‘Mr. IPO’ on SpaceX’s Historic Listing, Anthropic’s Repelling Data Retention Policy

欢迎各位来到《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11日，星期四。

It is Thursday, June 11.

在开始今天的节目之前，我想跟大家提一下我们最新订阅者调查的结果：随着AI模型和智能代理越来越强大，有60%的受访者认为，目前黑客比防御方更占优势。

Before we get to today's show, I want to flag for you the results of The Information's latest subscriber survey on the state of cybersecurity as AI models and agents become increasingly powerful, 60% of readers who responded to our survey say they think hackers have the advantage over defenders right now.

调查还显示，读者们对工作或公司可能遭遇的网络安全威胁感到担忧。

The survey also shows that readers are anxious about cybersecurity threats to their work or organization.

五分之四的受访者表示，他们现在的担忧比去年这个时候更强烈了。

Four in five readers said they have more concerns today than they did this time last year.

我们已经把完整的调查结果发布在网站上了。

We have posted the full survey results on our website.

你们可以去theinformation.com查看。

You can check it out at theinformation.com.

今天节目中，SpaceX的上市就在明天，倒计时一天，市场会怎么反应？我们请来了杰伊·里特，人称‘IPO先生’。

Today on the show, we are T minus one day away from the historic SpaceX IPO, and with so many questions on what the market reaction is going to be, we're going to bring on Jay Ritter, also known as Mr.

IPO先生。

IPO.

他马上就会上线和我们聊一聊。

He'll be on the show in just a minute.

接着我们会分析Anthropic新出台的数据保留政策，为什么让一些客户感到不安，还会聊聊关于Xbox的最新报道。

We'll then unpack how Anthropic's new data retention policy is scaring off some customers, and we'll also cover some reporting that we have about Xbox.

最后，随着企业为AI算力砸下重金，Palantir正试图扮演中间人的角色。

And finally, as companies break their budgets paying for AI compute, Palantir is making the case to act as an intermediary.

关键问题是，这真能帮大家省下钱来吗？

The big question, of course, is will that actually help anyone save money?

节目内容很精彩，咱们直接开始吧。

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

SpaceX明天就要上市了。

SpaceX is going public tomorrow.

这次上市有望成为史上规模最大的IPO，而能把这个时刻说清楚的人，恐怕没人比佛罗里达大学IPO计划负责人杰伊·里特更合适了。

It is poised to be the biggest IPO ever, and few people can put this moment in context better than Jay Ritter, director of the IPO initiative at the University of Florida.

他也被称为‘IPO先生’。

He is also known as Mr.

IPO先生。

IPO.

杰伊，欢迎来到节目。

Jay, welcome to the show.

很高兴你能来。

It's great to have you here.

早上好。

Good morning.

我得问问你，你是不是也穿着纽约尼克斯队的蓝橙色球衣，就像我这样？

So I have to ask you, are you wearing blue and orange for New York Knicks colors like I am?

不是，我其实支持圣安东尼奥队。

No, I'm actually rooting for San Antonio.

我一向不支持纽约的球队，但佛罗里达大学的颜色是橙色和蓝色。

I always root against the New York team, but the University of Florida colors are orange and blue.

明白了。

Okay.

行吧。

All right.

那我就这么定了。

Well, I'll take it.

就在这充满期待的早晨，我接受了。

I'll take it on this exciting morning here.

好的。

Okay.

明天就要上市了。

So we've got this IPO tomorrow.

大家最关心的问题是，你觉得这股票会怎么表现？

The big question everyone's wondering, how do you think it trades?

你觉得明天开盘会怎么样？

How do you think it'll trade tomorrow?

不一定会上涨，但可能性非常大。

It's not guaranteed to go up, but it's very likely to.

一般来说，七成五的IPO首日都会跳涨，但这次的概率比平常还高。

On average, 75% of IPOs jump on the first day of trading, but this has a higher probability than normal.

听说机构投资者的需求超出了配额四倍，散户投资者也至少超了三倍。

The word is that institutional investors have oversubscribed it by a factor of four, retail investors by a factor of at least three.

对机构投资者来说，需求是供给的四倍并不稀奇。

For institutional investors, it's not at all unusual to have demand exceed supply by a factor of four.

但这次不同的是，我觉得很多需求是真实的，不是那种明明只想买两万股却报了十万股的情况。

But what's unusual here is I think a lot of the demand is true demand, rather than asking for 100,000 shares when you really only want 20,000 shares.

特别是有报道说，一些大型机构，比如道富环球之类的，可能能拿到高达一百亿美元的股票配额。

In particular, it's been reported that a few big institutional holders, I would guess State Street Global and some others, are going to be allocated as many as $10,000,000,000 worth of shares.

我敢肯定，他们不可能真申请了四百亿的额度。

And I'm sure they didn't ask for $40,000,000,000 worth.

所以你说的‘真实需求’，能具体解释一下吗？

So when you say true demand, I mean, help us understand this.

意思是，很多人下的单子远超过他们实际想买或者能负担的数量？

Is the idea that people often place orders that are much larger than what they expect or even in some cases can afford to pay?

通常都是这样吗？

Is that typically how it works?

没错。

Exactly.

机构投资者经常申请的股数远多于他们实际能拿到的。

Very frequently institutional investors ask for a lot more shares than they expect to get.

而且还有一个问题，他们是长期持有者，还是短期炒作者？

And also, there's a question, are they going to be buyers or renters?

也就是说，他们是打算长期持有吗？

That is, are they going to be buy and hold investors?

还是拿到股票后马上抛售，赚差价？

Are they going to quickly sell the stock, flip it?

如果他们倒手卖出，就一定得有别的买家来接盘。

If they flip the stock, there has to be some other buyer to buy what's being sold.

那你觉得，这些机构投资者到底是长期持有者，还是短期炒作者呢？

And so, what do you think the answer to that question is in terms of whether or not these institutional investors are buyers or renters?

你对此有什么看法？

What's your sense there?

我觉得，那些大规模的配售，确实有一部分是给像道富环球这样的长期持有型投资者的。

Well, I think some of these really big allocations are indeed to buy and hold investors like State Street Global.

但因为像纳斯达克的QQQ这类指数很早就纳入了这只股票，一些原本可能首日就抛售的对冲基金，这次可能不会急着出手了。

But also because of the early index inclusion with NASDAQ's QQQ, among others, I think that some of the institutions hedge funds that might otherwise flip on the first day are not going to be flipping on the first day.

他们会等上五天、十五天，等各个指数把这只股票加进去，到时候指数基金买入时，自然会有接盘的人。

They're going to wait five, fifteen days until some of the indices add the stock to the index, and there'll be suppliers on the day that some of the index investors are buying.

所以我觉得，虽然明天像往常IPO一样交易会很活跃，但对冲基金的倒手行为不会像平时那么多，除非股价暴涨。

So I expect, while there's going to be a lot of trading activity tomorrow, as is typical with an IPO, there won't be as much flipping by hedge funds as normal unless the stock jumps a lot.

我们这次的指数纳入，正好赶上了加速生效的节点。

We have sort of the split decision acceleration of the inclusion into the indexes.

纳斯达克说他们会纳入，标普却表示不会采取这种做法。

You've got NASDAQ saying we will, S and P saying we're not going to take that approach.

你觉得在是否应该加快把这些公司纳入指数这个问题上，哪种做法更合理？

What do you think the right answer is in terms of whether or not these it should be accelerated to include in the indices?

我觉得纳斯达克和标普都做出了正确的决定。

I think both NASDAQ and S and P made the right decisions.

很多投资者都希望今年晚些时候，SpaceX、Anthropic 和 OpenAI 如果在纳斯达克上市，能尽快被纳入 QQQ 指数。

A lot of investors want SpaceX later this year, Anthropic and OpenAI if they go public on NASDAQ, to be included in QQQ.

但另一方面，标普500指数追踪的资金规模太大了，如果标普过早纳入，那恐怕是个错误。

On the other hand, the S and P five hundred has so much money indexed to it that I think it would have been a mistake for S and P to add it too early.

所以你的意思是，纳斯达克做出这个决定，主要是为了吸引公司上市？

So you're saying NASDAQ made the right decision just in terms of them wanting to get the listing basically?

你是这个意思吗？

Is that what you're saying?

这家公司本来大概率也会选择纳斯达克。

The company probably would have gone with NASDAQ anyway.

但粗略估算，追踪标普500的資金量大概是QQQ的30倍左右。

But in ballpark numbers, roughly 30 times as much money is indexed to the S and P 500 as to QQQ.

把一个规模较大的指数，哪怕相比之下很小，加进去，我觉得完全没问题。

Having a large, but in comparison, index added, I think it's perfectly okay.

但标普500要是加的话，会按流通股权重，而不是总市值来算。

But with S and P 500, although they would have been adding it with float weight rather than the total market cap.

可那里头资金实在太多了，要是马上加进去，需求激增可能会造成短期的价格波动。

Still, so much money is indexed there that that boost in demand might result in some temporary price dislocations if they had done it right away.

对。

Right.

这周科技股下跌了。

Now tech stocks are down this week.

你觉得有多少是因为投资者在腾仓位，好给SpaceX留位置？

How much of that do you think is investors making room in their portfolios for SpaceX?

可能有一点吧。

Probably a little bit.

一些散户投资者腾出个人账户里的资金来购买SpaceX股票，成了主要的卖方。

In that with some retail investors freeing up money in their personal accounts so that they can buy SpaceX stock resulted in the main sellers.

但我认为，对很多大型机构来说，更大的问题是，不少科技股的估值涨得太快、太高，有点儿泡沫了。

But I think for a lot of big institutions, the bigger issue has been have the valuations on a lot of the tech stocks risen so fast and so far that there's a bit of a bubble going on.

你刚才提到了散户的配置和散户交易者。

You mentioned the retail allocations here and the retail traders.

我想知道，你觉得这种趋势对IPO的影响有多大？感觉散户参与IPO的比例在上升。

I wonder what impact you think these seem It feels like retail allocations are increasing for IPOs.

你手上可能有更准确的数据，但单从感觉上来说，确实觉得越来越多公司开始重视这一点。

You probably have more data on whether or not that's the case, but just anecdotally, I mean, it feels like this is something that more companies consider.

你认为这对未来的IPO会带来什么样的结构性影响？

What structural impact do you think that would have on IPOs going forward?

我觉得这很可能只是个特例。

Well, I think this is going to be pretty much a one off thing.

对绝大多数IPO来说，大约95%的份额都会分配给机构投资者。

That for almost all IPOs, 95% or so gets allocated to institutional investors.

而零售投资者的配额中，很大一部分都进了主要承销商的私人财富管理账户。

And of the retail allocation, a lot of it goes to the private wealth management accounts of the main underwriters.

所以普通散户能分到的股份很少，甚至可能一点都没有。

So mom and pop retail investors don't get many shares, if any.

我觉得这种情况大概还会持续下去。

And I think that's probably going to continue.

SpaceX因为有埃隆·马斯克这个因素，算是个特例，毕竟特斯拉的散户比例异常高。

I think SpaceX, with the Elon Musk factor, is a bit of exception where Tesla has an unusually large proportion of retail investors.

我想这也是SpaceX这次决定要给散户留一大块配额的原因。

And I think that's one of the reasons that SpaceX decided that they want a big retail allocation here.

至于其他IPO，我觉得这种因素不会起那么大的作用。

For other IPOs, I don't think that factor is going to come into play as much.

杰伊，你研究IPO这么多年了。

Jay, you've studied IPOs for so long.

你能不能想起过哪一次IPO，像这次这样几乎成了对一个人、一位领袖的全民公投？

Can you recall an IPO that was as close a referendum to one person and one leader as this one ever before?

没有。

No.

没有。

No.

2012年Facebook上市时，马克·扎克伯格已经很有名了，但那只是一个公司。

When Facebook went public in 2012, Mark Zuckerberg was already well known, but that was one company.

上亿人拥有Facebook账号，所以它的零售知名度非常高。

Tens of millions of people had Facebook accounts, So it had incredible retail name recognition.

那次IPO确实给了零售投资者很大比例的股份，大约25%。

That was an IPO that had a very big retail allocation, about 25%.

但Facebook的零售配售情况有点不一样，公司原本并不打算给零售投资者这么多股份。

But that Facebook retail allocation was a bit different in that the company did not intend to have such a big retail allocation.

结果机构需求不如预期，承销商就决定把更多股份分配给零售投资者。

But when institutional demand turned out to be a little bit disappointing, the underwriters decided to give more shares to retail investors than otherwise would have been the case.

而这一次，是公司自己从一开始就想要大力向零售投资者倾斜。

Here, it's the company that wanted the big retail allocation upfront.

这周确实有媒体报道，机构需求非常强劲，零售投资者的配额可能会被小幅削减。

And indeed, there have been some press reports this week that the institutional demand is so strong that the retail allocation actually might get cut a bit.

这其实牵涉到我想问你的另一个问题，就是零售配额，我们常常把它和网络情绪混为一谈，对吧？

And that sort of connects to another question I wanted to ask you about, which is retail allocation, we sort of conflate that with the online sentiment, right?

比如Reddit上的讨论区，还有推特，确实能反映出大家的情绪。

I mean, we've seen these Reddit forums sort of capture the mood, certainly on Twitter as well.

我心里一直有个疑问：网络情绪到底在多大程度上推动股价波动？零售配额又能在多大程度上影响市场？

There's a question in my mind, to what extent online sentiment actually drives stock movements and to what extent the retail allocation can actually move markets?

对于小公司来说，这种情况是可能发生的，对吧？

I mean, smaller companies, it can happen, right?

但我很好奇，你有没有研究过相关数据，看看这种影响是被高估了，还是被低估了？

But I wonder, have you studied any data to the impact of this and whether or not it's overestimated or underestimated?

我并没有专门研究过这个问题。

I haven't studied this directly.

不过你可以想想像AMC和游戏驿站这样的迷因股。

But think with meme stocks like AMC and Gamestop.

对，GameStop。

Yeah, GameStop.

在那里，Reddit上的散户需求推动了股价波动，跟公司基本面完全没关系。

There, Reddit, retail demand was moving the price around unrelated to fundamentals.

但SpaceX可不是什么迷因股。

Here, SpaceX is not a meme stock.

特斯拉也不是迷因股。

And Tesla is not a meme stock.

可能会有一些散户受Reddit帖子影响，但我觉得这并不是眼下散户需求的主要驱动力。

There might be some retail investors that are influenced by Reddit threads, But don't think it's the main thing driving the retail demand here.

没错。

Right.

而且我觉得，如果一家公司规模这么大，还把它叫迷因股，那散户得掀起多大的风浪才能真正影响股价啊。

And strikes me that again, if the company is so big, you can call it a meme stock, the larger the company is, I mean, the more movement you would need by the meme stock traders to actually sway a difference in the stock price.

所以这可能正是你想表达的一个关键点。

So perhaps that's one point that you're hitting on here.

杰，你研究IPO这么久了，我想请你总体谈谈。

Jay, I want to ask you broadly, just you've studied IPOs for so long.

过去三年，科技公司IPO有哪些趋势？你对未来有什么预期？

What trends are you seeing in tech IPOs broadly over the past three years, and what are you expecting to come?

OpenAI、Anthropic、谷歌的Gemini，还有其他一些大型语言模型，确实对很多软件公司构成了威胁。

Well, OpenAI and Anthropic and Google's Gemini and some other AI, large language models, are definitely a threat to a lot of software companies.

所以我们看到，无论是风投市场还是公开市场，很多软件公司都表现疲软。

So what we've been seeing is weakness both in venture capital markets and in public markets for a lot of software companies.

这一点大家都有共识：AI具有颠覆性。

So this is something where everybody agrees AI is disruptive.

它就是下一个大趋势。

It is the next big thing.

虽然AI相关公司正从中受益，但其他一些科技公司的商业模式却因此受到冲击。

But while AI related companies are benefiting from this enthusiasm, it's actually harming the business models of some other tech companies.

因此，科技公司或其他领域的IPO并没有出现大规模的激增。

And consequently, we're not seeing an enormous surge in tech IPOs or other IPOs across the board.

如果回溯到四十六年前，也就是1980年，油价当时涨得厉害。

If we go back forty six years ago, 1980, oil prices went up a lot.

石油股的表现非常出色。

Oil stocks did very well.

当时石油和能源板块在标普500中的市值占比一度达到了30%。

The oil and energy sector got to be 30% of the S and P 500 in terms of market cap.

那个行业一片火热，而其他行业的IPO活动却没那么活跃。

That sector was really booming while other industries weren't booming in terms of IPO activity.

我觉得现在的情况也差不多，有些其他行业并没有从中受益，但投资者对AI相关公司的热情确实非常高涨。

And I think we're seeing the same thing now, that some other industries, some other verticals aren't benefiting, but there's definitely a huge investor enthusiasm for AI related companies.

你觉得在OpenAI和Anthropic即将上市的过程中，人们最忽略的是什么？

What do you think people aren't paying enough attention to as it relates to the OpenAI and Anthropic IPOs coming down the pipeline?

这三家巨头公司，个个都是顶尖企业。

Well, with all three of the mega IPOs, these are great companies.

但企业本身优秀，并不等于它们未来一定会是好的投资标的。

But great companies don't necessarily mean they're going to be great investments on a point forward basis.

这三家公司上市时的估值都高得离谱，不管是万亿美金以上，还是市销率来看都是这样。

All three of these companies are going public at very high valuations, whether we're talking about a trillion dollars or more, or just their price to sales ratios.

像SpaceX这样市销率超过90的公司上市，未来必须把收入翻上好几倍，变得极其盈利，才能配得上今天的高估值。

When companies are going public, such as SpaceX at a price to sales ratio of over 90, the company has to grow its revenue enormously and become immensely profitable in the future to justify this high valuation today.

这事也不是没可能。

It could happen.

私人和公开市场给这些公司这么高的估值，背后肯定有原因。

There's a reason that private and public markets are giving these companies such high valuations.

但要让公司收入，尤其是利润增长到足以支撑一万八千亿美金估值的水平，需要太多条件同时满足了。

But a lot of things have to go right for the company to have its revenue and especially its profits grow to the enormous levels that are needed to justify 1,800,000,000,000 valuation.

你刚才提到了私募市场。

Now, you mentioned the private markets.

我们这档节目之前也聊过，私募市场有多赚钱，有多愿意给这些公司注资，让它们长期保持私有状态。

We have talked on this show about how lucrative the private markets have been and how willing they have been to fund some of these companies and keep them private longer.

你觉得这种平衡会不会开始往五六年以前的情况倒回去？

Do you see that balance tipping at all back to maybe what we saw five, six years ago?

我们看到很多公司都上市了。

We saw a lot of companies go public.

如果一家公司选择上市，说明它保持私有的时间变短了。

If they're going public, it means they're staying private for shorter periods of time.

你觉得未来几年这种平衡会逆转吗？公司会不会又得靠上市来筹集所需的资金？

Do you see that balance flipping at all in the years to come where companies need to start to go public again to access the amount of capital they need?

不太可能。

Not really.

风险投资领域现在有太多资金涌入了。

There's so much money that has flowed into venture capital.

对于后期阶段的公司，共同基金、主权财富基金这些机构都愿意掏钱支持。

For later stage companies, mutual funds, sovereign wealth funds, others are willing to finance the companies.

无论是公开市场还是私募市场，美国的资本体系都特别强大，愿意为有前景的公司注资，哪怕它们现在还没盈利。

American capital markets, both public markets and private markets, are amazing in the ability and willingness to finance companies with a promising future, whether they're currently profitable or not.

这也是科技行业和其他行业不一样的地方。

And this is where the tech business is different than other industries.

开餐厅要是现在不赚钱，那什么时候才能赚？

For the restaurant business, if you're not making money now, when are you going to?

但科技领域有些行业已经证明，公司可以通过网络效应实现赢家通吃，或者因为前期投入巨大、规模效应明显，

But certain verticals in the tech space have demonstrated that a company can dominate a market either due to network effects where it's winner take all or because of huge upfront costs and huge economies of scale, where there's not going to

根本

be

不会有竞争对手来拉低利润空间。

competition leveling the profit margins.

所以，这类公司未来有可能获得巨额利润，而且不用担心竞争蚕食利润。

And so the company can potentially earn enormous profits in the future without competition eroding them.

英伟达、微软、谷歌这些公司都已经证明了这一点。

NVIDIA, Microsoft, Alphabet, others have demonstrated this.

这种现象在其他行业根本没出现过。

And this just has not happened in other industries.

这就是为什么风投愿意给那些目前亏钱、但前景光明的科技公司注资。

This is why the venture capitalists have been willing to finance money losing tech companies with a potential bright future.

超大规模云服务商的资本支出非常庞大，但它们筹钱根本不是问题。

The capital expenditures for the hyperscalers are enormous, and they're not having any trouble at all coming up with the money.

你是说，因为成为唯一赢家的机会太大了，所以私人市场的人才愿意承担这种风险吗？

You're saying because of the opportunity to become one of the sole winners is so great, that's the risk that people are willing to take in the private markets?

没错。

Right.

和其他没有进入门槛的行业不同，那些行业竞争激烈，导致未来的利润率很难提升。

Unlike other industries where there aren't barriers to entry, you've got competition that keeps the profit margins in the future from being big.

但训练一个大型语言模型的成本高得惊人。

But training a large language model is incredibly expensive.

把巨型火箭送入太空，再让助推器回收降落以降低成本，这过程复杂到极点。

Sending huge rockets into space, catching the booster on the way down to lower costs, it is incredibly complicated.

对。

Right.

说到这个，杰伊，我最后一个问题。

And so on that note, Jay, last question for you.

你觉得SpaceX那些近乎天方夜谭的登月目标，连带着它的IPO，会引发一波新的‘登月式’上市潮吗？

Do you think that the moonshot ambitions, quite literally, of SpaceX and its IPO, do you think that will usher in a new wave of moonshot IPOs?

这只是一个特例，就像你之前说的，还是说我们会看到更多公司带着五年、十年甚至更久都不赚钱的计划去上市，就为了把轨道数据中心送进太空？

Is it a one off, as you suggested earlier, or will we see a lot more companies trying to go public with plans to not generate any cash flow for the next five, ten years, however long it takes to get orbital data centers into space.

对。

Right.

生物技术行业本来就是个需要很长时间才能研发出新药的模式。

Well, the biotech industry has a business model where it takes a long time to develop a new drug.

无论是私募市场还是公开市场，都完全愿意投资那些短期内不可能靠新药盈利的生物科技初创公司。

Private markets and public markets are perfectly willing to finance, start up biotech companies that have no prospect of earning revenue from a new drug for five or ten years into the future.

在科技领域，能这样做的细分领域其实不多，但一旦有机会赚到巨额利润，现在砸再多钱也值得。

With tech, there's a limited number of niches where this is going on, the potential ability to earn enormous profits can justify huge expenditures now.

对。

Right.

这话说得很有道理。

Well, that's a good point.

生物科技行业确实就是这种模式，还没做出产品、没有任何收入就上市，这种情况很常见。

And certainly biotech is an industry where that model, where you go public before even having a product or any revenue at all, that's been a common story.

所以，尽管SpaceX是史上规模最大的IPO，但至少这个时间线对某些投资者来说可能并不陌生。

So in some cases, despite SpaceX being the biggest IPO ever, it's at least a timeline that might be familiar to some investors.

Jay，非常感谢你来做客。

Jay, I want to thank you so much for coming on.

这次聊得非常愉快。

It was a wonderful conversation.

以上是佛罗里达大学IPO研究中心主任杰伊·里特的发言，

That is Jay Ritter, Director of the IPO Initiative at the University of Florida, and Mr.

这里是TITV《IPO观察》节目。

IPO here on TITV.

现在不少企业都在测试Anthropic公司推出的Claude 3.5 Sonnet模型，但也有人没有急着跟风用上这个模型。

As companies continue to test Anthropic's Claude Fable five, some people are choosing not to jump at the model as quickly.

微软就是这类观望企业里最受关注的那家。

Microsoft is one of the biggest names in that latter camp.

我的同事亚伦·霍姆斯在今天的人工智能简报中详细报道了这个决定。

My colleague Aaron Holmes covered that decision in detail in today's AI agenda newsletter.

我想请他来跟我们分享一下他了解到的情况。

I wanna bring him on to share with us what he knows.

亚伦，欢迎再次回到节目。

Aaron, welcome back to the show.

很高兴你能来。

It's great to have you here.

能来我很高兴。

Happy to be here.

为什么微软不像其他人那样急着采用Claude Fable Five？

Why isn't Microsoft jumping at Claude Fable five the way everyone else is?

这不只是微软一家的情况。

Well, it's not just Microsoft.

我听说有不少客户都在推迟使用Fable，尤其是涉及关键任务的时候。

I've heard that a number of customers are holding off on using Fable, especially for mission critical tasks.

这是因为Anthropic在Fable的使用协议里加了一条不太寻常的条款，说你们输入到这个模型里的所有数据，他们都会保留三十天。

That's because Anthropic has this kind of unusual clause in the the usage agreements with Fable where they say, you know, we, Anthropic, will hold on to all of the traffic that you put into this model, for thirty days.

Anthropic解释说，这么做是为了监控用户如何使用模型，特别是确保没人拿它来做违法或恶意的事情。

And Anthropic says that's a necessary step in order to keep track of how people are using the model and specifically to make sure that nobody is using it for illegal or malicious purposes.

但与此同时，很多客户因此犹豫了，因为通常来说，像微软这样的公司，或者那些处于受监管行业的企业，对内部数据的存储和使用都有非常严格的规定。

At the same time, that's giving a lot of customers pause because, you know, typically, especially companies like Microsoft or companies that are in regulated industries have their own strict policies about where data that they're using internally is going and how it's being stored.

目前，大家对Anthropic的这项政策到底会对这些客户造成什么影响，还存在不少困惑。

And there's currently a good deal of confusion around exactly what Anthropic's policy would mean for those customers.

所以他们的想法是，万一真发生了什么恶意攻击，Anthropic可以在三十天内回溯这些数据，做个事后分析，看看到底出了什么问题。

And so the idea here is basically if any kind of malicious attack does happen, Anthropic can go back to that data that it's sitting on within a thirty day period and do a bit of an autopsy, I guess, on what happened, study it essentially.

没错。

Right.

而且Fable本质上就是被削弱版的Mythos，而我们知道，Mythos这个模型原本是专门用来搞黑客攻击、进行网络渗透的，能力超强。

And yeah, Fable is essentially a neutered version of mythos, which as we know, is this model that is supposed to be incredibly good at hacking and and basically doing cyber attacks.

现在Mythos只对少数测试人员开放，而Fable却是公开可用的。

And right now, Mythos is only available to a select number of testers, whereas Fable is widely available.

从Anthropic的角度来看，他们在Fable上设置了重重限制，防止它被用来发动这类攻击。

And, you know, from Anthropic's perspective, they've put a lot of guardrails onto Fable to stop it from being used for such attacks.

但可以肯定的是，还是会有黑客试图用它来做坏事。

But, you know, it's safe to assume that some hackers are still going to try to use it for malicious ends.

所以Anthropic希望确保这种情况不会发生。

So Anthropic basically wants to be able to make sure that that isn't happening.

如果真发生了，他们也需要留下一些证据，也就是这三十天的使用数据。

And and if it is to be able to have, you know, some receipts in the form of this thirty day of of traffic data.

那Anthropic是把这部分数据存在Azure或AWS这样的云服务器上吗？

And is Anthropic storing this data on a cloud servers like Azure or AWS?

这事儿其实挺复杂的。

So it's very complicated.

通常来说，过去Anthropic会给客户选择‘零数据保留’选项，意思是，你通过亚马逊或谷歌云输入模型的数据，完全不会离开你的环境，他们也不会另存一份。

Essentially, you know, in most cases, Anthropic previously, you know, would let customers have this zero data retention option, which means, like, any data that you're feeding into their models on a cloud like Amazon or Google is not leaving your environment, and they're not storing it anywhere else.

但微软的Azure情况有点不一样。

It's a little bit different for Microsoft Azure.

Anthropic 一直表示，使用 Azure 的客户必须同意他们的数据可能会离开 Azure，我认为这主要是因为 Anthropic 在 Azure 上起步较晚，使用的 Azure 服务器还不够多。

Anthropic has always said that customers of Azure do have to agree that their data might leave Azure, and I think that's mostly just because Anthropic is newer on Azure and they haven't used as much, of Azure servers to run their models.

不过，根据这项新政策，亚马逊已经提醒客户：如果你在使用 Fable，那么这三十天内保存的数据，并不能保证会留在你原本为其他 AWS 服务设置的同一地区或相同的安全防护范围内。

However, you know, now with this new policy, Amazon actually warned customers that if you're using Fable specifically, this data that's saved within the thirty day window is not guaranteed to stay in the same region or in the same, you know, security guardrails that you've already configured for your other AWS usage.

这可能会带来问题，因为像欧洲的 GDPR 这样的法律明确规定，不能将欧洲客户的个人数据转移到欧洲以外的地方。

And that can be a problem because, you know, there are laws, especially Europe's GDPR that say, you know, you can't move European customers' data out of Europe.

所以，如果一家公司有欧洲客户，并想用 Fable 处理涉及这些客户的数据，现在就必须仔细评估这样做会不会违反相关法规。

And so if, you know, a company is dealing with customers in Europe and and trying to use Fable for any tasks involving those customers, they now have to to go and figure out whether doing so would run afoul of of those laws.

所以这真的非常复杂，我接触过的公司都在努力搞清楚这些规定到底意味着什么，才敢开始使用这个模型。

So it it really is very complicated, and the companies I've spoken to are are trying to figure out exactly what it means before they start using this model.

是的。

Right.

所以我的理解是，他们虽然用的是 AWS，但 AWS 并不能保证那三十天内存储的数据，会像客户原本期望的那样受到同样的保护。

I guess so what I'm hearing from you is basically they are using AWS, but AWS cannot guarantee that the data that's being stored in that thirty day window, that they can't guarantee that it's stored the same way that customers may have initially hoped for type of thing.

没错。

That's right.

说白了，这些政策的意思是，只要你使用这个模型，就等于同意Anthropic按照自己的条款来存储你的数据。

And essentially, the way that these policies are worded say that you know, by using this model, you're basically agreeing that this data is going to be stored by Anthropic under Anthropic's terms.

我觉得，在云服务领域，这种做法还挺少见的。

And, you know, I think that's somewhat unusual in the world of cloud services.

通常来说，客户才有权决定自己的数据该怎么存、怎么用，而不是由供应商说了算，这大概就是现在有些人开始犹豫要不要用这个模型的原因。

Typically, the client is the one who gets to decide, like, exactly how their data is being stored or used, not the vendor, which I think is why there's now some move to pause on using this model.

所以，因为数据保留政策，大家对Fable有所顾虑，你开头提到微软和其他一些客户，这种情况普遍吗？

So this hesitation to use Fable because of the retention policy, You talked about Microsoft and a couple other customers at the start.

有多普遍？

How widespread is it?

我觉得，主要是一些必须遵守大量法律法规的公司才特别谨慎。

I think it's mostly concentrated in companies that have to comply with a lot of laws and regulations.

我们听说，一些大型法律AI公司也不敢轻易用，毕竟有律师-客户保密特权这类法律，他们得确保，如果用Anthropic的模型，绝不会触犯这些规定。

We've heard that some of the big legal AI companies are hesitant to use it because there are, you know, laws around attorney client privilege, and they wanna be able to guarantee that any, you know, usage of their products, if if they're using Anthropic models, don't run afoul of those.

还有医疗行业的人，他们也得应对严格的隐私法规。

There's also, you know, people in the medical industry who have to deal with privacy laws.

同时，我也听说一些科技公司的员工，他们只是用这个来写代码或者做类似的编程任务，对他们来说，这个问题没那么让人担心。

At the same time, I have heard from some folks specifically at, you know, tech companies who are just using this for writing code or or, you know, similar kind of coding tasks who told me that they are not too worried about this.

毕竟，只要用第三方服务商，多多少少都得信任对方怎么存储数据，而使用Fable带来的好处，往往能盖过这些潜在顾虑。

And, you know, anytime you're using a third party vendor, there's gonna be some amount of trusting that vendor and and how they store data and and that the the benefits of using Fable outweigh those potential concerns.

所以我觉得，这种犹豫并不是普遍现象。

So I I don't think that this is a universal hesitation.

那个群体的人，是不是觉得Fable实在太好了，干脆就忽略掉数据保留政策的问题？

That latter camp, are are are they in the camp that are sort of saying that, hey, Fable is so good that we are willing to kick aside any kind of retention policy issues?

是的。

Yeah.

据我所知，大家现在才刚开始试用，但Fable在编程任务上的表现，明显高出一大截。

Essentially, I mean, from what I've heard, people are just starting to test it out, but Fable is seemingly a, you know, a step function better at coding tasks.

我们昨天在节目里也提到过，说它确实很厉害。

I mean, we've we've heard oh, yesterday on the show, said it's really good.

是的。

Yeah.

确实是这样。

It it it it is.

我的意思是，从基准测试来看，它明显比Opus强很多，不过我们还得等等看大家最终的共识是什么。

I mean, like, on the benchmarks, it definitely outperforms Opus significantly, and we we have to, you know, kind of wait and see exactly what the consensus is around that.

但早期的数据表明，Fable是一个能力非常强的模型。

But early early data shows that Fable is an extremely capable model.

你觉得这个故事最后会怎么发展？

Where do you think this story ends up?

你接触的那些人，觉得Anthropic会缩短数据保留周期吗？

Do the people that you talk to, do they believe that Anthropic may reduce the data retention window?

我都不确定他们会不会降到零，但也许十天会更现实一些。

I don't even know if they would bring it to zero, but maybe ten days is a more willing characteristic.

他们觉得这种情况会发生，还是觉得客户会慢慢接受三十天的保留期？

Do they think that's going to happen, or do they think that customers will become okay with the thirty day retention?

这件事最终会怎么收场？

How does this sort of still may end up playing out?

我的预测是，像微软这样的大客户，Anthropic 应该能谈出个解决方案——哪怕不能完全公开数据具体存在哪里，至少也能给出一些保证，缓解这些初期的担忧。

My prediction is that, you know, especially for the biggest customers like Microsoft, for example, I think Anthropic will be able to work something out and, you know, essentially, if not show customers exactly how and where the data is being stored, you know, may be able to make assurances that will assuage some of these early concerns.

但对于那些得不到 VIP 待遇的中型客户来说，这事儿就没那么容易了。

But I think that'll be a little harder for, you know, the long tail of midsize customers who don't get the white glove treatment.

最终还得看这些客户的法务部门怎么判断：Anthropic 给出的说明，到底能不能真正打消他们的顾虑。

And, you know, I I think it's really gonna come down to whether the the legal departments at these customers decide that whatever communication they're getting from Anthropic is enough to put these concerns to rest or not.

对。

Right.

亚伦，在你走之前，我想快速问你一下，你昨天报道的关于 Xbox 的消息。

Aaron, before we let you go, I want to ask you quickly about some news you reported yesterday on Xbox.

你了解到微软这个部门现在到底是什么情况？

What did you find out about the current state of that company or that that division within Microsoft, I guess, I should say?

是的。

Yeah.

Xbox 目前财务压力挺大的，正计划大规模裁员，同时在其他方面也大幅削减开支。

So Xbox is in a bit of a financial tough spot, and they are now planning significant layoffs and also cost cutting in other areas.

他们昨天公开表示，首席执行官阿卡什·帕什里卡说，Xbox的利润率因硬件成本上升以及游戏开发支出增加而被压缩。

And they actually said publicly yesterday, CEO, Akash Pasricha, said that Xbox's margins have been compressed by rising costs of hardware as well as increased spending on game development.

她还提到，他们的利润率已降至约3%，这意味着截至本月结束的财年，公司几乎只是勉强盈利。

And she said that their profit margins have shrunk to around 3%, which means they're barely eking out a profit as of the the fiscal year that's ending this month.

因此，夏尔马不仅对员工表示，也公开说明，这种情况不能再继续了。

And so, you know, essentially, Sharma told staff and also said publicly is that this can't continue.

收入一直在下滑。

Revenue has been falling.

成本却在不断上升。

Costs have been rising.

我们必须扭转局面。

We need to right the ship.

所以我们预计会看到裁员和其他成本削减措施随之而来。

And so we're we're expecting to see layoffs and other cost cutting as part of that.

你有没有了解到，AI是否在这些成本上升中起了作用？比如硬件成本上涨，是不是因为主机用了更先进的芯片？

Did you get any indication as to whether or not AI is playing a role here in terms of those rising costs, maybe even rising hardware costs, maybe there's more advanced chips that are being used in the consoles?

我的意思是，AI确实可能带来影响，但不是像其他公司那样，用AI取代员工的工作。

I mean, that's a way that AI could affect it, not in the way that we've seen at other companies, which is that AI is doing the job of the workers.

他们讨论过AI吗？

Was AI at all in the conversation here?

有提到。

Yeah.

Sharma没直接谈AI，但整个游戏硬件、计算硬件、服务器硬件领域，内存成本都在普遍上涨。

Sharpen didn't really mention AI specifically, but we have seen sort of across the entire world of gaming hardware, computing hardware, server hardware that, you know, memory is getting more expensive, across the board.

图形处理器原本是为游戏设计的，现在却被广泛用于AI运算。

And and so we're obviously graphics, processing units, which originally were were kind of invented for video games and now are used widely for AI.

这肯定加剧了硬件成本的压力。

So that's definitely contributing to this kind of hardware, you know, crunch.

而且我觉得，地缘政治的大环境也在一定程度上影响了这种情况。

And and I think we're also seeing some of that be impacted by just, like, the the broader geopolitical situation.

不过上周Akash Pasricha确实提到，AI在某些方面其实能帮游戏开发节省成本。

At the same time, you know, one thing that Akash Pasricha did say last week is that she thinks AI can, in some ways, help save costs when it comes to game development.

她说，AI现在有了新的用法，能提升游戏画质，还能降低开发成本。

And, you know, she said that there are new ways to use AI to improve graphics in games and maybe cut down on development costs.

她昨天在宣布财务困境时并没有明确提到这点，但我认为这在两方面都至关重要。

So she didn't really say that explicitly in the announcement yesterday about the the dire financial position, but I think it's definitely part of the equation on both sides.

你几个月前给《信息》写过一篇周末深度长文，讲的是阿莎·夏尔马。

You wrote a a weekend big read for the information a couple months ago now, I think, on Asha Sharma.

能不能再跟大家回顾一下，她对管理这个部门的整体目标和思路是什么？

Just remind folks of what her overall ambition and approach to managing that unit is going to be.

到目前为止，我们看到她的首要任务，是重新赢得Xbox玩家和员工的信任。

I think so far we've seen that, you know, her her biggest priority is to sort of win back the trust of Xbox fans and and also staff at Xbox.

有意思的是，她刚上任时就说，自己的职责不只是提升利润率。

And it it's kind of interesting because originally, you know, she came out the gate saying, my mandate here is not just to improve margins.

她还希望Xbox成为市场上最棒的主机，拥有最出色的游戏，并且回馈那些一直支持Xbox的玩家。

I I also want Xbox to be the best console on the market and to have the best games available and to, you know, reward the people who have been loyal to Xbox.

所以这一周我们看到，她稍微调整了之前的说法，同时公开解释了为什么必须裁员——尽管她最初的重点只是想做好Xbox，而不是盯着利润。

And so I think what we're seeing this week is her slightly walking that back, but also kind of explaining publicly why cuts are necessary, you know, despite her her early focus on just wanting to improve Xbox and not worry about margins.

我觉得他们现在的财务状况很清楚地说明，总得做出一些取舍。

I I think that the financial state they're in makes it clear that, like, you know, something has to give.

说得对。

Great.

艾伦，非常感谢你今天来做客。

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

这位是艾伦·霍尔姆斯，我们《信息》杂志负责微软报道的记者。

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

像Palantir这样的老牌软件公司，正试图抓住用户对OpenAI和Anthropic等公司AI工具费用飙升的不满情绪。

Established software companies like Palantir are trying to capitalize on the growing frustration over the soaring costs of using AI tools from providers like OpenAI and Anthropic.

我跟我们的AI记者布拉顿聊了聊Palantir的这个策略，以及客户们的反应。

I spoke with our AI reporter, Bratton, on Palantir's pitch and how customers are reacting to it.

下面听听他们的对话。

Here is that conversation.

劳拉，欢迎再次回到节目。

Laura, welcome back to the show.

很高兴你来参加节目。

It's great to have you here.

嗨。

Hey.

好的。

Okay.

Palantir的首席执行官亚历克斯·卡普一如既往地对企业在AI上的开支该不该花，发表了尖锐的看法。

Palantir CEO Alex Karp, he has some choice words, as usual, about how companies should and shouldn't be spending on AI.

他到底在说什么？

What is he saying?

没错。

Yeah.

他基本上是说，客户不应该直接去找大型语言模型提供商，也就是Anthropic和OpenAI这类公司。

He's basically saying that customers shouldn't go directly to large language model providers, ostensibly referring to Anthropic and Open AI.

而应该找Palantir这样的中间商，让他们帮着在任何模型之上搭建应用。

And instead, they should go to intermediaries like Palantir that can help them build applications on top of any model.

这样他们就能在不同模型之间切换，如果某个模型后来更占优势，也能灵活调整，针对自己的具体需求做优化。

So that they can, you know, if if one model prevails over another, they can switch between those models and gear them towards whatever custom purpose they're they're looking to do.

他是想说，直接找大模型提供商更贵，所以通过中间商反而能省钱吗？

And is is he making the point here that it's more expensive to go directly to the model providers so that they can make it cheaper?

还是说，其实重点是你们能帮他们更快找到更合适的模型？

Or is it really just, hey, we'll help you find a better model faster?

我觉得是两方面都有。

You know, I think it's twofold.

他确实提到过，如果你直接找大模型提供商，最后只会觉得钱包被掏空。

He definitely says in his quote, you know, if you go to a large language model provider, you'll go home feeling poor.

不过说实话，这个说法有点站不住脚，因为Palantir自己的服务也贵得离谱。

I think that that's kind of moot point because Palantir itself is really expensive.

他们的方案价格根本不低。

Its offer is really expensive.

但我认为他真正想表达的是，直接找大模型提供商可能效率不高，而像Palantir这样的公司，能基于你的全部数据，帮你打造真正贴合业务需求的定制化应用。

But I think what he's really trying to argue is that going directly to the large language model providers might be less effective than going to a company like Palantir that can help you build some sort of custom application for a specific business purpose on top of all of your data.

他是在哪里说这些话的？

Where did he make these comments?

他是在上周旧金山郊外一场专属客户活动上说的，我刚好参加了。

So he made it at this exclusive customer event outside of San Francisco last week that I was able to attend.

我觉得他抓住了一个关键时机——客户开始担心自己消耗的令牌数量，以及用了多少AI。

And I think what he's doing is really seizing on this moment where customers are beginning to worry about the tokens that they're burning through, how much AI how much AI they're using.

我们看到账单一路飙升，尤其是Claude代码使用量激增之后。

And, you know, we're seeing bills go through the roof, particularly as Claude code usage surges.

企业现在对用什么软件工具、用在什么地方，变得越来越挑剔了。

And companies are starting to be pickier about which software tools they use and for what purposes.

所以他想表达的是，Palantir能帮他们更精准地把AI用在真正需要的业务目标上，而不是直接去找模型提供商，结果被绑死在一个模型上。

And so he's trying to make the argument that Palantir will help them aim AI towards the purpose they're looking to use it for better than if they go to a model provider and, you know, potentially get locked into that one model.

你之前提到过，说Palantir本身也很贵，这点其实没什么意义。

And I mean, you hit on this point earlier saying it was a moot point because Palantir is quite expensive.

他们的服务到底有没有真正帮客户降低AI成本？还是说，听起来也还是一样贵？

Is their offering, does it show any evidence of saving them costs on AI, or is it still also a very pricey offering, it sounds like?

是的

Yeah.

我接触过的客户都表示，Palantir 帮他们提升了市场份额，或者增加了收入、降低了成本。

I think customers I've spoken to have said that Palantir is helping them either achieve greater market share of, you know, whatever market they're in or, helping them grow revenue or save costs.

我觉得挺有意思的。

So, you know, I thought it was funny.

我跟一位高管聊过，他说 Palantir 确实挺贵的。

One executive I spoke to, was like, well, Palantir is pretty expensive.

那你该怎么说服别人呢？

So you know, how are you making this argument?

他回答说，这东西确实有价值——我觉得这话挺有意思，也挺有说服力的。

And he was like, it's actually valuable, which I thought was a kind of funny point and sell.

但事实确实如此。

But it is true.

客户们也告诉我，不管是解决供应链问题，还是帮他们应对复杂的合规要求，Palantir 都很管用。

And customers do tell me that, you know, whether it's solving supply chain issues or helping them helping them sift through regulatory requirements.

有个律师事务所叫柯克兰和埃利斯，在客户活动上分享过，他们把五亿美元预算中的一部分用来在Palantir平台上开发一个定制的AI应用，能让律师在几分钟内完成私募交易文件的起草，而不是原来耗上好几天。

You know, there was one corporate law firm, Kirkland and Ellis that presented at the customer event, and they said that they're using part of their, you know, dollars 500,000,000 budget to build a custom AI application on Palantir that can help its lawyers draft transaction documents for private fundraising deals in minutes rather than days.

单看起草文件这件事，可能觉得没什么大不了，但对这种大型律所来说，这其实省下了巨额成本。

And, you know, that might not seem like a big deal if we're thinking about just drafting documents, but that's actually a massive cost savings for a big corporate law firm.

现在，Palantir的发展势头相当迅猛。

Now, Palantir is on a bit of a rocket ship right now.

我很好奇，从整体来看，你觉得这家公司在未来可能会遇到什么挑战？比如当需求逐渐趋于平稳的时候。

I wonder, as you look at that business as a whole, what are the questions that you think that business could inevitably face down the road here as perhaps demand starts to normalize a little bit?

还是说，这种高速增长会一直持续下去？

Or is it going to continue being a rocky ship forever?

是的。

Yeah.

我觉得值得提一下，Anthropic预计本季度收入能达到110亿美元，而Palantir预计全年收入大约是77亿美元。

I mean, I think it's worth calling out that Anthropic has projected that it will see $11,000,000,000 in revenue in the current quarter, whereas Palantir has projected, you know, about 7,700,000,000 in full year revenue.

所以单从今年的预测来看，Anthropic的收入已经超过了Palantir。

So Anthropic is already kind of sore and past Palantir in terms of revenue, at least in terms of what the companies are projecting, you know, this year.

不过我觉得可以说，Palantir 因为与美国政府以及海外政府的一些有争议的合作，在媒体上备受关注。

But I think the argument can be made that Palantir, you know, it's definitely highlighted in media for its, you know, controversial work with the US government and governments abroad.

但正是这些政府合作，反而吸引了很多高度监管的行业来与 Palantir 合作。

But that very same work with governments is drawing in highly regulated industries to work with Palantir.

所以，不管一家公司持什么政治立场，Palantir 能拿到这么多敏感的政府合同，本身就是一个巨大的吸引力。

So, you know, wherever you fall in the political spectrum as a company, the fact that Palantir has these really sensitive contracts with governments is a big draw.

我认为，这也是支持 Palantir 未来能成功的一个重要理由。

And I think that that's one bullish argument you can make for Palantir to succeed.

不过，随着企业自研软件的成本不断下降，现在用像 Claude 这样的编程工具自己开发软件也变得越来越容易。

However, as the cost to develop in house software from different, you know, enterprise is goes down, you know, it's easier to use whatever, you know, Claude code coding tool to develop your own software.

也许 Palantir 会像其他软件公司一样，面临越来越大的价格压力。

Maybe Palantir will come under the same pricing pressure that we've seen other software firms come under.

很好。

Great.

Laura，非常感谢你今天来参加对话。

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