时，大家都觉得他完全疯了。

I remember when he pitched like the Starlink robots, GPUs in space vision about, what was this, like twelve months ago for the for the GPUs in space.

但他走的每一步，每一份财报，都让我们开始看到一个不同的图景在逐渐成形。

And everyone thought he was absolutely insane, but every step he takes, every earnings report he takes, we're starting to see a different picture start to form.

我不是说他现在已经做到了。

And I'm not saying he's there yet.

我想说清楚这一点。

I just wanna be very clear.

我觉得我们离那一步还远着呢，但这是朝着正确方向迈出的一步。

Like, I don't think we're anywhere near there, but this is a step in the right direction.

而且我觉得，听了昨天的电话会议，看了很多报道之后，我觉得他可能已经做到了。

And I think that, you know, after listening to the call yesterday and after seeing a lot of the reporting, I think he might be there.

但当然，大家担心的是明天发生的解锁事件。

But of course, something that people are worried about is the unlock that is happening tomorrow.

我不知道你怎么看，Josh，但有一千亿美元正在涌入市场。

And I I don't know if you have any thoughts on this, Josh, but, like, a $100,000,000,000 is flooding the market.

对。

Yeah.

嗯，我是说，肯定会有大量股票在流动，因为SpaceX目前也是市场上被做空最多的公司之一。

Well, I mean, there's gonna need to be a lot of shares that are moving because SpaceX is also among the most shorted companies in the marketplace right now.

所以它有很多空头头寸。

So there's there's a lot of short positions on it.

有很多人

There's a lot of people

看跌。

that are bearish.

有很多

There is a lot

解锁事件正在发生。

of unlocks happening.

这些解锁最终是利好还是利空还不确定，但我的预期是，我其实不太在乎。

Whether or not those unlocks become bullish or bearish is TBD, but my expectation is, like, I I don't really care.

作为长期投资者，比如，可能接下来几天会下跌，可能接下来一周会下跌，甚至可能接下来一个月会下跌。

Like, as an investor over a long period of time, like, okay, maybe it sells off for the next couple days, maybe it sells off for the next week, maybe the next month.

但说实话，这些数字相当惊人，你也能理解为什么人们会感到害怕。

But, I mean, the numbers are pretty amazing, and you could see why people are getting scared.

你提到了资本支出。

You mentioned the CapEx.

资本支出比人们预期的要高出39%。

CapEx was 39% higher than what people thought.

实际支出接近185亿美元，而NASA的年度预算大约是250亿美元，所以他们一个季度就花掉了几乎整个NASA的预算。

It came in at just under $18,500,000,000, NASA's, like, annual budget is 25,000,000,000, so they spent almost the entirety of NASA's budget in a single quarter.

这也意味着，他们每赚1美元收入，就花了2.35美元。

That also means that they spent $2.35 for every dollar that they made in revenue.

那你不得不问，这些钱都花到哪里去了？

And you have to ask the question, like, where is that going?

为什么他们要花这么多钱？

Why are they spending so much?

因为这听起来简直像疯了。

Because this sounds like lunacy.

就像你投资了一家餐厅，他们每晚只赚1万美元，却要花掉2.3万美元。

It's like if you have imagine a restaurant, you're investing in a restaurant, and they come and they spend $23,000 a night, even though they're only making $10,000 a night.

你会说，这到底是怎么回事？

You say, what on earth is going on here?

这完全说不通。

This makes no sense.

但然后他们告诉你，等等。

But then they tell you, wait a second.

我们正在建设新的餐厅。

We're we're building out new restaurants.

我们在打造更高利润率的菜单项目，这有点像SpaceX在数据中心方面做的事情。

We're building a higher profit margin menu items, and that's kind of what SpaceX is doing on the data center front.

所以，我认为很多人忽略的一个小细节是，他们已经积压了大约500亿美元的已签约收入。

So one of the smaller things that I think a lot of people overlooked is the fact that they already have a backlog of about $50,000,000,000 of revenue that's already signed.

就像，甚至没有人需要打电话，他们就能收到500亿美元，仅仅因为他们有能力建设这些数据中心。

Like, no one even has to pick up a phone, and they're collecting $50,000,000,000 just because of their ability to build these data centers.

这就是我们稍微聊聊数据中心的地方，它们确实是一种不公平的优势。

This is where we kinda talk about data centers for a second, which is they're, like, very much an unfair advantage.

在财报电话会议上，我听了他们的首席财务官，我觉得他叫布雷特·约翰逊。

On the earnings call, I was listening to their CFO, Brett Johnson, I think his name is.

他说，所有的资本支出都不一样。

He said that, like, all CapEx isn't the same.

有些资本支出比其他更赚钱。

Some CapEx is much more lucrative than others.

而就SpaceX而言，这几乎是我们听过的最赚钱的。

And in the case of SpaceX, it's about as lucrative as we've heard.

很多公司，比如你投资一栋公寓楼，建了这么大一片房地产，大概要十五到二十年才能回本。

A lot of companies, let's say you invest in an apartment building and you build this huge piece of real estate, it takes about fifteen to twenty years to pay itself back.

SpaceX的数据中心在一年之内就回本了。

SpaceX's data centers paid themselves back in a single calendar year.

所以从投资资本回报率来看，这简直是一个不可思议的套利机会，感觉他们有一种不公平的优势，因为他们是唯一能快速高效建造这些数据中心的人，然后还能把那些巨大得难以置信的合同卖给那些做不到同样事情的公司。

So in terms of return on invested capital, it's like this unbelievable arbitrage opportunity where it feels like they they have this unfair advantage being that they're the only ones that could build these data centers quickly and efficiently, and then go off and sell these huge unbelievably large contracts to companies that can't do the same.

而我最喜欢财报电话会议上的那句话，绝对是最喜欢的一句，就是马斯克被问到关于地面数据中心的时候。

And my favorite quote from the earnings call, my favorite one by far is what Elon was asked about terrestrial data centers.

这是在他说完太空数据中心之后说的，我们一会儿再聊那个。

This is after he concluded about the space data centers, which we'll get to in a minute.

但说到地面数据中心，他说这就像纽约洋基队去打小联盟比赛一样。

But the terrestrial data centers, he's like, this is like the New York Yankees playing in a little league.

感觉他们在执行这些事情上的能力简直是天壤之别，而且他们的投资资本回报率非常高。

It feels like they're just night and day difference in terms of their ability to execute on these things, and their return on the invested capital is huge.

你根本买不到Blackwell GPU，没得商量。

You can't buy a Blackwell GPU, period.

所以这就是为什么人们愿意支付溢价。

So that's why people are paying premiums.

事实上，如果你现在签了Blackwell的合同，你会尽量延长合同期限，因为如果续约的话，费用会翻倍，因为这种计算资源的成本实在太贵了。

In fact, if you are currently signed to a Blackwell contract, you are trying to extend the length of that as much as you can because if it renews, it's going to renew at double the cost because the cost of this compute is so expensive.

唯一能像他们那样快速大规模建设的就是SpaceX。

The only people that are able to build it at scale as quick as they do is SpaceX.

而且，投资这么大，回报周期只有十二个月，你很快就能明白为什么他们的收入模型在未来四年左右会朝着接近一万亿的方向发展，而不是现在的水平。

And I mean, return on invested capital on an investment this large being twelve months, you could very quickly see why their revenue models are guiding towards something closer to a trillion than where we currently are over the next four years or so.

这简直难以置信。

It's it's unbelievable.

我们先单独说说GPU这一点。

Just to focus on the the GPU point for a second.

对吧？

Right?

我觉得很多人都在担心。

Like, think a lot of people are worried.

那些唱衰AI泡沫的人说，泡沫主要是在GPU的资本开支上形成的。

The the doomers for the AI bubble say that the bubble is primarily formed in GPU CapEx specifically.

所以，所有这些投入在NVIDIA GPU、AMD GPU上的数万亿美元，不管是什么，都是泡沫在膨胀的地方，因为这些机器最终会贬值。

So all these trillions of dollars that is being committed and spent on NVIDIA GPUs, AMD GPUs, whatever that might be, is where the bubble is inflating because these machines will eventually, you know, depreciate.

一旦token成本下降，它们就不会再有需求了，等等等等。

They won't actually be in demand once the cost of tokens go down, etcetera, etcetera, etcetera.

我们从所有主要超大规模云服务商的每一份季度财报中看到的情况，他们是投资最多的，包括SpaceX，或者除了SpaceX之外。

What we've seen from every single quarterly earnings report from all of the major hyperscalers, which are the ones that are investing the most, including or aside to SpaceX.

所以你看亚马逊、谷歌，他们都在报告创纪录的收入数字。

So if you look at Amazon, Google, they're all reporting record revenue numbers.

事实上，他们所有的云服务平台，比如Azure、Google Cloud、AWS，不仅收入在增长，而且我估计每个季度复合增长率大概在40%到45%。

In fact, all their cloud service provider platforms like Azure, like Google Cloud, like AWS, not only is their revenue increasing by like, I think it's like 40 to 45% compounded every single quarter, by the way.

那可是每三个月啊。

That's every three months.

我觉得大家可能没意识到这有多疯狂。

Don't think people understand how absolutely insane that is.

不仅收入在复合增长，而且他们的利润率还在扩大。

Not only is it compounding their revenue specifically, but their margins are expanding even more.

所以他们赚的钱还能赚更多的钱，这生意简直好得离谱。

So they're making more money on the money that they're earning, which is just like an insanely good business.

所以马斯克看到这个情况，就说，我的GPU比你们谁都多。

So Elon saw this and was like, well, I have more GPUs than any of you.

而且我建数据中心的速度也比你们都快。

And I'm building these data centers faster than any of you.

我们变成像新云服务那样的公司，这很合理。

It would make sense for us to become like a Neo Cloud service.

这就是为什么他要把价值高达500亿美元的AI算力卖给Anthropic、谷歌，以及未来可能出现的其他客户。

And that's why he's selling upwards of $50,000,000,000 of AI compute to the likes of Anthropic, Google, and whoever that might be going forward.

所以他看到了这是一个利润非常丰厚的业务。

So he's seen that it's a hugely profitable business.

而且，他昨天发布的一个重大公告让我特别喜欢，他公开表示，我们将只与Jensen合作。

And what I love about one of the major announcements that he made yesterday is he came out and said, we are gonna be exclusive to Jensen.

我们将只与NVIDIA合作。

We're gonna be exclusive to NVIDIA.

他原话是这么说的：未来，我们决定只基于NVIDIA构建，因为我们觉得Vera Rubin架构是最好的架构。

And verbatim, he quotes, going forward, we've decided to build exclusively on NVIDIA because we think that the Vera Rubin architecture is the best architecture.

我们认为这是最好的AI计算机，所以我们将只与NVIDIA合作。

We think it's the best AI computer, and so we're going to be exclusive to NVIDIA.

而最疯狂的部分是他承诺要购买的东西。

Now the craziest part about this is what he committed to buying.

他基本上要在明年年底前把现有或SpaceX现有的GPU规模翻一倍。

He is basically going to double his existing or SpaceX's existing GPU fleet by the end of next year.

我怎么说都不够，这尝试简直太疯狂了。

I cannot emphasize enough how crazy of an attempt that's going to be.

那钱就全进NVIDIA的口袋了。

That is gonna be in NVIDIA's pocket.

假设他真做到了，对吧？

Let's say he pulls this off, right?

假设到明年年底他真达到了累计10吉瓦的GPU。

Let's say he gets to the 10 gigawatts cumulative worth of GPUs by the end of next year.

那光是SpaceX一家，到明年年底就会给NVIDIA带来2000亿美元的收入。

That is $200,000,000,000 from SpaceX alone in NVIDIA's pocket by the end of next year.

现在咱们现实点看。

Now let's be realistic.

比如说，他现在大概在2到2.5吉瓦的水平。

Let's say like, you know, he's at two, two and a half gigawatts.

到今年年底，他会达到三吉瓦。

He'll get to three by the end of this year.

假设到明年年底，他能达到大约六吉瓦。

Let's say he gets to like six gigawatts by the end of next year.

那相当于给英伟达多赚一千亿美元。

That is an extra $100,000,000,000 in NVIDIA's pocket.

他会成为英伟达的头号客户，而且他对黄仁勋做出了巨大的承诺。

He's gonna be his number one customer, and he's taking a massive commitment with Jensen.

这很有意思，因为就在几天前，Anthropic 和 AMD 签了一份大单。

And this is interesting because a few days earlier, Anthropic signed this major deal with AMD.

他们也在和其他云服务提供商签大单。

They're signing major deals with other cloud service providers.

埃隆基本上就是说，黄仁勋，如果你能保证给我分配额度，我们就会拿下这些算力。

Elon has basically said, Jensen, if you can guarantee me the allocation, we will take the compute.

这就是我对你未来的承诺，即使我们把它们发射到太空也一样。

And that is my commitment to you going forwards, even when we launch them into space.

对吧？

Right?

所以我非常看好他打造的云计算和算力业务。

So I'm just massively bullish on the cloud business, on the compute business that he's built.

我觉得就算其他事情都黄了，埃隆的生意依然非常赚钱，而市场现在严重低估了这一点。

I think even if everything falls to the wayside, Elon still has a very profitable business, and people are discounting that massively right now.

这个交易我也觉得挺有意思的。

This deal is pretty interesting to me too.

这是因为我们看到了很多这类加速芯片。

It's because we're seeing a lot of these accelerator chips.

比如训练芯片，还有谷歌的TPU芯片，大家都在关注这块。

We're seeing, like, the training chips and the the TPU chips from Google, and everyone's kinda focused.

就连OpenAI也在研发自己垂直整合的芯片组。

Even OpenAI is building their own vertically integrated chipset.

有趣的是，SpaceX却走了另一条路。

And it's interesting to see SpaceX going the other direction.

他们说，嗯，我们会在一定程度上自己造芯片，但也会和英伟达合作，用他们的新架构。

They're saying, well, we are gonna build our own chips to some extent, but we're also gonna work with NVIDIA to use their new architecture.

我们聊了很多关于这个新的Blackwell时代模型有多不可思议。

And we've talked a lot about how this new Blackwell era of models is unbelievable.

Vera Rubin时代的模型可能会非常出色，而SpaceX将独家使用它，这非常看好，因为我觉得这对双方都有好处。

The Vera Rubin era of models is going to probably be, like, remarkable, and SpaceX will be using that exclusively, which is very bullish because then I assume it it works both ways.

对吧？

Right?

就像英伟达得到了一个绝佳的展示机会，可以向全世界展示。

It's like NVIDIA gets this amazing showcase to share with the world.

比如说，嘿。

Like, hey.

看看我们的GPU能做什么。

Look what our GPUs can do.

看看它们部署得有多快。

Look how fast they can be deployed.

而且SpaceX团队在软件架构和硬件架构的开发上，能和英伟达团队有更紧密的反馈循环，我觉得这肯定会变成两家公司之间非常棒的合作关系。

And the SpaceX team gets a much more tighter feedback loop with the team when it comes to development of the software architecture, the hardware architecture, and I assume that's going to turn into, like, a pretty nice partnership from both of these companies.

我是说，两位创始人都很欣赏对方。

I mean, both founders love each other.

两人都处于巅峰状态。

Both are at the top of their game.

看到这一切发生，真的非常令人兴奋。

It's it's very exciting to see this happen.

而且初步计划，Ejaaz，你之前提到他们计划部署多少吉瓦的电力。

And the tentative plan, Ejaaz, you were talking about how many gigawatts they were planning to deploy.

在电话会议上，埃隆最初说在2027年底前达到10吉瓦。

On the conference call, Elon initially said 10 by the end of twenty twenty seven.

然后当人们追问时，他说到2030年左右，大约20吉瓦，最低也有15吉瓦。

And then when people pressed him by the end of the decade, around 2030, he is guiding towards roughly 20 gigawatts of power with a floor of 15 gigawatts.

所以他就说，好吧。

So he's like, alright.

就算一切都不顺利，我们至少也能达到15吉瓦。

Well, even if everything goes wrong, we can at least get to 15 gigawatts.

作为参考，我们之前聊过很多次，但一个吉瓦的电力大约相当于80万户家庭的用电量。

For reference, we've talked about this a lot, but, like, a singular gigawatt of power is equivalent to roughly 800,000 homes.

这相当于一个城市的电力，而他们计划在未来四年内建15个这样的项目？

It's like cities worth of power, and they're planning to do 15 of these in the next four years?

这速度真的太快了。

Like, that's that's really fast.

另外，黄仁勋在部署吉瓦级能源方面也有自己的一套经验法则。

The other thing is, I mean, Jensen has his own kind of rule of thumb when it comes to deploying gigawatts of energy.

他说，每吉瓦大约能带来400到500亿美元的年收入。

He says you can guide for around 40 to $50,000,000,000 of annual revenue per gigawatt.

所以，如果SpaceX到本十年末真的能卖出15吉瓦的容量，光这一项就值6000到7500亿美元的收入，这还不包括这家公司的其他支柱业务，比如太空和连接服务。

So if SpaceX does get 15 gigawatts of sold out capacity by the end of the decade, that alone is worth 600 to $750,000,000,000 of revenue, and that doesn't include the other pillars of this company, which are space and connectivity.

所以，即使你忽略其他所有业务支柱，这个看涨的理由也令人难以置信，但除此之外，还有像太空中的AI数据中心这样的支柱，这些又是和黄仁勋以及英伟达一起建设的。

So there is this, like, unbelievable bull case even if you disregard all of the other pillars of the business, but in addition to this, there are these pillars, like AI data centers in space, which again, they are being built with Jensen Huang and NVIDIA.

我觉得他们那个AI卫星的想法特别有意思，基本上就是在地球上建什么，在太空里也建什么。

And the idea I found this really interesting for their AI one satellites is they're basically gonna build the same thing on Earth as they are gonna be in space.

唯一的区别就是他们会给这个家伙装上太阳能板，然后把它发射到太空去。

The only difference is they're gonna put some solar panels on this bad boy and send it off into space.

所以这种从硬件到软件的垂直整合，不管是地面还是太空，都做得非常出色。

So this vertical integration of the hardware stack as it relates to terrestrial, as it relates to space, it's just remarkable.

如果他们真的能大规模把载荷送上太空，那简直不得了。

And then if they actually figure out how to get payload into space at scale, oh my god.

你把数字一加起来，就会想，等等。

Like, you start to put the numbers together, you're like, wait a second.

他们真的能做到。

They actually could do this thing.

而且，如果他们四年内把地面数据中心和15吉瓦的电力都搞定，那股价会涨到什么程度？

And, like, what is the stock gonna be trading at if they get terrestrial data centers, 15 gigawatts online in four years?

再加上他们开始大规模往太空送东西，这绝对是个巨大的机会。

And also they start putting mass into space at scale, like, this is gonna be a pretty huge opportunity.

我听到很多听众和看这个节目的人最大的批评是，这个愿景、这个计划太疯狂了，根本不可能实现。

The biggest critique, that I hear from a lot of you listeners and that watch this show is that this vision, this plan is just too insane to pull off.

说实话，我也同意你们的看法，但你也得看看那些真正能让公司赚钱的证据，这些钱能帮助埃隆和SpaceX实现一些更疯狂的愿景。

And honestly, like, I agree with you, but you also have to look at the evidence of things that can actually make the company money that will help fuel Elon and SpaceX getting to some of the more crazier visions.

对吧？

Right?

其中一个主要的业务就是计算收入，他一直在展示这个，说，嘿，这个能赚钱，而且我们会是最大的赌注，对吧？

And one of the main ones is this compute revenue business that he's kind of showcasing and saying, hey, this thing makes money and we're gonna be the biggest bettors on this, right?

我们一定能做到。

We're gonna be able to pull this off.

如果他能跟黄仁勋和英伟达锁定这些算力，而且黄仁勋那么喜欢埃隆·马斯克，那他肯定能搞定。

And if he's able to secure that capacity with the likes of Jensen and NVIDIA, and Jensen loves Elon Musk so much, he will be able to pull this off.

现在，简单解释一下这个计算业务对埃隆意味着什么，这不仅仅是通过卖算力给竞争对手来赚钱。

Now, just to kind of translate kind of like what this compute means for Elon, it's not just a way to kind of like earn money by selling compute to competitors.

他也在用这个来训练世界上一些最顶尖的AI模型。

He's also very much using this to train some of the best AI models in the world.

当然，你可能会觉得这很好笑，但如果你把时间线拉长到接下来的六到十二个月，你会发现，不管是什么模型、什么架构、用什么权重，唯一不变的就是算力。

And of course, like you might listen to that and like laugh, but if you kind of like span out the timeline over the next kind of six to twelve months, the only thing that is consistent across every single model, different model architecture, whatever, like, you know, different weights that you use, It is compute.

预训练需要更多的算力。

You need more compute to pre train.

后训练需要的算力要多得多。

You need so much more compute to post train.

然后，你还需要算力来以完美的速度和服务质量把模型服务给用户。

And then you need compute to serve this to people at like perfect speeds and quality of service.

你就是需要更多的算力。

You just need more compute.

要做到这一点，你需要考虑两件事。

And in order to do that, there's two things you need to look at.

你不仅需要足够的资金来购买GPU，还需要足够的资金来购买最好的GPU。

You need enough capital to buy the GPUs, but you also need enough capital to buy the best GPUs.

Tuakesh写了一篇很棒的文章。

Tuakesh had an amazing blog post.

不知道你有没有读过那篇文章，Josh。

I don't know if you read it, Josh.

他基本上是说，GPU越好，能解锁的AI模型就越强，这意味着NVIDIA基本上会把GPU卖给那些出价最高的人。

He basically said the better the GPUs get, the better AI models these GPUs can unlock, which means that NVIDIA will basically sell these GPUs to the highest bidder.

那出价最高的会是谁呢？

Who are the highest bidders gonna be?

我喜欢这个例子。

I like this example.

有两家公司，对吧？

There's two companies, okay?

公司A想用10万块GPU来削减他们公司30%的行政工作量。

Company A wants to use a 100,000 GPUs to cut down a bunch of, let's say administrative workload at their company by 30%.

这影响真的很大，对吧？

That is a really meaningful impact, right?

他们会省下很多钱，还能赚几百万美元。

And they'll save a ton of money and they'll be able to generate millions of dollars.

太好了。

Great.

但这里有个问题。

But there's a problem.

B公司想要同样的GPU来训练模型，目标是治愈癌症，对吧？

Company B wants those same GPUs to train and cure cancer, right?

在一个月内。

Over a month.

他们知道，如果能用他们的数据跑一次训练，就能治愈癌症。

And they know that if they can do a one training run with their data, they'll be able to cure cancer.

谁会出价更高？

Who's gonna bid more?

肯定是B公司。

It's gonna be company B.

所以这些GPU非常非常稀缺。

So these GPUs are like very, very scarce.

说到底，拥有最多资本的人会买下那些GPU，训练出最好的模型。

And at the end of the day, the people that have the most capital will buy those GPUs and train the best models.

他们会训练出最好的功能、产品、服务，不管是什么。

They'll train the best features, products, services, whatever that might be.

而埃隆走在这所有事情的最前面。

And Elon is at the front of all of this.

甚至不是Anthropic，而是他现在已经拿到了最多的分配。

It's not even Anthropic or, but now he has secured the most allocation.

他现在拥有最大的GPU储备。

He has the largest arsenal of GPUs right now.

所以可以说，他有可能创造出最好的AI模型，Grok 4.6、Grok 4.7，还有Grok 5，今年年底就会出来。

So it goes to say that he could potentially create the best AI model, Grok 4.6, Grok 4.7, and Grok five is gonna be coming out by the end of this year.

他还说会用SpaceX的所有数据来训练，这从技术上来说，会是最好的工程AI模型。

And he said it's gonna be trained all of SpaceX's data, which would technically make it like the best engineering AI model.

所以有很多事情都在酝酿中，正在发展，我们还没完全看到，但了解一下总是好的。

So there's a lot of these things that are kind of like in the oven, they're cooking and we don't quite see them right yet, but it's good to be just generally aware of.

我不知道，我就是对六到十二个月后的前景非常乐观。

I don't know, I'm just very optimistic about where this could be in six to twelve months.

是的，他们有很多

Yeah, there's lot

高质量的尝试机会，即使其中一个彻底失败了，还有很多其他项目进展顺利。

of really high quality shots on goal that they're taking, where even if one completely and catastrophically fails, there's still many others that are doing well.

而且听到一家太空公司谈论地面数据中心，真是挺有意思的。

And it's funny hearing a space company talk about terrestrial data centers.

他们正在积极尝试把这些东西送入太空，而这些东西实际上是在试图自我毁灭。

They're talking about, like, they're actively trying to send these things into space that are actually trying to blow themselves apart.

然后你就会想，把这些东西放在地球上多容易啊，而且大气层也好处理。

And it's just like, how easy is it that you could just drop these things on Earth, and it's easy, and the atmosphere is easy to handle.

这就像回到棒球的比喻，就像你拿着加重球棒上场，然后换成普通球棒，而普通球棒就是地面数据中心。

And and it just seems like going back to the baseball analogy, it's like when you go up to bat with, like, a weighted bat, and then you use, like, a normal bat, and a normal bat is the data centers that are terrestrial.

那些可能会用于训练。

And those will probably be used for training.

当我们考虑训练时，延迟很重要，那些可能用于训练。

When we think about training where latency matters, those are probably used for training.

卫星层有下行链路延迟，那可能更多用于推理，延迟要求没那么高。

The satellite layer where there is, like, downlink latency happening, that's probably more for inference where latency matters a little bit less.

但我们很快就会看到星链星座的发射，最快明年，星链版本三会随着下一艘星舰一起发射。

But we're gonna start to see this Star Mind constellation launch as soon as next year with Starlink version three being launched as soon as the next Starship.

所以下一次星舰发射的计划是延续上次的做法，把星链B3卫星送入轨道。

So the plan for the next Starship launch is to follow-up what they did with the previous launch, which was get these Starlink b three satellites into orbit.

现在他们真的要部署这些卫星，把它们送入轨道，开机，让它们工作，这就涉及到星链引擎的部分，那部分也相当令人难以置信。

Now they're actually gonna deploy them, put them in orbit, turn them on, make them work, and that takes us to the the Starlink part of the engine, which is pretty unbelievable as well.

这项服务的增长曲线几乎就是一条纯粹的指数曲线。

The growth curve of this service is the about as pure of an exponential curve as you can get.

他们的订阅用户同比增长了一倍。

They doubled their subscribers year over year.

现在他们已经有1200万用户了。

They are now at 12,000,000.

他们这个季度就新增了170万用户，总共1200万，增长曲线简直像垂直的曲棍球棒一样陡。

They just added 1.7 of those 12,000,000 this quarter alone, so it is just a vertical hockey stick.

我觉得这是财报里最被低估的部分之一。

And I think this is one of the more underrated parts of the earnings report.

格温·肖特韦尔对此说了很多，也说得不多，这关系到他们能否部署一个类似威瑞森或AT&T那样的网络。

Gwen Shotwell, she had a lot or a little bit to say on this as it relates to the ability to deploy a network similar to a Verizon or AT and T.

目前地球上大约有55亿互联网用户。

I mean, currently, there's on Earth, there's about 5,500,000,000 users of the Internet.

所以他们打算成为其中大部分用户的网络管道，彻底消除信号盲区，并实现直接面向消费者的服务。

So they are talking about being the pipeline for a large majority of that, and totally removing dead zones and making it direct to sell.

而且星链的移动业务，我觉得真的很有意思。

And the Starlink mobile play, I think, is really interesting.

我记得去年我们录了一期节目，他们当时买了固定量的频谱，准备在条件成熟后开始部署。

I remember last year, we recorded an episode where they bought a fixed amount of spectrum to start deploying once they have.

他们的计划似乎是继续购买这些专用频谱，直到拥有足够的带宽来真正参与竞争。

Their plan seems to be that they are going to continue to buy these dedicated pieces of spectrum until they own enough bandwidth to truly compete.

而且他们还有那些地面基站，能处理大量带宽，在人口密集的地方也能正常工作。

And then they have these ground base stations, which are able to kind of handle a lot of the bandwidth and work in higher density areas.

格温·肖特韦尔在电话会议上只说他们拥有基础性的竞争优势，没有透露具体的秘密和策略。但我预计，到明年这个时候，我们的iPhone上就会有相当高带宽的互联网连接了。

And Gwen Shotwell, all she said on the call is that they have a foundational competitive advantage and won't really go into the, like, secrets and strategies that they're gonna go with, but what I expect is that this time next year, on our iPhones, we will have, like, pretty high bandwidth Internet connectivity.

这么快就能实现，对我来说是个非常看好的信号。而且，谁会不想用一个在世界任何地方都有良好网络、没有信号盲区的服务呢？

And for that to happen that quickly seems like a very bullish thing to me, and it's just like, who wouldn't sign up for a service that has good Internet anywhere in the world with no dead zones?

我过去两周刚去了国家公园。

I just went to National Parks the last two weeks.

那里完全没有信号。

I had no service at all.

唯一能让我保持联系的就是偶尔出现的星链终端，但直接到手机的服务会非常棒，而且这个业务增长得飞快。

Only thing connected me was Starlink terminals here and there, but direct to sell would be such an awesome thing, and the business is growing so quickly.

这里补充一下，我有时候连续六到十二个小时都联系不上乔希。

For context here, I couldn't reach Josh for, like, six to twelve hours at a time.

那简直太疯狂了。

It was absolutely insane.

就像，冒着我的短信被烧掉的风险。

Like, would risk the burning of my texts.

被红杉和巨杉包围着。

Surrounded by sequoias and redwoods.

我以为他，你知道，受了什么致命伤之类的。

I I thought he, you know, suffered some kind of fatal injury or something like that.

我当时，那是什么电影来着？

I I was what was that movie?

就是那个一百小时什么的，哦，你在外面迷路了。

The the hundred hours or whatever where Oh, you get lost out there.

对。

Yeah.

对。

Yeah.

他在外面迷路了。

Where he gets lost out there.

我当时就想，天哪，Josh没事吧？

Was like, damn, is is Josh okay?

听着，关于Starlink这块，我觉得核心观点很简单：在这个AI模型和AI代理全天候在GPU上运行、不断干活的年代，你需要的是可靠性。

Listen, on the Starlink side of things, I think the thesis put very simply is in a world where you have these AI models, these AI agents running 20 fourseven on GPUs, getting work done, you need reliability.

而在当今世界，可靠性是建立在互联网之上的。

And reliability sits in the modern day world on the internet.

所以你不能让线路出问题。

Now you can't have lines going down.

也不能有连接故障。

You can't have connectivity issues.

更不能让T-Mobile把线路搞砸了。

You can't have T Mobile messing up some of their lines.

那么，有哪些最稳定的方式能把互联网传下来呢？

What are some of the most consistent ways of beaming internet down?

嗯，可以通过卫星，对吧？

Well, you can do it via satellites, right?

而且我也希望我的服务商能提供稳定的连接。

And I would love to have consistency with my provider as well.

说实话，如果我现在能用得起星链，而且能把它用在手机套餐上，我肯定会换过去，因为那样的话，我在飞机上或者其他地方都能用。

Like to be honest, like if I could have an affordable Starlink connection right now on my phone mobile package, I would move to that because then I could just use it on a plane or whatever that might be.

所以我觉得这不仅仅是一时的趋势。

So I don't think that this is just a temporary trend.

我觉得这就像AI一样，每个人都会想要稳定、强大、快速的互联网连接。

I think this is something similar to AI where everyone's just gonna want consistent, strong, fast access to the internet.

而星链显然是个很明显的赢家。

And Starlink is a very obvious win.

这又回到了我之前说的，马斯克涉足了很多不同的领域，而且这些领域确实都在起作用。

And it goes back to my earlier point, which is Elon has his hands in a lot of different things and they're actually working.

它们各自都运行得很好，而且它们之间也能很好地协同工作。

They're working well individually, but they're working well together as well.

我们会在未来的季度财报中看到这一点。

And we're gonna see that in future quarterly earnings report.

对。

Right.

Ejaaz，你提到了智能体。

Ejaaz, you mentioned agents.

说到智能体，我们得聊聊节目的赞助商LEDGER，因为如果你在用AI智能体开发，又担心安全问题——其实很多人都该担心，毕竟最近不少AI模型都突破了它们的沙盒限制。

And speaking of agents, we have to talk about a sponsor of the show, LEDGER, because if you are building with AI agents and you're worried about security, which a lot of people should be I mean, we've had a lot of AI models recently that have broken out of their sandboxes.

这有点吓人。

It's a little scary.

一个拥有不受追踪权限的智能体确实非常吓人。

An agent with untracked access is very scary.

所以Ledger有一个三步流程来解决这个问题：智能体提议，人类批准，然后Ledger签名器执行。

So Ledger has this three step process to solve this in which an agent proposes, a human approves, and then a Ledger signer enforces.

这项技术兼容Cloud Code、Codex、Cursor，这些我们常用的工具，而且它是开源的，可以完全审计。

This technology works with Cloud Code, Codex, Cursor, all of the things that we are used to using, and it's also open source to be fully audited.

基本上，他们用的是叫Ledger智能体栈的东西，包含四个开源工具。

Basically, they use this thing called a ledger agent stack, which has a series of four open source tools.

有钱包，有企业解决方案，还有多重签名，这样你的任何交易都不会被入侵，而且你基本上可以正常地跟它对话。

There's a wallet, there's an enterprise solution, there's a multisig so that you don't get compromised for any of your transactions, and you could basically just talk to it normally.

你可以说“重新平衡我的钱包”，它会帮你完成所有计算和交易，同时让你全程批准并参与其中。

You could say rebalance my wallet, and it'll do all the math and do the transactions on your side while letting you approve and be in the loop the entire time.

所以我想特别感谢LEDGER，提供了一种负责任的方式来构建AI代理。

So just want to shout out LEDGER for a responsible way of building with AI agents.

非常感谢。

Thank you so much.

你可以在下面的描述中找到链接。

You can find the link in the description down below.

在我们结束这一集之前，Josh，我们需要谈谈这一切的看跌情况。

So before we wrap up this episode, Josh, I we need to talk about the the bear case of all of this going on.

我们在之前的节目中已经提到过这一点。

Now we've mentioned this on a previous episode.

目前可供人们交易的SpaceX股票供应非常有限。

There is a very constricted supply of SpaceX stock that is available for people to currently trade.

最主要的批评意见是，明天市场会被大量股票涌入冲垮。

And the number one bit of critique is it's gonna flood the gates tomorrow.

实际上，当你听到这期节目的时候，应该就是今天，6号，价值一千亿美元的SpaceX股票正式上市交易。

Actually, as you're listening to this episode, it should be today on the sixth, dollars 100,000,000,000 worth of SpaceX shares goes live.

大家心里最大的疑问是，这些股票谁来买？

And the big question in everyone's mind is where's the buying power coming for this?

因为要想维持当前股价，必须有人把所有这些股票都买下来，对吧？

Because to keep the price at its current level, you need someone that is buying all of those shares, right?

而且我们看到的情况，可能也反映在今天的股价走势上，就是大家并不认为会有人接盘。

And what we've seen, and maybe this is reflected in the stock action today is people don't believe that that will be the case.

至于他们考虑的是短期还是长期，那就是另一个问题了。

Now, whether they're thinking over a short term period of time, whether they're thinking of a long term period of time is another question.

如果你是我们节目的常客，你应该知道我和Josh通常是从中长期的时间维度来思考的。

If you're common listeners of this show, you will know that Josh and I think on pretty mid to long term time horizon.

所以长期来看，我们对SpaceX还是比较看好的。

So we're kind of bullish in SpaceX over the long term.