Meta Debuts Muse Spark 1.1, Blue Origin to Raise $10B, Cursor Develops AI Agent

欢迎各位收看《The Information》的TITV节目。

Welcome everyone to The Information's TITV.

我是阿卡什·帕什里卡。

My name is Akash Pasricha.

今天是6月10日，星期五。

It is Friday, June 10.

在开始今天的节目之前，韩国芯片厂商SK海力士今天将在纳斯达克上市，这是它在美国的首次亮相。

Before we get to the show today, South Korean chipmaker SK Hynix is set to begin trading on the Nasdaq today in its US debut.

这将成为投资者如何看待内存芯片短缺行情的一个重要风向标。

It is a new major proxy for how investors are thinking about the memory chip crunch.

关于这件事的更多报道，我们会在官网发布，欢迎大家关注。

We'll have more coverage of that on our website, so check it out.

今天节目中，我们要深入聊聊Meta最新发布的AI模型更新。

Today on the show, we're going to unpack Meta's latest AI model update.

Muse Spark 1.1。

Muse Spark 1.1.

初步评价不错，但够不够强呢？

Early reviews seem strong, but will it be strong enough?

我们今天就来一探究竟。

We'll get to the bottom of that.

我们还会在本周的《编辑精选》里，详细分析这周发布的所有新模型的定价策略。

We'll also break down the pricing angle of all the new models that were released this week in our weekly edition of the Editor's Cut.

我们也会深入了解一下蓝色起源的新一轮融资，以及这对太空竞赛意味着什么。

We're also gonna understand more about Blue Origin's new funding round and what that means for the space race.

节目最后，我们会独家披露 Cursor 如何在开发自己的 AI 代理，直接对标 Anthropic 的 Claude Cowork。

And we'll close out the show with our exclusive reporting on how Cursor is developing its own AI agent to compete directly with Anthropics Claude Cowork.

今天内容很精彩，咱们马上开始。

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

Meta 发布了 Muse Spark 1.1，这是专为复杂编程和自主任务优化的升级版 AI 模型。

Meta announced the release of Muse Spark 1.1, its upgraded AI model made for complex coding and autonomous tasks.

之前这家公司总被觉得在模型竞赛中落了后，但马克·扎克伯格已经大力投入，想扭转这个局面。

The company has seen the company has been seen as a bit of a laggard in the model race, but Mark Zuckerberg has invested heavily into fixing that.

现在加入我们讨论这些的是公民银行的董事总经理安德鲁·布恩。

Joining me now to discuss all this is Andrew Boone, a managing director at Citizens.

安德鲁，欢迎再次做客我们的节目。

Andrew, welcome back to the show.

很高兴你能来。

It's great to have you here.

很好。

Great.

谢谢邀请我。

Thanks for having me.

好的。

All right.

那我们来聊聊 Muse Spark 吧。

So let's talk about Muse Spark.

你对它有什么看法？

What were your reactions to it?

我认为这标志着未来五年对Meta的认知发生了一个质的飞跃。

I think it's a step function change in terms of how you think about Meta over the next kind of five years.

昨晚有报道称，Meta打算把它的前沿模型API当作一个独立业务来运营，实现自给自足。

Last night, I think there was reporting out that Meta wants to treat its API of its frontier models as a standalone business that can support itself.

如果你想想成为前沿实验室需要哪些基础条件——能源、数据中心、令牌，然后这些最终能催生出什么样的智能，其实就清楚了。

If you think about that and, you know, kind of what are the building blocks for becoming a frontier lab, right, it's really energy, data centers, tokens, and then kind of what does that create in terms of intelligence.

Meta已经具备了实现这一目标的所有基础条件。

I think Meta has the building blocks to be able to go there.

关键在于他们能创造出多高水平的智能，以及这种水平跟OpenAI或Anthropic相比会是什么样子。

It's just a question of what level of intelligence are they able to create and what does that look like versus an OpenAI or Anthropic.

在这方面，Meta已经做得相当不错，正在努力缩小差距。

To that level, Meta has done a really good job of trying to catch up here.

他们大大加快了前沿模型的发布节奏。

They've accelerated what is the release schedule of their frontier models.

我觉得昨晚的更新信息相当到位。

I think last night was a pretty good update.

总的来说，我们对Spark的前景非常看好，这很可能为Meta开辟一条全新的业务线。

Overall, I think we're pretty positive about what Spark looks like and really the creation of what could be a new potential business line for Meta overall.

这挺有意思的，因为Meta多年来尝试过各种方式进军企业市场，但效果一直没达到大家预期。

This is kind of interesting because Meta has tried for many years and many different ways to develop an enterprise business, and it hasn't quite caught on, I think, the way many would have hoped.

我记得有个叫Clara的高管，当初就是来专门做企业业务的。

I'm remembering Clara, she was an executive that came in and tried to build an enterprise business that way.

所以这次听起来，把Meta当作广告以外的公司来卖，希望比以往任何时候都大。

So this sounds like there's more hope than ever for selling the business as something other than advertising.

是的。

Yeah.

咱们倒过来看一下。

I mean, let's go backwards.

他们收购了一家客户关系管理软件公司。

So they acquired a CRM that was customer.

他们还推出了AI商业助手这类产品。

They've done things like AI business agents.

我觉得他们给中小企业提供的这些工具本身是有道理的，但本质上还是为了推动广告业务。

I think all of the tools that they're offering to SMBs, it it makes sense, but it's all in in kind of it's to help the advertising business.

如果仔细想想，这其实是个开发者工具。

And so, if I think about what this is, I mean, this is a developer tool.

Meta现在瞄准的是一个完全不同的市场，过去企业服务和广告一直是绑在一起的。

Like, this is a whole different segment that Meta is going after, where enterprise used to be hand in hand with advertising.

我不是说否定，我觉得Spark最终还是会演变成专门服务于广告的AI模型，而AI整体的进步也确实会利好广告业务。

Don't get me wrong, I do think that Spark will get distilled down into advertising specific models, and the benefit of advancements in AI overall will help the advertising business.

但眼下Meta正在打造的，是一整套完全不同的产品，光是这一点就非常有意思。

But this is an entirely different product set in terms of what's Meta building, and that in and of itself is just incredibly interesting.

我想问你，你觉得这种向企业市场拓展的策略明智吗？因为一年前，马克·扎克伯格还说他的愿景是‘个人超级智能’，那更多是面向消费者的设想。

So, I want to ask you about if you think this is a smart strategy, because AI and Meta, if you go back maybe a year ago, Mark Zuckerberg, he talked about personal super intelligence as his vision for how AI makes its way into daily life, and that was more of a consumer vision, I think, that he was painting.

不管我们有没有真正实现个人超级智能，智能眼镜确实在不断进步，Ray-Ban那种显示设备未来只会越来越好。

Whether or not we've gotten to personal super intelligence, the glasses are getting better, there's no doubt about it, and the display Ray Bans, things will get better over time.

但在个人超级智能还在研发阶段的时候，同时进军企业市场，你觉得这步棋走得对吗？

But do you think that it's smart to diversify here with enterprise while the personal superintelligence thing is still being developed?

我觉得这能分散风险。

I think it diversifies the bet.

所以如果你想想Meta提供模型API意味着什么，这里面其实有不少战略上的退出路径。

And so there's a bunch of strategic kind of off ramps if you think about what the sale of an API for Meta's models means.

一方面，这是否意味着Meta可以更多地投资在数据中心、能源协议等方面，因为他们有了另一个可以利用算力的渠道。

And so one is, does this mean that Meta can then invest more in terms of data centers, energy agreements, everything else because they have another outlet in terms of what they could utilize the compute for.

我觉得这是其一。

I think that's one.

其二是，能不能把训练模型的成本分摊到更多产品上？

Two of which is, can you amortize the cost of training models over more products?

我觉得这也是对的。

I think that's also true.

对吧？

Right?

我们之前聊过广告的事。

We talked about advertising earlier.

我 definitely 认为，未来会有一些元模型通过眼镜分发出去。

I definitely think there's gonna be some form of meta models that get distributed via the glasses.

这个话题可以聊聊，但它可能就是下一个计算平台。

Can talk about that, but that could be the next compute platform.

你想想 OpenAI 这周发布的语音功能，语音交互一直在变强。

If you think about what OpenAI released this week in terms of voice, voice continues to get better.

现在这种交互方式，是不是已经算得上更成熟的用户体验了？尤其是当 Agendaq 作为后台真正执行动作的时候。

Is that now more of a capable UX in terms of consumers interacting with models in a new way with Agendaq as the back end that really implements what is the action?

所有这些都特别有意思。

Like, all of that's really interesting.

但关键在于，现在你有了更多产品来分摊这些成本。

But the idea here, big picture, is that you now have a wider array of products to displace that cost.

对吧？

Right?

如果训练模型现在动不动就要几十亿美元，那为什么不多卖一个产品，把成本分摊出去呢？

And so if training models is now billions of dollars, why wouldn't you sell one more product that you can allocate expenses to?

你有没有估算过，这个API业务对公司来说可能有多大？

Have you modeled out how how big this API business could be for the company?

我真的说不准。

I I truly don't know.

对吧？

Right?

这得看开发者工具最终会变成什么样。

Because it's gonna be a question of kind of, like, what developer tools look like.

现在我们还处于私密预览阶段。

And right now, we're in private preview.

对吧？

Right?

所以现在还太早了。

And so, like, it's it's very early.

咱们来聊聊，你可能会让团队回去做的那个分析工作？

Let's talk about the an Is exercise that you you might ask your team to to to go back and do?

我的意思是，这个业务真能做大到需要我们详细测算它对公司总收入的影响吗？

Mean, is is this gonna be that big a business that you're gonna need to model out sort of the that that impact on the top line for the company?

是的。

Yeah.

我们昨晚其实做了一个简化测算。

I mean, we we took a cut out last night.

对吧？

Right?

就是简单假设一下，如果Anthropic的收入达到300亿美元，而且保持大约十倍的增长速度，那会是什么样子？

And it was as simplistic as just saying if Anthropic does, you know, 30,000,000,000 and they're on a pace of what's been basically 10 x growth, like, what does this look like?

Meta能不能占到Anthropic收入的5%？

Can Meta be 5% of Anthropic revenue?

对吧？

Right?

光是这5%就已经是个百亿级别的生意了。

Like that five percent's you know, that's a billion dollar plus business right there.

没错。

Right.

我觉得这并不夸张，尤其是Meta现在这么激进的定价策略。

I don't think that that is that much of a stretch and especially with the aggressive pricing that Meta is throwing out there.

但要想在开发者调用API的份额上胜出，你真得站在智能技术的最前沿。

That being said, you really wanna be at the frontier of intelligence to win what is the the share of API calls that are being placed by developers.

Meta的智能水平正在提升，这进一步印证了超级智能实验室的方向是对的。

And so, like, the trajectory of Meta's intelligence is improving, and this was a really good reinforced view that Superintelligence Labs is in the right direction.

不过，要预测2028年市场份额会是什么样，真的很难，毕竟两年后模型会发展到什么地步，现在谁也说不准。

But it's really hard to tell you what share going to look like in 2028 as we think about where the models end up in two years, right?

这个市场变得太快了。

The market's just moving way too fast.

这件事还有另一面：随着这些模型陆续发布，我们做过一些调查，发现员工的屏幕监控、数据采集这些事，其实被大量用于模型训练。

There's another side to this story, which is that as the models have been released, we've done some reporting on the extent to which employees have their screens being recorded and stuff like that, the extent to which that's being used to train the models.

大家心里都清楚，这部分数据至少占了训练数据的相当大一部分。

I think it's sort of understood that that was at least a decent sized chunk of the training that happened.

还有一个说法是，如果你在Instagram上有公开资料，就得主动选择退出，不然你的内容就会被用来生成AI内容。

There's also the narrative here that you have to opt out, I think, on Instagram of having if you have a public profile, you have to opt out of having your content being used to create AI generated content.

我想问的是，Meta采取的这种做法，背后到底意味着什么？

And you know, I guess the broader question I have on all this is this is the approach that Meta has taken.

他们确实非常激进地在提升自己的工具性能。

They have taken a very aggressive approach to making their tools better.

你觉得这种做法最终会不会反噬他们？

Do you see this coming back to bite them at all in in any way?

这难道就是我们现在玩的这个游戏吗？

Is this just the game that we're in?

你怎么看？

What do you think?

我根本不觉得普通用户会在意数据隐私问题。

I I fundamentally don't think that consumers care about data privacy.

从用户的行为来看，我根本没看到有人会因为数据问题就停用某个服务。

I just don't think that you've seen any sort of inclination across consumer behavior that tells me that they're gonna turn off XYZ service because of data.

确实有一小部分人对此特别大声抱怨，但总体来看，我觉得这根本不是个问题。

There's a small handful of people that are very vocal about that, but at large, I just don't think that's a thing.

不过咱们先退一步，想想像Meta或者YouTube背后的Google这样的媒体公司，到底拥有什么核心资产，再对比一下OpenAI、Anthropic这些公司。

Let's step back though and talk about what is the asset of what is a media company like Meta or honestly Google with YouTube, and think about that within the context of OpenAI, Anthropic, and others.

对吧？

Right?

由于用户群体的独特性，这两家公司手里的数据集其实非常特别，值得深入思考如何用这些数据来训练模型。

There's a really interesting dataset that ends up being unique to both of those players because of the user base they have and how to actually think about training models with that information.

所以，咱们再把话题拉得更广一点。

And so, you know, again, let's let's bring this out more broad.

这其实凸显了一个关键问题：Reddit的数据到底有什么独特之处？今年下半年他们的数据授权就要到期了，我们该怎么看待这种专有数据的价值？

It really highlights what is the uniqueness of kind of Reddit's data who should be up for renewal in the back half of this year as we extend out, like, what is the thought of proprietary data and how important is it?

我觉得这非常重要，因为所有这些AI实验室都在拼命收集一切能拿到的数据，只为了提升自己的模型。

I think it's pretty important as all of these labs try to suck up everything they possibly can to improve their models.

咱们先说说Google吧。

Let's go to Google though for a second.

是你先提到他们的。

You brought them up.

他们在模型竞赛中似乎稍微落后的，但他们在应用层面有不少优势。

They seem to have fallen behind a little bit in the model race, and they have got a lot of stuff going for them with the application play.

那个编程模型，我觉得大家对他们在编程领域的表现有点失望。

The coding model, I guess people were a little bit underwhelmed, I guess, by where they're sit in the coding race.

但我预测，他们会强势回归。

But my prediction is that they're going to come back in a big way.

我们说的可是几个月内的巨大转变。

Look, we're talking months of step change here.

我今天说的这些，一个月后可能就不一样了，所以他们也没落后太多。

What I was saying today will change a month from now, and so it's not like they're terribly behind.

我觉得他们在模型竞争中会强势反弹。

I think they're going to come back in a big way in the model wars here.

这就是我的判断。

That's my prediction.

你觉得呢？

What do you think?

普遍的预期是，也就是我们接触的市场各方，大家觉得下个模型会在今天之后一周发布。

The broad expectation, right, of kind of like the market, everyone that we're talking to, it seems like everyone else that's out there, is that the next model gets released a week from today.

对吧？

Right?

所以感觉下星期就要出来了。

So it feels like it's coming next week.

关键问题是，谷歌这次能带来多大的突破？

And the question is gonna be, what is the step function change for Google?

短期内看股价波动确实有点荒谬，但市场已经因为Meta推出了一款明显更强大的模型，给它的股价带来了大约10个百分点的上涨。

Again, trading is somewhat silly to look at on a near term basis, but like market has rewarded Meta by about you know, 10 percentage points of of stock price appreciation just from what is a updated model that appears to be a significant improvement.

我觉得谷歌也遭到了市场的类似惩罚，因为大家普遍觉得Gemini的发展慢下来了。

I think Google has been public punished similarly by market participants because there's been a view that Gemini is is slowing down.

确实已经有人离开了。

You've definitely seen some people leave.

你提到的那些人离开DeepMind的事，应该就是你们这边的反馈吧。

Think that's your guys' reporting in terms of people stepping away from DeepMind.

是的。

Yeah.

但总体来看，别小看了谷歌。

But overall, you can't count Google out.

他们拥有海量的计算资源。

They have so much compute.

还有数据。

Have the data.

没错，他们也有数据。

Mean, they have the data too.

就像你说的，YouTube就是个例子。

Like you said, I mean, YouTube Exactly.

他们什么都有。

They they have it all there.

说实话，这个问题我真不知道答案。

And I've actually I I don't know the answer to this.

我倒是想过谁在数据上更有优势，是Meta还是YouTube，毕竟它们手里的数据量太大了。

I've thought a little bit about who has the upper hand in data, whether it's Meta or YouTube, because they they have so much at their fingertips.

我感觉还是谷歌占上风，不过你怎么看？

My hunch says it's still Google, but I mean, what do you think?

我觉得肯定是谷歌。

I I would just assume that it's Google.

对吧？

Right?

YouTube简直就是个巨大的数据宝库。

I mean, YouTube is just this tremendous database.

我在想以前有个数据说，每分钟上传的视频时长高达五亿小时之类的。

I'm trying to think of what the old stat was that it's like, you know, five hundred million hours of video is uploaded every minute or something.

是的。

Yeah.

这数字简直太惊人了。

It's it's something insane.

对。

Yeah.

我刚才在想，Instagram 和 Facebook 上到底有多少用户。

Was I was trying to think about how many how many people are on Instagram, Facebook.

不过其实也不太重要，毕竟还有私密账号之类的。

I guess they doesn't that matters, you know, private accounts and stuff like that.

所以也许吧。

So maybe it's

但更重要的其实不是用户数量，而是内容本身。

not And it's less about that as much as it is about the actual content itself.

对吧？

Right?

我觉得 YouTube 上的内容特别有价值，比如怎么解决同步问题这类实用内容。

And I I would argue that I think the content within YouTube where it's like, how do I fix a sync is very valuable.

对吧？

Right?

没错。

Right.

所以你有一个专门针对某个主题的视频。

So like you have something that is a dedicated video that is pointed at a subject.

我觉得这和我们在Instagram上看到的那种社交内容不太一样，那种内容我们虽然很喜欢、也很互动。

I think that's a little bit different than the social media content that we very much enjoy and engage with that ends up being on Instagram.

对。

Right.

我最后问你一个问题。

Last question for you.

所以，考虑到这些，你认为Meta会继续以他们目前的投入速度增加资本支出吗？

So, with all this in mind, are you expecting Meta to continue increasing CapEx at the rate that they've been investing in?

我们昨晚上调了资本支出的预测。

We took up our CapEx numbers last night.

所以如果

So if

多少呢？

anything How much?

到2027年，我们的资本支出将达到两千亿美元。

We're now at 200,000,000,000 for 2027.

你去跟那些买方投资者聊聊，看看其他人是怎么看的。

And so you talk to kind of buy side investors, you talk to kind of where everyone else is.

我觉得2027年的上限可能在22.5亿美元左右，甚至可能达到25亿美元。

I I think the upper range is now probably $2.25, maybe $2.50 in terms of what 2027 could look like.

Meta还在投资租赁项目。

This is Meta is also investing in leases.

对吧？

Right?

这些支出属于表外投资，不算在资本支出里。

And so that's off balance sheet, non CapEx type investment spend.

重点是，这规模太庞大了。

Point being is it's massive.

你想想，明年预计要新增的七千兆瓦，对吧？

You you got the the I think it was the seven gigawatts that are expected now for next year.

对。

Right?

如果你粗略算一下，每千兆瓦投入五十亿美元，

If you've just ballpark that and you say, hey, look, it's 50,000,000,000 a gigawatt that you end up putting in.

但这不是一年内完成的。

Again, that's not in a one year period.

我们说的是巨额资金，这些资金正在转化为产能，未来不仅能推动广告业务，还能促进API销售。

But we're talking about a lot of money that, again, is being turned into capacity that should be able to help fuel not only advertising business, but also API sales in the future.

所以我不确定2027年是不是峰值年，对吧？

So I don't know if '27 is the peak year, right?

这可是天文数字的资金。

Like, that's a lot of money.

到2027年，我们的自由现金流已经变成负数了。

And we're now at a negative free cash flow kind of number for '27.

大概会降到四千亿左右。

Like, we're it gets it's down $40,000,000,000 give or take.

是的。

Yeah.

那

How

这个速度是多少呢？

much Is do that the rate

到2028年会怎么变化？

move to '28?

我不清楚。

I don't know.

这个下降速度，是比其他超大规模云服务商更快还是更慢？

Is the rate of that is that faster or slower compared to the other hyperscalers?

Meta算不上超大规模云服务商，但我们可以说

Meta is not a hyperscaler, but we can

你看，亚马逊2026年的支出大概有两千亿美元。

see I on mean, Amazon is 200,000,000,000 for 2026.

谷歌2026年差不多是1850亿。

Google is like 185 for '26.

所以现在Meta在2026年和2027年的水平上，跟他们比起来差距不小。

And so we're at Meta now being at twenty twenty six levels and twenty twenty seven.

我觉得亚马逊和谷歌明年肯定得上调他们的资本支出预期。

I can't see any way that you don't have Amazon and Google both tick up their CapEx expectations for next year as well.

对吧？

Right?

AI领域的资本支出周期现在还远没结束。

Like, the AI CapEx cycle is very much still alive.

而且你看，Grok这周还发布了新模型，性能不错，成本还低。

And if anything, see Grok also put out, like, a model this week that's good and cheap.

我觉得现在有五家超大规模云服务商都在全力投入AI前沿模型的竞争。

I think you have five hyperscalers that are all going full bore in terms of the AI frontier model race.

这意味着他们的支出还会继续上涨。

And that means spend is going higher.

很好。

Great.

非常感谢你今天来做客，安德鲁。

Well, I wanna thank you for coming on, Andrew.

这位是公民银行的董事总经理安德鲁·布恩，正在TITV为您直播。

That is Andrew Boone, managing director at Citizens here on TITV.

这周发布新模型的不止Meta一家。

Meta wasn't the only company to release a new model this week.

OpenAI和xAI也推出了重磅新品。

OpenAI and xAI also had big releases.

价格是很多人关注的关键差异点，尤其是在这些模型之间。

Pricing is a big differentiating factor a lot of people have been paying attention to with these models.

关于这一点，我们请来驻旧金山的首席记者杰森·迪恩和专题编辑尼克·温菲尔德，为大家带来本期《编辑精选》。

For more on that, I wanna bring on our San Francisco bureau chief, Jason Dean, and features editor Nick Wingfield for this week's edition of The Editor's Cut.

欢迎两位。

Welcome to you both.

很高兴你们能来。

It's great to have you here.

能来真是太好了。

Good to be here.

好的。

Okay.

杰森，我先从你开始。

Jason, I wanna start with you.

你对这周这些模型的发布有什么看法？

So what did you make of all the model releases this week?

你最关注哪些方面？

What were you paying attention to?

嗯，这周有不少动静，但最值得注意的是，价格竞争明显加剧了。

Well, a bunch of stuff, but I would say one of the more interesting facts is the price competition is clearly intensifying.

比如GBT 5.6这个模型，按每个token算，价格大概只有原来的一半，具体要看是输入还是输出。

So, the the GBT 5.6 model, is about half on a per token basis, depending if you're talking about input or output.

还有独立分析指出，如果按性能来算，成本甚至只有三分之一左右。

And there's some independent analysis that puts the cost sort of on a performance basis at about a third.

这跟Anthropic的Fable 5相比是这样。

That's both relative to, to Fable five from Anthropic.

另外，Meta这次也第一次推出了面向开发者的模型API。

And then you have, Meta coming in for the first time offering an API for developers for its models.

以前他们一直只做开源。

You know, it was doing all open source.

现在也开始想靠让开发者用他们的模型来收费了。

Now it's trying to get in this game of charging people to develop using its models.

我们还没看到对这个模型的详细分析，不过扎克伯格已经放话了，他们在定价上一向很激进，主打价格竞争。

And, we haven't seen the same levels of analysis of the model yet, but, Zuckerberg came out and said, they tend to be very aggressive on price and and compete there.

所以我觉得，对Anthropic来说，这还构不成大问题，毕竟现在需求依然旺盛，尤其是前沿领域，很多人想用最新的模型做开发和其他工作。

So I think it is it's you know, this is not like a a huge problem for Anthropic yet because they still have so much demand and people there are there's a lot of demand at the cutting edge, people wanting to use the bleeding edge models and coding in other areas.

但从中期到长期来看，价格战只会越来越激烈。

But medium to long term, think that the the price wars are only gonna get more intense.

没错。

Right.

尼克，这周有什么让你特别留意的吗？

Nick, what stood out to you this week?

有几件事。

A couple things.

第一，我总觉得我们现在正处在Uber上市前的那个阶段，那时候它靠补贴打车服务来抢出租车的生意。

The first is I really feel like we're in that phase before Uber went public when it felt like the company was subsidizing ride shares in order to win business from big taxi.

你还记得那时候Uber打车便宜得离谱吗？

Do you remember those days when Ubers were

对。

That

现在这种API定价方式，感觉就跟当年Uber上市前的阶段很像。

kind of feels like the phase we're in right now potentially with this API pricing.

现在打Uber根本没有任何折扣了。

And now there's absolutely no discount at all to get an Uber.

贵得离谱，简直是全世界最贵的服务。

It's like the most expensive thing in the world.

对，完全没错。

Yeah, totally.

另一点是，Meta现在正陷入一场信息战。

The other thing is this is kind of an information war that Meta is in.

再没有什么比扎克伯格这几年打破在Twitter上的沉默、亲自发帖推广这个模型更能说明问题了——我们现在都叫它X了。这明显是在向意见领袖传递信号，也就是那些股东们，当他们试图从OpenAI和Anthropic手里抢市场时，必须争取他们的支持。

Nothing speaks to the importance of influencing kind of a lead opinion on this than Zuckerberg breaking his Twitter fast over the past few years and actually posting and promoting this model on Twitter, or X we're calling it now, it's really clear that they need to send a message to influencers, let's call them shareholders, as they embark on this journey of trying to win business from OpenAI and Anthropic.

他们得塑造公众舆论，尤其是精英阶层的看法，因为这场仗会打得极其昂贵。

They need to shape public opinion, elite opinion, because it's going to be a very, very expensive battle.

没错。

Right.

说到X平台，尼克，这周埃隆在X上看起来态度挺支持的。

Well, speaking of X, Nick, Elon was in a pretty supportive mood this week on X, it seemed.

他还在夸Anthropic呢。

I mean, he was praising Anthropic.

然后扎克伯格发帖的时候，你说，他不推广还能咋办？

And then when Zuckerberg tweeted, you know, I think he was I mean, he basically, why wouldn't you promote it?

他提到马克·扎克伯格也在这个平台上。

You know, saying Mark Zuckerberg's on the platform.

但这周他对那些朋友的态度明显更支持了，这可不像是他平时的风格。

But he very much taking a more supportive tone this week in terms of all of his buddies, which is sometimes not the tone that he takes.

对。

Right.

Anthropic现在是客户了吧？

Well, Anthropic is a customer now, right?

或者说是SpaceX的客户。

Or a client of SpaceX.

从这个角度出发，你倒是能说得通。

You can kind of explain that from that position.

不过呢，也许他最近不知怎么的，心态变得宽厚多了。

But yeah, maybe he's feeling in a more magnanimous mood now for some reason.

是的。

Yeah.

杰森，我回头再跟你聊。

Jason, I want to come back to you here.

在你之前，我们刚和公民组织的安德鲁·布恩聊过，谈到了Meta现在推行的企业战略。

So we were talking with Andrew Boone from Citizens before you guys about the enterprise strategy that Meta is going for here.

过去，Meta在广告之外，一直很难真正把企业业务搞起来。

Meta has struggled in the past to really turbocharge its enterprise business outside of selling ads, really.

API业务倒是个挺有意思的突破口。

The API business is an interesting opportunity here.

你觉得这能成为公司一项重要的收入来源吗？

Do you think that this becomes a significant revenue business for the company?

这终于会是他们一直期待的突破吗？

Is this finally the big hit that they were waiting for?

我觉得这里有很多理由值得怀疑。

I think there's plenty of reason for skepticism there.

我的意思是，

I mean,

这根本就不像是他们的基因里会有的东西。

it just doesn't seem to be in the DNA.

当然，也不是说人就不能重新塑造自己。

And, you know, that doesn't mean people can't reinvent themselves, I suppose.

但他们过去在企业领域，还有虚拟现实等方面，已经试过很多次了。

But they've been trying, in enterprise and in other areas with, virtual reality.

结果呢，这些都成了偏离核心广告业务的昂贵分心事，而广告业务本身非常赚钱，还正从人工智能中获益良多。

And these have proved, like, costly distractions from the core advertising business, which is a great business and and is benefiting a lot from AI.

至于要跟Anthropic和OpenAI去竞争企业级AI产品和模型，还非得站在技术最前沿，我觉得这根本没什么道理。

I I just you know, to try to compete with Anthropic and OpenAI in enterprise AI products and models and stay at the bleeding edge there doesn't make a whole lot of sense to me.

我觉得，想成为云服务提供商——这也属于外界传闻和报道中他们对企业业务的野心之一——同样有点不切实际。

I think the idea of becoming a cloud provider, which is also part of the sort of rumored and reported enterprise ambitions there, also seems like a stretch.

不过，埃隆·马斯克已经证明了，你不需要自己建云，也能做成大生意，比如把显卡租给Anthropic和其他急需算力的公司，而不必在公司内部搞一套完整的enterprise基础设施。

That said, Elon Musk has sort of demonstrated that you can do big deals without for for cloud, you know, and rent out your your GPUs to Anthropic and and others who are desperate for capacity right now without building out a whole sort of enterprise infrastructure inside your company.

所以，或许他们能抓住一个机会，短期内靠闲置算力赚上几十亿美元，毕竟这些算力还没完全用起来，谁都知道他们在砸钱扩建计算能力，说不定建得比用得还快。

So perhaps there's an opportunity there for sort of short term billions of dollars in revenue, or excess capacity that they haven't grown into yet because the Lord knows they're spending lots and lots of money to build out their compute and maybe building that faster than they can use it.

但要说Meta真能转型成一家有分量的企业级公司，我觉得还是太勉强了。

But the idea of, like, Meta turning into a serious enterprise player seems like a stretch to me.

明白了。

Okay.

尼克，你对这个看法有什么不同意见吗？

Nick, do you have an alternate opinion on that?

我其实对这个方向更乐观一点。

I I'm actually a little more bullish on this.

我觉得归根结底，Meta手上有这么多人才，对吧？

I think that we are just look, at the end of the day, Meta has all this talent, right?

他们只差一个模型，就能让所有人都想用Meta的模型，尤其是在编程领域。

We are one model away from everyone wanting to use Meta's model, certainly in coding.

我觉得他们做得到。

I think they could do it.

尼克，你怎么看？

Nick, what do you think?

明白了。

Okay.

任何一家公司，尤其是大型科技公司，都面临单一技能的瓶颈问题。

Well, there's a real one trick pony issue when it comes to any company, any big tech company.

亚马逊很多年前就证明了，他们能从消费业务拓展到企业市场，推出AWS。

Amazon proved many years ago that they could branch out from the consumer business, get into enterprise with AWS.

我们当时都觉得那是特例。

We all thought that was a one off situation.

后来谷歌经过多年努力，也成功做到了。

Then along comes Google after years and years of trying, and they managed to pull it off too.

所以我觉得这对Meta来说并不是什么不可能做到的事，只是没那么简单。

So I don't think there's anything inherently impossible about this for Meta, but it's not easy.

但如果你是马克·扎克伯格，你大概会看看你的广告业务最强劲的竞争对手Google在云计算方面取得的成就，然后心想：我也能做到。

But if you're Mark Zuckerberg, you probably look at what Google, which is your closest peer in the advertising business has done in terms of Google Cloud and say, I can do that too.

没错。

Right.

杰森，我们回到你刚才的问题。嗯，好吧。

Jason, coming back to you, mean, was well, okay.

你刚才想说什么

What were you going

来着？

to say?

我

Go I

觉得这话说得挺在理，不过尼克，你刚才也提到了，目前只有这两家公司成功做到了这一点。

think it's a fair point, but Nick, you also just mentioned the only two companies that have pulled that off.

微软长期以来一直想同时做消费级和企业级业务，但基本上没成功。

Microsoft's been trying to be both a consumer and enterprise company for a long, long time now and is basically failing.

历史上，他们在消费端唯一真正成功的只有Xbox，可现在这个项目也乱成一团，他们还不知道该怎么收拾。

The only real success in the consumer side historically was Xbox, and that is now a total mess that they're trying to figure out what to do with.

我觉得谷歌这家公司一直以来都比较擅长尝试各种不同的方向。

I think also Google just as a company has a longer history of being fairly effective at trying a lot of different things.

当然，他们的核心业务和赚钱的主要来源毫无疑问是广告。

I mean, obviously, their core business and the thing that made them money has been advertising for sure.

但你看看他们其他那些尝试，很多都没成功，但也有一些确实成了。

But if you look at all of the other bets, many of which didn't pan out, but some of which did.

比如Waymo，至少在技术和市场推广上已经取得了成功。

Waymo, for example, panned out at least in terms of of technological and market, distribution success.

当然，它现在还不能赚钱。

It isn't yet a moneymaker, obviously.

所以，我觉得，是的，这事儿是有可能的。

So, I I think, yes, it is possible.

至于Meta，这么多年来我们根本没看到它有这种能力的迹象。

I don't know that we've seen anything in the rather long history now of Meta to suggest that it has that ability.

而且，这一切在多大程度上是一种干扰呢？

And also, to what degree is all this a distraction?

我说的不是开发模型这件事。

And I'm not talking about building models.

我觉得开发模型确实有它的价值，但要是想全面进军企业市场，搞销售、客户服务、市场推广这一套，而不是专注在核心广告业务上，那就本末倒置了——广告的机会可是在不断增长的。

I think there is a utility to building model, but to try to be a player and build out all of the things that go with an enterprise business in terms of sales and customer service and go to market, etcetera, as opposed to focusing on the benefits you can reap in the core advertising business, which is only going you know, the opportunity there is only growing.

考虑到他们过去在这些领域从未证明过自己能成功，把资源投到这些事情上，机会成本可能很高。

You know, there may be an opportunity cost in in the resources they're putting into those other things given their lack of a track record in proving success.

尼克，我想聊一下这周的另一则新闻。

Nick, I want to turn to another story this week.

蓝色起源这次首次融资，而且不是小数目。

So Blue Origin is raising money for the first time, and it's not a small funding round.

他们估值1300亿美元，融资额高达100亿美元。

It's $10,000,000,000 at a $130,000,000,000 valuation.

我觉得那是在融资之前的事。

I believe that's before the fundraise.

关于这一轮融资是怎么推进的，我们了解多少？

What do we know about how this funding round is coming together?

我报道的是，其中40亿美元来自CO2，这是他们首次引入外部投资。

Well, what I reported is $4,000,000,000 of that is coming from CO2, so that's their first outside investment.

当然，过去25年来，他们一直从家族内部筹钱。

They have, of course, been raising money throughout their history, their twenty five year history from their family.

二十五年？

Twenty five?

我都没意识到他们已经这么久了。

I didn't even realize it was that old.

天啊，Blue Origin居然已经运作了二十五年？

Man, twenty five years Blue Origin has been kicking it?

快二十六年了。

Almost twenty six.

整整二十五年了。

Quarter of a century.

二十五年前，他们就开始搞太空火箭了吗？

Was it always space rockets leaving 25 ago?

是的，没错。

Yeah, yeah.

很多人忘了，其实他们在SpaceX之前就搞出了可重复使用的火箭。

People forget this, but they actually had a reusable rocket before SpaceX did.

他们之前已经成功演示过了。

They actually demonstrated it before.

这两家公司，这二十五年来一直在坚持干。

These two companies have been kicking it around for a quarter of a century.

这整个期间，杰夫·贝索斯一直在掏钱支持。

Jeff Bezos that whole time was funding it.

早期的时候，投入没后来那么大，但具体贝索斯这几十年到底花了多少，外界其实不清楚，但肯定是以十亿甚至上百亿计的，说不定吧。

Now in the early days, was not as expensive to fund as it later became, but we don't really know how much Bezos has spent over the history of it, but certainly billions, tens of billions, I don't know, possibly.

但终究有一天，Origin 必须成为一个有收入、最终有利润的真正企业。

But at some point, Origin has to become a real company with revenue and ultimately with profits.

我觉得我们现在离这个目标越来越近了，这也就是他们现在要融资的原因。

I think we are getting closer to that now, and that's why they're raising money.

当然，这里还有其他一些因素。

I mean, there's some other factors here as well.

SpaceX 的上市显然已经向投资者展示了太空领域的巨大潜力，引发了热烈反响。

The SpaceX IPO clearly has shown and generated a lot of excitement among investors for space.

这对 Blue Origin 融资来说，应该也是一股顺风。

That probably is a tailwind for Blue Origin in terms of raising money.

他们手头还有几个合同，比如那个叫 TerraWave 的太空数据中心业务，他们正努力推进。

They also have contracts and they've got a data center in space business called TerraWave, I think, that they're trying to push along.

你知道，贝索斯大概也不想一直自己掏钱补贴了——他到现在已经投了二十亿美元，这点要说明白，但他并没有打算撤资。

You know, Bezos probably wants to stop subsidizing this to the same He's putting $2,000,000,000 in this, to be clear, so he's not backing away from funding it.

所以杰森，你觉得他们未来几年会考虑上市吗？

So Jason, I mean, what do you think is do you think they try to go public in the next couple years?

这就是最终目标吗？

Is that the end game here?

你怎么看？

What do you think?

我觉得这个可能性确实存在。

I can certainly see that possibility.

依我看，就目前而言，这一轮融资完全合理，正如尼克说的那些原因。

I mean, I think for now, this funding round makes a lot of sense for the reasons that Nick said.

不过他们要想追上SpaceX，还有很长的路要走。

They have a long way to go though to catch up to where SpaceX is.

说真的，这挺让人惊讶的。

I mean, it's striking.

尼克刚才提到，他们可是第一个造出可重复使用火箭的。

Nick was recounting, you know, they were first to build the reusable rocket.

他们已经存在二十五年了。

They've been around for a quarter century.