OpenAI发布红色警报，苹果AI人才大洗牌，广告迎来新黄金时代？| 2025年12月2日

欢迎大家来到Informations TITB。

Welcome everyone to the Informations TITB.

我是阿卡什·帕什里查。

My name is Akash Pashricha.

今天是12月2日，星期二。

It is Tuesday, December 2.

我们今天为大家准备了一场精彩的节目。

We have got a great show planned for you today.

首先，萨姆·奥特曼因谷歌等竞争对手对ChatGPT日益增长的威胁，在OpenAI宣布进入'红色警戒'状态。

First up, Sam Altman is declaring Code Red at OpenAI over the rising threats to chat deep chat GBT from competitors like Google.

稍后我们将与报道此消息的人工智能记者和主编杰西卡·莱森一起讨论这个话题。

We will discuss that with our AI reporter who broke the story and our editor in chief Jessica Lesson in just a few moments.

然后我们将深入探讨苹果公司的人工智能重组。

We'll then dive deep into the AI shakeup at Apple.

这家科技巨头的AI负责人已离职，他们正从微软引进一位新成员。

The tech giant's AI leader has departed the company, and they're bringing in a new guy from Microsoft.

我们将讨论他将在公司做些什么。

We'll talk about what he's going to do at the company.

接着，我们将讨论我们独家发布的关于Marvell与Celestial AI进行高级收购谈判的报道，这笔交易价值可能超过50亿美元。

We're then talking about some exclusive reporting that we published around Marvell's advanced talks to buy Celestial AI, a deal that could be worth more than $5,000,000,000.

这在芯片界是一件大事。

It is a big deal in Chipland.

最后，我们将与Highbeam的联合创始人探讨AI与广告的话题，Highbeam是一家金融科技公司，其客户包括许多快速增长的消费品牌。

And we will finish with a conversation around AI and advertising with the cofounder of Highbeam, a fintech company that counts many fast growing consumer brands as customers.

这是一场精彩的节目，让我们马上进入正题。

It is a great show, and so let's get right on into things.

OpenAI已进入红色警报状态。

It is code red at OpenAI.

萨姆·阿尔特曼已经发出警报。

Sam Altman has sounded the alarm.

他正在号召所有人回到各自的战斗岗位。

He is calling everybody back to their battle stations.

昨晚深夜，该信息率先报道了OpenAI首席执行官向公司发出了一份惊人的备忘录，称ChatGPT正面临来自谷歌等竞争对手日益加剧的威胁，此时至关重要。

Late last night, the information was first to report that the OpenAI CEO sent out a stunning memo to the company declaring it was a critical time for ChatGPT as competition from Google and the likes becomes a bigger threat.

现在加入我们的是我们自己的斯蒂芬妮·帕拉佐洛，她率先报道了这一消息，以及主编杰西卡·莱森。

Joining me now is our very own Stephanie Palazzolo, who broke that story and editor in chief Jessica Lesson.

早上好，两位。

Good morning to you both.

很高兴见到你们。

It's great to see you.

斯蒂芬妮，萨姆·阿尔特曼昨晚的备忘录里说了什么？

Stephanie, what was in this note from Sam Altman last night?

是的。

Yeah.

在我们能够报道的这份备忘录中，萨姆实际上宣布了他所谓的“红色警报”。

So in this memo that we were able to write about, essentially Sam declared what he calls a code red.

因此，他告诉员工，由于面临谷歌等竞争对手的更多挑战，他希望他们优先改进ChatGPT。

So he told employees that he wants them to prioritize improving chat TBT as, OpenAI faces more competition from rivals like Google.

因此，这包括改善模型行为、个性化、图像生成、速度和可靠性。

So what that kind of consists of are things like improving model behavior, personalization, image generation, speed, reliability.

为了实现这些目标，他承认公司必须降低其他一些工作重点。

And to do all those things, he did admit that the company will have to deprioritize some other work streams.

这包括广告和AI代理。

So this includes advertisements and AI agents.

我认为真正引人注目的是，这实际上是公司首次承认有意在ChatGPT中推出广告，而这一话题一直颇具争议。

And I think what really stood out here is that this is actually one of the first times that the company has even acknowledged wanting to do ads in ChatGPT, which has been quite a controversial topic.

当然，为了将这一切置于背景中，这一切都发生在几周后，竞争对手如谷歌和Anthropic发布了新模型，这些模型在某些领域实际上已超越了OpenAI的模型。

And then of course, to kind of put all this in context, this all comes weeks after we've seen new models from competitors like Google and Anthropic that have actually outperformed OpenAI's models in certain areas.

对。

Right.

现在，这是他们首次承认广告计划，但至少在最近几周内，我们已经多次听到OpenAI高管谈到这些公司面临的威胁，以及公司可能在某些方面放缓的事实，对吧？

Now, it's the first time that they've acknowledged the ads effort, but it's not the first time, at least in recent weeks, that we've heard from OpenAI executives about the threats that these companies face and the fact that the company might be slowing down in some ways, right?

对。

Yeah.

所以，在我同事埃琳上周撰写的一篇报道中，正如你所说，这并不是萨姆·阿尔特曼第一次承认人工智能行业日益激烈的竞争。

So in another story that my colleague, Erin, I wrote about, I think last week, we did write that, you know, like you said, this isn't the first time Sam Altman has kind of acknowledged the rising competition in the AI industry.

上个月，在谷歌发布Gemini 3之前的几周，他也曾在公司内部发布了一份备忘录，承认：嘿。

Last month in the kind of weeks leading up to Google releasing Gemini three, he also released a memo within the the the company kind of acknowledging like, hey.

谷歌很快就要推出一款非常优秀的新型模型，但我们相信，我们所下的赌注将帮助我们在长期竞争中胜出。

There's this really, really good new model coming out from Google soon, but we believe that we're making bets that are going to help us win in the long run.

对。

So Right.

他说，公司的氛围将会很艰难。

He said that he said that the vibes the vibes were gonna be rough.

对吧？

Right?

我的意思是，他确实这么说的。

I mean, that that that's what he said.

对。

Yeah.

我和斯蒂芬，顺便说一句，斯蒂芬，你们的故事太棒了。

I and Steph by the way, Steph, great great stories.

独家新闻太精彩了。

Great scoops.

昨晚我读到很晚都没睡。

Kept me up late reading last night.

但这变化也太大了。

But but what a shift.

对吧？

Right?

我的意思是，当那份备忘录刚出来时，上面写的是：坚持原定路线，团队。

I mean, when that first memo came out, it was the memo was stay the course troops.

我们有很多艰难的事情要做，但要保持专注。

We have so many hard things to do, but stay focused.

而现在是红色警报。

And now Code Red.

我认为，有趣的是，尽管竞争正在加剧，但似乎有一种救世主般的竞赛心态，各方都在争相超越彼此的模型和产品，而我认为这实际上更重要。

And I think, you know, what's fascinating to me is I mean, yes, the competition is intensifying, but there seems to be this sort of like messianic quests for one upping each other on models and product, which I actually think is more important.

但这种观念，我们被卷入了一场由OpenAI和其他公司推动的竞赛，而在我看来，所有模型都在不断变好，最终都会变得非常出色。

But this idea, you know, we're caught into this race that OpenAI and others are kind of pulling us into when it also seems to me that all the models are getting better and they're all going to get really good.

也许我们更应该关注人们如何使用这些产品以及他们的使用行为。

Maybe we should be focused more on how people are using these products and the engagement behavior.

因此，这一点让我深受触动。

So I was really struck by that.

这是我最近一直在思考的一个更广泛的主题。

Just a broader theme I've been thinking about.

整个2025年，我们一直在谈论这场AI战争，而OpenAI和Altman似乎一直在煽动这场战争。

For this whole 2025, we've been talking about the AI war and OpenAI and Altman have kind of been the ones stoking it.

而Gemini则推出了一些酷炫的功能和产品，却没有被卷入同样的竞争漩涡，并且正在消费者市场取得进展。

And then you kind of Gemini rolling out with some cool features and products that kind of haven't been sucked in in the same way and they're getting inroads with consumers.

所以我在想，Steph，你的看法是什么？这种战争心态真的有必要吗？你认为为什么在这个时刻，这么多领导者都转向了这种思维？

So I kind of wonder, Steph, your thoughts, Is this war mentality really Why do you think so many leaders are turning to that in AI in this moment?

意思是——这就是你的想法。

Mean- That's what you think.

正如你提到的，这确实像是一种来回拉锯，每隔几周就有一个实验室发布新模型，接着下一周另一个实验室又发布一个声称略胜一筹的模型。

As you mentioned, it is very much like this kind of back and forth where every couple of weeks you see a lab release a new model, then the next week another lab releases a model that they say is slightly better.

所以，我们确实感觉陷入了一个循环，只是不断小幅改进模型，而正如你所说，我们更应该思考基于这些模型构建的实际产品，以及这些产品是否真正有用、人们是否在使用它们。

So it does kind of feel like we're stuck in this loop of just slightly improving models versus as you're kind of saying, thinking about the actual products that are being built on them and whether those are actually useful and people are, you know, using those products.

我认为，在某种程度上，萨姆昨天发出的备忘录实际上更侧重于产品，因为他确实提到了我们正在开发的新模型。

I think in some ways, you know, yesterday's memo that Sam put out is actually a bit more of a focus on product because he does he does mention, you know, we have new models that we're working on.

它们会非常出色，其中一个就是我们今天早上写过的那个。

They're gonna be really good, which, you know, one of them is one that we wrote about this morning.

但特别的是，这份备忘录还非常聚焦于ChatGPT的改进，而这显然是OpenAI主要的盈利产品。

But specifically, the memo is also super focused on ChatchBT and improving that, which is obviously kind of OpenAI's main money making product.

所以我认为，这两份备忘录恰恰体现了你所说的这种张力：一方面，是不断竞相发布新模型的竞赛；

So I think like these two memos kind of show this tension that you're talking about where on one hand you have this constant race to release new models.

另一方面，也不能忽视像ChatGPT这样的产品，必须确保其增长仍在持续，人们仍在使用这些产品。

On the other hand, you can't forget about products like ChatGPT and making sure that that growth is still happening and people are still using those sorts of products.

所以，显然，他们大部分收入都来自这里。

So that's where they make a lot of their money from, obviously.

对。

Right.

对。

Right.

我的意思是，杰西卡，我想听听你的看法，就是关于ChatGPT，哦，抱歉。

I mean, Jessica, that's what I wanted to get your take on was, know, ChatGPT and oh, sorry.

OpenAI已经全面铺开了各个方向。

OpenAI has really spread itself in every direction.

我们已经 extensively 撰写了关于他们探索的一系列产品的文章，包括浏览器和智能代理。

We've written extensively about this suite of products that it has been exploring with browsers, with agents.

它还有社交媒体动态。

It's got the social media feed.

我的意思是，你认为这家公司是否把自己搞得过于分散了？

I mean, do you think that the company at all stressed itself too thin?

那么它如何平衡产品开发与Steph所说的底层模型必须具备竞争力这一点呢？

And how does it balance the product thing with what Steph's saying, which is that the underlying model has to be competitive?

Kosh，这是我从投资者那里最常被问到的问题，包括OpenAI的现有投资者和尚未投资OpenAI的人。

Know, Kosh, this is the number one question I get from investors, current investors in OpenAI, and people who haven't invested in OpenAI.

他们审视这家公司。

And they look at the company.

他们实际上并不问我它将如何继续融资，虽然我觉得他们或许应该问问这个。

They actually don't ask me how it's gonna keep raising money, which I think maybe they should ask me.

但就是这个疑问。

But it's this question.

昨天我还在办公室里和一位知名投资者就这个话题进行了讨论。

I was sitting with a well known investor yesterday in our office talking about this very topic.

你看。

Look.

一方面，这是科技公司一个古老且普遍存在的难题。

On one hand, it's a tie a model as old as time with tech companies.

你种下百万颗种子，总有一些会开花。

You plant a million seeds, some flowers bloom.

你知道，谷歌在其历史上一再分散了自己的精力。

You know, Google has spread itself thin again and again in its history.

对。

Right.

我们在这个播客中讨论过，这就是他们所采用的策略。

So and and we've talked on this podcast and the information about that's the playbook that they're using.

但与此同时，现在的情况并不寻常。

At at the same time, these aren't normal times.

我认为，人工智能的杀手级产品究竟是什么，这是一个重大问题。

And I think there is a big question over what is the killer product in AI.

是独立的聊天机器人助手，不断往里面堆功能？

Is it the standalone chatbot assistant and cramming more into it?

还是将智能更广泛地嵌入到所有消费者平台中，比如我们已经占据主导地位的邮件、社交和商业应用？

Is it a more embedded distributed intelligence across all the consumer surfaces where we already have very dominant players like email, like social, like business apps?

所以我认为这对OpenAI来说也是生死攸关的。

So I think that is existential for OpenAI too.

我倾向于认为专注是好的，但我也明白，为了招募人才，你需要让很多花朵绽放并从中学习。

I tend to think focus is good, but I also understand to recruit people, you need a lot of flowers to bloom and learn.

所以我认为这取决于情况。

So I I think it depends.

但显然，阿尔特曼正在努力凝聚团队。

And but clearly Altman is trying to rally the troops.

我的意思是，我觉得这些动员口号、这种战争语言确实是刻意设计的

I mean, I I just think these rallying cries, this war language has really been designed

是为了让这种共同性变得普遍吗？

to Is have a bit this common?

这种语言代码并不罕见吗？

Is this language code not rare?

普遍。

Common.

我的意思是，你可以回顾硅谷的历史，看到许多这样的标语海报。

I mean, you could go back in the history of Silicon Valley and see many posters with such slogans.

但再次强调，当我展望明年这场人工智能之战时，我始终回归到这样一个框架：我知道，这种战争框架——媒体、每个人都被卷入其中——其设计初衷是为了筹集资金和吸引人才。

But, again, framework I keep coming back to as I look to next year in this AI battle is, you know, I think this war framework, which everyone's been sucked into, the press has been sucked into, everyone has been sucked into, has been designed to raise capital and attract talent.

它并不一定能在产品层面决定胜负。

It isn't necessarily what is gonna win on the product side.

我最近和一位经常去中国的人聊过。

And I was talking to someone recently who spends a lot of time in China.

我问他，中国人工智能的现状如何？

And I was saying, what's the state of play in AI?

我也怀疑所谓中美人工智能对抗的说法是否成立，因为我们都将获胜。

And I'm also skeptical that this China versus US in AI makes any sense because we're both gonna win.

两个国家都将拥有前沿的人工智能技术。

And both countries are gonna have cutting edge AI.

是的，我们应该创新。

And yes, we should innovate.

是的，我们美国应该尽可能做到最好。

And yes, we should be as far as we can in The US.

但我觉得，认为这种情况不会发生是毫无道理的。

But the idea that that's not gonna happen doesn't make sense to me.

但这位投资者说，你看。

But this investor said, look.

在中国，他们专注于打造产品。

In China, they focus on building products.

你不会听到深兰、阿里巴巴、腾讯和字节跳动谈论在AI战争中谁更胜一筹。

You don't hear Deepsea and Alibaba and Tencent and ByteDance talking about one upping each other in the AI war.

他们非常非常关注人们如何使用AI。

They're very, very focused on how people are using AI.

最后我想说，我觉得你备忘录中最有趣的一点是，降低广告的优先级。

And last thing I'll say and stuff, I think one of the most interesting things in your memo, yes, deprioritizing ads.

我的天，这想法真大胆——把收入来源放在次要位置。

I mean, what an idea, deprioritize the revenue stream.

投资者可能会对这一点或收入来源提出一些问题，当然这不是唯一的收入来源，但代理是一个重点。

Investors may have some questions about that or a revenue stream, obviously not the only one, but agents.

而且我觉得，当我回顾今年的时候，你也知道我一直在做这件事。

And I think, you know, when I also look back on this year, you tell I've been doing that.

你知道，我们在代理领域进展到哪里了？

You know, where are we in agents?

我们看到了多个数据点，显示人们对代理的期望正在降低。

We're seeing multiple data points of lowering expectations around agents.

现在我并没有像斯蒂夫那样深入细节，去了解代理只是企业采用AI的一个方面。

And now I don't I I'm not in the weeds the way Steph is to know, you know, agents are one front in business adoption of AI.

它们并不是全部故事，但毫无疑问，从OpenAI到Salesforce等公司一直大力推动的正是这个故事。

They're not the whole story, but they were certainly the story that companies, everyone down the line from OpenAI to Salesforce on and on have been pushing.

我认为这也是另一个值得关注的点。

And I think that's another thing to watch.

我们本周早些时候，也就是一天前，看到了Thrive和OpenAI的公告，宣布与Thrive的控股公司合作，旨在加速企业采用AI以及代理的应用。

We saw the Thrive and OpenAI announcement earlier this week, a day ago, about a partnership with Thrive's holding company designed to accelerate maybe business adoption and agents too.

但在这里，阿尔特曼主动指出这并不是优先事项。

But here we have Altman actively calling it out as not a priority.

所以我认为这也是值得关注的一点。

So I think that's something to watch as well.

对。

Right.

嗯，这曾经是个大新闻。

Well, it was a big story.

斯蒂夫，我知道你昨晚很晚才睡，现在还在路上。

Steph, I know you were late last night and you're on the road.

你实际上似乎总是在路上时捕捉到一些最重要的新闻，我觉得。

You actually you you seem to get some of your biggest stories on the road, I think.

所以，我的意思是，我们当然喜欢你在这里，但请继续在路上奔波，因为每次你带来的新闻都越来越精彩。

So, I mean, look, we like having you here, but stay on the road because the stories get bigger and better every time.

谢谢你，斯蒂夫，来参加节目。

Thank you, Steph, for coming on.

感谢杰西卡加入我们。

Thank you, Jessica, for joining us.

我们非常感谢。

We really appreciate it.

我们很快就会再见到你们两位。

We will see you both very soon.

好的。

Okay.

苹果也在努力竞争人工智能领域，正在做出调整。

Apple is also shaking up things as it tries to compete in AI.

自2018年以来一直担任苹果首席人工智能高管的约翰·加纳德里亚即将离职。

John Ganandria, Apple's top AI exec since 2018, is stepping down from the company.

与此同时，公司任命了前微软和谷歌高管阿马尔·苏布拉马尼安，负责领导苹果部分人工智能研究工作。

Meanwhile, the company has tapped a former Microsoft and Google executive, Amar Subramanian, to lead certain parts of Apple's AI research efforts.

现在加入我们的是亚伦和亚伦。

Joining me now is Aaron and Aaron.

亚伦·蒂利报道苹果，亚伦·霍姆斯报道微软。

Aaron Tilly covers Apple and Aaron Holmes covers Microsoft.

亚伦和亚伦，欢迎来到《亚伦秀》。

Aaron and Aaron, welcome to The Aaron Show.

很高兴你们两位能来。

It's great to have you both here.

我会严格用你们的姓氏来称呼你们，就像我们在高中时那样。

I am strictly going to refer to you both by last name, just like we did

公平。

Fair.

在高中时。

In high school.

虽然没人能念对我的姓，所以我用了阿卡什这个名字。

Although nobody could pronounce my last name, so I went by Akash.

蒂利，我们先从你开始。

Tilly, let's start with you.

我们先说说新闻。

Let's start with the news.

谈谈你对加农迪亚辞职的反应。

Walk us through your reactions to Ganondria stepping down.

这让你感到意外吗？

Was this a surprise to you?

是的。

Yeah.

这根本一点都不意外。

So this wasn't a surprise at all.

我认为这件事早就该发生了。

This had been, I think, quite a long time coming.

无论是公司内部还是外部，大家都早有预期。

This had really been expected by both folks internally and externally.

我的意思是，主要原因在于简·安德里亚难以融入苹果公司，也未能制定出一套连贯的AI战略。

And I mean, primarily that's because Jane Andrea struggled to fit in and work out some sort of cohesive AI strategy at Apple.

尤其是今年早些时候，经过全面升级的Siri——那个更个性化、更具对话能力的Siri——被推迟了，这对苹果来说是个重大失误，因为他们之前已经承诺过这一点。

The especially earlier this year, the delay of an overhauled Siri, the big sort of more personalized conversational Siri was delayed, and that was like just a big blunder for Apple after promising this.

所以他多年来一直挣扎，迹象早已显现。

So he'd been struggling for years, and the writing had been on the wall for a while.

好的。

Okay.

迹象早就已经很明显了。

And the writing had been on the wall.

这并不是苹果在人工智能领域看到的第一次人员离职，对吧？

This is not the first departure we've seen at Apple with respect to AI, right?

不，他们的基础模型团队已经有多人离开。

No, there's been a number of folks from their foundation model team.

这个团队原本负责开发他们自己的ChatGPT版本，即他们自己的大型语言模型。

This is the team that would be building their version of sort of a ChatGPT, their own large language models.

该团队的负责人彭如明今年早些时候离开了苹果，加入了Meta，并带走了大量团队成员。

And their head of that team, Ru Ming Peng, left earlier this year for Meta, and he took a lot of his team with him.

随着他的离开，还有大量其他人员前往其他人工智能实验室。

And with his departure, there was also a large number of other folks leaving for other AI labs.

因此，苹果人工智能团队中一些非常关键的人物一直在陆续离开。

So there's been sort of a long unraveling of some very important folks from Apple's AI team.

亚伦·霍姆斯，我们来聊聊这位新人。

Aaron Holmes, let's talk about the new guy.

这位新人是谁？

Who who is the new guy?

他来自微软。

He comes from Microsoft.

他在那里做什么？

What did he do there?

是的。

Yeah.

所以苏普拉马尼亚曾在微软的MAI团队工作，这是由穆斯塔法·苏莱曼领导的内部团队，旨在训练他们自己的AI模型，理论上可以替代他们在Copilot产品中使用的OpenAI模型。

So Supramania was working on Microsoft's MAI team, which is their, you know, in house team led by Mustafa Suleiman that's trying to train their own AI models, that in theory could, you know, swap out for OpenAI's models in their copilot products.

他们主要一直依赖OpenAI的模型，但他们希望变得更加自主。

They've mostly been primarily reliant on OpenAI's models, but they wanna be a bit more self reliant.

但他在微软的任职经历有点难以评估。

But his tenure at Microsoft is a little bit hard to judge.

首先，他几个月前才加入公司，当时被从谷歌DeepMind挖过来的。

For one, he, you know, only joined the company a few months ago when he was poached away from Google DeepMind.

据我了解，他当时负责微软模型的后训练工作，也就是在模型完成初始训练后进行微调的过程。

And, you know, my understanding is that he was working on post training Microsoft's models, which is kind of the the process of, you know, tweaking the models after they've had their original training run.

但自那以后，微软虽然透露了一些他们的模型，却并未公布详细的技术规格或模型卡片，这让AI界很难准确判断他们训练和后训练工作的成效如何。

But since then, you know, Microsoft has kind of teased some of their models, but they haven't actually published, you know, detailed specs or model cards, which makes it a little hard, for for people in the AI community to know exactly how successful their their training and post training have been so far.

所以他只在那里待了几个月。

So he was only there a couple months.

那他是个超级明星吗？

So is he like a superstar?

我的意思是，这到底是好事还是坏事？

Like, what I mean, is this a is that a a good thing or a bad thing?

很难说清楚。

Kinda tough to tell.

你知道，在加入微软之前，他在谷歌DeepMind，我认为那里被普遍视为人工智能研究的黄金标准，尤其是对于那些有经验构建模型和AI原生产品的人来说。

You know, before Microsoft, he was at Google DeepMind, which I think is seen as basically one of the the gold standards of AI research, especially if you want people with experience, building, you know, models and and AI native products.

所以我认为这确实给他带来了顶尖AI人才的光环。

So I think that gives him certainly, like, the the sheen of top AI talent.

我认为他在谷歌期间参与了Gemini应用的开发，这个应用在过去几年中经历了多次迭代。

And I think while he was at Google, was working on the Gemini app, which has had a lot of different iterations over the past couple of years.

但可以肯定的是，这影响了微软决定招募他的决定。

But it it's safe to say that that played into the decision to to recruit him at Microsoft.

好的。

Okay.

蒂莉，他将在苹果做什么工作？

Tilly, what's he going be working on at Apple?

他并没有完全接替盖纳德里亚的职位，对吧？

Is he he's he's not totally taken over Gennandria's role, right?

是的。

Yeah.

他并不是让·安德烈亚的直接替代者，因为让·安德烈亚会直接向CEO汇报。

He's not he's not a one to one replacement with Jean Andrea because the Jean Andrea would report directly to the CEO.

这次的新领导并不会直接向CEO汇报。

This time the new the new leader, he's not reporting to the CEO.

他向克雷格·费德里吉汇报，也就是他们负责软件的负责人。

He's reporting to Craig Federicki, their head of software.

所以这实际上意味着在突出人工智能以及将领导者提升到公司高层方面有所降级。

So really a downgrade in terms of like highlighting AI and bringing the leader to the upper echelon of the company.

他还负责更小一部分的人工智能战略。

He's also overseeing a smaller subset of the AI strategy.

也就是基础模型团队和研究型AI，即AI研究团队。

So the foundation model team and research AI, the AI research teams.

而其他部分则被拆分到了公司的不同部门，比如运营和服务。

And while the rest of the group we've broken off up into different parts of the company, like the operations and services.

所以它实际上被拆分和分散了，没有前任AI领导者那么有影响力。

So it's really split up and divided and is not as powerful as the former AI leader.

对。

Right.

提醒我们一下，你一开始谈到了Siri的延迟，但现在这个故事进展到哪一步了？从你交谈的人那里了解到苹果的情况如何？

And remind us, you talked a little bit about the delays in Siri at the start, but where are we in that story and how are things going over at Apple from the people you're talking to?

是的，有一位新的领导者负责重新改造Siri，彻底革新Siri，让整体AI产品战略重回正轨。

Yeah, there's a new leader of that effort to rehaul Siri, overhaul Siri and get the overall AI product strategy back on track.

他们目标是明年，可能是春季到夏季。

They're targeting Apple's targeted next year, probably spring to summer.

他们将发布这些产品的改进版本，并与外部大型语言模型合作，而不是内部开发。

They'll be releasing the sort of improved versions of these products and partnering with external LLM instead of developing something internally.

最有可能的是，谷歌的Gemini将成为产品的一部分。

Most likely Google Gemini will be the one part of this of the product.

所以苹果将

So Apple will

会逐渐从内部开发转向外部合作，但长期来看仍可能回归内部开发。

be kind of moving on from something internal, but probably long term doing something internal.

很好。

Great.

艾伦和艾伦，很高兴你们能来。

Well, Aaron and Aaron, it's great to have you here.

一如既往，我喜欢《艾伦秀》。

I love the Aaron show as always.

唯一缺席的艾伦是负责报道谷歌的吴艾伦。

The only Aaron that's missing is Aaron Wu, who covers Google.

我相信用不了多久，我们就能把你们三位都请到现场，那样就能让播客听众彻底惊喜了。

And I I think it's just a matter of time before we get all three of you here, and then we can really throw the podcast listeners for a spin.

感谢你们两位的到来。

Thank you both for coming on.

我们很快再聊。

We will talk to you very soon.

谢谢。

Thanks.

好的。

Okay.

据《信息》独家报道，美满电子正在就以数十亿美元收购芯片初创公司Celestial AI进行深入谈判。

Marvell is in advanced talks to buy chip startup company Celestial AI in a multibillion dollar deal that is according to a new exclusive report from the information.

我想邀请报道此新闻的记者安妮莎·加尔迪齐来帮我们详细解析一下。

I want to bring on Anissa Gardizzi, one of the reporters who broke the news to help us break it all down.

嗨，安妮莎。

Hey, Anissa.

你怎么样？

How are doing?

嗨，阿卡什。

Hey, Akash.

我很好。

I'm great.

你怎么样？

How are you?

我很好。

I'm good.

我们来谈谈这笔交易吧。

Let's let's talk about this deal.

Marvell 并不是一个我们经常听到的公司。

Marvell is not a company that we hear too much of.

不过，它是一家相当大的公司。

It's a pretty big company, mind you.

它实际上是一家大型芯片公司。

And it's it's it's actually you know, it's a it's a big chip company.

我们本应该更多地讨论他们。

We should be talking about them more.

他们想收购的是哪家公司？

Who is the company that they're trying to buy?

关于这笔交易，我们了解多少？

What do we know about the deal?

昨晚，我们报道了Marvell正在与一家名为Celestial AI的初创公司进行深入谈判。

Last night, we reported that Marvell is in advanced talks to buy a startup called Celestial AI.

我们听说这笔交易可能价值数十亿美元，并可能在今天就宣布。

And we're hearing that the deal could be for several billion dollars and potentially announced as soon as today.

所以，这就是这笔交易。

So so that's the deal.

我认为Marvell考虑收购Celestial是有道理的，但他们是一家非常早期的公司。

And I think it makes sense, that Marvell would be looking at Celestial, but they're a very early stage company.

所以大家都很好奇，为什么Marvell会在这个阶段进行这笔交易。

So I think everyone's very curious why why Marvell is is doing a deal at this stage.

我想谈一谈这一点，但你先退一步想想？

So I I do wanna talk about that, but but do you just take a step back?

Marvell是谁？Celestial AI又是谁？

Marvell, who is Marvell and who is Celestial AI?

因为他们不是英伟达和AMD。

Because they're not NVIDIA and AMD.

对。

Right.

是的。

Yeah.

你可能听到关于Marvell的消息少一些。

You might hear about Marvell a little bit less.

但重要的是，Marvell从事定制ASIC业务，类似于博通。

But, you know, what's important about Marvell is that they are in the custom ASIC business similar to a Broadcom.

因此，他们与超大规模云服务商合作。

And so they work with companies such as hyperscalers.

你知道，亚马逊是他们的客户之一，他们帮助其他公司制造定制芯片，包括AI芯片。

You know, Amazon is one of their customers, and they help other companies make custom chips including AI chips.

此外，他们还有一个网络业务。

And then they also have a networking business.

我认为这就是Celestial发挥作用的地方。

And I think that's sort of where Celestial comes in.

好的。

Okay.

而且，你看，Marvell的股票今年下跌了大约20%。

And this, I mean, Marvell stock is is it's down around 20% this year.

所以，他们没有像其他芯片公司那样受益于AI的利好吗？

So, I mean, what, they haven't benefited from the AI tailwinds that that the rest of the chip companies have?

他们落后了吗？

Are they falling behind?

是的。

Yeah.

我认为他们还是受益了一点。

I think they've benefited a little bit.

但就股价而言，市场对AI加速器以及任何可能与NVIDIA竞争的产品都太过兴奋了。

But in terms of the stock price, there's just been so much excitement on AI accelerators and anything that could potentially be competitive with NVIDIA.

我认为博通因为与谷歌的关系，以及人们对他们的TPU的兴奋，已经吸走了大量关注。

And I think Broadcom has sucked a lot of energy out of the room given its relationship with Google and you know, how excited people are about their TPUs.

另外一件事是，顺便说一下，上个季度，Marvell报告了其ASIC业务环比持平的增长预期。

And then another thing was that, you know, last quarter, by the way, Marvell reports today, but last quarter the company projected flat growth in the ASIC business quarter over quarter.

所以我认为这可能是人们对Marvell不太兴奋的原因之一。

And so I think that's one of the reasons that maybe people haven't been as excited about Martbell.

对。

Right.

我想聊聊你上周写的另一篇文章，但先快速把这里说完。

I want to get to another piece that you wrote last week, but very quickly, just to finish up here.

你预计2026年会看到更多这类芯片收购吗？

Are you predicting more of these types of chip acquisitions in 2026?

我的意思是，感觉这些芯片公司为了保持竞争力，正在努力尝试，我们几个月前还讨论过这些芯片初创公司以及它们获得的融资。

I mean, it it feels like as these chip companies try to start try to stay competitive and we talked a lot about these chip startups a couple months ago and, you know, the funding rounds that they were getting.

但我总觉得这是不可避免的，对吧？

But I sort of see this as an inevitability, right?

是啊，很难说。

Yeah, it's hard to say.

我的意思是，这是一笔非常早期的收购。

I mean, I think this is a very early stage acquisition.

在光子互连领域，有许多公司还处于无收入、无产品的阶段。

There are tons of companies in the photonics interconnect space that are also pre revenue, pre product.

因此，如果一家大公司战略性地收购其中一家，将非常有趣地看到这如何影响整个光子互连领域——到目前为止，这个领域一直非常火热，这也很好地解释了为什么如果你相信人工智能和AI集群的增长，就会投资于此。

So to have a big strategic sort of take one of those players out, will be very interesting to see how that affects the rest of the, photonics interconnect space, which you know to date has been pretty hot and makes a lot of sense of why you would invest in that if you believe in AI and AI clusters growing.

但你知道，我们还看到英伟达在几周前收购了Fabrica，所以我认为，毫无疑问，战略投资者正在接触所有公司。

But you know, we also saw NVIDIA buy and Fabrica a couple of And weeks so I think, you know, I'm sure strategics are reaching out to all companies.

而且，我们也听说有些公司拒绝了收购要约，我们将来或许会就此撰写文章。

And, you know, we've also heard about some companies turning down offers, which we'll hopefully, write about in the future.

但我认为，像英伟达、博通、美满电子这样的战略玩家，甚至像Meta这样的超大规模云服务商——它们最近也完成了一次芯片收购——都在积极进入市场，并与这些公司接洽。

But I think the strategic players like Nvidia, Broadcom, Marvell, even the hyperscalers, like Meta had a recent chip acquisition, They're all definitely in the market and talking to these companies.

对。

Right.

好吧，让我们聊聊AI生态系统中的其他领域。

Well, let's talk a little bit elsewhere in the AI ecosystem.

在AI芯片之后，人们最常谈论的就是AI数据中心和基础设施。

Next to AI chips, the next thing people love to talk about is AI data centers and infrastructure.

你上周写了一篇关于当前状况的专栏，我觉得非常棒。

You had a bit of a current state of play column last week that you wrote that I thought was great.

它稍微谈到了AI数据中心公司的整体氛围以及这些设施的建设进展。

It sort of talked a little bit about the mood and the atmosphere for AI data center companies and the progress on the construction of these facilities.

目前数据中心领域的整体情况如何？

What's the vibe right now in data center land?

是的。

Yeah.

压力非常大，因为有太多公司设定了极其紧迫的时间表，来建造非常复杂、大规模的AI数据中心。

It's high, high pressure because there are so many companies that have set out on these crazy fast timelines to build very, very complex large data centers for AI.

这些数据中心的建设成本高达数十亿美元，而且你知道，未来几年他们计划投入的资金还会更多。

These are the kind of data centers that companies are paying billions of dollars for, and you know plans to spend even more than that, in the coming years.

但目前生态系统中的氛围是，时间节点正在逼近。

But but the mood in the ecosystem right now is that, okay, timelines are approaching.

我们必须按时完成这些截止日期。

We do have to hit these deadlines.

你知道，这是一个涉及物理移动的行业，需要让实际的人员前往项目现场以按时完成工作。

And you know, this is a industry that involves moving physical things, like getting physical people to your projects to get things done on time.

而这一点，正如我们所见，正变得颇具挑战性。

And that, you know, we're seeing is proving to be challenging.

我认为这本就在预料之中。

That I think was to be expected.

但考虑到这里的巨大投入，这无疑是当前的热门话题。

But given the stakes, you know, here, it's it's definitely the talk of the town, would say.

一些云公司已经开始就这个问题发声。

And there are some cloud companies that are starting to make some noise about this.

是的。

Yeah.

因为云公司的收入直接与其数据中心上线并向客户出租计算资源的时间相关。

Because cloud companies, their revenue is directly correlated to when they can turn their data center online and start renting out compute to their customers.

你提到的具体例子是，CoreWeave 上个月在财报电话会议上表示，由于其一家数据中心供应商的延迟，他们下调了本季度的收入预期。

And the specific example you're referring to is, is CoreWeave on its earnings call, last month talked about how they were adjusting revenue, expectations in the quarter downward, because of a delay at one of their data center providers.

因此，延迟确实开始产生影响。

And so, delays do start to have an impact.

当你处于 CoreWeave 这样的规模时，如果你没有在预测中考虑到这些延迟，就可能不得不报告这些延迟。

And like when you're at the scale of a core weave, you know, you might have to start reporting delays if you don't forecast that into your projections.

对。

Right.

没错。

Right.

你认为这是由于截止日期设定得太现实，还是因为人们根本无法足够快地推进？

Do you think this is more a case of the deadlines being too realistic, or is it just that people cannot move fast enough?

在这场跷跷板的两端中，你认为目前哪一端是更重要的因素？

Which of these two ends of the seesaw, I guess, do you think are the bigger factor here right now?

是的。

Yeah.

我感觉人们正在非常迅速地行动。

I get the sense that people are moving very fast.

对。

Right.

与正常时期相比。

Compared to like like normal times.

我的意思是，几个月内就把一个数据中心建起来。

I mean, you know, getting a data center stood up in months.

我的意思是，这看起来相当快。

I mean, that seems pretty fast.

对吧？

Right?

没错。

Exactly.

我觉得，总的来说，一切都以创纪录的速度进行着，没错。

Like, I think, like, on the whole things are going at record speed and right.

你知道，如果没有像AI这样推动所有人如此迅速争夺业务的东西，人们通常不会这么快行动。

You know, people are moving faster than they typically would if there wasn't like something like AI that was pushing everyone to sort of win business so, so quickly.

但确实，公司正遇到一些实际的限制，不过大家并没有太恐慌。

But yeah, I think there are just some physical realities that companies are are running into and no one is panicking too much.

但这里的影响是，延迟可能会影响你与客户的合同或你的收入预期。

But the impact here would be that delays could affect your contracts with customers or your revenue guidance and expectations.

因此，市场或许也开始意识到，那些受物理条件限制的公司可能会因这些物理限制而遇到问题。

And so the market is maybe also starting to realize that, you know, these companies that are bound by physical constraints might run into issues due to the physical constraints.

所以，我最后问你一个问题，当我们转向思考2026年时，你认为这一切会如何发展？

And so last question for you, as we sort of shift ahead to thinking about twenty twenty six, how do you think this all plays out?

在你周围的人中，大家对这些公司有什么样的预测？

What are some predictions that people in your orbit are are talking about with respect to these companies?

有人告诉我，他们预计明年延迟将对客户合同产生影响。

One thing that people are telling me that they expect for next year is that delays will have an impact on customer contracts.

因此，根据你签订的合同，延迟可能会影响客户为他们同意购买的产品支付的金额。

And so depending on the contract you struck, having a delay could impact, how much your customer pays you for what they agreed to buy.

所以，如果你出现延迟，客户可能会从每小时每块GPU支付3美元降至支付2美元，这完全改变了经济模型。

So, maybe if you have a delay, someone is going from paying you $3 for the GPUs per hour to paying you $2 per GPU per hour, and that completely changes the economics.

因此，我非常感兴趣的是，寻找更多这样的例子：延迟虽然常见且可预期，却可能对云合同的经济性产生某种影响。

And so I'm I'm very interested in in finding out more examples there where a delay which, you know, is kind of common and expected could actually have some sort of impact on the economics of a cloud contract.

所以，这就是我明年听到的情况。

So that that is what I'm hearing for next year.

很好。

Great.

安妮莎，感谢你前来做客。

Well, Anisa, thanks for coming on.

我们非常感谢你的分享。

We really appreciate it.

我们很快再和你聊。

We'll talk to you again soon.

谢谢。

Thanks.

好的。

Okay.

AI公司争相寻找在平台上投放广告的方式，这可能对那些试图找到最佳营销渠道来接触客户的品牌和公司产生重大影响。

The rush of AI companies trying to find ways to advertise on their platforms could have big implications for brands and companies that are trying to find the best marketing vehicles to reach their customers.

现在加入我们的是Highbeam的联合创始人兼首席执行官萨米尔·谢吉尔，Highbeam是一家帮助New Yorker Nowhere、Cuts和Bird Dogs等品牌应对电子商务复杂性、管理现金流的公司。

Joining me now is Samir Shergil, the co founder and CEO of Highbeam, a company that helps brands like New Yorker Nowhere, Cuts and Bird Dogs manage their cash flows with all the complications of e commerce.

萨米尔，欢迎来到节目。

Samir, welcome to the show.

很高兴你能来。

It's great to have you here.

非常感谢你邀请我，阿卡什。

Thanks so much for having me, Akash.

我们来谈谈广告吧。

So let's talk about advertising.