Microsoft’s New Coding Model, Apple’s New Push for AI on Devices, Snowflakes Shares Jump

欢迎各位收看《信息》频道的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5月28日，星期四。

It is Thursday, May 28 today.

首先，我们将通过一场圆桌讨论，深入分析Snowflake和Salesforce的季度财报。

First up, we will break down Snowflake and Salesforce's quarterly results with a panel discussion.

我们还会探讨Meta在AI订阅服务上的布局，以及《信息》独家报道的——该公司正成立一个新团队，专门推动企业采用其AI工具。

We're also going to dive into Meta's push into AI subscriptions alongside the information's exclusive reporting that the company is launching a new unit focused on driving business adoption of its AI tools.

接下来，我会请来我们的苹果记者，聊聊他独家报道的内容，包括苹果重新发力本地设备端AI，而不是依赖云端。

I'll then bring on our Apple reporter to talk more about his exclusive reporting, Apple's renewed push for AI that runs on devices instead of the cloud, and we'll wrap with a look at Microsoft's Build Conference happening next week and the big announcements that the firm has planned.

最后，我们还会展望下周即将召开的微软Build大会，看看这家公司准备了哪些重磅消息。

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

节目内容很精彩，咱们马上开始。

Snowflake shares jumped yesterday after the company exceeded its own revenue forecast and also raised guidance for the year.

公司还宣布与亚马逊云服务达成了一项60亿美元的交易。

The company also announced a $6,000,000,000 deal with Amazon Web Services.

与此同时，Salesforce下调了2027财年的现金流预期。

Salesforce, meanwhile, lowered its cash flow forecast for the 2027 fiscal year.

为了更深入分析这些财报，我请来了D公司的董事总经理吉尔·卢里亚。

To break down more of the results, I want to bring on Gil Luria, Managing Director at D.

A.

A.

戴维森，以及BEP研究公司的创始人兼首席执行官本·波拉迪安。

Davidson, and Ben Pouladian, Founder and CEO of BEP Research.

欢迎两位的到来。

Welcome to the both of you.

感谢你们参与这次对话。

Thank you for joining us.

我想先问问，你们觉得Snowflake的财报中哪些地方最值得关注？

I What want to start with stood out to you about the Snowflake results?

他们赢了多少。

How much they're winning.

今年软件行业压力很大，连像Snowflake这样我们原本就知道表现不错的公司都受到了影响，但他们实际的增长远超我们的预期。

Software has been under a lot of pressure this year, which impacted even stocks that we knew were winning like Snowflake, but they're winning a lot more than we thought they would be.

从30%增长到34%，这种加速太出色了。

So accelerating growth from 30 to 34% is fantastic.

这反映出一个现实：对于想部署AI的企业来说，Snowflake已经变得不可或缺。

It's reflecting what's happening, which is Snowflake is absolutely critical for companies that want to deploy AI.

如果你的数据没放在数据仓库或数据湖里，很多AI能做的事你根本做不了。

If you don't have all of your information in your data warehouse, in your data lake, you can't do a lot of the things that AI does for you.

所以企业都在把Snowflake当作核心平台来采用。

And so companies have been embracing Snowflake as a platform.

除此之外，Snowflake昨天还推出了一个全新的编码工具，直接内嵌在平台里，让客户能用他们的Snowflake积分来写代码。

Now above and beyond that, Snowflake talked yesterday about the fact that they now have a new coding tool that works within Snowflake, within the platform, that allows their customers to use their Snowflake capabilities, their Snowflake credits to write code.

这看起来有点像Cursor，对吧？

So it looks a little like Cursor, right?

这并不是他们自研的前沿模型，但基于前沿模型开发的这个编程工具，在过去几周乃至几个月里取得了巨大成功，这也是他们收入加速的原因之一。

It's not their own frontier model, but it is a coding tool based on frontier models that has had tremendous success just over the last few weeks and months, which is part of how they've accelerated revenue.

所以这是一份非常出色的成绩单。

So it was a fantastic result.

接着，他们还和亚马逊达成了长期合作协议。

And then they added that long term deal with Amazon.

亚马逊是他们最重要的合作伙伴。

Amazon is their main partner.

大多数Snowflake客户都在使用AWS。

Most Snowflake customers are on AWS.

之前那份合同已经到期，但他们成功地和亚马逊续签了。

And they've had a deal that was, that, ran out and they were able to renew with Amazon.

这之所以关键，是因为它确保了他们的毛利率能稳定在百分之七十五左右。

And the reason that's important is that it locks in their gross margins in the mid seventies.

所以我们现在清楚了，他们的收入增长正在加速，同时毛利率也牢牢锁定在非常高的水平。

So we now know that they're on an increasingly accelerating revenue trajectory with gross margins locked in at a very high rate.

所以本，这个AWS的交易挺有意思的。

So Ben, that AWS deal was pretty interesting.

这个交易中一个值得注意的地方，是他们打算利用的Graviton芯片。

One of the notable things apart about that deal is the Graviton chip that they are planning to leverage.

你能谈谈你对Graviton芯片的看法吗？

Can you talk a little bit about your view on the Graviton chip?

这对AWS来说显然是个大胜利。

This seems like a big win for AWS.

这交易算不上惊天动地，但确实不小，对吧？

It's not a monstrous deal, but it's a pretty big deal, right?

是的。

Yeah.

在计算和智能代理AI领域，亚马逊在ARM架构的CPU集群方面确实领先，这就是Graviton芯片。

In the in the land of compute and agentic AI, Amazon really leads on having a large fleet of CPUs based on the ARM cores, which is the Graviton.

这是他们多年持续迭代出来的一款芯片。

It's a multigenerational chip that they put together.

他们把这芯片做得很好，运行成本也特别低。

So they've built it really well, and it's they run it very cheaply.

所以对于他们想用智能工作负载调度的任务来说，这芯片特别合适。

So it works well for these tasks that they wanna orchestrate with the agentic workload.

从投资的角度来看，投资者关注的就是这一点。

So that's kind of what investors are looking at from a investing perspective.

亚马逊拥有大量Graviton芯片，现在通过Snowflake和其他公司，它们能很好地变现了。

Amazon has a large fleet of Graviton, and now they can monetize it pretty well now with Snowflake and other companies.

在这方面，这对他们来说确实效果非常好。

So it it it's really working well for them in that aspect.

吉尔，如果你看Snowflake的估值倍数，我今天早上看到的是，它大约以9.6倍的未来营收在交易。

So, Gil, if you look at the multiple for Snowflake, I mean, it's trading around, I think, 9.6 times forward revenue is what I saw this morning.

在经历大幅上涨之后，现在可能更高了。

It's probably up higher than that after the big jump.

你觉得这个倍数还能涨到多高？会不会回到几年前的水平？

How high do you think that multiple can get, and do you think it gets anywhere close to what we saw maybe a couple years ago?

2021年高峰期的时候，Snowflake的市值一度达到收入的100倍。

Well, at the peak in 2021, Snowflake traded at a 100 times revenue.

对。

Yeah.

我觉得我们不会再回到那种水平了。

I don't think we're going

我也不觉得我们会回到那种地步。

back I don't to think we're going there.

但话说回来，现在的软件行业，只要你领先、收入增速在加快，估值倍数根本不是问题。

But having said that, what's happening right now in software is if you're winning, if you have accelerating revenue growth, multiple is not an issue.

你去看看Palantir、Datadog、Shopify就知道了。

Ask Palantir, ask Datadog, ask Shopify.

这些公司的收入和现金流倍数都高得吓人，因为它们是真正的赢家。

Those multiples are very high on a revenue and cash flow basis because these are the companies that are winning.

这些公司都有加速增长的收入，而绝大多数软件公司根本达不到这一点。

These are the companies that have accelerating revenue growth, which is not the case for the vast majority of software companies.

大多数软件公司正被人工智能挤出市场。

Most software companies are being crowded out by AI.

那些真正因人工智能而表现更好的少数公司，目前估值倍数根本不是问题。

Those handful of companies that are actually doing better because of AI, the multiple doesn't seem to be a restricting factor right now.

本，你对估值倍数怎么看？

What's your take, Ben, on the on the multiple?

你觉得这个倍数能涨到多高？未来可能是什么样子？

Do you have a view on how high the multiple should go, could go, you you know, what that could look like?

我觉得吉尔说得对。

I mean, I I agree with Gil.

现在的软件行业正在出现巨大的分化。

There's a huge bifurcation in what software is and how it's going.

那些能有效利用自身算力、融入智能、采用类似Datadog按使用付费模式的公司，才是真正抓住了AI增长机遇的赢家。

I think companies that are able to harness and have availability of their own compute in the right way to add their intelligence and are based on consumption models like the Datadog are the ones that are really winning the AI acceleration.

赢家会脱颖而出，输家则会彻底掉队。

And you'll see the the winners rise to the top and the losers really fall to the bottom.

我认为，如果这种加速势头能持续下去，利润率也保持得住，估值倍数就会进一步上升，市场也会重新评估。因为我觉得，大多数华尔街投资者对很多软件公司都太悲观了，简直是一棍子打死。

And I think if that acceleration is maintained and the margins are there, you'll get higher multiples and a rerating because I think most of the street investors have been very pessimistic on a lot of the software names and have thrown out the baby with the bathwater.

明白了。

Okay.

那吉尔，说到软件的不同细分领域，咱们聊聊思爱普吧。

So, Gil, so speaking of different segments of software, let's go to Salesforce now.

他们这次的财报电话会风格焕然一新。

They had a new look to their earnings call.

我看到他们用了挺高级的麦克风。

They had some fancy microphones I saw.

整体感觉更像播客了。

They had a little more of a podcast vibe.

但投资者的反应并没有因此被带动起来。

Wasn't enough to make investors jump in terms of the reaction.

你从思爱普的业绩里，最注意到的是什么？

What stood out to you from Salesforce's results?

是的。

Yeah.

这种播客形式挺有意思的。

The podcast format is fun.

真希望更多公司都能采用这种方式。

Actually hope all companies It's go to a lot more.

让内容更丰富了。

It adds a lot of texture

我得跟你说，吉尔，他们确实得这么做。

to I their have to tell you, Gil.

咱们不妨先聊聊这个，挺有意思的。

Maybe we can all bond about this for a second.

我觉得特别奇怪，电话会议的音质居然差成这样。

It is so strange to me that the conference calls are notoriously like the worst audio quality in the world.

不过Netflix在这方面做得还不错。

Mean, Netflix is pretty good at it.

但像亚马逊、微软、谷歌这些公司，他们的音频听起来就像是对着耳机上的小麦克风说话一样。

But like Amazon, I mean, Microsoft, Google, I mean, some of the audio, it's like they're talking into their little headphone microphones.

是的。

Yeah.

而且还没有视频，什么都没有。

And without video and anything like that.

所以，咱们可以跟着这个趋势走，但别学Salesforce，他们的预期本来就很低，结果也只是勉强达标。

So, yeah, let's let's follow that trend, but let's not follow how Salesforce is doing because expectations were very, very low, and they barely eked them out.

这个季度表现不错，主要是因为他们最近几周通过大规模加速回购回购了11%的股份。

So the quarter itself was good mostly because they bought back 11% of their shares in the last few weeks through a massive accelerated buyback.

但他们在第二季度的指引却低于预期，而这时候他们还告诉我们增长正在加速。

But then their guides for the second quarter was below expectations at a time where they're telling us that acceleration is coming.

对吧？

Right?

现在他们的有机收入按固定汇率计算，增长率已经降到了大约7%。

They've now decelerated to about a 7% organic constant currency revenue growth rate.

这才是关键数字。

That's the number that matters.

他们说明年就能重回两位数增长。

And they're saying that they're going to be back in double digits by next year.

真很难想象这怎么能做到。

It's just very hard to imagine how that's going to happen.

即使预期很低，估值也不高，股价才10倍现金流。

So even with very low expectations and a very low multiples, a company trading at 10 times cash flow.

即便如此，他们还是让人失望了。

Even with that, they were still able to disappoint.

本，我知道你不覆盖Salesforce，但你对他们的市场主导地位怎么看？毕竟客户很难彻底替换掉他们的系统。

Ben, you know, I know you don't cover Salesforce, but I wonder if you have a view at all on the incumbent strength that Salesforce has and this idea that it's very hard for people to tear out their Salesforce systems outright.

所以某种程度上，他们规模大本身确实是个优势。

And so, you know, to some extent, the fact that they are so big gives them a bit of an advantage.

我想知道，你接触的那些人里，这种 incumbency 优势现在对他们有多重要？

I mean, wonder in the people you talk to in your orbit, how much does that incumbent strength matter in right now for them?

是的。

Yeah.

我从来没投资过Salesforce。

I mean, I've never been a Salesforce investor.

我一直远远关注着马克·贝尼奥夫，还有他们做的那些收购。

I've always watched Mark Benioff from afar and then the the multiple acquisitions that they've made.

大家都管他叫SaaS救世主。

And everyone calls him SaaS Jesus.

我觉得他确实做得不错。

I think he does a good job.

他确实把这事做得挺到位的。

I think he does a he does a good job of that.

但问题在于，你不能一边说我们是核心系统，一边搞AI代理，还拿威廉姆森家居当例子。

But at the same time, the the challenge is is that you can't run around saying we're a system of record, and we're doing agents and look at think of William Sonoma.

你一边用代理，一边砸钱请马修·麦康纳拍广告，还有各种其他花销。

You're using agents and spending all this money on Matthew McConaughey commercials and, like, all this other stuff.

我的意思是，他们确实挺

Like I mean, they're pretty

有趣的。

fun.

这些广告确实挺

They're they're

有趣的。

they're fun.

对吧？

Right?

吉尔，这些广告确实很有趣。

Gil, they're they're fun ads.

但与此同时，你还是得回归产品本身。

Like, at the same time, you really need to get down to the product.

如果你把这笔钱花在自建算力上，掌握更多主动权，而不是拿去回购股票和其他花销，会怎么样？

What if you spent that money and actually bought, like, your own compute and controlled some of your own destiny as opposed to spending on buybacks and, like, all this other stuff.

我觉得投资者对此非常谨慎，不像过去十年那样愿意给他信任了。

And I think investors are very wary about that, and they're not giving him the benefit of the doubt which they gave the last decade.

这一点从股价和市盈率的表现上就能看出来。

And you see it in the performance of the stock and the multiple.

他们真的不太满意。

Like, it it they're really not happy

你觉得Salesforce未来会更倾向于垂直整合吗？

with Gil, do you think that do do you think Salesforce vertically integrates a little more looking ahead?

我挺喜欢麦康纳的。

Well, I love McConaughey.

我觉得他片酬太高了，但那些广告我确实喜欢。

I don't think he's worth what they're paying him, but I do love those ads.

是的，没错。

I I yeah.

我说的不是营销这一块的上游整合。

I'm talking maybe not upstream in terms of marketing.

我说的是下游，比如模型和基础设施方面。

I'm talking downstream in terms of models and infrastructure.

对。

Yeah.

说实话，他们现在处境挺棘手的，很多软件公司都面临类似问题，但我还是愿意说句公道话，Salesforce 其实算是受AI冲击较小的，因为他们跟客户的关系实在太深了。

I mean, they're in a tricky situation that many software companies find themselves in, but I would actually put Salesforce, I'll be kind to them and say, they're actually on the safer side of the AI impact because they are so entrenched with their customers.

几乎每家公司都在用 Salesforce，客户的资料、业务数据，全都在上面。

Everybody has an installation of Salesforce, and it's how they have all their data relating to their customers and a lot of other aspects of their business.

这种状况短期内根本不会变。

And that doesn't seem to change.

我们管这个叫‘系统性记录平台’。

That's what we're referring to as system of record.

这种状况短期内根本不会变。

That doesn't seem to change.

所以未来很可能是，Salesforce 成为人类和智能代理共同交互的平台。

So there is a future where Salesforce is the platform where humans and agents interact.

这带来了额外的价值。

And that is value added.

他们依然能继续经营这项业务。

And they can continue to manage that business.

但这并不是一个增长很快的业务。

That's just not a very high growth business.

我们能留住现有客户，但在这些客户的业务里，很难实现快速增长。

That's really we will keep our existing customers, but we won't be able to grow very fast within those customers' business.

所以他们一直在找办法，怎么和第三方智能代理合作。

That's why they're struggling to find a formula to work with third party agents.

一开始他们说，要向使用第三方代理的人收费。

Now they're talking initially, they said they charge people to use third party agents.

现在他们又在谈无界面和API接入，但还没做到。

Now they're talking about headless and API access, and they're not there yet.

他们还没找到解决方案。

They haven't figured it out.

但有人要把他们的 Salesforce 系统换掉，把它移除，这种可能性其实很低。

But the the likelihood of somebody replacing their Salesforce installation, taking it out is quite low.

确实很低。

Pretty low.

所以我们确实有个持续性的收入来源，这玩意儿是有价值的。

So we do have an an annuity, a perpetuity that's worth something.

我觉得与其回购股票，本之前提到过，他们以前挺爱收购的。

Now I would argue that instead of buying back stock, Ben mentioned that they used to be acquisitive.

但他们以前的收购都挺糟糕的。

They used to make terrible acquisitions.

他们曾经在收入峰值的时候，以二十倍的市销率收购软件公司。

They used to buy software companies at the peak at 20 times revenue.

现在他们完全可以挑一批优质、正在成长的公司，以低于二十倍现金流的价钱买下来，让这些收购变得很有收益，重新激活业务。

What they could be doing now is looking at this very rich, target list of companies that are good companies that are growing, that they could buy at less than 20 times cash flow and make them very accretive and revive the business.

比如哪些公司？

Like what?

比如哪些？

Like what?

你是说哪些？

That's Like what?

Braze、Zeta、Plavio、Amplitude。

Braze, Zeta, Plavio, Amplitude.

这些都是各自领域里的顶尖产品，还在保持两位数的增长，如果Salesforce能把它们收购下来，通过自己的渠道销售，第一年就能提升利润，帮助公司加快有机收入增长。

These are all very good category leading products that still have double digit growth that if Salesforce was to incorporate them and sell them through their distribution, they could be year one accretive and help them accelerate organic revenue growth.

过去不该收购的时候，他们倒是收购了。

So at the time they shouldn't have made acquisitions, they did.

现在该出手收购了，他们反而变得谨慎了。

At a time where they should be making acquisitions, they're a little more gun shy.

所以你的意思是，给马克·贝尼奥夫的建议是：嘿。

So you're saying, you know, the message to Marc Benioff is, hey.

别去追那些花哨的东西。

Don't worry about the flashy stuff.

你知道，与其担心那些花哨的东西，不如把重心放在真正提升产品力上。我是说，少炒作AI概念，多展示底层能提供的实际功能和价值。

You know, worry about really making the offering more, well, I guess less about the AI, more about sort of what you can offer under the hood and offer more value that way.

顺便说一句，这正好也是我们在节目里一直讨论的话题。

Which by the way is exactly what we've been talking about on the show.

你知道，我们请过一些嘉宾来分享他们是如何跟软件供应商谈判合同的。

You know, we've had folks on the show who would help negotiate these contracts with software vendors.

客户们的反馈通常是：你卖给我一堆AI概念。

And the message that customers have is, You're selling me all this AI stuff.

我是说，我还在努力从已有的订阅里挖掘出更多价值。

I mean, I'm still trying to get value out of the subscription that I already have.

所以回到你刚才说的，Gil，如果我理解得没错的话，意思是我们应该先把产品体系搭建好，而不是把精力放在Agent Force或者某个可能根本不具备竞争力的模型上。

And so to your point, Gil, if I understand you correctly, it's like, let's worry about, you know, building out that offering rather than focusing on agent force or a model that might not even be competitive.

没错。

Right.

你需要打造的是代理们所使用的平台和载体。

What you need to create is the venue and the platform that the agents use.

现在越来越清楚，企业会从Anthropic和OpenAI那里购买他们的智能体。

It's now becoming increasingly clear that companies are gonna buy their agents from Anthropic and from OpenAI.

他们不会从软件供应商那里买。

They're not buying them from their software vendors.

软件供应商能做的，只是让自己的平台对这些智能体更友好、更好用。

All the software vendors can do is make their platform more usable for those agents.

这才是提升产品价值的方式，增加更多的功能和使用场景，让智能体能和人类配合，为客户创造更多价值。回到你刚才说的，这样订阅服务就显得更有价值了，而不是去跟OpenAI和Anthropic竞争。过去几年他们一直在这么做，但显然行不通。因为像Anthropic和OpenAI这样的前沿大厂，肯定会有最强的能力和最好的智能体服务。

That's how they can make their product more valuable, create more surface area, more functionality that those agents can use in conjunction with humans to add more value to their customers, to your point, so the subscription is now worth more, as opposed to try to compete with OpenAI and Anthropic, which is what they've been doing for the last couple of years and clearly does not work because those companies, the Frontier Labs, are going to have the best capabilities and the best agentic offering.

你只需要把你一直擅长的事做得更好，也就是更好地适应这股新的智能体应用浪潮。

All you can do is just do what you've been doing better at accommodating this new wave of agentic use.

现在有个非常贴切的类比，就是互联网。

There's a really good parallel right now, which is the web.

对吧？

Right?

早期的万维网，也就是互联网，主要是供人类使用的网页，而且这些网页也是专门为人类设计的。

The the worldwide web, the Internet used to be web pages that only humans would use, and they were designed for humans.

现在，这些网站越来越多地被聊天机器人和AI代理使用，而不是人类。

Right now, increasingly, they're being used more by chats, by by bots, by AI agents than they are by humans.

这意味着网站必须做出调整，同时兼顾人类和代理两种使用场景。

And that means that these websites have to adapt to that and have these two facets, a facet that deals with humans and a facet that deals with agents.

没错。

Right.

这正是软件公司现在必须经历的转变。

That's the evolution software companies need to go through right now.

明白了。

Okay.

我想快速转个话题。

So I want to pivot quickly.

我想聊聊昨天爆出的那些关于Meta的新闻，有不少条。

I want to talk about the Meta headlines that we saw coming out of yesterday, there were a number of them.

本，我想先听听你的看法。

And Ben, I want to get your reaction to them first.

他们正在试点一些AI服务的付费订阅模式。

I mean, saw they are piloting subscriptions, paid subscriptions for some of AI services.

我们还看到马克·扎克伯格对股东承认，也许进军云计算领域是个可行的方向。

We also saw the news that Mark Zuckerberg sort of acknowledged to shareholders that, hey, maybe a cloud computing play is a possibility.

第三个新闻我一时想不起来了，好吧，咱们先说这两个，我马上就想起来第三个了。

And then the third headline, I'm blanking well, okay, let's start with those two, and then I'll remember what the third one is in a second.

是那个前线部署工程师项目。

It's the forward deployed engineers.

前线部署工程师。

Forward deployed engineers.

对，就是这个。

That's what it was.

不好意思。

I'm sorry.

独家报道称，Meta正计划推出前线部署工程师项目，这我完全没料到。

Our headline, the information exclusively reported that Meta is looking to implement a forward deployed engineer program, which I didn't see coming.

不过本，你怎么看？

But Ben, what do you think?

我觉得，现在搞前沿部署工程师就是实现AI的流行做法。

I think forward deployed engineers, the du jour way to implement AI.

显然，Palantir 是第一个提出这个概念的，现在 NVIDIA、Salesforce 这些公司都这么说：你想用好AI？我们派工程师过去，帮你梳理业务流程。

Obviously, Palantir was the first to really coin that term, and now NVIDIA and other companies like Salesforce, everyone is saying that if you wanna implement the best practices of AI, let us let us drop in some of our engineers, and we'll we'll see how your business processes are.

然后我们就能把AI系统专门适配到你的实际使用场景里。

And we'll be able to tailor the the AI to your usages.

说到你刚才讲的新订阅模式，这其实和那10%的裁员完全对得上。

And, you know, going back to what you were saying about the new subscriptions, I mean, it dovetails well with the the 10% layoffs.

我觉得扎克伯格看到的是股价表现不如预期。

I think what Zuck is seeing is an underperformance in his stock.

他觉得投资者认为他花在算力上的钱太多了，得想办法把成本收回来。

He doesn't he thinks investors think he's spending too much money on compute, and he's gotta be able to really pay for it.

我觉得Facebook、Meta或者扎克伯格的一个优势在于，他本人和公司并不以创新著称，但他们手里握着互联网上最优质的虚拟资产——WhatsApp、Instagram，还有Facebook本身。

I think one of the the benefits that Facebook and or Meta or Mark Zuckerberg has is that I don't really see him as an innovative type of person or or a company, but they own some of the best virtual real estate on the Internet with WhatsApp, Instagram, and obviously Facebook.

所以他总能稳稳收租，同时不断尝试各种可能奏效的点子。

So he's always gonna be collecting rent, and he's gonna be taking shots on whatever works.

这不过是他又一次的尝试。

And this is like another shot.

我不确定美国真正的五百强企业会不会用这种产品。

I I don't know if actual fortune 500 companies in The United States are gonna use this type of product.

我肯定他以前也这么干过，比如搞过类似Slack的项目，还有别的东西。

I'm I'm sure he's done this before with other he tried to take over he had a, like, a Slack program and other stuff.

我一直有个观点，认为Meta通过WhatsApp来做企业工具。

But I've had this thesis that Meta with its business tools through WhatsApp.

我想，不管他们会不会用Manus，只要能搭上这个平台……

And I guess if they if if they use Manus or not with the with that

如果这事真能成，你知道的，

If if it goes through, you know,

咱们就这么说吧。

let's yeah.

要么就让这些代理来负责吧。

Let's Or the agents have it.

在世界其他地方，比如中东、非洲或印度，WhatsApp 就是客户关系管理系统。

In other parts of the world, like The Middle East or Africa or India, WhatsApp is the CRM.

很多人靠它做生意，这一点很明显。

People run businesses through that, and you can see

所以，也许人们其实是愿意为此付费的。

So so maybe I mean, well, maybe people pay for it.

吉尔，你觉得呢？

Gil, what do you what do you think here?

我的意思是，Meta 打企业市场的核心故事到底是什么？

I mean, the enterprise push the story of Meta is what?

我觉得这已经是他们第十八次尝试推出企业级产品了。

This is, like, the eighteenth time I think they've tried for an enterprise product.

我们之前见过 Clara Shee 来了，就是那位前 Salesforce 高管。

We saw Clara Shee come in, that was this former Salesforce executive that came in.

我想她现在在公司担任顾问角色。

She has now, I think, taken on an advisory role at the company.

你觉得这是Meta进军企业市场的契机吗？

Do you think this is the enterprise moment for Meta?

我觉得目前的情况是，Meta的增长速度远超其他科技巨头。

So I think where where we're at is Meta is growing a lot faster than the other mega caps.

别忘了，他们增长幅度高达百分之二十多，而其他巨头才刚到个位数到十几个百分点，这还只是靠卖广告呢。

Let's not forget they're growing high twenties when the other mega caps are are growing low double digits to to teens, and that's just by selling ads.

所以现在，他们也说了：等等，

So here we are, and they're saying, you know what?

我们得为这些计算资源付钱了。

We need to pay for all this compute.

不如也把竞争对手用的那些办法拿过来用一用。

Let's turn on some of the levers that our competitors are turning on.

谷歌就有很多订阅服务。

So Google has a lot of subscriptions.

我们目前没有。

We don't have any.

我们可以为重度用户增加功能，把AI做成订阅服务，因为市场现在已经习惯了谷歌、OpenAI和Anthropic的做法。

We could turn on some subscriptions by offering heavy users more functionality, by offering our AI by subscription because the market is now conditioned to that by Google OpenAI and Anthropic.

所以我们现在有了新的盈利方式。

So we have new monetization tools.

顺便说一句，如果我们建了这么多数据中心却没用上全部算力，完全可以转手卖出去。

And by the way, if we're not using all this huge capacity that we've created by building data centers, we can just turn around and sell it.

对吧？

Right?

埃隆在SpaceX上市前就是这么干的。

This is what Elon just did ahead of the SpaceX IPO.

他手上有大量的算力资源。

He had all this compute capacity.

他说，我现在用不了这么多。

And he said, I'm not using all of it right now.

那我干脆把这产能以每年一百五十亿美元租给Anthropic怎么样？

So how about I rent that out at $15,000,000,000 a year to Anthropic?

Meta也处在同样的境地。

Meta is in the same position.

只要他们没把算力全用上，就能转手租出去。

As long as they're not using all of their compute, they can turn around and rent it out.

至于进入企业市场，像本说的，已经有四名工程师在做了，这领域早就人满为患了。

Now participating in the enterprise space with four deployed engineers, to Ben's point, that's that's already crowded.

对吧？

Right?

Palantir在这方面做得比谁都强。

Palantir does it better than anybody.

OpenAI现在也正往这个方向切入。

OpenAI is getting into that.

Anthropic和其他一些软件公司也在试图挤进这个市场。

Anthropic, and and other software companies are trying to get into that.

我不确定他们是否真能在这方面成功，但他们现在确实有大量算力闲置，完全可以高价出租，毕竟算力现在太值钱了，这又是一笔额外的收入。

So I'm not sure that that's necessarily where they'll have success, but they do have a lot of compute capacity they could turn around and rent out on at a very high rate right now because the compute is so, so valuable, and they have extra monetization.

再说，他们本来增长就比谁都快，现在又多了订阅收入和算力出租这两块额外的盈利来源。

Again, they're already growing faster than anybody, and now they have this extra monetization on subscription, extra monetization on renting out compute.

而且订阅模式还能帮他们解决其他问题。

And subscription solves other problems for them.

别忘了，在欧洲，他们没法像在美国那样做广告，因为隐私法规限制太多。

Let's not forget that in Europe, they can't serve ads in the same way that they do here because of privacy restrictions.

所以他们在欧洲的盈利远不如在美国。

So they monetize in Europe a lot less than they do in The US.

突然间，他们就可以说：嘿，不如这样。

All of a sudden, they can turn around and say, you know what?

既然在欧洲没法好好做广告，那我们就直接对所有用户收使用费。

If we can't sell ads effectively in Europe, we'll just charge everybody to use these platforms.

没人会因为要付订阅费就放弃使用 Instagram。

Everybody nobody's gonna drop Instagram because of a subscription.

是的。

Yeah.

我是说，对于Meta派出的现场工程师，咱们能不能聊聊，这个产品到底可能长什么样？

So, I mean, for deployed engineers from from Meta, can you just let's just talk a little bit about, like, what this offering could even look like.

说实话，我不太清楚这个企业级产品到底是什么。

I mean, it's not it's not clear to me what the enterprise product even is.

本提到了WhatsApp，所以我想，也许一种方式是帮企业搭建他们的WhatsApp线上商店？

I mean, look, Ben mentioned WhatsApp, and so I guess, I don't know, maybe one way to do this is helping a business set up their WhatsApp storefront.

我们说的指的是这个吗？

Is that what we're talking about here?

或者，你认为这些现场工程师在Meta能真正发挥什么作用？

Or what role do you think these forward deployed engineers could actually play for Meta?

这个词其实来自Palantir。

So again, the term comes from Palantir.

咱们先说说Palantir是干什么的。

So let's talk about what Palantir does.

是的。

Yeah.

Palantir 就是做企业级产品的。

Palantir does.

他们有个企业级产品，对吧？

Have an enterprise product, right?

我的意思是，你确实需要人帮忙部署这套系统。

Mean, I could see you need help setting it up.

没错。

Right.

这正是 Meta 想切入的市场。

That's what Meta wants to get into.

去跟企业的 CEO 和 CFO 说：你们在数据报表上有个大问题。

The market of going to the CEO and CFO saying, you have this huge problem around reporting.

我们要做的，就是用我们的 AI 模型来解决。

What we're gonna do is we're gonna use our AI model.

我们会派福特的工程师上门，帮你们用我们的AI模型解决那些复杂的数据报告和供应链问题，就像Palantir能为你们做的那样。

We're gonna send you, Ford deployed engineers, and they're gonna help you with our AI model solve your very complicated reporting and supply chain problem and and and do what Palantir would have done for you.

这样一来，你们就是在购买我们的算力，使用我们的模型和资源。

And that way, you're buying our compute and and using our model and using our capacity.

这正是他们的目标。

That's their goal.

再说一遍，这市场已经很饱和了，他们根本不可能做得比Palantir更好。

Again, it's already a crowded market, and and there's no way they do it as well as Palantir.

Palantir。

Palantir.

没错。

Right.

很好。

Great.

谢谢你们两位来参加这次对话。

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

这位是来自D公司的吉尔·卢里亚。

That is Gil Luria from D.

A。

A.

我是戴维森和本·波拉迪安，正在TITV为您直播。

Davidson and Ben Pouladian here on TITV.

谢谢。

Thank you.

谢谢。

Thank you.

苹果正打算重新聚焦于能在其设备上运行的AI模型。

Apple is looking to renew its focus on running AI models that could run on its devices.

这一消息来自我的同事亚伦·蒂利的独家报道。

That is according to a new exclusive report from my colleague Aaron Tilley.

我想请他上线，详细说说他的发现。

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

亚伦，欢迎再次回到节目。

Aaron, welcome back to the show.

很高兴你今天能来。

It's great to have you here.

是的，回来感觉真好。

Yeah, great to be back.

那么我们现在知道些什么呢？

So what do we know?

苹果公司一直以来，这都是他们的关键优势。

So Apple has, you know, this has been a key competitive advantage for them.

就是他们自主研发的芯片方案。

It's their internal chip program.

在这方面，他们只会继续深化投入。

And as part of that, I mean, they're going to only continue to deepen this.

在下一次的WWDC上，他们会把这个纳入他们的战略规划。

And at the next WWDC, they will include that as a part of their strategy.

其中一部分是把谷歌 Gemini 那个很大的模型压缩一下，放到他们的设备上，因为这是他们未来 AI 战略要用的模型。

And one part of that is sort of distilling a very big version of the Google Gemini model onto their device because that's the model they're going with for their future AI strategy.

所以他们拿到了授权，会把模型缩小，以便在设备上运行。

So they've gotten the license rights to do that and they will shrink it down to run on device.

而且他们也是少数真正有能力做到这一点的公司，靠的是他们独特的内存架构和自研芯片。

And they are sort of one of the key players that's really capable of doing something like that with their their own sort of unique memory architecture, their chips.

所以这对他们即将到来的 WWDC 来说是个很大的优势。

So that will be a key advantage to them for this upcoming WWZ.

所以这里的想法是，他们正在闭门操作。

So the idea here is that they are behind closed doors.

他们正在悄悄地进行很多调整。

They're sort of moving a lot.

他们现在更侧重于在设备本身部署计算能力，而不是依赖云端计算，对吧？

They're focusing a lot more on installing compute on the actual devices themselves rather than relying on compute, right?

对，与其依赖云端，或者像Meta那样在数据中心建设上投入数千亿美元，他们选择了另一条路。

Yeah, rather than relying on cloud or making or spending hundreds of billions on the data center build out, like we see with companies like Meta.

他们还会继续扩大这一优势。

And, you know, they'll be looking to further that that that sort of advantage.

他们正在寻找收购目标，帮助他们缩小模型规模，让模型在设备上运行得更高效。

They are looking for acquisitions to help for companies that can help them shrink down models and make them run more efficiently on device.

未来几年，这一趋势只会越来越明显，肯定会在他们下一次开发者大会上充分展示。

And so this will only continue to expand, you know, in the years to come, and certainly it will be on display at their next upcoming developer conference.

这还挺有意思的。

So this was kind of interesting.

在你的报道里，你引用了一位从事设备端计算的人，他说，大家现在这么重金投入数据中心其实是个误区，边缘计算或者设备端计算，那才是真正的未来。

In your story, you quoted someone who is working in the realm of the on device compute, and they basically made the claim that, hey, it's kind of a mistake that everyone is investing into data centers so heavily because edge computing or on device computing, I mean, that is really the future.

现在，这个技术领域的声音是不是变大了？

Is that segment of technology, is that getting louder right now?

有没有更多投资人开始往这些初创公司砸钱？

Are there more investors pouring into those startups?

有没有更多创业者把目光投向这个方向？

Are there more founders looking to it?

你能跟我聊聊，现在这个领域有多火吗？

Like, just talk to me a little bit about how buzzy that category is right now.

是的，我觉得确实还谈不上主流，但热度已经很明显了。

Yeah, it's certainly, I think, very you know, it's not I wouldn't call it mainstream yet.

目前大家还是铆足了劲，想打造全球最强的计算能力。

I think there a there is still this sort of fever pitch like effort to just build the biggest sort of compute in the world.

这种思路现在还是主流趋势。

I think that is still very much in the zeitgeist.

但下一代的重心，是如何更高效地利用资源，这已经成了最前沿的议题。

And this this sort of next generation of we need to be efficient with how we use our resources is really at the forefront.

我觉得，目前只有那些最前沿、最有远见的人才在认真考虑这个问题。

I think just the most sort of like, you know, forward thinking kind of folks are thinking about this.

这更多还是集中在一些更专业的AI应用场景上。

And it's not like it's it's definitely for kind of more, I think, specialized AI.

到了企业级产品或者消费级产品，反而更适合做本地化部署，直接在设备上运行。

And when that comes to something like enterprise or or consumer product, you can kind of more specialize and run on device.

但我觉得，确实还没到主流的程度。

But I think I think that's certainly think it's, yeah, definitely not mainstream yet.

明白了。

Okay.

最后一个问题。

And last question for you.

你之前也报道过，苹果正打算更大规模地使用谷歌云来支持自己的云服务。

I mean, you also had reporting on how Apple will be looking to Google Cloud in an increased capacity for its own cloud services.

我们之前聊过苹果对隐私的重视，那就有个疑问：如果苹果要建自己的私有云，会不会有风险？

We had a discussion about the emphasis on privacy that Apple has and sort of the question mark around, hey, if it's Apple's private cloud, know, are there any risks there?

那这个说法和苹果正在推进的边缘计算之间，该怎么平衡呢？

How does that story then balance with the edge computing that the company is going after?

是的。

Yeah.

苹果现在正在授权使用Gemini的一个版本，但完整的模型根本跑不了本地设备。

So Apple's licensing a version of Gemini and the full model just can't run on device.

但即便如此，这仍然完全做不到。

Still, that's still really not possible.

模型的参数量高达万亿级别。

It's in, you know, trillions of parameters the size of the model.

所以他们必须依赖一定程度的云端服务。

So they are they're going to have to rely on some amount of cloud for this.

他们自己的内部系统根本行不通。

And their own internal stuff doesn't work.

虽然用的是自家芯片，但还是跑不动这个模型。

It's running on their own chips and it's not capable of running this.

所以，和谷歌合作使用Gemini，这也是协议的一部分。

So they are getting, you know, this is part of the deal with the Gemini and Google.

他们因此获得了使用谷歌云服务的机会。

They do have access, an opportunity to rely on Google Cloud.

据我们了解，他们最近还启用了英伟达的机密计算功能，借助英伟达芯片的加密技术，确保在云端也能维持隐私保护。

As part of that, I my our understanding is that they recently turned on capabilities with NVIDIA just for confidential compute, so Apple can sort of maintain that privacy guarantee in the cloud with with NVIDIA's encryption features on their chips.