BlackRock Weighs $10B Investment in SpaceX IPO, Microsoft’s GitHub Alarm, Genspark COO on AI Apps

欢迎大家收看信息TITV。

Welcome everyone to the Information TITV.

我是阿卡什·帕什里卡。

My name is Akash Pasricha.

今天是5月18日，星期一。

It is Monday, May 18.

接下来这一周内容非常丰富。

We've got an action packed week ahead of us.

有几件事值得你们关注一下。

A couple things to put on your radar.

谷歌正在举办年度开发者大会IO。

Google is hosting its annual IO Developer Conference.

Meta计划启动大规模裁员，涉及约10%的员工；NVIDIA则将在周三收盘后公布第一季度财报。

Meta is scheduled to begin a big round of layoffs that will account for about 10% of its total headcount, and NVIDIA is reporting its first quarter results on Wednesday after the closing bell.

这周对于埃隆·马斯克来说也很关键，SpaceX的星舰预计将在周三发射。

It's also a big week for Elon Musk with expected launch of SpaceX's Starship scheduled now for Wednesday.

如果这次成功，那对SpaceX即将进行的IPO将意义重大。

If that goes well, it would mean big things for the upcoming SpaceX IPO.

当然，这次发射在2026年就已经推迟了至少三次，咱们就看看它到底能不能顺利升空吧。

Of course, the launch has been delayed at least three times in 2026 alone, so we will see if things take off or not.

今天节目中，Information独家报道了微软高管对GitHub的担忧。

On the show today, the information published exclusive reporting about Microsoft executives sounding the alarm over GitHub.

稍后我们会详细聊聊它最近频繁出现的故障问题。

We'll talk about the outages that it's been experiencing shortly.

我们还拿到了关于黑石可能对SpaceXIPO感兴趣的独家消息。

We also have exclusive reporting around BlackRock's potential interest in the SpaceX IPO.

接着我会采访AI平台Genspark的联合创始人。

I'll then talk to the cofounder of AI platform Genspark.

我们会聊聊定价、竞争、模型，当然还有投资回报率。

We'll talk about pricing, competition, models, and of course ROI.

最后，我们会用来自我们生成式AI数据库的独家数据，总结一下这个领域收入集中度的情况。

And lastly, we will wrap the show with some exclusive data from our generative AI database on how concentrated revenue is in the sector.

这期节目会很精彩，咱们直接开始吧。

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

GitHub 给微软带来的问题越来越多了。

GitHub is causing Microsoft more and more problems.

平台上的使用量激增，但问题和宕机也跟着多了起来。

Activity on the platform is surging, but so too are issues and outages.

我请来了我们的微软记者亚伦·霍姆斯，今天他刚发布了一篇独家报道，内容就是这些细节。

I want to bring on Aaron Holmes, our Microsoft reporter, who published an exclusive story with those details today.

亚伦，欢迎再次做客我们的节目。

Aaron, welcome back to the show.

很高兴你今天能来。

It's great to have you here.

几周前你刚来过我们节目。

We had you on the show a couple weeks ago.

你当时提到 GitHub 出现过一些宕机问题。

You mentioned that GitHub had some problems with outages.

你今天发了一篇更深入的报道。

You published a deeper dive on the platform today.

自从我们上次聊过之后，有什么新情况吗？

What's changed since we last spoke to you?

是的。

Yeah.

简单来说，我了解到GitHub最近的宕机问题已经引起了它一些大客户的关注，有些客户甚至直接向微软投诉，并且拿到了因服务中断而产生的退款。

So essentially, I have learned that these outages that GitHub has been suffering are basically resonating with some of its biggest customers, you know, and heard that some of those customers have actually been complaining to Microsoft and getting essentially refunds on some of the time that there were outages.

这已经开始影响到GitHub的营收了。

That's starting to obviously impact some of bottom line at GitHub.

除此之外，我还了解到，GitHub的管理层正在全力应对来自Claude Code和Cursor等新兴工具的激烈竞争。

And on top of that, I think I also just kind of learned more broadly that the the executives who lead GitHub are trying to really push to fend off the rising competition that GitHub Copilot is seeing from Claude code and from Cursor.

目前来看，他们显然在全力以赴提升微软自家的AI编程模型，好让GitHub Copilot能建立起更强的竞争壁垒，对抗OpenAI和Anthropic这样的对手。

And that, at this point, looks like, you know, an all out effort to improve Microsoft's own AI coding models so that GitHub Copilot has more of a moat compared to competitors like OpenAI and Anthropic.

我们稍后会详细聊聊这些方面。

We'll get into all those parts of the story shortly.

我想回过头来看看这些中断的根本原因。

I want to go back to the fundamental cause of the outages.

我们清楚这些中断到底是怎么回事吗？

Do we know why these outages are occurring?

造成这种特殊状况的因素有很多。

There's a lot of different factors that seem to be creating this unique situation.

公司自己提到，GitHub上的流量在增加，这主要是因为人们在使用AI编程助手，它们上传和修改代码的速度和频率远超人类。

So the company itself has pointed to rising traffic on GitHub, which is, you know, being driven by people using AI coding agents who can effectively upload and edit code on GitHub much more rapidly and and more frequently than a human ever could.

但并不是所有流量增长都能带来新收入，因为GitHub对上传代码或用第三方AI助手编辑代码这类基础操作并不收费。

Not all of that increase in traffic is leading to new revenue, though, because GitHub does not charge for basic things like, you know, uploading code or or editing code with your own third party AI agent.

与此同时，我们也知道GitHub正在从谷歌云迁移到Azure，公司表示这次迁移带来了一些问题。

At the same time, we also know that GitHub is moving from, you know, Google Cloud to Azure, and that migration has caused some issues according to the company.

GitHub内部最近也经历了大量人员变动。

There's also just been a lot of churn at GitHub.

很多之前负责管理的高管，要么调到了微软的其他部门，要么干脆离开了公司。

A lot of the the leaders, that were previously leading it have moved to other parts of Microsoft or out of the company entirely.

所以我觉得，这些因素凑在一起，让 GitHub 面临了相当棘手的状况。

So I think that all of these things are kind of coming together to create a somewhat challenging situation for GitHub.

不过公司表示，他们正在对平台的开发和托管方式做根本性的调整，理论上应该能缓解之前频繁出现的宕机问题。

But the company has said that they are making some fundamental changes to how they develop and host the platform that in theory should help with the outages that they've been seeing.

我之前还真不知道，GitHub 居然是微软旗下的，而且这么长时间一直用的是谷歌云？

So I didn't realize this, but but GitHub, is owned by Microsoft, they were actually using Google Cloud this whole time?

他们一直没用 Azure？

They weren't using Azure?

是的，他们一直混合使用多家云服务商。

Yeah, they've used a mix of different cloud providers.

微软 2018 年收购了 GitHub，当时他们本想尽快迁到 Azure，结果实际耗时比公司预想的长得多。

They were acquired by Microsoft in 2018, and originally wanted to get on Azure pretty quickly, but it ended up taking just a lot longer than the company had planned for.

但今年，他们正式下定决心，要把迁移 Azure 的工作彻底完成。

But this year, they are officially trying to make the Azure migration happen fully and once and for all.

对。

Right.

你这个故事里有个挺有意思的地方，亚伦，你提到 GitHub 的领导层现在有点模糊，说不上谁真正掌舵。

So one of the interesting parts about your story, Aaron, is you also talked about who's leading GitHub, or rather, there being sort of a bit of a not a void, but a question mark as to who's actually leading.

现在到底是谁在管那边？

Who's in charge over there?

是的。

Yeah.

目前负责 GitHub 的最高主管是杰·帕里克，微软几年前从 Meta 把他挖过来的，专门负责所有内部开发工具类的业务。

So the most senior executive leading GitHub is Jay Parikh, who Microsoft hired for Meta a couple of years ago and specifically has been leading all of the kind of internal businesses that, that basically sell developer tools.

所以 GitHub 这块自然就交给了帕里克来管。

So GitHub kind of became a natural fit for Preak to take over.

不过去年秋天，前 GitHub 首席执行官托马斯·多姆克离开了微软。

However, last fall, former GitHub CEO, Thomas Domke, left Microsoft.

从那以后，GitHub 就一直没再设专门的 CEO。

And since then, there hasn't been, you know, a dedicated CEO just on GitHub.

另外还有好几位高管也相继离开了。

There's also been some other big departures.

阿卡什·帕什里卡之前负责 GitHub 的产品，或者说是负责包含 GitHub 在内的核心 AI 团队，今年早些时候他离职去当了 Xbox 的 CEO。

Akash Pasricha was leading product at GitHub or or within the core AI org that includes GitHub and then left to become the Xbox CEO earlier this year.

朱莉娅·卢辛之前主管微软的开发者部门，也 oversees GitHub 的工程团队，她刚刚宣布退休。

Julia Lusin, who was leading the developer division at Microsoft, also had some, you know, oversight of engineering at GitHub, she just announced her retirement.

所以现在确实处于一个过渡期，很多原本在 GitHub 担任要职的人，离职后都去了新的岗位。

So it is definitely a period of transition where a lot of the most senior folks who have been leaving GitHub have moved into new roles.

那我想弄清楚，微软现在到底在采取什么措施来应对这些服务中断？毕竟他们的流量大得多。

So then I just want to understand then what Microsoft is doing then to address these outages, because they have a lot more traffic.

我们得提醒一下，GitHub 的核心业务是让开发者上传代码，共享代码仓库，我想是这么个意思。

And again, we should remind folks here, GitHub is in the business of basically people who develop code, they upload their code, and it's a place that they can sort of share access to these code repositories, I think it's called.

现在这个环境里，生成代码变得容易多了。

And so, we are in an environment where it's a lot easier to generate code.

上传的代码越来越多，流量越来越大，服务中断也更频繁，再加上领导层空缺。

There's a lot more code being uploaded, more traffic, more outages, lack of leadership.

GitHub 到底在做什么来解决这些问题？他们跟你怎么说的？

What is GitHub ultimately doing to solve the problem, or what do they tell you they're doing?

是的。

Yeah.

而且问题还不只是宕机。

And it's not even just the outages.

我们还知道，GitHub的利润率最近受到了压力，部分原因是他们对GitHub Copilot采用了按席位收费的模式。

We also know that GitHub's margins have come under pressure recently, in part because were serving GitHub Copilot on a per seat pricing model.

由于越来越多的人在使用AI工具，却没有为此支付更高的费用，导致上个季度的利润率被压缩了。

And the fact that more people were using the AI tools without necessarily paying more to use them, compressed margins last quarter.

所以，他们不仅要设法解决让客户不满的宕机问题，最近还全面调整了定价，基本上提高了GitHub上大多数AI功能的价格。

So not only are they trying to, you know, sort of stop the the outages that have annoyed customers, they also recently overhauled pricing to essentially raise prices for most of the AI features on GitHub.

正如我在报道中提到的，管理层也在探索新的方式，根据第三方AI代理使用平台的量来调整收费。

And as I reported in the story, you know, leaders have also been looking into potential new ways to to scale how much they charge customers with how much third party AI agents, use the platform.

显然，他们不希望以一种进一步激怒客户的方式来做这件事，目前还不清楚他们最终会采取什么方案，但尽快调整定价、避免为平台上这些昂贵的AI工具买单，已经变得非常紧迫。

Obviously, they don't want to do that in a way that would further alienate any customers, and it's not clear exactly what course they're going to take there, but it has become pretty urgent to overhaul pricing to make sure that they're not subsidizing the use of these costly AI tools on their platform.

当你跟GitHub的客户或者平台使用者交谈时，你觉得这件事的利害关系有多大？

When you talk to GitHub customers then, or people using the platform, how high are the stakes here?

因为你的报道里还有一层意思，你提到了GitHub面临的竞争压力越来越大。

Because another layer of your story is you talk about the competition that is mounting against GitHub.

你之前上过这个节目，聊过OpenAI打算开发自己的GitHub替代品。

You've been on the show previously talking about how OpenAI was thinking about developing their own rival to GitHub.

在你的报道中，你还谈到了某些大型AI实验室可能对GitHub构成的威胁。

In your story, you talk about the threats that even some of the larger AI labs could pose to GitHub.

你知道，这种平台是不是客户会说，嘿，我直接换一家算了？

You know, is this a type of platform where customers are saying, Hey, I'm just going to switch.

这对他们来说是很简单的迁移，还是说这种迁移会比较困难、速度也会慢一些？

It's very easy for me to go, or is this kind of a harder migration that might be a bit slower?

我了解到一些中小客户曾跟我提过，他们正在考虑换平台。

I think we've seen some smaller customers that told me that they were considering switching.

另外，HashiCorp 的联合创始人 Mitchell Hashimoto 也说过，他要把自己的开源项目 Ghostty 从 GitHub 搬走。

We also saw Mitchell Hashimoto, the co founder of HashiCorp, say that he was moving his open source project, Ghosty, off of GitHub.

不过，当我跟大客户聊的时候，他们说要是把存在 GitHub 上的代码库搬到 GitLab 这样的竞争对手那里，成本会非常高，而且过程也挺让人头疼的。

However, when I talk to larger customers, they do say it would actually be very costly and kind of annoying to move their code bases that they keep in GitHub into a competitor like GitLab.

大公司自己托管代码仓库其实挺简单的。

It's pretty straightforward for a large company to also host its own code repository.

像Meta这样的公司，早就这么做了好多年了。

That's what companies like Meta, for example, have done for years.

这个选项确实存在，但对大客户来说，现在自己搞一套还是挺麻烦的。

And and I think that that option is there, but currently, you know, doing that would be a bit of a pain for large customers.

目前我还没听说哪家大企业正在主动迁出GitHub。

So I I haven't heard of any large enterprises that are actively moving out of GitHub.

我觉得，迁移的代价高，某种程度上暂时帮GitHub稳住了客户。

I think that to some degree, the switching costs involved with moving away are, you know, keeping GitHub from bleeding customers for now.

所以如果GitHub能解决最近频繁宕机让人头疼的问题，应该还能留住这些大客户。

So I think that, you know, if GitHub is able to sort of staunch the the flow of people being annoyed by recent outages, they are likely gonna be able to hang on to those big customers.

不过，开发者越来越多地用Cursor或者Clawed Code这类工具个人开发，代码可能就不再存到GitHub，而是留在本地或其他地方。

However, you know, the more that developers use tools like cursor or clawed code on an individual basis means they might not be necessarily storing that code in GitHub versus just storing it locally or elsewhere.

这也是GitHub现在面临的一个实实在在的隐患。

That's another real concern that GitHub is facing now.

太好了。

Great.

好的，阿伦，非常感谢你来做客。

Well, Aaron, want to thank you for coming on.

这位是阿伦·霍尔姆斯，我们《信息》杂志的微软记者。

That is Aaron Holmes, our Microsoft reporter, here at The Information.

贝莱德正在考虑对SpaceX的IPO进行大规模投资。

BlackRock is considering a major investment in SpaceX's IPO.

这是根据我们的金融副首席记者科里·温伯格的独家报道。

That is according to exclusive reporting from Cory Weinberg, our deputy bureau chief of finance.

我想请科里来跟我们分享一下他的了解。

I want to bring on Cory to share with us what he knows.

科里，欢迎再次来到节目。

Cory, welcome back to the show.

SpaceX IPO方面又度过了忙碌的一周。

Another big week in SpaceX IPO land.

只要S1文件一公布，我们都会翘首以待。

Whenever the S1 drops, we are eagerly waiting.

关于贝莱德，你打听到什么消息了？

What did you find out about BlackRock?

贝莱德显然是个老牌资产管理公司，投过很多IPO，对SpaceX这事儿特别上心。

BlackRock is obviously a classic, asset manager that invests in a ton of IPOs, and they've been pretty hot on the SpaceX train.

我一直试着联系我的消息源，想搞清楚到底谁是这笔交易的主要买家。

I've been trying to talk to my sources to understand, you know, who are the big buyers of this deal?

SpaceX打算融资七百五十亿美元，这数字最近一直被提到。

SpaceX needs $75,000,000,000, or that's what it's been sort of talking about raising.

这将是历史上规模最大的IPO，至少是第二名的两倍以上。

That's the biggest IPO in history by by sort of at least a factor of two.

对于那些经历过不少IPO的人来说，这种规模的交易里，单笔投资超过十亿美元的案例少之又少。

And from people who've been around IPOs for a while, billion dollar plus investments in a deal like this are rare.

这种级别的投资，整个行业里掰着手指头都能数得过来。

They you you can count them on two hands once over.

我从消息源那里听到的是，如果能拿到份额，贝莱德至少在考虑投入一百亿美元参与这次IPO。

And so what I've heard been hearing from sources is BlackRock has at least been discussing investing as much as $10,000,000,000 in this IPO if they can get their hands on it.

我之前报道的范围是五到一百亿美元。

5 to 10,000,000,000 is the range that I reported.

具体金额可能会随着临近上市时间、股价波动而调整，但关键在于，贝莱德已经全力投入了。

That could change as we get closer to deal time depending on the price, but the upshot is is BlackRock is all in.

那他们为什么这么坚决呢？

And and why why are they all in?

背后的逻辑到底是什么？

What's the calculus here?

从黑石CEO公开谈论马斯克的言论来看，他们的考量是：马斯克在特斯拉上已经一次次证明了回报能力，现在SpaceX又要登陆公开市场了。

The calculus, at least from if you if you talk if you listen to the CEO of BlackRock's public comments on on Elon Musk, the calculus is Elon has delivered returns time and time again with Tesla and and now coming with with SpaceX coming to the public market.

这其实已经说明了一切。

And that, you know, sort of tells the whole story for them.

今年早些时候，拉里·芬克在达沃斯论坛上把马斯克请上主舞台，一开场就对在场的金融精英们说：看看这个人带来的复利回报有多惊人。

Larry Fink brought Elon Musk onto the main stage at Davos earlier this year, and he started his conversation with Musk by telling, you know, the sort of financial elite audience, look at the compounded returns that this man has has brought to bear.

你想想，要是所有养老基金当初都在特斯拉上市时跟着投资，现在回报得多可观。

Imagine if if all the pension funds sort of had invested alongside Tesla when it went public, just how good their returns would be.

所以，黑石的首席执行官在筹备这场大型金融活动时，几乎是在为马斯克站台。

So you have a CEO of BlackRock almost talking up Musk as he's planning this huge financial event.

如果黑石投入五十亿到一千亿美元，这笔钱进去之后，会不会引发连锁反应？

So if BlackRock puts 5,000,000,000, 10,000,000,000, I mean, put money in, would there be ripple effects?

考虑到黑石的分量，这会不会向整个投资界传递出更广泛的信号？还是说，这本质上只是黑石CEO和马斯克之间的私人交易？

I mean, would this send a broader message then to investors given BlackRock's stature, or is this really just a Is this a down Is this a deal between the CEO of BlackRock and Elon Musk?

这笔投资的赌注到底有多高？

How high are the stakes here for this investment?

我接触的一些买方机构人士，也就是投资圈里的人，正在努力弄清楚，哪些是这次IPO新增的主要买家。

I think sources on the buy side, sort of in the investment world that I talk to, are trying to discern who are the big incremental buyers of this IPO.

既然知道黑石要投一大笔钱，这本身就会给市场发出强烈的买入信号，尤其是因为有些业内人士担心，很多大型共同基金其实早就通过私募市场持有了一部分SpaceX股份。

And knowing that sort of BlackRock is in for a big sum does send a strong demand signal to the rest of the market, especially because some bearish commentary that I've heard from sources is sort of this idea of, well, so many of these large mutual fund complexes already own SpaceX in some form from the private markets.

比如富达投资——这种本该积极参与IPO的大型基金——已经持有大量SpaceX股权了。

Fidelity Investments, one of the largest mutual funds that you would expect to invest in an IPO, for instance, already has a humongous SpaceX stake.

没错。

You Right.

聊了点相关的情况。

Talked about some of story.

意思是黑石现在有点落后了。

It was like this idea that BlackRock is almost playing catch up here.

对。

Yeah.

黑石在SpaceX这场游戏中确实落后了。

BlackRock is playing catch up in the SpaceX game.

他们过去八年左右确实私下投过一些，但跟富达比起来，投资规模小得多，跟贝莱德、富兰克林邓普顿这些大型公募转私募投资者相比，也远远不及。

They have invested, you know, privately for I think the last eight years or so, but their stake they haven't invested nearly as much as Fidelity for sure, and then also not as much as some other large public crossover investors like Bailey Gifford and Franklin Templeton.

科里，我想给听众简单介绍一下，等S-1文件一公开，大家都会抢着去看，这些文件内容相当丰富。

Cory, I just want to give people a little bit of a primer here on whenever the S-one does flip and people are going to rush to it, and these documents are sort of quite meaty to get through.

你翻到第一页的时候，最先看到的是什么？

What is the first page, as you're flipping through?

我的意思是，你打开S1文件后，第一眼会看哪里？

I mean, where are you going to go first in the S1?

你觉得最值得留意的部分是什么？

What's the most interesting part that you're going be looking for?

我觉得有几个地方值得关注。

I think there's a few different places.

其中一个，建议大家认真读一读S1里的管理层讨论部分。

One is I think people should actually read the kind of management discussion section that are in these S1s.

这是每个IPO文件里，公司用来阐述自身业务愿景的地方。

This is an area of every IPO document where companies get to explain sort of their vision for the business.

这部分能提供很好的背景参考。

And that's useful framing.

这一段挺长的，全是段落，看着密密麻麻，但真的值得花时间看。

So I wouldn't, it's sort of this long section, there's paragraphs, you know, many paragraphs, and and it can look quite dense, but it's worth worth reading.

我觉得最值得看的，其实是像‘关联方交易’这样的部分，这些内容会揭示马斯克旗下公司、内部人士和SpaceX之间有哪些重大交易。

The juiciest parts, I think, will be sections like related party transactions, which sort of are are supposed to show where have there been significant deals between Musk's companies, between insiders and and SpaceX.

这些地方通常是一些投资者发现交易混杂问题的预警信号。

You know, those are usually some areas where investors will find red flags in terms of sort of intermingling of deals.

另外，大家还会特别关注高管薪酬部分，尤其是因为这是马斯克——我们之前报道过，他有几笔巨额的‘登月式’股权激励，只要他能把SpaceX送到火星，或者往太空发射海量算力，就能拿到更多股权。

And then you also have, I think people are gonna fixate on the sort of compensation, executive compensation section, especially because it's Musk, because as we've reported, he has a couple huge moonshot grants that give him a lot more equity as he if he brings SpaceX to Mars and if he launches an insane amount of compute into space.

我觉得这几个部分是最值得关注的。

So I think those are kind of gonna be a couple of juicy sections.

但总的来说，我发现大家很少完整读这些文件，真该抽出一小时，静下心来好好看一遍。

But overall, I find people don't read these documents enough in full, and, you know, set aside an hour and just, you know.

一小时？

An hour.

你真能在一个小时里看完？

You get through it in an hour?

真的吗？

Really?

嗯，

Well,

如果你能在一小时内读完整个S1文件，那说明内容真的很多。

if you're getting through a whole S1 in an hour, mean, there's a lot.

SpaceX的知识产权部分，我可能得花比这更长的时间，不过……

The SpaceX IP, it might take me a little longer than that, but

第一遍读。

First read.

我觉得吧，对，第一遍读花一个小时是可以的，然后再回头细看。

I think, you know, yeah, you do a first read in an hour, and then, you know, come back to it.

太好了。

Great.

好的，Cory。

Okay, Cory.

非常感谢你来做客。

Well, I want to thank you for coming on.

这位是Cory Weinberg，我们《信息》杂志的金融部副主编。

That is Cory Weinberg, our Deputy Bureau Chief of Finance, here at The Information.

如果你住在纽约市，很可能在地铁里见过 Genspark 的广告。

If you live in New York City, chances are you may have seen subway ads for Genspark.

这是地铁里最响亮的广告活动之一。

It is one of the loudest campaigns on the trains.

今天早上我还真看到了一个。

Saw one this morning, actually.

今天我想请 Genspark 的联合创始人兼首席运营官温桑来分享一下他们公司的战略。

I want to bring on the co founder and COO of Genspark, Wen Sang, to share more about his company strategy.

Wen，欢迎来到节目。

Wen, welcome to the show.

很高兴你能来。

It's great to have you here.

早上好，Akash，很兴奋能来到这里。

Good morning, Akash, and excited to be here.

我看过地铁广告了，对吧？

So I've seen the subway ads, okay?

我看到那个广告里有个地铁线路的示意图，展示的是你从一个想法出发，最终生成一份PPT的全过程。

I see it's, you know, it's got the, there's the mock diagram of the train, and it's emulating the path that you go from a single idea to generate a PowerPoint deck.

所以Genspark就是做AI生成PPT的吗？

Is that what Genspark is all about, generating AI PowerPoint decks?

没错，这确实是我们的一个方向，但远不止这些，Akash。

Well, it is what we're about, but it goes way beyond that, Akash.

Genspark的核心是为全球超过十亿的知识工作者打造AI助手。

So essentially what Genspark is, we build AI agents for the 1,000,000,000 plus knowledge workers.

现在整个世界正经历根本性的转变。

So the world is going through fundamental shifts right now.

过去，我们人类就是劳动力的主体。

In the past, we human beings, we are the workforce.

我们就是推动生产的引擎。

We are the production engine.

但要实现真正的商业目标，我们不得不花大量时间做那些琐碎、重复的杂活。

For us to achieve real business goals, we have to go through the busy work, all the grunt work.

你以前得一个字一个字地写文档，一像素一像素地做演示文稿，一个单元格一个单元格地建财务模型。

You craft documents word by word, you build presentations pixel by pixel, you build financial models sell by sell.

但有了Genspark，你就不用再这么干了。

Well, with Genspark, you no longer have to do that.

每个人都能像贾米·戴蒙一样，直接做高阶决策。

Everybody gets elevated to the level of a Jamie Dimon.

你见过贾米·戴蒙埋头打字做演示文稿吗？

You don't see Jamie typing hard to get presentations done, right?

你只需要一群分析师在你口袋里跑腿，替你把琐碎活儿都办了。

You just have a fleet of analysts running in your pockets, and they run around, they get the busy work done.

Genspark要做的，就是这个。

That's what Genspark is all about.

但贾米·戴蒙是有人替他做这些事的，他有团队帮他处理。

But Jamie Dimon has people doing it for him, He has people doing it for him.

没错。

Yes.

我们把AI助手装进你的口袋，替你处理这些琐事。

We put AI agents in your pocket to do those work for you.

明白了。

Okay.

好的。

Alright.

所以就是那些繁琐的活儿。

So so it's the busy work.

就是，你知道的，表格工作。

It's, you know, it's spreadsheets.

还有各种文档。

It's documents.

以及PowerPoint演示文稿。

It's PowerPoint presentations.

我明白整个套件是怎么运作的了。

I see the whole suite.

我想问的是，我昨天刚用了 Claude。

My question is when I I was just using Claude yesterday.

我是说，你可以把 Claude 装进 Excel 里。

I mean, Claude has You can install Claude into Excel.

而且我觉得 Gemini 现在也有他们自己开发 PPT 的方式。

And I think Gemini now, they have their own way of developing PowerPoint decks as well.

那你们打算怎么跟那些大型 AI 实验室竞争呢？他们很容易就能像你们一样开发出属于自己的应用。

So how is it that you are going to compete with the large AI labs, which very easily can develop their own applications the way you have?

没错，阿卡什，确实如此。

True, true, Akash.

显然，这是大家都认同的方向，对吧？

Clearly, this is the direction everybody believes in, right?

就连微软、谷歌这样的大型科技巨头也是如此。

So goes with even big tech giants like Microsoft, Google, and so on.

所以我们都看到了让AI帮忙处理那些琐碎工作的价值，这样我们人类就能退一步，去思考、去制定战略，发挥创造力。

So we all see the value of asking the help from AI to get all the busy work done for us so that we human beings can take a step back and then think and be strategic and creative.

但 Genspark 的独特之处在于，我们推出了全球首个混合智能体技术架构。

But when it comes to what is so special about Genspark is we released the world's first mixture of agents tech architecture.

这意味着我们集成了 70 多种最先进的模型。

What that means is we sit on top of 70 plus state of the art models.

我们与 OpenAI、Anthropic、Google 以及所有开源权重模型合作，通过编排调度，在合适的时间为合适的任务调用最匹配的模型，从而充分利用这些不同模型的优势。

We work with OpenAI, Anthropic, Google, and all the open weight models, and we orchestrate the, we call on the perfect model for the right task at the right time in order to take advantage of the strength of all of these different models.

因为正如大家所知，这些模型各有所长。

Because as you know, all these models are good at different things.

它们各有优势，重要的是不要偏信某一家，而是能同时调用所有模型来完成任务。

They have different strengths and it is important to not get biased by one of them, but being able to summon all the models at once to get onto your tasks.

所以我们明白，不仅要让个人用户，也要让我们的企业客户能一次性接触到最强大、最新的模型，而不是让他们觉得被锁定在某一家模型供应商那里。

So we understand the importance of allowing not only the individuals, but also our enterprise clients to have access to the greatest and latest models all at once, instead of them feeling that being locked into one model provider.

而且当别人推出新模型时，他们也能立刻用上。

And now when someone else launches a new model, they just can't have access to it.

这就是让我们极具

That is Let extremely

我问你个问题。

me ask you a question.

你们已经取得了相当亮眼的成绩，我得说，一开始我们没提，但你们上一轮估值是12.5亿美元，对吧？

You guys have put up some big numbers, and I should say we didn't mention it at the outset, but I mean, the last valuation you guys got was one point two five billion, I believe.

那是在2025年秋天的事。

This was in the fall of twenty twenty five.

你看着我，是说这轮估值是最新数据吗？

You're looking at me not Was the latest round?

最新一轮其实是16亿美元。

The latest was actually 1,600,000,000.

16亿美元，哦，抱歉，是我记错了。

1,600,000,000, okay, my apologies.

那16亿美元，最近的年收入是多少？

So 1,600,000,000, and the latest revenue figure is what?

我们在十二个月内，从零增长到了2.5亿美元的年收入。

So we grew from zero to $250,000,000 annual run rate within twelve months.

那是在三月底的时候。

That was toward the end of March.

就是三月。

The March.

月底。

End

我想问的是，像你们这样收入增长这么快的公司，你们有多大的把握说这种增长是可持续的？

my question for you is, companies like you that grow their revenue profile so quickly, how confident are you that this is sustainable revenue?

我听说很多公司一开始只是试用，预算也是临时的，短时间内客户数量猛增，但流失风险也很大。

I mean, hear about these experimental budgets, these trial periods, you get a lot of companies quickly, and then there's the churn risk.

你们是怎么应对这个问题的？

What are you doing to prevent that?

你们现在有客户流失的情况吗？

Are you seeing any churn?

和所有公司一样，Akash，每个人都会有一些客户流失。

So just as any company, Akash, everybody, everybody has certain amount of churn.

所以很自然，一开始有人发现了它，后来有些人就决定继续用下去。

So it's only natural that, you know, you get discovered early on and then some people decided they wanted to keep using it.

也有些人觉得这并不太适合他们。

Some people decided this is not necessarily right for them.

但对我们来说，关键在于，增长就像野火一样迅速蔓延。

But for us, what is important we found is that the growth is like a wildfire.

我们才刚刚起步。

We're just getting started.

整个AI革命，我们现在还处在非常早期的阶段。

We're all in a very early stage of this whole AI revolution.

不过我们发现，输出的质量才是最重要的。

But what we found is, you know, output quality is essentially the most important thing.

我的意思是，人类其实挺聪明的。

What I mean by that is we human beings are actually pretty smart.

我们很容易就能判断，自己是不是在浪费算力，教AI却得不到想要的结果。

It's pretty easy for us to tell, am I wasting my time burning tokens, teaching AI and not getting what I want?

或者就一个提示，事情就解决了。

Or it's one prompt, the job is done.

所以我们开发了一套技术，不仅能接入七十多个顶尖模型，还自研了一百五十多种工具，确保大语言模型能拿到真正派得上用场的工具，毕竟它们就像大脑一样。

So we built a technology, not only it has access to 70 plus SOTA models, but we've also built out 150 plus in house tools to ensure that the large language models, they would have the right tools to use in order to do the real world work because LLMs are just like brains.

它们没有手，也没有脚。

They don't have arms and legs.

所以我们给它们装上了手和脚。

We built arms and legs for them.

有了这套系统，用户不仅喜欢我们的产品，还用它赚到了远超我们收费的收入。

With that, we have seen users loving our product and taking advantage of the product, making a lot more money than they pay to us.

说到底，输出质量才是关键。

And essentially, yeah, the output quality is the key.

换句话说，截至三月底的两亿五千万美元年经常性收入，还在持续增长。

In other words, the $250,000,000 end of March in annual recurring revenue, has continued to go up.

你根本没看到任何下滑的迹象。

You haven't seen that dip at all in any way.

我觉得我们才刚刚起步。

I would say we're really just getting started.

你可以这么想，全球有超过十亿的知识工作者，从自由职业者、银行家到分析师，还有大量在企业里工作的人。

Think of it because, I mean, there are over a billion knowledge workers that goes to solo entrepreneurs, bankers, analysts, also, you know, a lot of people working in the enterprise environment.

如今，如果拿AI和互联网的影响力相比，AI还差得远呢。

And today, if you compare the impact of AI versus Internet, it's way not there yet.

它还没到真正爆发的时候。

It's not clutching yet.

你们是怎么应对模型价格上涨这个问题的？

How are you guys dealing with the issue of rising prices for the models?

平均下来，就是70个模型，对吧？

Mean, so 70 models, right?

而且，成本还是很高。

And yet, it's expensive.

费用还在不断上涨。

It's getting more expensive.

我们已经做过这方面的报告了。

We've done reporting on this.

你们是应用层公司，模型收费多少，你们就得照单全收。

And you're an application layer company, you are sort of subject to the prices that the models will charge you.

我猜这肯定对你们的利润率造成了影响。

I imagine this is having an impact on your margins.

不确定你们有没有把这些成本转嫁给客户。

Don't know if you're passing these costs on to your customers.

你笑了。

You're smiling.

我觉得这肯定是你们最关心的问题。

I feel like this must be a top of mind issue for you.

听我说，阿卡什，大家都愿意为Frontier Labs多付钱。

Well, look, Akash, everybody's paying me premium to Frontier Labs.

这没得说。

There's no question.

而Frontier Labs也要向英伟达支付高价。

And the Frontier Labs would pay premium to NVIDIA.

这不过是现实罢了。

That's just the state of the world.

但实际情况是这样，Akash。

But the reality is this, Akash.

你知道，GVD 4.5是个不错的模型。

Know, GVD 4.5 is a fine model.

对吧？

Okay?

它有支持，长期支撑着ChatGVD的运行。

It's supported, it powered chat GVD for a long, long time.

很多人都喜欢它。

People love it.

但现在大家都用5.5了。

Today everybody's on 5.5.

为什么呢？

Why?

而且4.5的成本已经降下来了。

And the cost of 4.5 has went down.

鼻子染了色。

Nose dyed.

但大家现在用的都是全新的模型。

But people are on completely new models.

这是因为所有AI应用公司都在拼命提升性能极限。

That is because all the AI application companies are trying to push to the ceiling, the performance ceiling.

并不是说这个模型就过时了或者没用了。

It's not to say that model is no longer valid or whatnot.

这本质上就像Uber和Lyft刚起步时的早期抢地盘，其他二十家公司也一样，都是这个套路。

So basically this is early land grab, like the early days of Uber and Lyft, so goes with 20 other companies, is the nature of it.

人们都在补贴推理成本。

People are subsidizing inference.

这毫无疑问。

There's no question.

我们为这些前沿实验室支付了高昂的费用。

We are paying premium to all of these Frontier Labs.

不过呢。

When But Yeah.

在你走之前，我再问一个问题。

Let let me ask one more question before you before you go.

当然，请说。

Yes, please.

我其实就想问你这个问题。

Because I I I did wanna ask you this question here.

你们怎么衡量地铁广告的投资回报率？

How do you measure the ROI of Subway ads?

你们的广告简直无处不在。

You guys are literally everywhere.

我的意思是，你们有没有什么数据，比如有多少人看到广告后实际转化了之类的？

I mean, like, do you are there metrics around how many people will see it in conversion or stuff like that?

还是说只是提升品牌知名度？

Is it just brand value?

我真是在哪儿都能看到你们的广告。

Like, I I see your ads everywhere.

你们怎么衡量这些广告的投资回报率？

How do you measure the ROI of these ads?

这是个好问题，阿卡什。

That's a great question, Akash.

我们有一整套广告和营销策略。

We have a full portfolio of advertisement and strategy.

在年收入达到一亿美元之前，我们一分钱都没花在广告上。

So before we hit $100,000,000 at a run rate, we didn't spend a single advertisement dollar.

我们相信，这其实是产品本身带来的口碑。

And then we believe it is product signal.

其实我们在二月买了超级碗的广告， halftime 和第四节都有投放，整整六十秒的黄金时段。

So we bought Super Bowl ads actually back in February, halftime, fourth quarter, we had a whole sixty seconds out of the Super Bowl.

后来我们觉得纽约和东京是知识工作者最集中的两个城市，就想让更多人知道我们。

And then we decided that New York and Tokyo, these two cities are the major hubs of knowledge workers, and we wanted that, you know, people to know about us.

我们知道我们做出来的东西很有价值，很多人喜欢我们，所以希望更多人能注意到我们。

We know we've built something that is valuable and a lot of people love us, and we want to make sure people can see us.

所以我们买了纽约地铁的广告位，因为每天有上千万人乘坐地铁。

So that's why we bought, you know, inventory from the New York subway system, because we know so tens of millions of

那回报率到底怎么样？

What's the ROI though?

回报率到底有多高？

What's the ROI?

至于谷歌和YouTube这些平台的广告，我们很容易就能测算出回报率。

So we measure the ROI a lot more easily with spending on Google and YouTube and so on.

但地铁广告，就像你提到的，主要是提升品牌认知。

But for subway ads, to your point, it's brand awareness.

当有大投资人打电话给我，说愿意给我们这样的公司投几亿美元，还提到‘我看过你们的广告，所以才联系你们’，那一刻才真正感受到。

It is when I get a phone call from a major investor who's willing to put in hundreds of million dollars to a company like us saying that, oh, I've seen your ads, that's why I'm calling you.

这种时刻你就知道，广告确实起到了作用。

Those are the moments that you know that, Hey, this did the job for us.

钱和钱是不一样的。

It's not like a dollar is a dollar.

明白了。

Got it.

好的。

Okay.

文，非常感谢你今天来做客。

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

聊得很精彩，我也很期待看到公司接下来的发展。

It was a great discussion, and I look forward to seeing how things shape up for the company.

这位是Genspark的联合创始人兼首席运营官文桑，欢迎收看TITV。

That is Wen Sang, the co founder and COO of Genspark here on TITV.

Anthropic 和 OpenAI 正在拉大自己与其余 AI 初创公司之间的收入差距。

Anthropic and OpenAI are widening the revenue gap between themselves and the rest of the AI startup field.

我的同事阿米拉·弗拉蒂和斯蒂芬妮·帕拉佐洛在上周末发布的周日洞察专栏中谈到了这一点。

My colleagues, Amira Fratti and Stephanie Palazzolo wrote about that in their Sunday Insights column that published over the weekend.

现在斯蒂芬妮加入我们，来为我们解读她分析的那些数据。

Stephanie joins me now to explain with us the numbers that she crunched.

斯蒂芬妮，欢迎再次做客我们的节目。

Stephanie, welcome back to the show.

跟我们说说你算出的数据，这些数字反映出 AI 初创公司整体收入的什么情况？

Tell us about the data that you calculate and what it tells us about AI startup revenue overall.

没错。

Totally.

这其实是我们在过去几个月、甚至可能快一年的时间里持续追踪的一系列报道的延续，我们一直在跟踪这些原生 AI 初创公司的收入情况。

So this has been a continuation of a series that we've been working on for the past couple months, or maybe at this point up to a year, where we're kind of tracking the revenue of these AI native startups.

这里主要有两点。

So I mean two things here.

先看积极的一面，这些原生AI初创公司的整体收入确实在增长。

Think first on the bright side, we do see kind of overall revenue from these AI native startups growing.

相比六个月前，收入上涨了112%，这显然非常可观。

So it's up 112% from six months ago, which is obviously pretty impressive.

但另一方面，如果你是投资了这些大型AI模型公司的投资人，可能会觉得不太乐观——这笔收入的很大一部分其实只来自两家公司：OpenAI和Anthropic。

I think maybe on the not so bright side, depending if you're an investor who's invested in these big AI model labs or not, a pretty big percentage of that revenue is coming from just two companies, OpenAI and Anthropic.

过去几个月，这两家模型公司的收入占比还在持续上升。

And that share of revenue that's coming from those two model labs has been growing over the past couple months.

现在它们的份额已经达到了大约89%，比六个月前高出4.5个百分点。

So right now it's at, you know, around 89% which is 4.5 percentage points higher than what we saw six months ago.

89%。

So 89%.

不过在谈这个之前，Stephanie，你对《The Information》的生成式AI数据库描述得太简单了。

By the way, before we do that, Stephanie, you are selling the information's generative AI database very short here.

我们追踪AI初创公司的收入已经快三年了，从《The Information Pro》上线就开始了，这一点我很清楚，因为两年前就是我在负责这项数据追踪。

Have been tracking AI startups revenue for not a year, not two years, but more nearly three years now since the launch of the Information Pro, and I know this, of course, because I was responsible for tracking said revenue two years ago.

这个数据库里有大量的数据，绝对值得一看。

So we've got a lot of data in that database that is definitely worth checking out.

不过我想再回过头来聊聊那个89%的数据。

But I want to come back to the 89% figure.

也就是Anthropic、OpenAI，再加上其他所有公司。

So it's Anthropic, OpenAI, and the rest of the sector.

这说明了什么？

I mean, what does this tell us?

是不是意味着这些初创公司根本没机会竞争？

Just that, I mean, these startups are never going to be able to compete?

我们已经看到，Anthropic和OpenAI一推出新功能，往往就把那些小公司整个产品线都给覆盖了。

I mean, we've seen Anthropic and OpenAI, they unveil features that end up sort of encapsulating the entire product portfolios that these smaller startups have.

从这些现象里，你看出AI领域这场竞争有什么更深层的信号吗？

What are sort of the broader signals that this is sending to you about the AI race?

是的。

Yeah.