Meta Raises CapEx up to $145B, Microsoft Copilot Sales Up 33%, Elon Musk Battles OpenAI Lawyer

欢迎各位收看《信息时报》TI电视节目。

Welcome everyone to the Information's TI TV.

我是肯·布朗。

My name is Ken Brown.

今天是4月30日，星期四。

It is Thursday, April 30.

今天为大家准备了一场精彩的节目。

We have a great show lined up for you.

我们会邀请分析师聊聊科技巨头的财报，包括微软、Meta、谷歌和亚马逊。

We're talking to analysts about the big tech earnings, Microsoft, Meta, Alphabet, and Amazon.

我们还会连线我们的AI记者罗基特·德鲁，他正在现场报道马斯克与OpenAI的庭审。

We'll also check-in with our AI reporter, Rocket Drew, who is covering the Musk OpenAI trial.

此外，《信息时报》独家发布了关于星链收入的深度报道。

Plus, the information published exclusive reporting about Starlink revenue.

接下来我们将邀请我们的马斯克特派记者来深入解读这件事。

We'll be joined by our Elon Musk reporter to discuss that.

节目最后，我们会看看英伟达芯片短缺的情况，以及AI开发者如何利用他们手头现有的显卡。

We'll wrap up the show with a look at the NVIDIA chip crunch and how AI developers are utilizing GPUs that they do have.

这期节目会很精彩，咱们马上开始。

It's going to be a fun show, so let's get into it.

周三下午，多家科技巨头发布了财报，包括谷歌、亚马逊和Meta。

We had a slew of big tech earnings Wednesday afternoon, including Alphabet, Amazon, and Meta.

接下来，我们会请杰富瑞的高级分析师布伦特·希尔为大家详细解读。

We're going to break it all down with Brent Thill, a senior analyst at Jefferies.

布伦特，我们先说说Meta，他们把2026年的资本支出预测上调到了最高1450亿美元，这让投资者有些担忧。

Brent, let's start with Meta, raising its twenty twenty six CapEx forecast to as much as $145,000,000,000 That appeared to worry investors.

股价因此大跌。

The stock's way down.

你对这个数字怎么看？

What did you make of the number?

确实，大家担心的是他们对下个季度广告业务的增长指引——预计增长会达到百分之二十八九，而后续指引又回落到二十出头。

Yeah, the worry is really the guide they gave on the ad business for next quarter growing high, high 20s and the guide into the mid 20s.

至于你提到的资本支出，他们上调了指引，但最终回报会在哪里呢？

And then to your point on the CapEx, the higher guide and ultimately, where's the return gonna come?

我觉得Meta之前那波AI投资确实提升了Facebook、Instagram和WhatsApp这些核心产品的表现。

I think we saw the first wave of AI investments in Meta improve the core properties of Facebook, Instagram, WhatsApp.

但投资者现在对下一阶段的投资持怀疑态度。

But I think investors are questioning this next leg of investments.

智能眼镜能带来回报吗？

Are the glasses going to pay off?

这些对核心产品的新AI投入，真能推动业务再上一个台阶吗？

Are the new AI investments in some of the core properties going to really take this to the next level?

还是说扎克伯格只是在砸钱搞些根本赚不回成本的项目？

Or is this Zuckerberg, you know, wasting money on projects that aren't going to produce the type of ROI.

所以当你把Meta和亚马逊、谷歌、微软放在一起比，看看他们AI投入的规模，再对比那些已经显现的订单和回报，你会发现Meta被抛售，其实是因为其他领域表现太亮眼、太受追捧了。

And so I think when you match Meta up against Amazon, Google, Microsoft, and you see the magnitude of their AI investments and then the associated bookings and return we're starting to see, this is also being sold off because of the relative outperformance and excitement that is going on in other areas of tax.

我觉得这些数字本身其实没那么差。

So I think the numbers itself weren't that bad.

只是跟整个板块的其他公司相比，无论是存储芯片、半导体，还是谷歌刚才公布的强劲积压订单数据，这些都在短期内削弱了投资者对Meta的信心。

It's just when you put it relatively to the rest of the group, whether it's memory, semis, what Google just said in their blowout backlog number, it basically weakened the meta case for investors in the short term.

明白了。

Got it.

我觉得马克·扎克伯格说他们没有明确的变现计划，这话确实不太妥当。

I think Mark Zuckerberg didn't do himself any favors when he said they don't have a precise plan for monetization.

这听起来确实让一些投资者感到不安。

That seemed to freak Yeah, a few I'd never people go

话虽如此。

out of saying that.

所以呢？

Was So

他们赚钱的路径是什么？

what is the path for them to make money?

你在他们的投资中会关注哪些迹象，来判断事情是否正在往好的方向发展？

What do you look for in their investments where you're going to see signs that things are working?

我觉得已经有一些迹象表明他们在进步了。

Well, I think there is signs that are working.

他们的广告业务是我们所覆盖的公司里增长最快的。

I mean, they have the fastest growing ad business that we cover.

他们的增长速度超过了谷歌。

They grew faster than Google.

也超过了亚马逊。

They grew faster than Amazon.

亚马逊的增长才刚到百分之二十出头。

Amazon grew low 20s.

谷歌在广告业务上确实做得不错，但增长最快的是Meta，所以我觉得他们执行得挺到位的。

And obviously Google's executing well on their ad business, but the growth was the highest at Meta, so I'd say they are executing.

我觉得投资者因为科技圈其他热点太多，已经基本忽略了这个故事。

I think investors have just lost effectively focus on this story given all the other excitement that's going on in the rest of our coverage of tech.

所以这可能是原因之一。

And so I think that's part of it.

我觉得你提到的第二点是，大家都看Meta，想知道下一个重大回报会从哪里来。

And I think the second part to your point is I think everyone looks at Meta and says like, what is the next big return that can happen?

会是来自广告收入的增长吗？

Is this going to come in ad driven revenue?

还是说他们会开源一个AI模型，然后卖给企业？

Is this going to happen where they can open source an AI model and sell it to enterprises?

我们之前也说过，目前还不清楚下一个收入引擎到底是什么。

I think we've said like, I think it's unclear exactly what the next revenue engine is.

根据首席信息官们的意见，他们非常希望看到Meta开源自己的AI模型，并推出配套的封装产品。

Our belief based on CIOs is that they would love to see Meta open source and launch their own wrapper, their own AI model.

纽约的一些金融服务公司已经告诉我们，他们非常愿意采用Meta的技术。

There's financial services companies in New York that have said to us that they would love to embrace Meta.

但他们现在还没有这方面的业务。

They don't have that business today.

长远来看，这会不会成为他们的一个收入来源呢？

Is that something that they could do as a revenue engine over time?

答案是肯定的。

The answer is yes.

可穿戴AI会怎么发展呢？

What's gonna happen with wearable AI?

我们可能算是个异类，但真的很看好他们在眼镜上的布局。

You know, we're big fans of what we might be an outlier, but we're fans of what they're doing with the glasses.

我觉得他们还有其他几个项目。

I think there's a handful of other initiatives.

不过我觉得这些项目都没谷歌云、AWS，或者微软旗下那些业务来得那么大、那么贴近眼前。

I just don't think they're quite as big and as in front of us as what you see with Google Cloud, AWS, what's happening in Microsoft with some of their properties.

所以我觉得这事儿还得再往后推，真正的投资回报率现在依然很模糊。

So I think this, again, is just, it's further down the road, and ultimately, the real ROI is still really opaque.

当扎克伯格说‘相信我们’的时候，大家心里都想：我们确实信你，当初你在广告业务上兑现了承诺，但接下来那个能复制这种成功的产品到底在哪？

And when Zuck says, you know, trust us, I think everyone is saying, well, you know, we trust you in the first wave, and it's showing up in the ad business, but ultimately, what's the next property that's gonna do this?

大家对这个问题其实都有点拿不准。

And I think everyone's kind of scratching their head a bit on that.

没错。

Right.

那咱们接着说点好消息吧。

Well, let's move on to the good news here.

咱们先聊聊谷歌。

So let's talk about Google for a minute.

你印象最深的是什么？

So what stood out to you most?

他们公布了不少亮眼的数据，但最让你惊讶的是哪一点？

There was a lot of nice numbers there, but what was the thing that really jumped out at you?

最让我意外的是谷歌云的增长提速了。

Yeah, what jumped out is the acceleration of Google Cloud.

增长率达到70%，而Azure接近40%，AWS也才28%左右。

70% growth versus Azure growing close to 40, and AWS growing close to 28.

他们的增长速度真是惊人。

So their growth rates are incredible.

第二点是市场份额的提升。

Second is market share gains.

他们连续两个季度在云服务上超过了亚马逊和微软。

So they gained market share for two quarters in a row against Amazon and Microsoft.

我觉得他们的未完成订单量简直惊人。

I think the backlog was mind blowing.

从第四季度到第一季度，他们的云业务未完成订单增加了2190亿美元，同比增幅高达400%。

I mean, the type of cloud backlog, they added $219,000,000,000 of backlog sequentially from Q4 to Q1, And their cloud backlog is up 400% year over year.

这些数字真的太夸张了。

I mean, these numbers are insane.

他们现在还开始向第三方出售TPU芯片。

And part of what they're doing now is selling TPUs to third parties.

其实大部分都是他们在向企业销售自己的AI基础设施。

Most of this is actually their AI infrastructure they're selling to corporations.

我觉得，从数据中心到应用，再到Gemini的这种端到端整合模式，现在特别对路。

So I think what's happening is this kind of vertically integrated model from, you know, the data center all the way to the apps into Gemini is really resonating.

我们一再提到，每一份CIO调查都显示，谷歌的势头越来越强；硅谷的高管们也都在说，他们公司用Gemini的频率越来越高，用谷歌基础设施的次数也越来越多。

And we have said this repeatedly that every CIO survey, Google's gaining steam, every independent conversation with executives in Silicon Valley, that they're all using Gemini more and more, their corporations are using their infrastructure more.

更让人惊讶的是，不只是谷歌云，他们的广告业务也表现惊人——搜索查询量创下历史新高，搜索收入增长接近20%。

And what's even wild to see beyond just the Google Cloud is that in their ad business, search queries were at an all time high and their search revenue grew close to 20%.

大家都以为AI会终结搜索，可他们告诉我，搜索量反而达到了前所未有的高度。

So while everyone thought AI was going to kill search, they're telling me queries are at an all time high.

我觉得，谷歌股价这波涨了8%，背后有好几个原因，今年它可以说是互联网板块表现最好的股票之一。

So I think this, you know, there are several factors that are driving Google up 8%, And it's one of the best performing stocks in internet this year.

这背后有多方面的原因。

And it's for several reasons.

一个是订阅业务。

It's a subscription business.

一个是广告业务。

It's the ad business.

一个是谷歌云的积压订单。

It's the Google Cloud backlog.

我觉得去年大家还觉得OpenAI在第一季度能彻底压倒谷歌，现在情况完全反过来了。

And I think last year, everyone thought OpenAI in Q1 was going to crush Google, and now it's reversed.

现在大家都觉得谷歌走在了前面。

Everyone thinks Google now is leading.

华尔街的情绪也因此发生了逆转。

And so the sentiment on Wall Street got twisted.

我们觉得，谷歌真正的实力和底层驱动力正在逐渐显现。

And we think that kind of the true power and the underlying engine of Google is becoming apparent.

是的。

Yeah.

不好意思。

Sorry.

这里的情况其实已经明朗了。

It's really untwisted here.

我想先跳过亚马逊，因为我们还有很多内容要聊。

I to want to Amazon just because we have a lot to cover.

那你总体来看，最让你印象深刻的是什么？

And so give me your top level take on what stood out there.

总的来说，AWS的收入符合预期，占28%，但他们的待执行订单情况和谷歌很像。

Yeah, top level, AWS was in line at 28%, but their backlog is, again, similar to Google.

他们的待执行订单简直爆了。

They're blowing the doors off on the backlog.

他们和Anthropic签了一笔一千亿美元的订单。

They signed a $100,000,000,000 deal with Anthropic.

他们的整体待执行订单同比将增长超过100%。

Their overall backlog will grow over 100% year over year.

亚马逊云业务的利润率也提升了。

Their Amazon AWS cloud margin improved.

大家本来以为，AI时代利润率肯定会下滑。

Everyone thought in the world of AI, margins were gonna be down.

但他们的利润率反而上升了。

Their margins are actually up.

他们的广告业务增长了二十出头，表现非常不错。

Their advertising business grew in the low 20s, which was really good.

至于消费业务，完全没有出现疲软的迹象。

And then the consumer business, there was no signals that the consumer soft.

他们的消费业务收入和利润率都超出了预期。

They had consumer revenue upside and margin upside.

这是近四年来，亚马逊首次实现15%的增长。

So this is the first time that we have seen Amazon grow 15% in basically almost four years.

按固定汇率计算，增长了15%。

So constant currency growth of 15%.

他们的利润率达到了13%，高于指引区间的上限，这表明在贾西的领导下，他不仅关注收入增长，也重视利润率。

And then they posted a 13% margin, which is one point above the high end of the guide and has showed under Jassy that he cares not only about revenue growth, but about margin.

正如我之前说的，让一位软件背景的CEO来掌管亚马逊很重要，因为我过去长期跟踪软件行业，知道这类高管通常都重视高利润率，而贾西显然也是如此。

And as I've said, I think having a software CEO in place running Amazon is important because in my history covering software, you know, executives care about high margins, I think chassis really cares.

所以这背后还是源于纪律性、运营效率，以及AWS的出色表现。

So I think it goes back to discipline, operational efficiency, AWS doing phenomenally well.

包括Anthropic在内的递延收入达到4640亿美元，增长了146%。

The RPO, including Anthropic, will be $464,000,000,000 or growing 146%.

这些数字简直太惊人了。

And these numbers are incredible.

和谷歌类似，订单 backlog 很强，说明AI真的在落地，这些AI公司都在往基础设施上砸钱。

And so similar to Google, great backlog, commitment that AI is working and these AI companies are going to their infrastructure.

哇，我们聊了这么多内容。

Wow, we covered a lot.

谢谢，刚才的是杰富瑞的高级分析师布伦特·希尔。

Thanks, That was Brent Thill, senior analyst at Jefferies.

微软第一季度收入增长了18%，比上个季度有所加快。

Microsoft revenue grew 18% in the first quarter, an acceleration from the prior quarter.

公司还首次公布了AI业务的总收入，同比暴涨了123%。

And the company also disclosed total AI revenue for the first time ever, reporting it was 123% higher compared to the same period last year.

现在邀请到RBC资本市场的软件股权研究主管里希·贾卢里亚，和我们一起来聊聊这些业绩。

Joining me now to discuss the results is Rishi Jaluria, Managing Director of Software Equity Research at RBC Capital Markets.

早上好。

Hey, good morning.

非常感谢您邀请我。

Thanks so much for having me.

不客气。

Pleasure.

对于微软本季度的表现，您最大的感受是什么？

So what's your biggest takeaway for the quarter with Microsoft?

是的，抛开股价表现不谈，我觉得微软在核心云业务、人工智能以及整体产品线上都在持续稳步执行。

Yeah, look, putting the stock action aside, I think what it's telling me is Microsoft is continuing to execute well on the core cloud, on AI, on their broader portfolio.

他们正在为长远发展打基础，比如 M365 Copilot 和 GitHub Copilot 这样的产品。

They're setting themselves up for success in the long term with things like, M365 Copilot, GitHub Copilot.

他们明显是在下长远的棋。

They're, they're definitely, I think playing the long game here.

我认为，就我所覆盖的公司而言，微软是人工智能领域布局最全面的公司之一，因为他们的产品线覆盖了整个技术栈，能从多个层面受益于 AI。

And I think this is one of the best positioned companies, at least in my coverage universe for AI because of the broad portfolio and the number of ways they can benefit from AI throughout the stack.

我对这些数字非常满意。

So I was, you know, very pleased with the numbers.

看到Azure增长加速，而且他们还提到接下来几个季度还会继续提速，这真让人振奋。

It was great to see Azure accelerate, for them to talk about further acceleration the coming quarter and then the two quarters after that as well.

我觉得这特别有希望。

That I think is very encouraging to see.

对，没错。

Yeah, exactly.

市场对这家公司的看法没你这么乐观，但咱们走着瞧吧。

The market is not as upbeat as you are about the company, but we'll see.

微软披露了超过90亿美元的AI收入。

Microsoft disclosed more than $9,000,000,000 in AI revenue.

这个数字有多重要呢？

How meaningful is that number?

好吧。

Okay.

我觉得这个数字意义重大。

I think it's incredibly meaningful.

而且，顺便说一句，这个定义可能还比较保守，因为过去很长一段时间，微软压根没把训练OpenAI模型的收入算进财报里，以后说不定会有变化。

And it's probably, by the way, a conservative definition, because if you recall Microsoft, for the longest time, and there's gonna be maybe changes with this, had not been recognizing revenue from training of OpenAI models in that, in in in anywhere in the P and L.

所以这部分其实也没算进去。

So that excludes a lot of that as well.

因此，这个数字很可能已经很保守了。

So it's probably a pretty conservative number.

现在主要还是来自OpenAI，可能还有一点Anthropic的贡献？

Now is it mostly OpenAI and now maybe a little bit of Anthropic?

是的。

Yes.

但这里面也包括了微软自研的AI服务。

But then there's zero and first party AI services in there.

比如GitHub Copilot，还有365 Hub Copilot。

There's, GitHub Copilot, F36 Hub Copilot.

我觉得微软在AI领域的强劲发展真的很棒，而且未来只会变得更好。

And I think it's great to see, you know, this really strong trajectory on the AI side for Microsoft, but I think that will only improve from here.

那具体到AI这块，从他们公布的数据能看清楚是什么在推动增长吗？还是说因素太分散了？

So on AI side specifically, so can we tell from what they disclosed exactly what's driving the AI growth or is it too many pieces here?

是的，因素太多了。

Yeah, it's too many pieces.

我们只能自己猜一猜，做个粗略的估算，但很明显，很大一部分增长都来自OpenAI的扩张。

We can do our own kind of guessing and do some napkin math and you know, guaranteed a lot of that is gonna be a function of OpenAI's growth.

这一点我们还是能大致推断出来的。

And we can kind of draw a little bit of a line to that.

不过他们也公布了，本季度有两千万用户在使用365 Copilot，比上个季度的一千五百万增加了五百万。

But look, they did also disclose that they had 20,000,000 m three six five Copilot users in the quarter, and that was up 5,000,000 from 15,000,000 a quarter ago.

我们或许可以对每个用户的平均收入做个估算，看看能推算出多少贡献。

And, you know, we could probably make some assumptions around ARPU there and how does that translate.

所以这些地方我们也能做个粗略的计算。

So we could do some napkin math in there.

要准确判断太难了，毕竟影响因素太多了。

It's hard to do it with any high degree of certainty, just given all the moving pieces here.

那个Copilot的数字挺让我在意的。

That Copilot number was interesting to me.

这反映出企业对AI的需求是什么情况？

What does that tell you about enterprise demand?

是的，我觉得这说明企业级AI的应用还处在非常早期的阶段。

Yeah, look, I think what it's telling us is we're still in the very early stages of enterprise AI usage.

这和我们外面做的调研也是一致的。

And this checks out with all the work that we've done out there.

我们看到AI热度一直很高，但说实话，现在企业里绝大多数AI应用——撇开AI实验室和谷歌、Meta那些不算——其实都还只是试点项目。

You know, for all that we're seeing out of, you know, AI interest and it's clearly, unyielding, but at the same time, most AI usage today in the enterprise, right, putting aside the AI labs and, you know, Google, Meta, etcetera, Putting that aside, like most of the AI usage today is a lot of pilots.

很多都是原型验证。

It's a lot of proof of concepts.

真正上线使用的，主要集中在软件开发、编程这类工作，还有内部和外部的客户服务。

And when it is deployed live in organizations, it's being used for software development and coding and that kind of stuff, as well as customer support, both internal and external.

但目前为止，AI还没有真正大规模渗透到知识型员工、生产力工具，以及销售和营销这些领域。

But it really hasn't made its way to the domains of the knowledge worker, to productivity, to sales and marketing in a major way yet.

我们当然都看到过一些非常出色的早期案例，但要说在企业中广泛普及，还远没达到。不过我觉得，这说明微软很可能是这个方向上的领头羊。

We obviously all see really great early success stories of that, but in terms of widespread enterprise adoption, it's not there yet, but I think it's telling us that Microsoft is probably one of the companies that's leading the charge on that.

明白了。

Got it.

这确实挺有意思。

So that's interesting.

我们都知道现在AI主要用在编程这些方面。

So I mean, we know about the coding and all that and where the use is now.

但这直接关系到公司的增长，对吧？

This gets right to the growth of the company, right?

等它真正开始影响知识型员工时，你认为未来几个季度，我们会看到哪些关键信号，表明 adoption 开始加速？

When it starts to hit knowledge workers, what do you think what are you looking for in the next few quarters for that that increased adoption?

是的。

Yeah.

我觉得我们最想看到的有三点。

Look, I think what we'd want to see is really three things.

第一点，从数据上就能看出来，就是付费的 M365 Copilot 用户在持续健康增长。

Number one, which, you know, we can see in the numbers is continued healthy growth in paid M365 Copilot users.

不过这里会有点复杂，因为我们现在开始看到 Microsoft 365 E7 的推出，它已经整合了 Copilot。

Now this is gonna get a little muddied because we're now seeing the introduction of Microsoft e seven, which includes Copilot.

实际上，它明天就正式上线了。

That goes GA tomorrow, actually.

我对目前看到的进展非常看好。

And I'm really encouraged by by what we're seeing in there.

但归根结底，M365 E7 里使用 AI 的用户数量，增长势头是不是依然强劲？

But but ultimately, like, is the amount of AI users within that m three x eight five Copilot growing at a healthy clip?

第二点，这种增长有没有反映在整体产品线的业绩提升上？

Number two, does this show up as overall acceleration on the offers line?

因为这才是最终的关键。

Because that's ultimately what matters.

你可不希望出现拆东墙补西墙的情况，搞得像在搞资源分配的游戏。

You don't wanna, you know, robbing Peter to pay Paul type of situation where it's a game of allocation.

你希望这能带动整体增长提速。

You want this to lead to overall growth acceleration.

所以我们希望看到类似这样的趋势。

So we'd like to see something like that.

第三点，虽然这不是财务指标，但我们很希望看到用户对Office Copilot技术本身的反馈越来越好。

And the third, which is now a nonfinancial metric, but would love to see, you know, improving customer feedback to the actual technology around Office Copilot.

大家都知道，初期用户反馈其实挺两极的，有好有坏，但最近确实有改善。

I think no secret that initial customer feedback was was pretty, you know, mixed to negative, and it has improved.

就在过去六个月里，我就看到它一直在往上走。

Even over the past six months, I've seen it tick up.

我很期待看到，随着E7的发布，还有像Critique和Counsel这类利用多模态技术的解决方案上线后，用户反馈会怎么发展。

I'm excited to see how customer feedback with the launch of e seven and solutions like critique and counsel that leverage multimodality, how those start to actually play out.

如果我们看到用户反馈持续提升，那我会更坚信，这些功能最终能推动使用量，进而为微软带来收入。

If we start to see improved customer feedback, that makes me feel even more bullish that they will be able to drive usage and ultimately that drives revenue for Microsoft.

对。

Right.

不，我们也注意到了，

No, yeah, we saw it too,

普遍的不满情绪。

general unhappiness.

昨晚那堆财报出来，想必让所有分析师都挺高兴的。

So we had this pile of earnings last night, which was I'm sure a joy for all the analysts out there.

微软跟竞争对手相比处于什么位置？

Where does Microsoft sit versus the competitors?

今天各家股票走势都不一样。

The stocks are all going in different directions today.

是的。

Yeah.

我觉得，所有主要竞争对手都在用不同的方式切入这个市场，对吧？

Look, I think that all of the major competitors have different angles by which they're attacking this, right?

我的同事布伦特·埃里克森在跟踪互联网公司方面做得特别出色。

And think, and okay, you know, my colleague, Brent Erickson does a fantastic job covering the the Internet guys.

我主要关注甲骨文。

I cover Oracle.

我们俩加起来，基本覆盖了所有主要的混合型技术领域。

So between us, we all cover the major hybrid skills.

但我觉得让微软处于独特位置的，首先是他们产品线的广度，我觉得这些都能从人工智能中受益。

But what I would say, you know, puts Microsoft in a very unique position is number one, the breadth of their portfolio that I think can benefit from AI.

对吧？

Right?

你当然看到Meta在谈怎么用人工智能提升用户参与度和广告效果。

I mean, you obviously see Meta talking about how they're using AI to benefit user engagement and advertising.

谷歌也在说，人工智能能怎么改变搜索体验。

Google's obviously talking about, you know, advertising and what can this do for search.

所以每个人都是从不同的角度在切入这个领域。

So So everyone's attacking it from a different angle.

甲骨文更偏向于一种纯粹的底层基础设施提供商。

Oracle definitely much more of a, I would say, more raw infrastructure provider.

微软则不同，它从基础设施起步，还延伸到了开发者层面和数据层面。

For Microsoft, it it starts at infrastructure, but it goes into the developer layer, goes into the data layer.

像微软Fabric和Cosmos DB这些工具，我们已经看到了不错的早期成果。

Like, we've seen some pretty good early success with tools like Microsoft Fabric and and Cosmos DB.

现在它又进入了生产力层，比如我们刚才提到的M265 Copilot。

It's going into the productivity layer as we're just talking over at, m two six five Copilot.

再往后，我觉得它会渗透到安全层面，比如安全版Copilot。

At some point, I think it starts to flow into, you know, the security layer with security Copilot.

甚至一些常被忽视的资产，比如游戏或领英，也能从AI中获得巨大收益。

And I can even make a case that, assets that get very ignored, gaming or LinkedIn can see major benefits from AI.

事实上，斯特凡妮刚才就提到领英已经从中获得了不少实际好处。

In fact, Stephanie was talking up some of the benefits from AI that LinkedIn is already seeing today.

所以我认为，微软的独特优势在于，它能利用多模态能力，结合与OpenAI早期的紧密合作，把AI全面融入整个微软产品体系。

So that's what I think ports Microsoft is a in a unique position is leveraging the multi model nature, leveraging the close early relationship they had with OpenAI, and then bringing out that throughout the entire stack of Microsoft suite.

我认为这让他在这个后AI时代占据了非常独特的位置。

I think that puts them in a very unique position in this post AI world.

这确实很有意思。

That is really interesting.

非常感谢你，里希。

So thank you so much, Rishi.

这位是来自RBC资本市场公司的里希·贾卢里亚。

That was Rishi Jaluria from RBC Capital Markets.

罗杰特·德鲁一直在法庭上全程报道埃隆·马斯克与OpenAI的庭审。

The information's Rocket Drew is covering the Elon Musk OpenAI trial from the courtroom all week long.

这是他为我们带来的关于马斯克交叉质询的最新进展。

Here's his latest update for us on Elon Musk's cross examination.

埃隆·马斯克在OpenAI律师的交叉质询中遭遇了严峻考验。

Elon Musk had a brutal cross examination session from OpenAI's lawyer.

马斯克已经连续第二天出庭，但这次轮到OpenAI方面对他进行提问。

Musk was on the stand for the second day in a row, but this time it was OpenAI's turn to question him.

他们抓住这个机会，向马斯克提出了许多尖锐的问题。

And they took that opportunity to ask some very hard hitting questions of Musk.

事实上，两人之间还出现了几次激烈的交锋。

In fact, there were some heated moments between the two.

马斯克指责律师提问具有误导性，问题太过模糊。

Musk accused the lawyer of asking misleading questions, asking questions that were too vague.

律师则说马斯克不配合，一再需要翻出旧文件，甚至他本人之前的证词记录，来提醒他之前已经回答过这些问题，或者至少帮他回忆起来。

The lawyer said Musk was being uncooperative and repeatedly had to pull up old documents or even the transcript from Musk's own deposition to point out that Musk had answered questions before or to at least jog Musk's memory.

这对OpenAI一方至关重要，因为这是他们第一次有机会让马斯克正面回应他们一直强调的观点——他们认为马斯克是个心怀不满的联合创始人，只要自己能掌控公司或技术，他就乐见OpenAI以盈利方式运作。

Now this was really crucial for OpenAI's side because it was their first chance to get Musk to speak to some of the points that OpenAI has been making all along, that they see Musk as a disgruntled cofounder, someone who was always content to have OpenAI operate in certain for profit ways as long as Musk was in control of the company or was going to be in control of the technology.

事实上，马斯克曾协助设立OpenAI的营利性子公司，甚至提议让OpenAI依附特斯拉，把它当作提款机。

And in fact, Musk helped set up a for profit subsidiary for OpenAI and went so far as to suggest that OpenAI could attach to Tesla as its cash cow.

所以OpenAI想表明，马斯克根本觉得离开后OpenAI就会完蛋，觉得如果没有他，OpenAI根本不可能成功。而正是因为OpenAI如今这么成功，他才心生怨恨，才提起这场诉讼。

So OpenAI showed that Musk basically thought he was leaving OpenAI for dead, that it didn't stand a chance without him, and if OpenAI had never been such a success story, then Elon wouldn't have been so resentful and wouldn't have brought this lawsuit in the first place.

这就是他们描绘出的马斯克的形象。

So that's the portrait that they painted of Musk.

另一方面，这种尖锐的质询也让马斯克有机会检验并完善他在这起案件中的立场。

Now on the other hand, this brutal questioning was an opportunity for Musk to stress test the refined theory, the ways he's refined his own position in the case.

比如，他已经找到了回答‘你为什么不早点起诉’这个问题的方式。

So for example, he's present he's he's found a way to answer the question, why didn't you bring this lawsuit sooner?

之前有人质疑，诉讼时效是不是已经过了。

Before there was a question about whether the statute of limitations had expired.

他说，我对OpenAI的怀疑是逐步加深的。

And he says, well, I only became suspicious of OpenAI in degrees.

这个过程分了好几个阶段。

There were phases to it.

没错，2017年我就有些疑虑了，但那时候只是担心未来可能违背承诺。

And yeah, sure, I had some suspicions back in 2017, but there were suspicions that a promise might be broken in the future.

直到几年后，我才确信承诺真的被打破了，或者至少我开始这么认为。

It wasn't until years later that a promise was actually broken or I came to believe that that was the case.

这就是他如何逐步完善自己论点的方式。

So that's a way that he's refined his case.

不过，经过这轮质询后，马斯克的形象并不太好。

Still, Musk came out of the questioning session not looking so good.

比如，在整个诉讼过程中，马斯克一直强调AI安全的重要性，说OpenAI在AI安全上掉链子，把利润放在了安全前面。

For example, Musk has had a lot to say about AI safety during the case, that he thinks OpenAI is dropping the ball on AI safety, that they've been deprioritizing it and instead prioritizing profits.

但当OpenAI的律师问他一些关于AI安全行业最佳实践的基本问题时，他却完全答不上来。

But when OpenAI's lawyer asked Musk some very basic questions about the industry's best practices around AI safety, Musk was clueless.

他们问他什么是安全卡或系统卡——这些通常是新模型发布时附带的文件，用来说明模型经过了哪些测试。

He was asked about safety cards or system cards, which are the papers and documents that often accompany a new model and explain the testing that went into that model.

但他根本不知道这些东西是什么。

And he didn't know what that even was.

同样，他还被问到OpenAI的准备框架，这是指导OpenAI大量安全工作的关键文件。

And similarly, he was asked about OpenAI's preparedness framework, which was which is a key safety document that governs a lot of OpenAI safety work.

马斯克还说，他从来没听说过这个框架，也没听过OpenAI其他任何相关规程。

Musk also said he had never heard of it or other protocols that OpenAI has in place.

所以，马斯克在这场质询中表现不佳，但他也借机澄清了一些立场，还解释了为什么他创立xAI时选择做营利性公司——这个问题一直笼罩着整个案件。

So Musk comes out of this not looking so hot, but he did have a chance to advance some of his positions, and he was also able to justify why he founded xAI as a for profit company, which has been an awkward question hovering over this whole case.

如果运营AI作为非营利组织真的这么重要，那为什么马斯克在2023年创立自己的公司时，没有直接成立一个非营利机构呢？

If operating know, if running AI as a nonprofit is so important, why didn't Musk found a new nonprofit when he came when it came time to found his own, you know, organization in 2023?

他对这个问题的回答是，非营利或营利都可以，但你必须选一个。

And his answer to that is he's okay with it being a nonprofit or a for profit, but you have to pick one.

在他看来，OpenAI这是想两头占便宜。

And the way he puts it, OpenAI was trying to have its cake and eat it too.

我们可以期待马斯克在这些话题上继续发表更多看法。

So we should expect to hear more from Musk on these topics.

他还要接受第二轮交叉质询。

He's due for a second round of cross examination.

马斯克之后，他的助手贾里德·伯奇尔预计将出庭作证。

After Musk, his associate, Jared Birchall, is expected to take the stand.

接下来是加州大学伯克利分校的计算机科学教授斯图尔特·罗素，最后是格雷格·布罗克曼。

After him is Stuart Russell, the UC Berkeley computer science professor, and finally, Greg Brockman.

所有这些，还有更多内容，都将在马斯克诉……的庭审中陆续展开。

So all of that and more is coming up in the Musk v.

阿尔特曼审判。

Altman trial.

《The Information》独家报道了SpaceX星链服务的单用户收入与其不断增长的用户基数之间的关系。

The Information published exclusive reporting about SpaceX's Starlink revenue per user relative to its growing customer base.

我的同事泰奥·韦特和瓦莉塔·鲍威尔撰写了这篇报道。

My colleagues, Theo Wayt and Valita Powell wrote the story.

现在泰奥加入我们，来分享他了解到的情况。

Theo joins me now to share what he knows.

嗨，泰奥。

Theo, what's the Hi, Theo.

最主要的发现是什么？

What's the big takeaway?

星链用户数量在增长，但每个用户的收入却在下降。

So Starlink is growing, but as it's growing, it's it's making less money from each user.

我们根据机密文件报道，2024年SpaceX星链的单用户收入是99美元，而去年是81美元。

So we we reported, based on confidential filings that, SpaceX, SpaceX Starlink's revenue per user was $99 in $2,114 in 2024, and $81 last year.

同一时期，用户数量增长了四倍。

Over the same time, the number of customers quadrupled.

他们赚的总钱数多了，但每个用户的收入却变少了。

So they're making a lot more money, but they're making less money from each customer.

那为什么会这样呢？

So, so why is that?

为什么每用户平均收入在下降？

Why is the ARPU going down?

Starlink 以前是面向富裕人群的高端服务，主要用户是住在小木屋或船上的人，但现在它已经转向大众市场，在美国郊区，价格有时已经能和康卡斯特或AT&T这样的传统有线网络竞争了。

So, Starlink has kind of gone from this niche service for rich people that with cabins and boats and things being the average user to a more, you know, mass market service that, you know, in suburban parts of The US, you know, is now competitive sometimes with traditional wired Internet from, you know, Comcast or AT and T.

但为了实现这一转变，Starlink 不得不大幅降价。

But to get there, Starlink has had to slash prices pretty aggressively.

几年前，美国最便宜的套餐每月大概120美元，现在最低只要50美元。

A couple years ago, the cheapest plan in The US was something like a $120 a month, and now the cheapest is is $50.

这彻底改变了整个业务模式。

So it's it's changed the the business quite significantly.

你之前写过摩根士丹利对星链2024年收入的预测。

You wrote about Morgan Stanley's estimates for Starlink revenue back in 2024.

他们预测得太高了。

They missed the mark.

他们在这个问题上到底错在哪里？

What did they get wrong about this story?

他们当时真的过于乐观了。

So they were honestly insanely bullish.

这根本说不通。

It doesn't really make sense.

当时其实已经有其他分析师觉得这不太合理，但摩根士丹利却认为星链每个用户每月能带来超过170美元的收入，对。

I mean, there were other analysts at the time even that didn't think it made sense, but they Morgan Stanley thought that Starlink was bringing in more than a $170 per month per user Mhmm.

那时候根本不是这样，现在也不是。

Which just wasn't true at the time and and isn't true.

现在倒是有可能了。

It is unless true now.

从他们的角度看，他们只是觉得Starlink的普通用户都特别富裕。

And I guess from their perspective, they just thought the average Starlink customer was a lot more affluent.

不知怎么的，他们还觉得这种情况会一直持续下去。

And for some reason, they thought that that would continue to be the case.

但在美国，Starlink正像我说的那样，转向大众市场了。

But, you know, in The US, Starlink is trying to go, like I said, to this mass market offering.

而在国际上，它正在进入那些人们根本没想到要花这么多钱上网的国家。

And internationally, it's expanding in countries where just people don't expect to pay that kind of money ever.

比如在欧洲，网络服务本来就便宜得多。

Like, you know, Internet service in Europe is cheaper.

更别说它还在不断进入非洲和东南亚的更多国家。

Not to mention they're, you know, entering more countries in Africa, Southeast Asia.

那些地区的消费者收入有限，对价格的预期也完全不同。

Like, consumers in those areas just have less money to spend and have different expectations.

所以我真不明白，摩根士丹利怎么会觉得这情况能一直不变。

So I honestly don't understand why Morgan Stanley thought that it would stay the same.

这根本说不通。

It doesn't make sense.

所以看起来价格压力已经出现了，而且照你这么说，情况还会更糟，对吧？

So seems like we've seen pricing pressure and it seems from your story like it's going to get worse, right?

今年晚些时候会有一个竞争对手的卫星服务上线。

There's a rival satellite service coming on later this year.

那这个情况现在怎么样？

What's what's happening with that?

亚马逊的Leo计划今年晚些时候推出。

So Amazon Leo is supposed to launch later this year.

亚马逊还没具体说终端设备或服务收费多少，但他们明确表示，这会是Starlink的直接竞品，价格上一定会很有竞争力。

Amazon hasn't detailed exactly how much they're gonna charge for the terminals or for service, But, you know, they're they're framing it as a competitor to Starlink that, you know, will be just competitive on price.

亚马逊在渠道分销上优势很大。

And Amazon has a big advantage in distribution.

你知道的，他们有Prime。

You know, they have prime.

他们可以把这个和Prime捆绑在一起。

They can bundle it with that.

他们有自己的电商网站，只要在首页放个Amazon Leo的推广，每分钟就有上百万人看到。

They have their ecommerce site where if you just put a Amazon Leo, you know, splash on the homepage, a million people are gonna see it every minute.

所以如果亚马逊能推出一个和Starlink差不多的服务，比如速度还行、体验也过得去，那对SpaceX的压力就会很大。

So if, you know, if Amazon can offer a service that's competitive with Starlink, you know, technically, like if if the speeds are okay and and the service is okay, it's it'll lead to a lot of pressure on SpaceX.

但这还只是个大大的假设，因为服务还没上线呢。

But that's still a big if because the service has yet to launch.

咱们聊聊Starlink的用户群体吧，个人用户和企业用户。

So let's talk about Starlink's customer base, individuals versus enterprise.

这两者有什么不同？现在趋势怎么样？

So how are they different and how is that evolving?

个人用户这部分，包括自己买终端装在家里的，还有那些用标准订阅方案的小企业。

So the individual business is both, you know, people that buy terminals and and put them on their house and smaller businesses that have their own, you know, standardized subscription plans.

这部分占了Starlink用户的大头。

And that's the majority of Starlink's customer base.

根据我们看到的数据，占比超过百分之六十，而且增长得更快。

It's over 60% according to the numbers we saw, and it's also faster growing.

这跟AT&T或者康卡斯特这样的公司差不多。

And that's, you know, comparable to, you know, AT and T or Comcast.

就是个人用户订阅，都是按一种通用的标准化方式来管理的。

Like, it's it's just individual subscriptions that are managed in kind of a, you know, generalized standardized way.

然后是企业客户，比如美国政府、外国政府、军方、航空公司、航运公司，这些大客户都跟SpaceX单独谈了合约，由他们来管理。

And then there's the enterprises, which are The US Government, foreign governments, the military, airlines, shipping companies, like all of the gigantic customers out there who have, you know, individually negotiated deals that are manages that are managed by SpaceX.

这也是个不小的生意，但规模没那么大，增长也没那么快。

That's also a pretty big business, but it is not as big and not growing as fast.

所以Starlink本质上还是个以个人用户为主的业务。

So Starlink is still kind of a primarily a consumer consumer business.

接下来有个小规模的IPO，可能让SpaceX估值达到一万亿美元。

So, we have this little IPO coming, potentially a trillion dollars for SpaceX.

这种单用户收入下滑的趋势，对IPO意味着什么？投资者该担心吗？

How does this declining revenue per user trend fit into the IPO, and is this something that investors should be nervous about?

我觉得，要说让投资者担心的事，这一条其实挺正常的。

You know, I think out of all the things that should make investors nervous or might, this is pretty normal.

这种公司越做越大、效率越来越高，又拓展到那些人均消费能力较低的国家，这种事根本不算意外。

Like, you know, as a business like this gets more efficient, gets bigger, expands in countries, you know, internationally where people have less money to spend, like, you know, this is this is not totally shocking.

收入其实还在涨，因为服务增长得特别快。

And, like, revenue is is going up because the service is growing quite quickly.

要说SpaceX里有什么值得担心的，这一条真算不上。

Like, out of all the things at SpaceX that you could, you know, potentially find concerning, this is, like, not really one of them.

顺便说一句，SpaceX自己也说，Starlink的效率在不断提升。

And, you know, for what it's worth, you know, SpaceX says Starlink is getting more efficient.

他们要发射更大的卫星，能服务更多用户。

They're gonna launch bigger satellites that can serve more people.

如果从普通商业角度看，这其实没什么奇怪的，但显然，它的估值跟其他卫星公司或电信企业的估值完全不在一个量级。

And so, you know, if you're thinking about it as a standard business, it's it's not that unusual, but obviously the valuation is totally out of proportion to how other satellite companies and telecoms are valued.

所以，关键看你拿它跟谁比。

So depends on what you're comparing it to.

Starlink上市最有趣的地方在于，它才是真正的盈利引擎，对吧？

The interesting thing about the IPO with Starlink is Starlink is like the profit driver, right?

这家公司里，大家眼下最关注的盈利点就是Starlink。

It is this business in there that people are looking at for, at least in the near term, the profits.

所以，是不是大家对它的关注有点过度了？

So, I mean, is there an outsized attention paid to this?

我觉得是的。

I think so.

他们正试图拓展这个业务板块，投入大量资金发展Starlink手机服务，这是一种全新的服务，用另一组卫星直接向手机传输信号，而不是通过地面终端。

Think that it's a segment that they're trying to grow, and they're putting a lot of money into Starlink mobile, which is kind of a separate service that uses a different set of satellites to beam service directly to cell phones rather than terminals.

我觉得未来几年，这项业务肯定会受到极大关注，因为SpaceX为了获取相关频谱，几乎花了200亿美元。

And, you know, I think that will get a lot of scrutiny in the next few years because SpaceX is spending almost $20,000,000,000 on buying spectrum for that.

但目前来看，它带来的收入微乎其微。

But as of now, it brings in very, very little revenue.

所以，任何能推动Starlink持续增长的举措，对这家公司都至关重要。

So think anything that they do to keep Starlink growing will be quite important to the company.

谢谢，西奥。

Thanks, Theo.

刚才那是西奥·韦特，我们《信息》杂志负责报道埃隆·马斯克的记者。

That was Theo Wayt, our Elon Musk reporter here at The Information.

现在，NVIDIA的服务器芯片供不应求。

Right now, there's a big scramble for NVIDIA server chips.

但《信息》的最新报道显示，即使AI开发者拿到了GPU，也不一定能充分发挥这些昂贵硬件的性能。

But recent reporting from The Information shows that even when AI developers have GPUs, they might not be getting the most out of that expensive hardware.

我的同事斯蒂芬妮·帕拉佐洛和阿尼萨·加尔迪齐在我们的《AI动态》专栏里写了这篇文章。

My colleagues Stephanie Palazzolo and Anissa Gardizzi wrote about this in our AI Agenda column.

现在请斯蒂芬妮来和我分享一下她的发现。

Steph joins me now to share what she knows.

斯蒂芬妮，先说说，开发者一般怎么判断自己的GPU有没有充分利用？

Steph, to start with, how do developers judge whether they're getting enough performance out of their GPUs?

这个问题可以从几个角度来理解。

So there's a couple of ways to think about this.