Why Alphabet Wants $80 Billion for AI, Twitch’s Ad Plan & Self-Aware AI Models

欢迎各位收看《The Information》的TITV节目。

Welcome everyone to The Information's TITV.

我是阿卡什·帕什里卡。

My name is Akash Pasricha.

今天是6月2日，星期二。

It is Tuesday, June 2.

今天的第一条重磅消息是，《The Information》独家披露了亚马逊正越来越依赖外部公司来为其媒体平台销售广告。

First up today, The Information published exclusive reporting about Amazon's increasing reliance on outside firms to sell ads on its media properties.

接下来我们会请我们的亚马逊记者来聊聊这件事。

We'll talk about that with our Amazon reporter.

然后我们会深入分析字母表公司计划发行800亿美元与人工智能投资挂钩的股票。

We'll then dig into Alphabet's plans to issue $80,000,000,000 in equity tied to AI investments.

我们还会梳理一下Anthropic和SpaceX上市的最新进展。

We'll also unpack the latest on the Anthropic and SpaceX IPOs.

今天早上我们还收到消息，Anthropic正在扩大其Project Glasswing计划，让更多机构能用上Mythos。

We also had news this morning that Anthropic is expanding its Project Glasswing program, putting Mythos in the hands of more organizations.

我们接下来会采访前CrowdStrike首席产品官，聊聊这则新闻。

We'll talk with the former Chief Product Officer of CrowdStrike about that news.

节目最后，我们会深入报道Cognition在AI编程竞赛中的新策略，以及为什么评估AI模型正变得越来越难——毕竟这些模型可能已经意识到自己正被监视。

And we'll wrap the show with some reporting on Cognition's new strategy in the AI coding race, and separately, why evaluating AI models could be more difficult given the models might know that they're being watched.

今天这期节目内容非常精彩，咱们马上开始。

It's going to be a great show, so let's get right on into it.

当OpenAI努力打造自己的广告业务时，像亚马逊这样的老牌科技巨头，早已转型为成熟的广告引擎，现在正寻找新方法来加速广告收入的增长。

While OpenAI tries to create its own advertising business, older tech giants who have now become legacy advertising engines like Amazon are finding new ways to keep their own ad sales growing fast.

我们的亚马逊记者佩洛夫今天发表了一篇深度报道，我想请她来跟大家分享她的发现。

Our Amazon reporter, Perloff, wrote a deep dive on that today, and I want to bring her on to share with us what she has found out.

Catherine，欢迎再次做客我们的节目。

Catherine, welcome back to the show.

很高兴你今天能来。

It's great to have you here.

能跟我们说说亚马逊广告业务的最新进展吗？

Give us the update here on Amazon's ad business.

最近怎么样？

How it going?

他们还是赚了很多钱，不过——

Well, they still make a lot of money, but, but-

那他们得继续维持下去。

They gotta keep it going then.

他们得继续维持下去。

They gotta keep it going.

我们报道的那些事，简单说，你一想到亚马逊，可能会想到他们网站上的广告，或者Prime Video，又或者周四晚上的橄榄球赛。

But on the news we reported, so basically, you know, like you think of Amazon, you might think of the ads on their website, you might think of Prime Video, you might think of Thursday Night Football.

他们的广告业务规模很大，而且非常多元。

Their ad kind of empire is pretty big, pretty varied.

不同业务板块的广告销售策略也不一样。

And the strategy for ad sales for some of those different properties is different.

我们报道的是Twitch，他们做的是游戏直播——不过也不只是游戏，虽然大家一提到它就想到游戏。

And so what we reported is that Twitch, which is their video game streaming it's not just video games, so that's what it's known for.

直播。

Live streaming.

直播。

Live streaming.

就是这样。

There you go.

我其实不太用Twitch。

So I'm not a big Twitch user.

看广告。

Watch ads.

对。

Yeah.

但比如说那种

But like that

这正是问题所在。

That's the problem.

我的意思是，虽然有人用，但没法在上面投放广告。

I mean, people use it, but they can't sell ads on it.

问题就在这儿，就是这个。

That's the, mean, that's the that's

关键在于此。

the issue.

对。

Yeah.

所以他们在和广告技术公司合作，帮他们在Twitch上卖广告。

So they're they're working with ad tech, companies to help them, sell ads on Twitch.

他们还跟广告技术公司合作，在两个你可能根本没想到亚马逊旗下的网站上卖广告——Goodreads和IMDb.com。

They've also worked with ad tech companies to sell ads on two sites you might have not even thought about Amazon owning, Goodreads and imdb.com.

他们也找过广告技术公司，想在Alexa上投放广告。

And they've also approached ad tech firms to sell ads on Alexa.

所以我觉得，他们实际上是在把自己定位成一个媒体广告的万能商城，你想买哪种广告，他们都能提供。

So I, you know, I think we're kind of taking a look at the fact that like they kind of position themselves as this everything store for media, where you could buy any type of advertising you want.

他们最近还进一步拓展了业务，开始为其他媒体公司如Spotify和Netflix卖广告，不过面对这么庞大的媒体帝国，有时候他们还是需要一些帮助。

They've even expanded recently more so into selling ads for other media companies like Spotify and Netflix, but sometimes they might need a little bit of help with all with this kind of sprawling media empire.

明白了。

Okay.

所以关键问题是，他们拥有这么多不同的广告渠道——Goodreads、IMDb、Twitch、Prime，还有你提到的Alexa。

So the big issue is so they've got all these different ad properties, Goodreads, IMDb, Twitch, Prime, as you said, Alexa.

整体来看广告收入表现不错，但细分到每个平台，情况就不一样了。

The ad sales in aggregate are doing quite well, but segment by segment, it's a bit of a different story.

那他们是怎么应对这个问题的呢？

How are they addressing this then?

他们有什么计划来推动广告增长，或者让各个板块的表现更均衡一些？

What is their plan here to kick start ad growth, or I guess make it more uniform throughout?

是的。

Yes.

而且说实话，我并不觉得这些其他业务做得不好。

And you know, I mean, I wouldn't say that necessarily these companies are doing these other businesses are doing badly.

主要是Prime和网站上的广告，对广告主来说更值得投。

It's just that Prime and, like, the ads on the website, those are kind of more must buys for advertisers.

但其他一些媒体平台就比较小众了，得看广告主具体想怎么花预算。

Whereas some of these other media properties, they're a bit more niche or depends on, like, you know, what you need to do with your advertising budget.

不过呢。

But yeah.

所以他们开始和外部广告技术公司合作，这一点其实挺值得注意的，因为另外两大广告巨头——谷歌和Meta——都不用外部广告技术公司。

So they are they're bringing these, outside ad tech firms, which, you know, I think is, you know, partially notable because the other two big ad businesses, Google and Meta, do not work with outside ad tech firms.

你根本用不了。

You can't use them.

你在Facebook、Instagram或者YouTube上，根本没法通过外部广告技术平台去买广告。

You know, you can't use outside ad tech to buy ads on Facebook or Instagram or, you know, YouTube.

而他们早就和这些外部广告技术公司建立了合作关系。

And they're used to they're they're partnered with these outside ad tech firms.

另外，Twitch还专门扩充了广告销售团队，这很有帮助，毕竟Twitch的广告模式对广告主来说不太好解释。

Also with Twitch, they've added new, salespeople to their sort of dedicated Twitch ad sales team, which can help because, Twitter is kind of hard to explain to advertisers.

他们根本不知道这是家社交媒体公司。

They don't know it's a social media company.

他们也不清楚这到底算不算一个流媒体平台。

They don't know if it's a streaming service.

他们担心广告会出现在直播里一些尴尬的内容旁边。

They're worried about their ads, you know, showing up against something embarrassing on a livestream.

所以配备更专业的广告销售团队会有帮助。

So having more kind of dedicated ad sales staff can help.

他们实际上投入了更多资源，至少我聊过的一个广告主说，他们把更多时间和推广重点放在了Twitch上。

They've actually devoted more, resources or kind of, like, the time and their pitch, at least to one advertiser I talked to, to Twitch.

他们正试图加强在Twitch上的投入，同时借助其他公司来推动其他一些平台的广告业务。

So they're trying to sort of, like, yeah, beef up some of the efforts, you know, with with Twitch specifically and then with some of these other, properties, you know, relying on sort of other companies to help them

来处理这件事。

with this.

我们知不知道他们具体在和哪些广告技术公司合作？

Do we know which ad tech companies they are working with at all?

这个在故事里没提。

We don't get into that in the story.

所以。

So

明白了。

Okay.

但你说他们用外部公司，比如Meta和Google，其实他们自己并不用外部广告技术公司。

But this idea that they're using outside firms, you said Meta and Google, for example, they don't use outside ad tech firms.

他们坚持用自家的系统。

They stick to their own infrastructure.

对。

Yes.

这里有什么利弊呢？

What are the trade offs here?

为什么一家广告巨头要去依赖外部公司，这会这么重要？

Why is this such a big deal for an ad giant to look to outside firms?

说到底，你就是想把广告卖出去，对吧？

I mean, at the end of the day, you just want to sell the ad, right?

所以不管怎么卖，也不会太糟。

So however you do it, can't be that bad.

我觉得关键在于控制权。

I think, like I mean, the big thing is, like, it's about control.

对吧？

Right?

如果你能自己掌控广告销售，就能决定定价，也能掌控所有伴随而来的数据，比如观众信息。

So if you can control your own ad sales, you can control the price it's sold at, and you can also control all of the data that comes, like, with it, which is like, you know, the audience data.

亚马逊作为广告商最有价值的资产，就是他们掌握的用户购买数据，这些数据正是来自他们在网站上投放的广告。

And I I Amazon's most valuable asset as an advertiser is, like, all the data they have on what people buy, and that's what they get from the ads that run on their website.

所以我觉得，他们短期内不可能把这么重要的资源交给外部公司，这可是他们的核心资产。

So I don't see them giving so much access to other firms, you know, for that anytime soon because that's sort of their crown jewel.

不过，对于其他一些媒体平台来说，这确实是个权衡的问题。

But, yeah, with some these other media properties, it's sort of a calculation.

你知道吧，我们靠广告能赚更多钱，但可能得让出一点控制权。

Well, you know, like, we can make more money on ads and maybe we'll lose a little bit control.

我觉得另一个成本可能是——这个看法因人而异，但他们最近一直在把自己定位成一家独立的大型广告技术公司。

I think the other maybe cost is, you know, and I, you know, maybe different people would see this different ways, but like they've been positioning themselves recently as a big ad tech firm in their own right.

他们一直在跟Netflix、Spotify这些公司合作，宣布各种伙伴关系，想让其他媒体公司把他们当成卖广告的技术后台。

They've been partnering with like, you know, announcing these partnerships with Netflix and Spotify and trying to get like lots of other media companies to be the technical infrastructure to sell ads.

所以如果你连自己卖广告都用不了自己的广告技术，那对其他公司来说，这个吸引力可能就没那么大了。

So if you're if you're saying, hey, I can't even we can't even really use our own ad tech to sell our own ads, it sort of makes the pitch to other companies less compelling for- Potentially.

对。

Yeah.

这其实就是个两难局面。

That's sort of the con.

而且你还会发现，亚马逊这么大一家公司，简直像是一家控股企业，说白了就是——

And then you'd also sort of highlight this idea that like Amazon is such a big company and it kind of almost functions like a holding company where it's like, yeah.

他们可能需要广告技术的帮助，好赚更多钱。

Might want the help of ad tech so you can make more money.

但另一方面，其实还有很多更宏观的考量。

But then, like, you know, there's kind of these broader kind of priorities.

我们在文章里也提到过，多年来他们对Twitch到底该不该用广告技术一直摇摆不定。

And and we talk about this in the article that they vacillated with Twitch over the years of whether they should use ad tech or not.

当初收购的时候，Twitch是用广告技术的，后来他们又给停了。

Like, when they acquired it, Twitch used ad tech and they shut it down.

现在他们又打算重新启用。

Now they're going back to it.

所以我觉得，这其实说明了他们面临很多相互冲突的目标，有时候为了推广，不得不做出取舍。

So I think, yeah, it kinda shows like there's a lot of competing priorities and, you know, you have to sort of sometimes make some trade offs if you wanna sell it.

人工智能会改变这一切吗？

Is is AI gonna change any of this?

这是个挺有意思的问题。

You know, it's an interesting question.

我觉得，AI嘛，说到底，它的目标——尤其是在广告技术领域——我们之前也聊过一点，就是尽量简化媒体购买这个特别复杂的流程。

I think well, AI, like, I guess, you know, to the extent like, think that, the kind of goal for AI, especially in advertising technology, and we've talked about this a little bit before, is to just simplify the meeting media buying process, which is very complicated.

但真正阻碍这种简化的原因是，像Meta、谷歌，还有亚马逊的大部分业务板块，都属于行业里说的‘围墙花园’。

What has been the biggest roadblock though, to that simplification is the fact that most big media companies like Meta, Google, and for most of its properties, Amazon kind of function in industry jargon like a walled garden.

它们不让其他广告技术公司接入自己的媒体资源。

Like they don't let other, ad tech companies access their media.

如果更多公司能像亚马逊那样，把围墙打开，AI就能更容易地整合媒体投放方案，快速在成千上万个渠道上完成购买，而不用一个个登录不同的平台去下单。

So if more companies took the approach like Amazon and sort of opened up their walls, it can make it easier for like an AI Asian, for example, to sort of put together a media plan and quickly buy across a bajillion media channels and, you know, not have to use all these different separate buying platforms to buy different media.

但我认为，AI在广告技术上的潜力，正被很多企业想自己掌控整个广告销售流程的意愿所拖累。

But I do think the kind of potential of AI and advertising technology is sort of handicapped by the fact that a lot of businesses wanna control the whole ad sales process themselves.

你能再解释一下‘围墙花园’是什么意思吗？

So the walled garden thing, just help me understand that.

你是说亚马逊不允许其他第三方广告技术软件购买它的广告位？

So you're saying that Amazon would not allow its ads to be purchased by other third party ad tech software.

你的意思是这样吗？

Is that is that what you're saying?

你是说发布商？

Publisher?

所以这个故事的重点是，他们正在向某些媒体平台开放。

So that's the the point of the story is that they're opening it up to some media property.

对。

Yeah.

但像Prime Video或者他们网站上的广告，这些还是像围墙花园一样，其他广告技术公司没法卖这些广告位，你也无法通过那些第三方平台购买。

But but to to others like Prime Video or like the ads on its website, like those function more like a walled garden in which like other ad tech can't sell that media, and you can't buy it via those other ad techs.

所以我觉得。

So I guess

也就是说，要让AI在这里真正发挥作用，让广告购买变得更简单，像Prime Video这样的媒体平台就得把广告购买流程向其他公司开放，本质上是这样。

that- So for AI to work here at all and make the advertising buying process easier, a media property like Prime Video would have to open up its ad buying process basically to other companies, essentially.

是的。

Yeah.

Meta和Google也是一样。

And I mean, so would Meta, so would Google.

这基本上就是我的看法。

That's kind of my opinion on it.

我觉得，那些在广告技术领域推销各种智能代理解决方案的人可能会说，即使没有AI，也能获得很多好处。

I think, you know, people who are trying to sell lots of different, like, agentic solutions in ad tech might say there's lots to be gained without it.

但我确实觉得，嗯，

But I I do think, like, the yeah.

如今购买媒体资源最复杂的地方，在于各方的商业关系，以及这些媒体公司都想最大限度地掌控局面的既得利益。

The most complicated part of buying media today is the business relationships and the kind of vested interest of all these media companies to have as much control as possible.

所以我不确定AI能不能解决这个问题。

So I don't know if AI can solve that.

很好。

Great.

嗯，没错。

Well, yeah.

AI的兴趣和商业利益并不总是一致的，但咱们得想办法解决。

AI interest and business interests are not always aligned, but I mean, we'll have to figure it out.

谢谢您，Catherine，来参加这次对话。

So thank you, Catherine, for coming on.

这位是凯瑟琳·佩洛夫，我们《信息》杂志的亚马逊记者。

Wanna that is Catherine Perloff, our Amazon reporter here at The Information.

谷歌昨天宣布，将通过股权融资800亿美元，用于投资人工智能基础设施和计算资源。

Google is raising $80,000,000,000 in equity, the company said yesterday, to invest in AI infrastructure and resources for compute.

消息公布后，股价下跌了。

The stock has been down on the news.

接下来我们请来 Mostly Media 的创始人 C.J. 古斯塔夫森，帮我们深入分析一下。

I want to bring on CJ Gustafson, founder of Mostly Media, to help us break it all down.

C.J.，欢迎再次做客我们的节目。

CJ, welcome back to the show.

很高兴你能来。

It's great to have you here.

见到你真好。

Great to see it.

好的。

Okay.

谷歌说他们想通过股权融资八百亿美元。

Dollars 80,000,000,000 in equity is what Google says it wants to raise.

投资者似乎并不买账。

Investors don't seem to like it.

CJ 对这事怎么看？

What does CJ think of it?

如果你退一步看，想买一样东西，比如算力，其实就三种选择，阿卡什。

Well, if you take a step back, if you want to buy something, in this case compute, you really have three options, Akash.

一种是用自己业务赚来的现金。

You can use cash you generate from your business.

谷歌拥有世界上可能也是历史上最强大的现金生成机器——广告业务。

And Google has quite literally the best cash generation machine in the world, maybe in the history of business, which is ads.

这个业务每个季度能产生大约四千六百亿美元的经营性现金流，也就是每三个月一次。

And this produces about $46,000,000,000 in operating cash flow per quarter, so every three months.

但在这四千六百亿里，有三千六百亿又被重新投入了资本支出，用来购买更多算力。

But 36 of that 46 gets plowed back into CapEx to go and buy more compute.

这样算下来，他们每年还有整整一百亿美元的自由现金流，可以随意支配。

And so that leaves 10,000,000,000 in free cash flow that they can do whatever they want with it.

所以，比较成熟的企业可能会分红，或者干脆把钱存着，留着应对不时之需。

So more mature companies may grant dividends, or they may just keep it for a rainy day in their bank account.

下一个选择就是借钱，也就是债务融资。

And so the next option is debt.

但债务不是想借多少就借多少的。

You can only take on so much debt.

他们已经发了很多公司债券，不想把负债权益比拉得太高，否则会影响那些优质债券的信用评级。

They've issued a ton of corporate bonds and don't want to get the debt to equity ratio so high that it actually hurts the rating on those pristine bonds.

第三个选择就是股权融资，也就是卖掉公司的一部分所有权。

And the third is equity or selling a portion of yourself.

这么做会稀释现有股东的持股比例，每个人分到的份额就变小了。

And this dilutes the current shareholders who now own a smaller chunk.

但这么做的初衷是，用这笔钱把整个蛋糕做得更大，让公司整体价值提升。

But the idea is that you're doing it to make the entire pie worth more via whatever you're funding.

所以你看到的，正是资本市场在这三个工具之间的平衡博弈。

So you're just seeing this balancing act in the capital markets between these three levers.

谷歌不想过度依赖债务，但也不想把现金耗尽。

Google doesn't want to overweight the debt portion, but they also don't want to run out of cash.

你的文章每周都从 CFO 的视角来写，受众也是 CFO，对吧？

So, mean, look, you write every week from the perspective you write for an audience of CFOs.

你站在决策者角度，分析他们如何权衡这些选择，你觉得这步棋走得怎么样？

You take the perspective of a CFO as someone who studies the calculus of these decision makers, I mean, you think it's a good move?

我觉得现在他们只能这么做。

I think it's just what they have to do at this point.

得把能用的手段都用上。

You have to use every lever at your disposal.

这种 IPO 之后再大规模增发的做法，以前真没见过，通常都是在公司还没上市、还是私有时才稀释股东权益，对吧？

And it's truly unprecedented how much this is as an add on offering post IPO, because usually you will dilute your shareholders when you're still private, right?

因为第一，私有公司很难大规模借到债。

Because, A, you really can't tap into the debt markets as much.

但我今天早上看了数据，阿卡什，有史以来第二大增发股票的公司是波音，发生在2024年。

But I was looking at the numbers this morning, Akash, and the number two largest follow on equity raise of all time was Boeing back in 2024.

他们筹了大约两千四百亿美元，之前是2008年的美国银行，大家都知道当时发生了什么，那笔钱差不多是两千亿美元，所以这个数字完全碾压了其他所有案例。

They raised about $24,000,000,000 And before that, it was Bank of America, which was in 2008, and we all know why that happened, and that was for just under $20,000,000,000 So this is like a number that dwarfs every other.

所以从资本市场和资源配置的角度来看，这么做是合理的。

So like from a capital markets perspective and like a resource allocation perspective, it makes sense.

但更关键的是，回到根本上，这对公司整体来说真的是正确的决定吗？

It's more just, I think, if you go back to the primaries of it, this the right decision for the company overall?

那你说说，你觉得其他科技公司也会跟着搞股票增发吗？

So tell me about, do you think other tech companies will follow suit here with equity raises?

这其实打开了大门，你看看谷歌现在做的，它的市值已经超过四万亿美元了，对吧？

I mean, it opens the door, because if you look at what Google is doing, their market cap is over $4,000,000,000,000 right?

他们本质上是在用这四万亿美元作为工具，去推动这些大型项目。

They're basically weaponizing that $4,000,000,000,000 to go out and fund these builds.

所以我觉得，只有那些估值特别高、能充分利用自身股票优势的公司才适合这么做。

So I would say only companies that have a really chunky valuation and are able to use their float to do so.

但真正有足够市值支撑这种操作的，其实没几家。

But they're kind of like only one out of couple who have a market cap large enough to sustain this.

股价已经跌过一回了，不过融资消息出来后，还是涨了不少。

And we already saw this stock go down already, but at least they're up a good amount on the air.

对。

Right.

我想问问你，昨天那条大新闻你怎么看？

I want to ask you about the other big news that came yesterday.

Anthropic 私下提交了上市申请。

So Anthropic confidentially filed to go public.

每次看到‘私下提交’这种说法都觉得挺逗的。

It's always funny when they say, We've confidentially filed.

说好是保密的，可这事儿他们自己到处宣传，生怕别人不知道。

The details are confidential, but the filing itself, they came out very loud about it.

从那份公告的措辞来看，他们似乎是想先把流程启动起来，但具体什么时候上市，还没拿定主意。

You know, the language, I guess, in this press release seemed to suggest that they wanted to get the things started, but they're sort of undecided when they actually will go public.

你知道吗，我们联合执行主编马丁·皮尔斯昨晚在他的通讯里提到，有些公司说‘已保密提交’，意思其实是根据之后的市场情况再做决定。

It's you know, our co executive editor, Martin Pierce, made the point last night in his newsletter that some companies, when they say we've confidentially filed, they say, Based on market conditions afterwards, we will make a decision.

但这次他们的说法更像是：我们就想先把流程启动起来。

Here, it was like, Yeah, we just want to get things started.

你觉得Anthropic明年上市的可能性有多大？

How likely do you think it is that Anthropic does go public in the next year?

你觉得这件事在多大程度上取决于SpaceX的上市？

And how much of it do you think depends on the SpaceX IPO?

我觉得它和SpaceX上市没什么直接关系。

I don't think it's truly linked to the SpaceX IPO.

要说有关联，可能更跟OpenAI的上市有关。

I think it would be more to the OpenAI IPO, if anything.

我觉得他们其实更想

I think they'd actually want

赶在他们前面上市

to beat them out of

打开大门，尽可能吸纳更多需求，因为这笔资金总得有个来源。

the gate to soak up as much demand as possible because that capital has to come from somewhere.

资金就像在池子里流动，总会流向最能发挥价值的地方。

Like, it travels in pools, and it's going go to its best use case.

我觉得他们今年年底前上市的可能性很大，毕竟刚完成了9650亿美元的H轮融资。后期公司通常会在IPO前做最后一轮私募融资，这样能给现有投资者一点折扣，相当于说：‘现在你来支持我们，我们不会让你亏。’

And I think it's very likely that they IPO before end of the year, because they just did that Series H at $965,000,000,000 And it's very common for a later stage company to do one final private fundraise right before IPO, because it gives their current investors a bit of a discount to say, Hey, you're underwriting the business now.

我不确定Anthropic在这段时间里会有多少变化。

I don't know how much is going to change between now and then with Anthropic.

也许只有在他们每个季度持续追加两千亿美元的情况下，才会有例外。

Maybe it's the only case if they keep adding, like, 20,000,000,000 each quarter.

你现在进场价格不错，但你也得参与IPO，帮我们把订单量撑起来。

But you're getting in at a good price now, but you also have to get in on the IPO to help us build the book out.

所以这对双方都有好处——那些后期的交叉投资基金，比如龙耳或测高仪，能以低于IPO价的价格拿到份额。

So it helps both parties where the final late stage crossover funds, like a Dragon Ear or an altimeter, they get a nice bite at it at a price that's lower than what they'll IPO at.

一旦上市首日股价大涨，他们早就盈利了，还能轻松覆盖掉这笔巨额资本的业绩提成门槛。

And on day one, if it pops, they're already in the money, and they clear their carry hurdle on this huge amount of capital they're deploying.

这样，Anthropic 就能完成他们必须建立的那部分投资单。

Then Anthropic gets to fulfill part of that book that they have to build out.

所以他们已经走了一半的路了。

And so they're already partway there.

所以，先进行一轮大规模私募，然后直接启动IPO，这很常见吧？

So this is common then, doing a large private funding round and then going straight into the IPO process.

这在这里一点都不奇怪。

That's not an anomaly here.

完全不是。

Not at all.

你回头看看Snowflake或者CrowdStrike，它们都是在IPO前几个月才完成私募融资的。

If you look back at a Snowflake or a CrowdStrike, they all raised private rounds a couple months before they IPO ed.

所以这其实是个成熟的做法，Akash，也能为估值设个底线。

So this is a tried and true method to also, Akash, set a floor on what the valuation will be.

你实际上是在说，这些眼光敏锐的重要投资人都认可了，底线就是965亿美元，对吧？

So what you're doing is you're saying, these important investors who are really smart said that the floor is $9.65, right?

这就是你的起点。

That's your jump off point.

所以我觉得这个数字在上市前非常关键。

And so I think that number is very important heading into the IPO.

说到Anthropic的业务，我们都等着看他们的S-one什么时候发布。

When you think about Anthropic's business here, we're all eager to see the S-one whenever it comes.

不过估计还得等一阵子。

It's still probably a while away.

再说一遍，这还只是保密申报文件。

Again, this is the confidential filing.

但当你想想他们业务的各个部分，我会去琢磨他们现有的各种成本结构和投入方向。

But when you think about all the different parts of the business, I sort of think about all the different cost structures that they have in place, all the different investments they're making.

你觉得他们业务的哪一部分最可能受到关注？

What part of their business do you think is most likely to get scrutinized?

我在想他们那些大型云服务协议。

You know, I'm thinking about these big cloud deals that they structure.

你有什么问题吗？

What questions do you have?

在IPO过程中，你最想知道哪些答案？

Are you eager to get answers to throughout the IPO process?

我想知道，如果他们不受算力限制——就像你刚看到的，他们刚签了那个X AI的单子——如果能自由使用足够的算力，他们到底能赚多少钱？

I want to know what their true capacity is to generate revenue if they weren't so constrained by, like you just saw that that x AI deal they just tapped into, if they weren't constrained by how much compute was out there, like how much revenue could they possibly do?

他们又签了多少长期协议来锁定这些算力资源？

And then how many long term agreements do they have to lock down that compute capacity?

对吧？

Right?

你其实是在衡量两个方面。

You're kind of measuring for two things.

假设一下，他们的引擎理论上能有多大？

Hypothetically, how big could the engine be?

但同时，又能给它加多少油？

But then also, how much gas can you put in it?

我觉得，唯一可能被质疑的就是，你们想达到这么高的收入目标，却没法把油箱加满，缺乏足够的算力来实现它。

And I think that's like the only thing that could possibly be scrutinized, that you are trying to get to these lofty revenue targets, but you're not able to fill up the tank and have enough compute to get there.

换句话说，这周SpaceX披露的信息显示，他们和Anthropic的协议其实延长了，或者更准确地说，他们刚公布这个协议的期限比之前大家以为的要长一些。

In other words, the disclosure that we saw this week from SpaceX, that it sounds like their deal with Anthropic, they've actually extended it, or maybe I mean, they just disclosed that it was a bit longer, I think, than what was previously thought.

听起来你对这个并不意外。

That didn't surprise you, sounds like that.

首先，他们是在转租波士顿的公寓吗？

Well, first of all, are they subleasing a Boston apartment?

我对这么大规模的交易，条款居然这么灵活，实在想不通。

I don't know of terms this fluid for such a large amount.

你这话是什么意思？

What do you mean by that?

一开始传出的消息是，任何一方都可以在九十天内随时终止协议，对吧？

Well, at first it came out that any party can terminate it within ninety days, right?

可这可是每个月12亿美元的承诺投入和支出啊。

So we're talking about $1,200,000,000 in commitment and spend per month.

如果按年计算，那一年的支出就超过一千四百亿美元，而且这还是在九十天内发生的。

If you annualize that, it's over $14,000,000,000 a year, and the same within ninety days.

结果他们又出来澄清，说其实前一百八十天是锁定的，不能随便取消。

So then they come out, people start to panic, and they say, oh, actually, the first one hundred and eighty days are locked in.

哦，原来是这样。

It's like, okay.

但我真不确定这个协议能不能撑满一年。

But I don't even know if this is gonna last for a year.

我们也可以从Anthropic的角度来看，对他们来说这当然是好事。

So, I mean, we could come at this from the Anthropic side, like, great for them.

他们至少锁定了接下来二百七十天的算力资源。

They they locked in some more capacity, I think, at least for the next hundred and eighty plus ninety days.

对吧？

Right?

整整二百七十天。

Like, two seventy days.

但你也可以从 xAI 的角度来想，这真的是收入吗？

But you could also come at it from the x AI, like, is this true revenue?

所以这件事其实两面都有道理。

So it kinda goes both ways.

嗯，我也不确定。

Well and I I don't know.

这让我更清楚地看到，情况真的变动得很厉害，就像你说的那样。

It seems to highlight to me just, like you said, how fluid the situation is.

我们都知道，这一切都取决于需求，而眼下真正的瓶颈是供给不足。

I mean, we know that all of this relies on demand, and then needing demand, and the game right now is we are supply constrained, basically.

我们手头已经积压了大量订单，说白了，只要哪一边情况一变，我们就得赶紧调整，这真让人捏把汗。

We basically have standing orders for And our I guess the calculus is, it might If the situation changes on either side, we got to be ready to adapt, which is scary.

说实话，这种状况去上市，我真想不出还有什么比这更不稳定的了。

I mean, that's Going into an IPO, I can't see how that's the most predictable situation to be in.

一点都不是。

Not at all.

完全不是。

Not at all.

在整个生态系统里，这种情况无处不在。

Throughout the ecosystem, it's pervasive.

不只是Anthropic和XAI的问题。

It's not just Anthropic and XAI.

如果你回溯到Cursor，他们整个计划的基础就是需要更多算力，对吧？

If you trace it back to Cursor, that whole deal is predicated on them needing more capacity, right?

在其他条件都相同的情况下，XAI肯定更愿意用自己的东西、自己的大模型，把数据中心用来做这个。

I'm sure all things equal, XAI would rather be selling their own stuff and their own large language models and using the data centers for that.

我觉得他们建数据中心的目的并不是当房东。

I don't think they built the data centers to be a landlord.

但大家都在抢购多余的算力，说白了就是不惜一切代价，只要能保证运转就行。

But everybody is buying up excess capacity, it's kind of like beg, borrow, and steal whatever you have to do in order to keep these things rolling.

CJ，我最后问你一个问题。

CJ, let me ask you one last question.

当你想到SpaceX、Anthropic和OpenAI的IPO时，大家一直在说，这三家公司都是巨无霸级别的上市。

Here, as you think about the SpaceX IPO, Anthropic, and OpenAI, I mean, the dialogue has been, these are three monstrous IPOs.

市场到底有没有能力消化得了这些新的融资需求？

To what extent does the market have capacity to absorb these new offerings?

这么大的公司，投资者得卖掉什么才能挤进去？

What do people sell to get into these companies, given how big they are?

你觉得潜在的痛点会在哪儿？

I mean, where do you think the pain could come from?

你觉得整体上，这会怎么改变企业的估值倍数结构？

Like, how do you think this overall could change the structure of the multiples that companies have?

这可不只是OpenAI、Anthropic和SpaceX各自孤立的情况。

Mean, this is not just Open Air Anthropic and SpaceX on its isolation.

你背后还有SaaS行业的崩盘在发生，对吧？

You have the SaaSpocalypse happening in the background, right?

还有芯片公司也在里面。

You have the chip company.

你觉得人们为了买这些公司的股票，得卖掉什么？

What do you think people sell to get into these companies?

如果你觉得Salesforce在3倍估值时不错，那1.5倍的时候你肯定会更喜欢。

Akash, if you like Salesforce at 3x, you're going to love it at 1.5x.

资金总得有来源，而且资金都是集中流动的。

Seriously, the capital has to come from somewhere, and the capital travels in pools.

要是把OpenAI、Anthropic和SpaceX都算上，每家融资500亿到1000亿美元，那总共就得有5万亿美元从别处抽出来，对吧？

And so if you take the OpenAI, IPO, Anthropic, and SpaceX, and if you assume they're each raising between 50 and 100,000,000,000 in funding, there has to be 500,000,000,000 to come from somewhere, right?

这些钱会从那些长期前景没那么稳的地方抽走，不管它们短期数据看起来多好看。

And it gets pulled from places that you don't have as much confidence in the long term viability of it, regardless of the short term metrics.

这让我特别担心，因为我一直在看净美元留存率，这其实是衡量收入持久性最好的指标。

And that's what's scary to me, because I was looking at net dollar retention rates, which I think is the best metric to measure durability of revenue.

在SaaS行业，每个细分领域里的留存率都下降了14到15个百分点。

And in SaaS, it's fallen in every quartile by between 1415%.

以前是130%，现在掉到了115%、114%。

So say it used to be 130, it went down to 115%, 114%.

每个季度的降幅都差不多。

And every quartile dropped by about the same amount.

自2022年以来，这些公司之间的差别其实没那么大。

The company since 2022, they haven't segregated all that much.

现在有更多值得信赖、长期增长潜力更大的投资方向，比以前强多了。

It's just there are better uses with more confidence in the long term outlook and ability to grow revenue than what we had before.

说白了，就是有更好的选择出现了。

So it's kind of like there's just a better option coming along.

对那些没有真正AI故事的公司来说，这挺吓人的。

And it's kind of scary for companies that don't have a true AI story.

但如果你相信这个AI故事，就得同时接受这些公司正在烧掉巨额资本的事实。

But if you buy into that AI story, you also have to buy into the incineration of capital that these companies are undergoing.

对，没错。

Right, right.

很好。

Great.

CJ，非常感谢你今天来做客。

Well, CJ, I want to thank you for coming on.

这位是CJ Gustafson，Mostly Media的创始人，欢迎来到TITV。

That is CJ Gustafson, founder of Mostly Media, here on TITV.

Anthropic今天宣布扩大Project Glasswing计划，该计划向部分机构开放其强大的Mythos模型。

Anthropic said today it is expanding Project Glasswing, the program through which it is giving select organizations access to its powerful Mythos model.

我们之前报道过，早期参与Project Glasswing的团队对Mythos模型的强大能力印象深刻，但成本也相当高昂。

We've reported on how expensive but how powerful early participants of Project Glasswing have found Mythos.

为了更深入聊聊这些，以及整个网络安全生态，我们邀请到JetStream Security的联合创始人兼CEO Raj Rajamani。

To talk more about all of that and the cybersecurity ecosystem at large, I want to bring on the co founder and CEO of JetStream Security, Raj Rajamani.

Raj，欢迎来到节目。

Raj, welcome to the show.

很高兴你能来。

It's great to have you here.

谢谢，Akash。

Thanks, Akash.

谢谢邀请我来。

Thanks for having me.

今天我们看到新闻，说Anthropic正在扩大Project Glasswing。

So we saw the news today that Anthropic is expanding Project Glasswing.

你看到这个消息时第一反应是什么？

What was your reaction to that headline?

我觉得Anthropic做得非常出色，不仅在技术上不断突破，营销方式也很到位。

I think Anthropic is doing a terrific job, not just in pushing technology boundaries, but also in terms of how they market it.

他们通过制造一定的专属感和稀缺性，让那些没被选入初始名单的机构都对它产生了强烈兴趣。

So by creating a certain level of exclusivity as well as scarcity, they're driving this form of factor amongst all the others that were not part of the original group that had access.

而且他们慢慢扩大范围，持续让Mythos保持话题热度，也进一步点燃了他们的营销引擎。

And by just slowly expanding it, they just keep Mythos in the news and they're just definitely fanning the flames of their marketing engine as well.

也许吧。

Maybe that's

但你说的这种情况，难道只是市场炒作吗？

But a you scenario, should it's just but market?

我的意思是，Mysthos真没让你觉得害怕吗？

I mean, does Mythos not scare you?

它并不可怕。

It doesn't scare me.

我觉得这只是一个自然的发展。

I think it's just a natural evolution.

过去几年里，每个模型都比前一个更强、能力更出色。

Every model over the last few years have been significantly improved or have greater capabilities than the previous one.

有Mysthos这样的东西，只是时间问题。

So it's just a matter of time that we had something like Mythos.

我敢肯定，未来还会有比Mysthos更厉害的模型出现。

And I'm sure there will be something that comes post Mythos that's even better than Mythos.

对吧？

Right?

所以我觉得这完全没什么不正常的。

So I don't think there is anything unnatural here.

这其实很正常。

This is just to be expected.

明白了。

Okay.

这还挺有意思的，拉吉，说实话，我原本以为Anthropic才是那个理性的一方。

So this is kind of interesting, Raj, because I mean, look, I sort of thought Anthropic was being the adult in the room here.

他们一直说：看，我们搞出了这个东西。

They were saying, Look, we've developed this thing.

太疯狂了。

It's crazy.

你知道吧？

You know?

这真该让你感到害怕。

Like, it should really scare you.

所以我们采取的是谨慎的态度。

And we're taking a cautious approach here.

听起来你是在说，这纯粹是个营销手段。

It sounds like the perspective you're bringing is it's just a marketing strategy.

不，我没说这只是营销。

No, I'm not saying it's just marketing.

我仍然和不少公司的人员保持着联系，

I still have friends with many of the companies that have

超过一家，

more than

我会跟他们交流看法。

one and I compare notes with them.

他们确实提到，这套系统能发现一些以前没注意到的问题，不管是识别出新的漏洞，还是把各种攻击方式巧妙组合起来，突破他们的防御。

And they do say that does find stuff that were not previously obvious, both in terms of identifying new vulnerabilities as well as chaining together new exploits or exploits in a novel way to actually get through their defenses.

所以这确实是真实的，但同时，你也得承认，他们在推这个产品的同时，营销做得也相当到位，正好赶上了上市的节奏。

So it's definitely real, but at the same time, you have to give them some credit for the great marketing that they've been using along at the same time in parallel driving into their IPO.

对。

Right.

说说你的新公司 JetStream 吧。

Tell me about JetStream, your new company.

你之前是 CrowdStrike 的首席产品官，对吧？

You were the former Chief Product Officer of CrowdStrike.

你现在在 JetStream 做什么？为什么不在 CrowdStrike 里做这个？

What are you building at JetStream and why couldn't you build it within CrowdStrike?

我们正在打造一个 AI 治理平台，之所以独立做，是因为我和乔治聊过，他建议我出来单干。

We are building an AI governance platform, and we are building it here because, you know, for various reasons, George and I had a conversation and he suggested that I do it outside.

不过我们得到了他们的支持和合作，包括 Falcon 基金、Georgia 投资者，还有 JetStream 本身。

But we are doing it with their backing and partnership, both Falcon Fund and Georgia Investors and JetStream.

那 AI 治理具体是指什么？

And so AI governance, what does that actually mean?

你们的产品到底是什么？

What is the product here?

这个产品能让你更清楚地掌握企业内部所有类型的 AI 应用，并且更好地控制它们。

The product is something that gives you much better visibility and control over all form factors of AI that are being used within your enterprise.

你看大型企业，阿卡什，会发现一个很有趣的现象：每个员工都成了公民开发者。

When you look at large enterprises, Akash, what you realize is that there are some very interesting trends like every user, every employee turning into a citizen developer.

AI系统带来的内部威胁，远比外部攻击要严重得多。

The insider threat of AI systems is far greater than external threats.

而且我们正在失去组织内的经验积累，因为自动化推进的速度，远远超过了我们记录和整理的能力。

And we are also losing institutional wisdom because the pace of automation increase is actually far greater than what we are able to document.

这些如果不解决，就会在企业想部署的AI系统和它们需要管理的治理框架之间，制造出巨大的信任鸿沟。

So these are things that unless addressed, creates a real trust gap between the systems that we are trying to deploy and the enterprises that are trying to govern them.

我们正在打造一个AI治理平台，就是为了填补这个鸿沟。

We are trying to fill that gap by building an AI governance platform.

抱歉，拉吉，我理解得对吗？你是说，内部威胁——比如我自己作为开发者开发的应用，可能出问题或者被攻击——这种风险，比外部黑客入侵系统的可能性要大得多？

Sorry, if I understand you correctly, Raj, is what you're saying that the threat from inside, meaning that if I'm a developer and I am creating these applications myself, the threat that I develop something that could go wrong or is susceptible to some kind of an attack, that threat is greater than the probability that you get an outside attacker coming in and hacking the system.

我的理解没错吧？

Is that what I'm understanding?

没错。

Yes.