SAP Blocks OpenClaw, Anthropic Talks With UK Chip Startup, Investors to Sell Billions for SpaceX IPO

欢迎各位收看《信息时报》。

Welcome everyone to the Informations TITV.

我是阿卡什·帕什里卡。

My name is Akash Pasricha.

今天是五月四日，星期一。

It is Monday, May 4.

今天节目的开场，我们带来一条新报道：软件公司SAP正试图阻止客户使用外部AI代理访问其应用中的数据。

We are kicking off the show today with new reporting, the information published about software firm SAP's efforts to block customers from using external AI agents to access data on their apps.

接下来，我们会请一位我们的AI记者，聊聊他们上周末独家披露的内幕。

We'll then bring on one of our AI reporters to discuss a scoop that they broke over the weekend.

Anthropic正在与一家英国初创公司洽谈，计划采购AI推理芯片。

Anthropic is in talks to buy AI inference chips from a UK startup.

随后我们会讨论，基金经理们正计划腾出资金，准备在埃隆·马斯克的SpaceX公司期待已久的上市时抢购股票。

We'll then discuss how fund managers are planning to raise cash to grab shares of Elon Musk's SpaceX when the company makes its eagerly anticipated debut.

最后，我们会与Gradient的管理合伙人聊聊，风险投资圈如何看待今年预计将密集出现的这些大型IPO。

And we will wrap the show in conversation with a managing partner at Gradient discussing how the venture capital world is thinking about all these big IPOs expected this year.

这期节目肯定会很精彩，咱们直接开聊吧。

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

企业软件公司正面临代理程序大规模抓取数据的挑战，数据保卫战愈演愈烈。

The corporate data wars are raging on as enterprise software companies grapple with agents pulling data on mass from their systems.

我的同事凯文·麦克拉林和劳拉·布拉顿报道了SAP为保护自家系统所采取的惊人举措。

My colleagues, Kevin McLaughlin and Laura Bratton wrote about that the dramatic lengths that SAP is going to protect its own house.

我请凯文来跟大家详细说说他了解到的情况。

I want to bring on Kevin to share more about what he's learned.

凯文，欢迎再次做客我们的节目。

Kevin, welcome back to the show.

很高兴你能来。

It's great to have you here.

谢谢邀请我。

Thanks for having me.

凯文，SAP那边到底发生了什么？

So what is going on at SAP, Kevin?

上个月，SAP发布了一份关于访问其API的政策文件。

So last month, SAP published a document, which was a policy document for accessing its application programming interfaces.

看起来这可能是不小心泄露的，因为他们根本没有配套的说明或解释。

It looked like it might have been an accident actually because they didn't really have much messaging behind it.

但这份文件的核心意思是：如果你是SAP的客户，想把SAP的数据连接到外部AI代理，就必须通过一个听起来像是SAP审批的流程。

But basically, the document says, if you are a customer of SAP and you wanna connect your SAP data with outside agents, you will have to do it through what sounds like an SAP approval process.

这个流程可能是某个特定产品，也可能是一种授权机制，但文件里的措辞实际上标志着我们在所谓的AI数据战或企业数据战中，局势进一步升级了。

It could be like a specific product or it could be some sort of authorization, but the language in the document basically amounts to an escalation in what we've been calling the AI data wars or the corporate data wars.

这在企业数据战里，确实是我们之前没见过的。

So this is a I mean, is a like we haven't seen before in the corporate data wars.

我们以前听过收过路费的情况，对吧？

We've heard about tolls, right?

但现在不一样了，你得主动申请我们的许可才行。

But this is, you actually have to seek our permission.

是的，到目前为止，我们看到的情况几乎正好是一年前，Salesforce旗下的Slack团队修改了API条款，让像Glean这样的公司更难用他们的Salesforce数据做企业搜索。

Yeah, so what we've seen so far, it's been almost exactly a year since we broke the story about Salesforce's Slack unit changing its API terms in a way that made it harder for companies like Glean to do enterprise search on their Salesforce data.

从那以后，我们还看到其他厂商采取了类似限流或API速率限制这样的措施，这些也常用于安全目的。

And since then, we've seen other measures taken by vendors like throttling or what's known as API rate limits, which are also used for security purposes.

但说实话，还没哪家公司像SAP这样，准备走到这一步。

But, yeah, no one has gone as far as what SAP appears to be ready to do.

这里头其实有几个细节值得注意。

And there's a couple of like nuances here.

一是SAP表示，不会向客户收取通过第三方访问数据的费用。

One is that SAP says it's not going to charge customers for accessing their data through third parties.

所以看起来，SAP会去找那些客户也在用的其他软件公司，说：咱们得重新谈现有协议，或者在AI时代签新合同。

So it sounds like it's going to be a situation where SAP goes to other software providers that its customers also work with and says, we're going to need to renegotiate our existing agreements or form new agreements for this new world of AI.

明白了。

Okay.

这确实挺有意思。

That's interesting.

那咱们具体聊聊实际操作层面。

So let's get a little tactical here.

你听到哪些公司在和他们进行这类讨论？

Mean, companies are you hearing that they are having these discussions with?

我们有没有听说过任何具体的客户对话？

Do we know of any specific customer conversations?

目前这一切都还很模糊。

It's all very murky at this point.

就像我说的，我觉得他们可能是不小心把这份政策文件发出来了。

And like I said, I've it feels like they might have accidentally posted this policy document.

在和SAP的交流中，他们提到已经和微软、谷歌以及亚马逊云服务这些公司有协议，允许客户自由使用他们的数据。

In conversations with SAP, they mentioned that they already have agreements with companies like Microsoft and Google and Amazon Web Services that allow customers to use their data freely.

我猜接下来我们会看到更多类似的协议，对象是其他软件供应商，而大多数这些公司其实SAP早就已经在合作了。

And it sounds like my guess would be that we're gonna see more of these types of agreements with other kind of software providers, many of whom I mean, most of whom SAP already works with.

所以我的直觉是，这些协议必须把AI代理纳入考量，而这一点在过去的合作里是不存在的。

So my my guess or my gut feeling is that the agreements are gonna have to take into account agents, which wasn't part of the existing agreements before.

SAP还提到，因为我们问了他们，这对我们和Anthropic、OpenAI的关系意味着什么？

SAP also said, because we had asked them, like, what does this mean for Anthropic and OpenAI?

至于Anthropic和OpenAI这两家公司，它们没能在发布前作出回应。

And those companies, Anthropic and OpenAI, didn't respond in time for publication.

但SAP提到，他们似乎正在和Anthropic至少就某项合作进行沟通。

But SAP mentioned that they it sounded like they're working with at least Anthropic on something.

SAP下周将举办年度客户大会。

So SAP has its annual customer conference next week.

我预计我们会听到更多关于这项新政策的消息，很可能还会公布一些与此相关的全新协议。

I expect that we're gonna be hearing a lot more about the new policy and very likely some sort of new agreements as part of this new policy.

你有没有跟SAP的客户聊过他们对这件事的反应？

Have you spoken to SAP customers about what their reaction is to this?

我接触过一些为SAP客户做许可咨询的专家。

I spoke with licensing experts that work with SAP customers.

我觉得这些变化太新了，而且SAP也没广泛宣传，客户们可能还没来得及认真思考其中的影响，但我和那些专家聊过，他们确实已经想得很清楚了。

I think the changes are so new, and like I said, we're not broadly communicated by SAP that I don't know the customers have really thought through the implications, but the licensing experts I spoke with certainly have.

其中一位专家告诉我，他们觉得SAP这明显是在比赛进行中突然改规则。

And one of them told us that it really looks to them like SAP is trying to change the rules mid game.

这里有几个方面。

Now there's a couple pieces here.

其中一个涉及代理访问SAP数据。

One of one has to do with agents accessing SAP data.

长期以来，SAP一直很担心客户把数据大量导出到其他平台，比如ServiceNow、Salesforce这些地方去做分析。

But SAP for a long time has been very concerned about customers mass exporting their data out into other platforms, ServiceNow, Salesforce, etcetera, in order to do analysis there.

值得注意的是，SAP在企业软件领域一直像个封闭花园，规则从来就不一样。

And it's important to note that SAP has always sort of operated in its own, what what people call a walled garden in the enterprise software world, meaning that the rules have always been sort of different for them.

SAP定下的规矩，向来就和其他公司不太一样。

The rules that SAP has laid out have always been somewhat different.

我觉得这也是其中一部分原因。

And so that's also part of this, I think.

现在，SAP终于第一次把AI代理这类事情的政策正式明确了。

It's just SAP is finally, for the first time, formalizing what their policies are going to be for things like AI agents.

凯文，我想稍微退一步说清楚点。

So Kevin, I want to step back here a little bit.

你说过，自从你最初报道那些像Atlassian这样的公司开始应对挑战以来，已经过去一年了。这些公司并不是在对冲风险，而是在努力保护自己的地盘，因为像Glean这样的初创企业正在崛起。

You said it's been a year since you initially broke that story about companies like Atlassian, all these companies trying to really hedge their not so much hedge their bets, but really protect their own house in a world where startups like Glean are coming into the picture.

我记得我们之前聊过这个故事的早期阶段，那时候企业软件公司都说：别担心，根本没什么大不了的。

I recall us having talked over the early parts of that story about how enterprise software companies were saying, Hey, look, there's nothing to see here really.

我想知道，最近几个月，这种讨论有没有什么变化。

And I wonder how you have seen that conversation change over the last few months.

我觉得SAP这次的举动，是我们见过最严厉的一次。

It feels like this move from SAP is the most stern move that we've seen.

所以尽管之前那些公司都说‘别在意’，但他们现在却采取了如此极端的措施。

And so I guess despite the companies having said, Don't worry about it, I mean, they're really going to dramatic lengths here.

是的。

Yeah.

我觉得我们该具体聊聊，公司到底要对哪些东西收费，因为这部分其实相当复杂。

I think let's talk about what specifically companies are going to charge for, because that part is really sort of it's pretty complicated actually.

但根据SAP上周，或者更准确地说，是上个月他们CEO克里斯蒂安·克莱因在财报电话会上的说法，他们要收费的并不是数据本身。

But judging from what SAP told us last week or or actually what SAP's CEO, Christian Klein, said on their earnings call last month, they are not it's not like the the data that they're gonna charge for.

他们管这叫知识图谱和本体论。

It's it's called things like knowledge graphs and ontologies.

要解释清楚这到底是什么挺难的，但简单说，现在厂商们说的‘上下文’就是指这个。

And it's really sort of hard to explain what that is, but basically, this is what vendors are now referring to as context.

所以，HubSpot和Workday的CEO也在财报电话会上提到，打算对AI代理访问这些上下文信息收费。

And so the hub the the CEOs of HubSpot and Workday have talked also on earnings about basically floating the idea of charging when agents access context.

上下文之所以重要，是因为如果代理只看到客户数据，根本不知道该怎么处理；它需要靠这些图谱和本体论来理解，比如这个客户具体下了哪笔单，订单是从哪个仓库发出的。

Now context is important because if an agent sees customer data, it doesn't know what to do with it at all, but it needs to know from these graphs and the ontologies, like basically can say this specific customer made this specific order and it was shipped from this specific warehouse.

就是这种极其细致的洞察，才能让代理真正干活。

The very granular kind of understanding that enables agents to do their work.

所以，厂商们接下来要收费的，就是这些东西。

And so that's the stuff that vendors are gonna start charging for.

我觉得SAP在财报电话会上说的，指的就是这个。

And I think that's what SAP was alluding to on earnings.

听上去他们不是直接收费，而是会和合作伙伴、软件厂商重新谈合作条款。

It sounds like they're gonna charge or not charge, but they're gonna form new agreements with their partners, their software partners.

通过这些协议，代理访问的问题将得到解决。

And through those agreements, this issue of agentic access is going to be addressed.

很好。

Great.

凯文，非常感谢你今天来做客。

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

这位是凯文·麦克劳克林，我们《信息》杂志的企业软件记者。

That is Kevin McLaughlin, our enterprise software reporter here at The Information.

Anthropic 正在寻找拓展算力资源的方式。

Anthropic is looking for ways to diversify its compute.

我的同事斯蒂芬妮·帕拉佐洛和阿尼萨·加德西上周末率先报道了 Anthropic 正在寻求支持的一家初创公司。

My colleague Stephanie Palazzolo and Anissa Gardesi broke news over the weekend about one startup that Anthropic is looking to for support.

我想请斯蒂芬妮来跟我们说说她了解到的情况。

I want to bring on Stephanie to tell us what she knows.

斯蒂芬妮，欢迎再次做客节目。

Stephanie, welcome back to the show.

很高兴你能来。

It's great to have you here.

谢谢。

Thanks.

能来也很开心。

Great to be here.

好的。

All right.

说说看，Anthropic 是怎么打算拓展算力来源的？

Talk to me about how Anthropic is looking to diversify its compute.

当然。

Totally.

上周末我和Anissa报道的是，Anthropic最近在和一家伦敦的初创公司Fractal接洽，这家公司正在开发专注于推理的芯片，Anthropic有意在明年某个时候开始采购他们的芯片。

So what Anissa and I reported over the weekend is that Anthropic has been recently in talks with a London based startup called Fractal, which makes inference focused chips to potentially become a customer of that startup's chips once they become available at some point next year.

虽然这短期内帮不上Anthropic解决算力紧张的问题——我们之前已经写了很多相关内容——但它未来能帮助公司进一步多元化算力来源。

So even though this won't really help Anthropic with its near term compute crunch, which we've been writing a lot about, it could really help the company further diversify its sources of compute in the future.

这在Fractal正在进行的融资轮次中也是一个很大的卖点，他们正试图筹集超过十亿美元的资金，估值超过十亿美元。

This has also been a really big selling point in Fractal's ongoing funding round that it's raising, where specifically it's looking to raise more than $100,000,000 in new funding at a $1,000,000,000 plus price.

这确实是个不错的筹码，能向潜在投资者表明，Anthropic明确有意向使用你们的产品。

Well, that's a pretty good contract to hang in front of potential investors saying Anthropic is very much looking to use us.

说说Fractal的技术吧。

Tell me about Fractal's technology.

像Anthropic这么大的公司，为什么会看上Fractal？

Why would such a big company like Anthropic be looking to Fractal?

当然。

Sure.

不过这里得先澄清一点，Fractal的芯片目前还没上市。

So I think the first big caveat here is that Fractal's chips are actually not available yet.

据他们首席执行官说，公司计划在今年下半年开始生产这些芯片，然后明年把它们部署到数据中心。

So basically the company's CEO has said that they want to start making those chips at some point in the second half of this year, and then they want to make those chips available in data centers next year.

所以某种程度上，这一切还都是假设性的。但Fractal的目标是打造专门用于推理的芯片——也就是实际运行模型的过程，而不是训练模型。

So a lot of this is kind of a hypothetical in some ways, but what Fractal wants to do is basically build these chips that are specialized for inference, which is the kind of process of actually running models versus training them.

所以，这些芯片理论上能比现在的英伟达芯片更快、更便宜地运行当前的大语言模型。

So essentially these chips would theoretically be able to run today's large language models faster and for cheaper than NVIDIA chips do right now.

这其中很大一部分原因在于Fractal芯片使用了一种特殊的内存，和Grok、Cerebras等其他芯片初创公司的内存类型类似。

And a large part of this is because of a special type of memory that Fractal's chips use, which is also similar to the type of memory used in other chip startups, chips like Grok and Cerebras.

这种内存比英伟达芯片常用的内存表现更好，因为它能更高效地在不同芯片之间传输数据，而数据传输正是如今支撑AI模型的GPU面临的主要瓶颈之一。

And so this type of memory works better than the memory that's that's typically used in Nvidia chips because it basically helps to move the data faster between different chips, which can be a really big bottleneck in terms of today's GPUs that used to power AI models.

我想问你一个问题，为什么Anthropic会选择初创公司，而不是那些已经拥有强大算力的大型云服务商？毕竟他们有英伟达。

I want to ask you a question here around why Anthropic would be looking to startups as opposed to the big cloud providers that have their We obviously have Nvidia.

我说的云服务商，指的是谷歌和他们的TPU。

When I say cloud provider, we're talking about Google and their TPUs.

你认为Anthropic为什么会选择一家初创公司呢？

Why would Anthropic be looking to a startup, do you think?

Anthropic其实早就已经是多元化算力来源的典范了。

Anthropic really has been the poster child already for diversifying its sources of compute.

除了英伟达的芯片，他们已经在使用谷歌的TPU，还有亚马逊的Tranium芯片。

It already uses, in addition to Nvidia chips, the TPUs from Google, well as Amazon's Tranium chips.

他们已经在多元化布局上做了不少工作。

It's already done a lot of work in diversification.

所以我觉得，现在他们主要是想为未来做准备，说白了就是……

So I think at this point it's really trying to kind of hedge for the future and say, you know.

未来一两年里，这些新兴初创公司很可能会推出突破性的芯片，他们想尽早跟这些公司建立联系。

We are going to have to really look at all these kind of up and coming startups that you know, they might be the ones to create the next breakout chip in, you know, the next year or two, and they wanna start building those early relationships with those with those startups early.

我知道已经有不少报道说，OpenAI 在考虑自研芯片。

I know we've we've had reports of OpenAI looking to build its own chips.

你觉得 Anthropic 有朝一日也会自己设计芯片吗？

Do you think Anthropic would look to build its own chip at one point?

我觉得他们肯定在认真考虑这件事。

So I think that's definitely something that they were thinking about.

有消息说，Anthropic 确实正在研发自己的芯片，就跟 OpenAI 以及谷歌、亚马逊、Meta 这些公司一样。

There have been some reports that Anthropic is actually looking to build its own chip, similar to what OpenAI is trying to do in many other model developers out there like Google, Amazon, Meta, they all have their own chips.

所以我觉得，Anthropic 在考虑自研芯片，这事儿是板上钉钉的。

And so I think this is, you know, definitely something that Anthropic is thinking about.

自己研发芯片显然是个巨大的工程。

It's obviously a very big undertaking to go and like build your own chip.

所以我觉得他们正在认真考虑各种策略——不管是自研芯片、跟初创公司合作，还是跟大型云服务商联手，总之就是想尽一切办法获取算力资源。

So I think they're just really trying to consider like all different strategies here, whether that's making their own chip, partnering with younger startups, working with the larger cloud providers, just really everything it can do to get any source of compute possible.

最后一个问题。

Last question for you.

你提到Fractal正在寻求融资。

You mentioned that Fractal is looking to fundraise.

能跟我们说说这家公司在整个芯片生态中的位置吗？

Just tell us a little bit about this company's place in the chip ecosystem.

关于这家公司的创始人或者背景，我们了解多少？

Do we know anything about the founders or the story of the company at all broadly?

当然。

Totally.

Fractal是一家总部位于伦敦的初创公司。

So Fractal, again, it's this London based startup.

这家公司一直比较低调，但成立于2022年。

It has been pretty under the radar, but it was founded in 2022.

我觉得有意思的是，正如我之前提到的，据我们了解，Fractal 正在利用这个潜在的合作来吸引投资人在融资轮次中的关注。

And so I think one interesting thing about this is, as I mentioned earlier, it from what we're hearing, Fractal is really using this potential deal to, you know, get a lot of investor interest during the the the funding round.

不过这里确实有很多不确定因素，对吧？

I do think that one interesting thing here though is that a lot of this is pretty up in the air, right?

即使 Anthropic 最终签了这份客户合同，他们至少还得等一年，甚至更久才能拿到芯片。

So even if Anthropic does move forward signing this customer contract, they actually wouldn't get the chips for, you know, at least another year, if not further.

而且，Anthropic 很可能也在和其他一堆芯片初创公司接洽。

And, you know, Anthropic is probably also talking to all sorts of chip startups.

所以我觉得，虽然 Fractal 确实在这个大生态里显得很引人注目，但我们还是得保持一点谨慎，毕竟一年两年后，如果这单真能成，最终会是什么结果，现在谁也说不准。

So I think like while this does make Fractal a very interesting, you know, player in this broader ecosystem, I think we have to definitely take this with a bit of skepticism as well, to see what's actually going to end up happening in a year or two whenever this, you know, if this deal even happens, what like, what is going to actually happen then.

很好。

Great.

谢谢您，Stephanie，欢迎来做客。

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

这位是斯蒂芬妮·帕拉佐洛，我们《信息》杂志的AI记者。

That is Stephanie Palazzolo, our AI reporter here at The Information.

SpaceX即将上市，让公开市场的投资者开始思考的不只是要不要买，而是能不能负担得起，毕竟它的估值已经涨得太高了。

SpaceX's IPO has public market investors thinking about not just if they will buy, but how they will afford to purchase the company's shares given how big SpaceX's valuation has grown.

我的同事科里·温伯格和瓦莉塔·波在上周末报道了这个故事。

My colleagues Corey Weinberg and Valita Poe covered the story over the weekend.

我想请我们的高级财经编辑肯·布朗来给大家深入分析一下这些情况。

I wanna bring on our senior finance editor, Ken Brown, to talk us through those dynamics a bit more.

肯，欢迎再次做客我们的节目。

Ken, welcome back to the show.

很高兴你今天能来。

It's great to have you here.

你好，阿卡什。

Hi, Akash.

肯，你能说说，公开市场的投资者现在是怎么考虑如何为SpaceX的IPO腾出空间的吗？

So Ken, tell me how are public market investors thinking about how they're going to make room for the SpaceX IPO?

这正是关键所在。

Well, that's just it.

你得腾出空间，对吧？

You have to make room, right?

因为这些人大多手头有点现金，但大部分资产都押在股票上了。

Because most of these guys, they have some cash, but they're heavily invested in stocks.

所以他们得卖掉一些别的东西。

And so they're going to have to sell something.

这事儿挺棘手的，因为SpaceX规模太大，投入的资金会非常可观。

And so it's a tricky thing because SpaceX is big and it will require a lot of money to go into it.

所以这就像一场平衡术。

And so, yeah, it's a dance.

很多人在琢磨，我是不是该筹点钱出来？

So a lot of them are talking about, do I raise money for this?

得先看看那七大科技股，它们才是我得优先考虑的对象。

Well, the MAG7, the other big tech stocks are what I have to start to look at.

这可能是个有点傻的问题，我不是专业的投资经理，如果在听节目的朋友里有专家，还请包涵——但为什么不能直接买一点点呢？

So this might be a bit of a stupid question, and I'm not an expert portfolio manager, so forgive me for those listening on the show who are, but why can't fund a manager why can't you just buy a little bit?

你知道的。

You know?

为什么非得腾出这么大一块仓位，去押一个有点投机、要等十年二十年才见成效的公司？

Why do you have to make room for a honking chunk of a company that is, a little speculative, a ten, twenty year trajectory?

他们干嘛非得在这家大公司上押这么重的注？

Why do they have to make big bets here on this big company?

这其实讲的是相对回报的问题。

Well, so this is all a story about relative return.

很多基金经理就是靠这个吃饭的。

And this is what a lot of these fund managers live and die by.

我跟标普500指数比业绩，要是跑输它，我就输了。

So I compete against an index, the S and P 500 say.

只要跑不赢指数，我就落后了。

If I lag the index, I'm behind.

如果我跑赢了指数，我就算打败了市场，成了英雄，还能吸引到更多资金。

If I exceed the index, I've beaten the market, I'm a hero and I take in money.

要想吸引资金，唯一的方法就是表现优异。

The only way to take in money is to perform well.

所以这正是他们的目标。

So that's the goal.

因此，不管你是主动型基金经理，还是跟踪某个指数，你唯一要做的就是表现好。SpaceX 就像你说的，会成为一个体量巨大的公司。

And so all these guys, whether you're an active manager pegged to an index, you're just trying to perform well, SpaceX is going to be a big, as you say, big honking company.

如果他们真的表现突出，而你却没买，或者只买了很少，那你就会落后。

If they really perform well and you don't buy or you buy very little, then you're going to lag behind.

这其实就是已经发生的事。

I mean, this is what's happened.

过去几年里，任何没持有‘七巨头’股票的基金经理，都跑输了市场，投资者很可能已经撤资了。

Any fund manager out there who did not own the Mag-seven for the past few years has lagged behind, and investors are probably pulling money out of their fund.

所以这对他们来说至关重要，如何策略性地应对、尽可能做到最好，是他们最关心的事。

So this is a big deal for these guys, and strategizing around it and how to do it the best possible way is really important for them.

那你就给我们详细讲讲这个逻辑吧。

And so just walk us through calculus here.

所以现在，如果SpaceX被纳入这些主要指数——就像我们之前讨论的那样——从基金经理的角度来看，问题就在于：我之前没买够SpaceX，不然指数本来还能涨得更高。

So now if SpaceX gets added to some of these major indices, as we've seen there have been a lot of conversations about, then from the fund manager's perspective, it's that, Well, I didn't own enough SpaceX basically because then the index would conceivably rise.

没错。

Right.

如果它真的被纳入这些指数，而所有指数看起来都在准备加仓，那你确实不得不跟。

So if it gets added to the indexes, and these indexes all seem like they're gearing up to add, then yeah, you really have to.

重点不是我喜不喜欢买这只股票。

It's not so much like, Do I want to buy this or not?

而是：我是想跟指数对着干，还是想顺着指数走？

It's, Do I want to bet against the index or I want to bet in favor of the index?

所以你要么轻仓配置SpaceX，要么重仓配置，这确实是个难题。

And so you either underweight or overweight SpaceX, and so it's a real challenge.

对很多基金经理来说，真正的挑战在于——这对大多数主动管理型基金都适用——他们正在亏钱。

The real challenge for a lot of these guys, and this is true for most actively managed mutual funds, they're losing money.

投资者正在撤资。

Investors are pulling money out.

这种情况已经持续一段时间了。

This has been true for some time.

他们把钱投进了指数基金。

And they put money in index funds.

所以，如果大家都在撤资，那筹钱买股票就更难了，因为你每天甚至每周都得卖股票来应对赎回。

And so again, if people are taking money out, it's even harder to raise money to buy stuff, because you're already selling stocks every day to meet redemptions or every week or whatever.

这对这些基金经理来说真是个难题。

So it's challenging a moment for these guys.

肯，科里和瓦莉塔的报道里提到，SpaceX为了吸引投资者，真是下了大功夫。

Ken, there was some great reporting in the story from Corey and Valita about the dramatic lengths that SpaceX is going to in order to woo investors.

报道开头讲的是，200名投资者从纽瓦克包机飞往德克萨斯。

The top of the story was this flight of 200 investors from Newark to Texas.

我不确定，SpaceX这种路演是特例，还是在IPO前的拉投资圈子里挺常见的？

And look, I don't is SpaceX this esque roadshow, or is this common in the land of IPO wooing?

这次IPO根本没什么普通的，对吧？

Nothing about this IPO is common, right?

所以一切都更宏大了。

So everything is bigger.

埃隆·马斯克这个人设更强大。

Elon Musk is the bigger personality.

太空和这些火箭本身也更引人注目。

Space and these rockets are a bigger thing.

还有人工智能。

AI.

我的意思是，这简直是一场疯狂的盛会。

I mean, it is a bonkers kind of event.

所以没错，一切都有点过头了。

And so yeah, everything is a little over the top.

市场影响也会同样非同寻常。

And the market impact is going to be over the top too.

所以我们现在关注的就是这个。

And so that's what we're watching.

之前在一周前的纽约，信息主办的‘融资AI：革命’大会上，我也稍微提过这一点。

Now, touched on this a little bit at the Financing AI, the Revolution Conference that the information hosted a week ago today in New York.

这个IPO，可能会很成功，也可能搞得很糟。

The idea that this IPO, I mean, it could go well, it could also go quite poorly.

其中可能出问题的地方多得数不清。

And there are a million ways that could happen.

我想问你的是，如果真成了个烂摊子，对AI生态和大科技公司会有什么连锁反应？

My question for you is if that happens, if if the IPO kind of turns out to be a bit of a dud, what do you think the the ripple effects of that would be on the AI ecosystem, on bigger tech companies?

股市是一回事，但我想听听更深层的影响。

I mean, the stock market is one thing, but talk about the implications.

确实如此。

Yeah.

我的意思是，SpaceX跟其他那些AI公司很不一样，它不只是做AI的，还搞航天、星链互联网，还有Twitter，现在又有了XAI。

I mean, so SpaceX is different than almost any of these other AI companies because it's a space company and they have a space based internet company and they have Twitter, what and then they have XAI.

XAI现在烧钱烧得厉害。

And XAI is burning through a ton of cash.

所以问题来了，市场会怎么看？

And so the problem is what will the market see?

他们会看重它的太空业务，喜欢还是不喜欢，还是别的？

Will they see the space business and that's what they like or they don't like or whatever?

如果IPO表现不好，很难判断到底是AI业务拖了后腿。

So it's going to be hard to discern whether if the IPO does poorly, if it went poorly because of the AI component.

有可能是。

It might.

你得等到下一批公司上市才能看清楚。

You won't really know until the next IPOs.

你看，那七个巨头几乎都跟AI息息相关，但它们也赚得盆满钵满。

I mean, look, the Mag-seven are basically almost all tied to AI or heavily tied to AI, but they also make a ton of money.

它们的利润高得吓人。

They are so profitable.

所以如果AI业务搞砸了，对他们会是不小的打击，但还能撑得住。

And so, AI goes badly, it's going to hurt them, but they will survive.

至于那些纯做AI的公司，谁说得准呢？

The AI specific companies, who knows, right?

他们可能会遭受重创。

I mean, they could really suffer.

甚至可能直接消失。

They could disappear.

这才是真正棘手的地方。

So, that's the tricky thing here.

没错。

Right.

换句话说，SpaceX的IPO虽然是第一个，但未必能代表市场对AI的整体态度，可能真不能。

In other words, the SpaceX IPO, although it will be the first one, it may not be the referendum on AI It may not be.

对，对，对。

Yeah, yeah, yeah.

这简直就是一堆乱七八糟的东西，对吧？

And it's just mishmash of stuff, right?

没错，没错。

Right, right.

这样一来，看下去反而更有趣了。

Well, makes it all the more interesting to watch.

肯，谢谢你今天来做客。

Ken, I want to thank you for coming on.

顺便说一下，我人在旧金山。

And by the way, I'm an SF.

你那边是纽约的演播室。

You're in the New York studio.

咱们换个位置了。

We're trading places.

正是如此。

Exactly.

我真的很期待以后再和你深入聊聊这个话题。

We I I look forward to chatting with you more about it.

这位是肯·布朗，我们《The Information》的资深财经编辑。

That is Ken Brown, our senior finance editor here at The Information.

随着SpaceX、OpenAI和Anthropic陆续准备上市，我们一直在关注一个问题：这一切会对风险投资产生什么影响？

As SpaceX marches towards its IPO, as does OpenAI and Anthropic, one question that we are waiting to answer is how all of this will affect venture capital.

今天来和我们讨论这个问题的是Gradient的管理合伙人达里安·谢拉齐。

Here to discuss that is Darian Shirazi, a managing partner at Gradient.

达里安，欢迎来到节目。

Darian, welcome to the show.

很高兴你能来。

It's great to have you here.

谢谢邀请我。

Thanks for having me.

我们上一段聊了SpaceX，以及它对整个AI生态可能产生的影响。

So we talked about SpaceX in the previous segment and how that could affect the AI ecosystem at large.

SpaceX其实是个各种业务混在一起的公司。

SpaceX is a bit of a mishmash of companies.

它搞航天。

It's space.

它也做人工智能。

It's AI.

当然，我们还有OpenAI和Anthropic，它们也正朝着上市迈进。

We, of course, have OpenAI and Anthropic marching towards their IPOs.

作为一位风投，你是怎么看这件事的？这对你们早期投资会有什么影响？

How are you watching this as a venture capitalist and how it could affect your world of early stage investing?

我觉得我们对未来的走向还是挺乐观的，主要是因为风投市场之前一直缺乏流动性。

Well, I think we're pretty positive and optimistic about where things are going, mainly because there's been sort of this lack of liquidity in the venture capital market.

现在这些公司要上市了，能给投资人和有限合伙人回笼资金，这样我们才能继续支持下一波创新浪潮。

And with these IPOs coming up, it's gonna be really nice to return capital to investors and LPs such that we can sort of feed the cycle of the next wave of innovation that's coming up.

我觉得这些上市公司的估值都定得太高了，不过AI领域里也有一些其他公司，比如Cerebras，他们今天刚公布定价，估计这周就挂牌，那个看起来定价还挺合理的。

I do think that a lot of these IPOs feel priced to perfection, so to speak, and there but there are some other APOs in the AI ecosystem like Cerebras that just, announced their pricing today or that they're gonna list probably this week, and that one seems to be well priced.

你知道，公司上市的目的就是融资、宣传品牌，同时也要确保投资者能赚钱，尤其是那些参与IPO的新投资者。

You know, the point of an IPO is to raise capital, market your company, and then also to make sure your investors are gonna make money, especially the new investors in the IPO.

所以对于那些定价过高的公司，我确实有点担心。

So I worry about that for the ones that are very highly priced.

但对其他一些公司来说，我觉得还有很大的增长空间。

But for some of these others, I think there's a lot of upside potential.

明白了。

Okay.

等一下。

Hold on.

所以‘定价完美’和‘定价合理’到底有什么区别？

So so price to perfection versus well priced.

你这话是什么意思？

What what's the difference there?

你能再解释一下吗？

What do you mean by all that?

所谓价格完美，我觉得，当SpaceX的市值达到1.75万亿美元，而年收入才二三十亿美元的时候，这个倍数就算在私募市场也高得离谱了。

Price to perfection, I think, you know, if you have a $1,750,000,000,000 market cap for SpaceX when the business is doing somewhere between 20 and 25,000,000,000 in revenue, I think that's a pretty rich multiple even in private

完美定价就是定价太高了，我只是想确认一下我有没有理解对这里的说法。

Perfection is is is rich in in in I just wanna make sure that I understand the the language here.

没错。

Exactly.

对。

Yes.

价格完美就是定价过高，可能被高估了，对参与IPO的投资人来说，上涨空间大概很有限。

Price perfection means rich, potentially overvalued, probably not a lot of upside for the IPO participants.

明白了。

Got it.

好的。

Okay.

那像OpenClaw和Anthropic这种定价合理的公司，你觉得估值倍数合适吗？

And then but well priced for OpenClaw and Anthropic, you you you think it's a it's a good multiple?

我的意思是，对于Anthropic来说，我觉得还有很大的增长空间。

I mean, for Anthropic, I think there's a lot of upside.

我们看到他们的营收一个月内就从300亿美元涨到了440亿美元。

We saw the business go from 30,000,000,000 in top line to 44,000,000,000 within a month.

我觉得，以一万亿美元的市值来看，考虑到现在的估值水平，这其实并不算高。

I think that, you know, at a trillion dollar market cap, that's actually not too bad given where multiples are today.

但如果我们看看其他公司，比如Cerebras，这周刚宣布要上市，其中一家看起来潜力很大，确实挺让人兴奋的。

But if we look at other businesses like Cerebras that announced this week that they're gonna list, like, one seems to be have a lot of upside potential in it, there's one that's quite exciting.

对。

Right.

我明白你提到的流动性问题。

So I I hear you on the liquidity piece.

这其实我们节目里聊过很多次了，还有风投们在这个全新的IPO环境下，想出的各种办法来释放流动性。

I mean, this is something we've talked about on the show a lot, and also the creative ways in which VCs have have shot liquidity in this novel IPO environment.

不过说实话，我很好奇，假如OpenAI和Anthropic都上市了，那将是市场对AI公司价值的一次关键检验。

But look, I mean, I do wonder, let's say OpenAI and Anthropic go public, those are the referendums of sort on AI companies and how the stock market reacts to it.

假设这些IPO没有按计划进行，或者在前六个月里，股价像Figma那样从上市初期的峰值跌下来，你觉得会怎样？

Let's say the IPOs don't go the way people plan, or let's say in the first six months, the stock price kind of falls from the pop that they initially see just like Figma did, for example.

我的意思是，你认为这会对有限合伙人（LP）如何向风险投资公司分配资金，或者风投公司所做的投资决策产生什么影响？

I mean, what impact do you think this has on the way LPs are thinking about allocating money to venture capital companies or even the bets that VC companies are making?

我认为，即使这些IPO表现不佳，且在上市后的前三个月内股价没有上涨动力，LP们仍然会对他们从这些交易中获得的巨额回报倍数感到满意。

I think even if these IPOs trade off and they don't actually see upward momentum in their stocks within the next first three months of the IPO, I still think LPs are gonna be very happy with the fact that they are returning such a large multiple or receiving such a large multiple back from the the investments they made in various firms that are in these deals.

所以我认为，这些公司能够成功上市本身就是一个积极的信号。

So I actually think that the fact that these get out at all is a positive sign for

我是说，从前瞻性的角度来看，也就是接下来三年早期投资领域新募集基金的情况。

I'm talking about sort of sort of forward looking in terms of, you know, the the next vintage of of funds that are raised, you know, the next three years of early stage investing.

你觉得这对那方面会有什么影响？

What impact do you think it has on that?

我觉得会有非常积极的影响。

I think it has a very positive effect.

我觉得我们会看到更多的有限合伙人重新跳回市场。

I think that we're gonna see a lot more LPs jump to get back into the market.

我们会看到更多早期阶段的风投公司能够顺利募资，因为有限合伙人手头充裕，这将带动大量早期公司获得资金支持。

We're gonna see a lot more venture capital firms at the early stage be able to raise money because the LPs are liquid, and that will yield a a ton more companies that are gonna get funded at the early stage.

我觉得软件领域更复杂的变化是，Cloud Code 已经颠覆了很多我平时经常接触的软件公司，毕竟这东西实在太有趣了。

I think that more complicated is what's happening in software is that, you know, Cloud Code has disrupted a lot of software companies that I myself spend a lot of time in Cloud Code because it's so fun.

现在你几乎能重新构建任何一家软件公司现有的产品。

And you can really rebuild anything that a software company has today.

我觉得这比IPO走势带来的影响更大——我们会看到大量新生态系统的投资涌现，创新程度将远超过去十年到十五年，那时候你只要达到几个SaaS关键指标，就能轻松融到下一轮钱。

And I think that's more of a problem than the IPOs will be is that we're gonna see investment in all new ecosystems, and it's gonna yield much more innovation than I think we had in the past ten to fifteen years where you just needed to hit a certain number of SaaS metrics to raise your next round of funding.

但不管IPO表现如何，我都非常乐观：风险投资市场将迎来更多流动性，更多基金将获得LP的资金支持，从而催生大量资金充足的初创企业。

But I'm very positive on regardless of how these IPOs go, we're gonna see a lot more liquidity enter the venture market and a lot more funds get funded by LPs, which will yield a lot more startups that are well funded.

你现在在Gradient主要投哪些方向？

What are you investing in in Gradient right now?

这是个好问题。

That's a good question.

过去六个月我们做了大量研究和主题分析，因为Cloud Code进步神速，我们预计它还会继续升级，很多原本看似有护城河的软件应用，其实已经不再安全了。

I mean, we have been doing a lot of work and thesis work in the past six months because Cloud Code has gotten a lot better, and our expectation is that it will get even better than it is today and that a lot of the software applications that are defend seemed defensible or had a moat.

比如Salesforce这样的公司，我们认为会因为Cloud Code这些技术的能力而从内部被颠覆。

Like, a company like Salesforce, for example, we think is gonna get sort of disrupted from the inside out because of the the capabilities of Cloud Code in these models.

未来我们会花更多精力投资那些跟软件无关的领域，比如我们投了一个AI对冲基金。

If you look at, the future is that we are spending a lot more time investing in businesses that are not software related, that are we invested in an AI hedge fund, for example.

我们还投了一家机器人公司。

We invested in a robotics company, for example.

我们也在投资一家人体组织工厂，对此非常看好。

We're investing in a human tissue foundry as well that we're very excited about.

这些都是崭新的前沿方案，但不是基于软件的。

And these are new cutting edge soft solutions that are not software based.

这感觉特别像2017年，我们刚起步的时候，正好是Transformer论文发布之后，那时候创新才是风投真正押注的东西。

It feels a lot like 2017 when we launched right after the transformer paper came out, and that was really an era where innovation was sort of what you underwrote as a venture capitalist.

有点像九十年代和两千年初的样子。

A lot like the nineties and the early two thousands.

而在2010年代，决定一家公司能不能融到资的，主要是财务指标。

And we went through this era in the twenty tens where it was really financial metrics that drove whether companies were gonna raise capital or not.

我们非常期待重新回归到为真正的创新提供支持，承担技术风险而非市场风险，专注于投资那些真正想打造颠覆性大类业务的顶尖技术创始人。

And we're excited to get back to underwriting true innovation, taking technology risk rather than market risk, and focusing on backing the best technical founders that really wanna build big category defining businesses.

所以我觉得我们的目光已经超出了软件领域。

So I think we're looking outside of software.

任何全栈式AI驱动、或用前沿AI方式颠覆软件以外市场的项目，都是我们的关注对象。

Anything that's full stack AI enabled or that is using AI in a cutting edge way to disrupt a market that's outside of software.

我觉得很多风投都在这么做，但很多人很吃力，因为他们首先来自金融背景，这让他们很难真正理解并支持技术型创始人和以技术为核心的企业。

And I think a lot of venture capitalists are doing that, but many are struggling because a lot of them have finance backgrounds first and foremost, which make it difficult for them to underwrite technical founders and technology first and foremost.

没错。

Right.

我再问你一个问题。

One more question for you.

我们《信息》杂志曾报道过，像通用催化剂这样的风投公司，为了帮投资组合里的企业获取算力、应对当前GPU短缺，已经付出了多大的努力。

We have reported here at The Information the lengths to which venture capital companies like General Catalyst for example, have gone to try to help their portfolio companies access compute and deal with the shortage of GPUs that there are in the market right now.

你们有没有为自己的投资组合做类似的努力，帮它们解决芯片短缺的问题？

Are you doing any similar type of negotiations on behalf of your portfolio to help your companies deal with the chip issue?

其实没有。

Not really.

原因是我们并没有投资那些试图打造前沿模型的公司，或者那些跟前沿模型实验室竞争的公司。

And the reason is that we're not investing in a lot of the frontier model contenders or front frontier model lab competitors.

我觉得现在很多风投都往那些试图训练大型模型的公司砸了大笔钱，这些模型需要大量的算力。

I think that a lot of venture capitalists today have poured large sums of money into companies that are trying to train new models that are large models that require a significant compute.

所以在算力或训练这方面，我们基本没怎么碰，也不太担心。

So on the compute side or the training side, we don't really see that, and that's not really a concern.

但在推理端，我们确实很担心，因为未来推理算力的供给肯定会跟不上需求。

On inference, that definitely is something that we are concerned about because there will be a lack of of inference capability and also supply of inference compute.

而且随着时间推移，这个问题会越来越严重。

And we think that's gonna become more and more acute as time goes on.

目前我们还没找到解决办法。

We haven't really come up with a solution.

好在我们是Lambda的最大投资方之一。

Thankfully, we are one of the largest investors in Lambda.

我们领投了他们的种子轮，所以能优先使用他们很多技术能力。

We led the seed there, so we have access premier access to a lot of capabilities there.

我们没什么特别的协议，但跟不少供应商关系都很近。

We don't have any special deals, but we really are close to a lot of the vendors.

我们基金的最大投资人之一也是谷歌。

We also one of our biggest investors in the fund is Google.

所以如果需要，我们能为旗下公司快速拿到算力，不过到现在还没必要谈什么特殊安排。

So we have access to really get compute for our companies if we need to, but we haven't had to cut special deals yet.

我觉得如果你在训练

I think that if you're training

你是在说你们投资组合里，比如Lambda和其他公司之间的一些特殊安排吧？

You're you're talking about special deals between, like, inside your portfolio lambda to

对。

Correct.

没错。

Correct.

我们还没谈过任何特殊安排。

We haven't cut any special ones.

也许以后得这么做。

We may have to

你会吗？

do you would?

你觉得你会吗？

Do you think you would?

我觉得将来我们可能不得不这么做。

In the future, I think we might have to.

是的。

Yes.

算力的供应，特别是推理层面，将来会变得非常紧张。

Is that compute the supply of compute, especially the inference level, is gonna be very problematic.

这跟石油市场的情况差不多——就是供应根本不够，成本会持续上涨。

It's it's almost like the same thing that's going on in oil markets is that you just don't have enough of it, and you're gonna continue to see costs go up and up.

这样一来，智能的成本也会随之上升。

And and what is that gonna do is the cost of intelligence will go up as well.

所以我们正在关注那些真正能优化推理效率的公司。

And so that's why we're looking at companies that can actually optimize inference.

我们投资过一家叫 SentML 的公司，去年被英伟达以高价收购了。

We invested in one called SentML, which NVIDIA acquired for a large sum last year.

我们一直在寻找各种方法，尽可能优化推理过程，但这个问题至今还没有解决。

And so we really are looking at ways in which we can optimize inference as much as possible, but this is an unsolved problem.

计算成本只会越来越高。

We're gonna see compute costs go up.

我之所以对大型云服务商和新兴云平台这么看好，就是因为这些公司一旦需要推理、训练或其他任何形式的算力，根本没别的地方可去。

It's And one of the reasons why I'm so bullish on a lot of the hyperscalers and the neo clouds is that there's just nowhere for these companies to go if they need access to inference or training or any form of compute.

我对您之前提到的那些特殊协议挺好奇的。

I am curious about these so called special deals that you talked about.

这种情况常见吗？

Is that common?