Meta可穿戴设备副总裁谈AI眼镜商业模式，OpenAI挖角苹果AI人才 | 2025年9月19日

欢迎各位收看TITV资讯。我是阿卡什·帕什里查特。今天是9月19日星期五。这是新闻频出的一周迎来新闻不断的收官日，今天我们为您准备了强大的嘉宾阵容。稍后节目中将连线Meta可穿戴设备部门负责人亚历克斯·希梅尔，请千万不要走开。

Welcome everyone to the Informations TITV. My name is Akash Pashrichat. It is Friday, September 19. It is a Newsy end to a very Newsy week, and we have got a great lineup of guests for you today. We've got Alex Himmel, the head of Meta's wearables unit coming on the show in just a few minutes, so do not go anywhere.

这场对话令人非常期待。我们还将剖析英伟达斥资50亿美元投资英特尔，再耗资9亿美元进行AI领域已屡见不鲜的人才并购。随后将由我们的埃隆·马斯克专题记者，讲述这位前石油大亨如何成为马斯克超级计算机计划的关键人物。节目最后将与苹果专题记者讨论OpenAI为自身硬件计划挖角苹果核心硬件人才的内幕。

That's very exciting conversation. We're also gonna break down NVIDIA's big move to invest $5,000,000,000 into Intel and then spend another $900,000,000 to do one of those acqui hires that have become all too familiar in the land of AI. We're going to talk all about that. We've then got our Elon Musk reporter to tell us about the ex oil man who has now become key to Elon's supercomputer ambitions. And we're going to end the show with a discussion with our Apple reporter about his story that talks about how OpenAI is going after key Apple hardware talent for its own hardware ambitions.

今天的节目内容非常紧凑，让我们直接进入第一个环节。想必各位已了解Meta本周发布的Ray-Bans智能眼镜，这是科技巨头首款面向公众销售的带显示屏智能眼镜，将于本月底上市。亚历克斯·希梅尔作为Meta可穿戴设备部门负责人，领导了该产品的研发团队。昨晚我与他进行了25分钟的精彩对谈，探讨如何从这款高资本密集型产品中实现盈利。

It is a busy, busy show today, so let's get right on into our first segment. You probably know all about the Meta Ray Bans display that the company unveiled this week. It is the first major smart glasses with a display from a big tech giant that will be going on sale to the public at the end of this month. Alex Himel is the executive leading Meta's wearables division, which means he oversees the group that made this product. And last night, I sat down with Alex for about twenty five minutes for a great conversation about how he plans to make money from this highly capital intensive product.

我们还讨论了其新研发腕带是否可能拓展健身功能，以及公司超级智能小组如何改变可穿戴设备团队的运作模式。这段访谈非常精彩，我很荣幸能为大家呈现。下面是我与亚历克斯·希梅尔的对话。

We also talked about whether or not he could ever see the new wristband that they've developed expanding its functionality into fitness. And I also asked him about how the company's Super Intelligence Group is changing how the wearables team functions. It was a great conversation. I'm excited to play it for you. Here is my chat with Alex Himmel.

亚历克斯·希梅尔，欢迎来到TITV。很高兴您能出席。

Alex Himmel, welcome to TI TV. It's great to have you.

谢谢邀请。虽然觉得穿着有些随意，但很荣幸参与节目。

Hey, thanks for having me. I feel a little underdressed, but it's great to be here.

说真的，比起着装，我觉得我的眼镜才是真正逊色的——它们可没你的这么酷。

Well, look, dress aside, I feel like, you know, my glasses are really the ones that are lacking right now. They're not as cool as yours.

谢谢。这些是新款的Ray Ban Meta智能眼镜，属于第二代产品。这是蓝色镜框配色版本，搭载了升级的硬件配置。

Thank you. These are the new Ray Ban Meta Glasses. This is generation two. These are the blue frame, colorway, and they've got the upgraded hardware.

不错。我很期待详细聊聊这款眼镜，我们有很多内容要探讨。你们这周在Meta Connect大会上发布了不少新品，我想大家现在已经见过这款眼镜了，应该也看过一些演示视频之类的素材。

Nice. Well, I'm excited to talk all about the glasses, and we have so much to get to. You unveiled a lot this week at Meta Connect. I think people have seen the glasses by now. They've got to see some of the demos and stuff like that.

所以我首先想深入了解你负责的可穿戴设备业务模式，以及这次发布的显示眼镜。众所周知这类设备的硬件盈利难度很大，能否从宏观层面谈谈你对这个产品的商业模式和盈利方式的思考？

So I really want to start by diving into the business model behind the wearables group that you lead, but then also the display glasses that you unveiled. I think it's pretty well understood that it's going to be difficult to make money on the hardware side of these units. And so I just wanna understand from you at a high level, how do you think about the business model and making money from this product?

目前我们的重点是打造优质产品，实现产品市场契合度，让用户爱用且日益感受到便利性。Ray Ban Meta眼镜的起点就是如此，马克提到过眼镜开发的三大原则。关键是即便关机状态下，这仍是副时尚舒适的眼镜——可以是太阳镜或光学镜片。然后我们再叠加智能功能。

Well, our focus right now is on just delivering a great product that finds strong product market fit that people enjoy using and increasingly find convenient and useful. You know, we started with the Ray Ban Meta Glasses, and Mark went through kind of the three principles of building glasses. But, I mean, the thing about these is even if they were powered off, they're still just a good looking, comfortable pair of glasses that's either you can wear sunglasses or optical glasses. Right. And then we add functionality to it.

同理，Medaray Band显示设备也遵循这个理念。我们专注让用户真正获得使用价值，通过优化取景器等显示功能，并不断加强AI能力来提升Ray Ban Meta眼镜体验。现阶段重点在此。从商业模式看，首代设备研发投入巨大，当资本支出平摊到第一代产量时，效益自然不如规模化量产后的情况...

And so similarly with the Medaray Band display, like, we're really focused on it just being something that people are finding value out of, and we're focusing on messaging and making the Raybea MetaGlasses just better with the viewfinder and things that display can deliver and increasingly AI. So that's the focus right now. Right. From a business model standpoint, you know, part of it is when you're launching a first generation device, you're pretty burdened with a lot of the investment with the research, r and d investment costs just to deliver that. And so whenever you're amortizing your capital expenditures and everything else over, you know, the generation one unit volume, it'll be less good than when you hit large scale, and you're able

没错。积少成多。

Right. To It adds up.

客户基数扩大后确实如此。随着良率提升，物料浪费和瑕疵品减少——大多数设备其实都达不到标准，这部分成本就会优化。

Customers more. It adds up. Yeah. And, you know, sales can generation as your yield gets better, so you're throwing away less materials and less devices that don't make it, most of them don't make it. So that's one part of it.

因此，随着我们制造工艺的成熟和规模扩大，我们预计经济效益会越来越好。

And so we expect the economics to just get better as our manufacturing matures and we hit higher scale.

那么这是否意味着...我想深入探讨第二部分，但这是否意味着在800美元价位，你们会亏本销售这些设备？

And then Does does does that mean at at I wanna get into the second part of it, but does that mean for at at $800, I mean, will you be selling these units at a loss?

我们认为眼镜和腕带组合定价7.99美元是合理的。我们如此定价是因为这符合该品类定位，也是我们规模化运营负责任企业的目标价位。我们对这个价格很满意。就像我们所有眼镜产品一样，我们并不追求单件产品初期就获得巨额利润，而是希望通过时间积累实现盈利。

Well, we so we think so $7.99 for the glasses and the band together. We think you know, we price it there because we think that's the right price for the category and where we think it will be for running a responsible business as we're hitting scale. So we feel good about the price. And, you know, similar to all the classes we've made, you know, we're not trying to make a massive profit on the per unit right out of the gate. We're trying to get there with time.

我们真正关注的是确保以合理价格和优质产品满足消费者需求。

We're really focused on just making sure that we're meeting people with a good price and a great product.

好的，明白了。

Right. Okay.

我刚才打断了你。你正在谈商业模式的延续性。请说说其他盈利方式。

So I I I cut you out. You were talking about you're continuing with the business model. Talk to me about the other ways to make money.

没错。软件业务是关键。归根结底我们是软件公司，虽然硬件方面也会保持利润空间。比如我们未提及的Ray-Ban Meta智能眼镜、Oakley Meta Houston和Oakley Meta Vanguard系列运动眼镜——如果你尝试过它们的慢动作视频功能，就会发现这些运动眼镜非常有趣。

Yeah. That was very good. So software. So at the end of the day, we're a software company, and we do expect that we will be making margins on the hardware side. You know, our our line of glasses that we're not talking about, which the Arabia Meta Glasses, the Oakley Meta, Houston's, the Oakley Meta Vanguard's, which those are extremely the sport glasses are extremely fun if you try this slow mo video.

是的。但我想说的是，仅硬件业务本身就已经是负责任的商业模式了。而我们作为一家软件公司，正在探索——这正是我们的专长——如何通过软件构建新的盈利模式。

Yeah. But, I mean, those are already a responsible business just on hardware. And then we're looking to you know, we're a software company. That's what we do. And so we expect to make a business model with software.

比如AI订阅服务就是一个例子，我认为未来还会有其他变现方式。一旦设备达到规模效应，我们持续增加软件功能、提升AI能力，相信会涌现更多商业机会。

And so AI subscriptions would be one example, and I think there'll be other ways that we'll be able to monetize with that. Once we're able to hit, you know, once devices hit scale and the we add more and more software, the AI gets better, we think there'll be good opportunities there.

明白。我想重点聊聊订阅服务，不过先快速确认下——你提到的其他设备，比如Oakley和没有显示屏的雷朋眼镜，听起来它们目前单件利润更高？

Right. So I wanna get on subs subscriptions, but very quickly, you talked about the other the other devices. So the Oakleys and the the Ray Ban without the display. So it sounds like those are profitable right now than on a on a per unit basis.

嗯，没错。让我看看...我的Oakley Meta眼镜放哪了...啊糟糕。

Mhmm. Yep. They are. We've got let me see what I got here. I got my Oakley Meta oh, shoot.

跑哪去了？哦，这是Oakley Meta的宣传视频。

Where'd they go? Oh, here's the Oakley Meta video.

找到了，就在这儿。

There go. There you go.

蓝色那款。昨晚参加Diplo跑团时就戴着这副，结果...对对，三十个人同时举着手机录像。

The blue ones. Was wearing these when I was doing the Diplo Run Club last night, and they were Right. Right. 30 people taking videos at once.

我还没参加过外交跑团，但我知道你是个跑步爱好者，那看起来超级有趣。其实我有点想深入了解超级智能这块，不过关于AI订阅服务的时间线，你是怎么考虑的？你觉得这些AI订阅大概什么时候能推出？

I haven't done the Diplo Run Club yet, but I know that you're a big runner, and so that that that looked like a ton of fun. You know, I I I kinda wanna get into the superintelligence side of this, but, you know, on the AI subscriptions, timeline wise, how how are you thinking about it? When when do you see those AI subscriptions rolling out potentially?

这个还得看情况。我们在Meta进行了重组，正在重新调整AI战略。Alexander、Matt和MSL团队都已就位。当然，我们对下一代Llama模型及其在智能眼镜上的应用潜力抱有很高期待。

I mean, we're gonna kinda see. You know, we've rebooted in Meta. Think we're revamping the AI efforts a little bit. You know, we've staffed up Alexander and Matt and the MSL team. And, obviously, we have big expectations for the next generation of the Lama model and what's able to be powered in the glasses from that.

我认为眼镜这类可穿戴设备是AI的最佳载体。它能看见你所见，听见你所闻，成为全天候的私人助手。当这些功能成熟时，自然会出现值得付费的高级功能——当然可能需要支付一定费用...

I mean, I think of the glasses or wearables as the best delivery vehicle for AI. I mean, this is a device that can see what you see, hear what you hear, and just be a personalized assistant that helps you throughout your day. And so when that functionality is ready, you know, I think there'll be features that that'll make sense to pay for or pay for, you know, some amount of

未来可能也会加入广告？

And maybe ad adver advertising too at some point?

我在Meta做了多年广告业务。广告只有在规模效应下才有意义，广告主不会为小众用户定制方案。在设备用户量达到足够规模前，这甚至不值得讨论。目前我们完全没有相关计划，这本质上是个规模门槛问题。

I mean, it's I mean, I worked on advertising for years at Meta. And, I mean, the thing about ads is it only makes sense when you have a large scale. Otherwise, it doesn't make sense to the advertisers don't wanna do custom formats that targeted at a small number of users. And so until we're hitting it's not even something worth talking about until we're hitting, like, a really large scale of device users. And so we have no plans in the works right now on that just because it's a you know, really a scale question before it's even interesting.

就像Threads现在也没投放广告，就是因为用户规模还不够。

Even, like, Threads doesn't have ads in it right now because it hasn't hit the scale yet.

明白。对了，你提到Alexander Wang和超级智能团队。作为可穿戴设备负责人，我想了解你们团队目前是如何与超级智能组展开协作的？

Right. Right. Well, so you you talked about Alexander Wang, and you talked about the super intelligence group. You know, one of the things I wanted to get an understanding from from you leading the wearables group is is how is your team, the wearables team, working with that super intelligence group right now?

我们正在紧密合作。眼镜上搭载的AI技术，目前雷朋眼镜的两大主要用途是作为手机的蓝牙耳机，以及捕捉那些你原本会错过的照片和视频。我们看到在运动场景中的使用率正在上升，这也是我们选择与Oakley合作的原因。但AI是第三大用途，其使用率也在增长。人们开始进行免提通话。

We're partnering closely. So we the AI that's on the glasses, and, you the glad the top two use cases for the glasses right now are for the Ray Ban Meadows are using them as a Bluetooth headset for your phone and for capturing photos and videos that you we would have otherwise missed, and we see increasing amount of usage in sports scenarios. That's why we leaned into the Oakley Medivangards. But AI is the third use, and we're seeing that increase. We see people doing hands free calling.

他们还用眼镜进行免提信息发送。目前最主要的AI应用场景还相对简单，就是即时提问。比如我六岁的儿子会不停地问我关于周围世界的问题。当我开车时——

We see them doing hands free messaging. The top AI use case is somewhat simple right now. It's just, you know, asking questions in the moment. You know, I've got a six year old son, and he has nonstop questions for me about the world around him. And as I'm driving

这套系统必须超级智能，连六岁小孩的需求也得满足。

for the suit it's super intelligent. Has has to has to satisfy the six year olds too.

他会问非常具体的问题。比如如果问我一个足球场能容纳多少人，回答‘几千人’根本打发不了他，我必须说出李维斯体育场的精确容量数字，否则他会一直追问。目前这个功能运行良好。眼镜上的MedAI体验是基于MSLT开发的模型，我们对其进行蒸馏和优化以适应设备形态。在眼镜上，如果收到冗长的文字回复会很烦人，所以我们做了大量工作来缩短响应内容。嗯。

He's got really specific questions, you know, and if you ask me how many people can fit in a football stadium, saying many thousands is really unsatisfying for him, and I've gotta tell him the exact number that fit in Levi's Stadium, otherwise, he will keep asking. And so we're seeing that work pretty well. And the experience the MedAI experience on the glasses is powered by you know, we take the models that the MSLT makes, and then we distill and we optimize them for the form factor that we have. You know, you want on the on the glasses, you want if you just got, like, a long response that would be in text, that's frustrating. So we do a lot of work to make the response shorter Mhmm.

然后

And then

进一步优化以适应眼镜的使用场景。

more optimized for what you'd want in the glasses.

所以实际上他们是在提供模型，并帮助你们调整这些模型，重新打包成适合——或者说能在眼镜上运行的模型规格？

So so they're actually they're actually you're they're actually giving you the models that and and they're helping you sort of adjust those models, repackage them into sort of the size of model that would fit, or I get not fit, but that would work with the glasses, essentially.

在某些情况下确实适用。这里涉及多个维度。我们采用了一种级联架构，有运行在眼镜上的模型，有运行在手机上的模型，还有运行在云端的模型。

Well, in some cases, fit. So, like, there's multiple dimensions to it. I mean, we have kind of a cascaded architecture where we have models that are running on the glasses. We have them running on the phone. We have them running in the cloud.

根据延迟要求、是否需要实时信息查询、以及未来是否需要代理功能或深度研究，架构会动态调整——无论是向眼镜还是手机发送指令。整体而言，我们的合作模式是与他们共同构建底层模型，因为如果想让AI在某些语言、领域或任务上表现特别出色

And depending on the latency requirements, depending on whether something requires a real time lookup of information, depending on in the future whether something will be agentic or do deep research, there's different the architecture is whether it's doing commands out to your glasses or the phone. I mean, it's it's architected pretty dynamically. But in general, the way to think about it is we're taking one, we're partnering with them to build the underlying models because if there are languages or domains or tasks that we want the AI to be especially great at

是的。

Yeah.

我们就会提供数据集和评估标准，帮助确保底层模型在这些方面表现优异。对吧。

Then we're providing datasets, providing evals. We're helping to make sure that the underlying model is great for that. Right.

听起来你们与这个团队合作非常紧密，而且这个团队本身也是新组建的。我想知道，鉴于新人才的加入，过去两三个月里他们在与可穿戴设备团队的合作方式或模型调整策略上是否有所改变？是否有针对你们合作做出的具体调整？

And so it sounds like you're working pretty closely with the group, and the group is obviously it's a newer sort of team that has come in. I wonder, has anything changed about their approach to how they've worked with the wearables team or how they've sort of adjusted the models in the last two to three months given the new talent? Have there been any adjustments that they've made explicitly working with you?

有件事我很欣赏——他们对可穿戴设备和我们开发的眼镜充满热情。不是说之前缺乏热情，但现在大家特别热衷于探索如何提升可穿戴设备的体验。Nat Freeman曾在我们顾问组任职，对产品路线图很熟悉。他加入后尤其热衷于推动可穿戴设备的发展。总体而言，他们带来了模型开发的新思路

Well, you know, one thing is I mean, they're one thing I love about them is they're very bullish about wearables and the glasses that we're making. And so they brought not saying there wasn't excitement before, but there is just a lot of excitement about see what we can do to make the wearables great. You know, Nat Freeman was on our, you know, in our advisory group for a while, so he was pretty familiar with with the road map. And one of when he came in, he was especially excited to figure out how to lead into wearables. But I'd say in general, you know, they've been great partners as we expect them to be, but they brought kind of a new they brought some new approaches to how to do model development Right.

在Beta阶段。这些新思路也影响了我们的合作伙伴关系，包括他们想要实施的模型训练方式、解决工程师和研究人员的瓶颈问题等。这些新方法让我们受益匪浅。

At Beta. And, you know, some of that spilled over into the partnerships we have just in terms of how they wanna do model training, what's holding engineers back or researchers back, and just there's some things are a bit of a new approach, and we've we've appreciated it.

那么，目前阻碍工程师和研究人员前进的因素是什么？

And what is holding engineers and researchers back right now?

我认为问题不在于我们仅仅试图最大化效率和生产力，比如每位加入的工程师都希望整天高效完成大量工作。所以我们正在分析，比如：当你入职时，需要多久才能完成设置？是否有可用的机器来训练模型？是否存在阻碍你快速进入高效工作状态的因素？

I'd say it's not one that means we're just trying to maximize efficiency and productivity for like, every engineer who who joins wants to have just be, you know, spend all day getting a lot of stuff done. And so we're just analyzing like, hey. When you when you arrive, how long does it take to get set up? Do you have, you know, machines you can use to train models? Do you have, like is there is there stuff that's in the way of just being fast and productive out of the gate?

所以这就像是一份待办事项清单，需要逐一解决。这与大约十六年前我刚加入时的情形并无太大不同，当时也是类似的情况。

And so it's, you know, it's just a list of things trying to trying to bang out. It's not not dissimilar to, you know, what was it, over sixteen years ago when I started here, and it was like, hey.

听起来MSL团队正在协助整个组织，或许也在帮助可穿戴设备团队变得更加灵活，让他们几乎一加入就能更快地推进事务。

So it it sounds like the MSL team is kinda helping the the you know, is sort of helping the organization and and maybe helping the wearables team actually be a little more scrappy with things and sort of get get things done a bit bit faster almost, or right out of the gate as they join.

听着，模型改进得越快，我们就能更快地提炼它们、缩小体积并降低延迟，这些都是我们迫不及待想要看到的好事。我们认为目前正在打造的可穿戴设备除了AI之外在其他方面也很出色，而且在AI领域正变得越来越强。未来一两年内，可穿戴设备将在AI领域变得极具竞争力，我们正为此做准备，确保能够有效竞争。

Hey. You know, the the faster the the models get better, and the the faster we're able to distill them and the smaller they get and the improvements in latency, this is these are all good things that we can't wait for. I mean, you know, we really think that you know, one of the great things about the the werewolves that we're building is they're great for stuff other than AI right now, and they're getting increasingly good at AI. And I think that one thing you'll see in the next one to two years, I think wearables will get really competitive in the AI space, and we're trying to, you know, get ready for that and make sure that we're able to compete well.

你如何看待与谷歌的竞争？

How do you think about competing with Google?

我只能猜测。我听过他们的演讲，但我...

I'd only be speculating. I mean, I've seen, you know, I've seen the talks they've given, but I

不，因为我是说，模型很棒。你知道，很多人认为这是它们的优势之一，就是模型运行得非常好，而且从根本上说，很多这些设备最终都是基于模型工作的。

don't because I mean, models are great. You know, a lot of people and that's that's one of the advantages that people say that they have is that the models work really well, and fundamentally, a lot of these devices wind up working on the models.

Gemini确实是一个令人印象深刻的模型，它的感知堆栈尤其出色，比如对事物的视觉识别能力，是的。那相当不错。他们有一个好模型。目前市场上还没有可穿戴设备，所以我们拭目以待。

Gemini is a Gemini is definitely an impressive model, and it's the perception stack is especially good, meaning, like, the visual recognition of stuff, which yeah. That's pretty good. They've got a good model. No wearable from the market yet, so we'll see.

所以你押注于你是第一个进入市场的，并且实际上给人们提供了可用的东西。

So you're banking on the fact that you're coming in first, and you're actually giving people something to work with.

你知道，我们……我认为实际上并不清楚眼镜是否会成功。很多公司在这方面有过几次虚假的起步，是的。或者尝试过但没有获得他们想要的关注度。我们也不例外，我认为Ray Ban Stories没有我们希望的那么成功，但我认为我们绝对展现了对这个领域最大的承诺。如果你看看，你知道，我们推出了五款设备，包括新款Ray Ban Meta眼镜、Oakley Meta Houstins、Oakley Meta Vanguard、Meta Ray Ban显示器和Meta Neural Band。

You know, we we've I think as it really wasn't clear if glasses were gonna be successful. You know, a lot of companies had kinda false starts there Yeah. Or or tried and didn't get the traction they were looking for. Us included, I think Ray Ban Stories wasn't as successful as we had hoped, but I think we've definitely shown the most commitment to the space. And if you look at you know, we launched we've launched five devices this half between the new Ray Ban Meta Glasses, the Oakley Meta Houstins, the Oakley Meta Vanguard, the Meta Ray Ban Display, and the Meta Neural Band.

显然，我们现在对这个领域有着相当实质性的承诺，并且正在推向市场。现在，我认为我们已经……

Like, clearly, we've had a pretty substantial commitment that is coming to market now in the space. Right now, I think we've

你没有被模型吓倒。那些模型没有吓到你。

You're not fazed by the by the model. The models don't scare you.

嗯，我……我认为我们的模型需要变得很棒。我认为AI需要变得很棒，才能真正实现我们和行业内其他人所设想的愿景。所以我们在这方面投入了很多精力，你知道，我们正在尽可能地在公司内部进行合作，因为这将是至关重要的。你已经开始看到一些迹象，比如能够识别你面前的物体，可以翻译你面前的标志。

Well, I I think it I think our models need to get great. I think the AI needs to get great for the devices to really deliver on the vision that we and others in the industry have. And so we're focusing a lot on it, you know, partnering you know, we're really partnering across the company as much as we can because that that is going to be critical. And you can start to see glimmers of it where, you know, you're able to recognize an object that's in front of you. You can translate a sign that's in front of you.

我最喜欢的功能，听起来很简单，但当我展示一个电话号码时，我可以说，看这个号码并拨打它。就是这些看似微小简单的事情，却能为你的一天增添点滴价值。我想你已经开始看到它的潜力了。

My favorite thing, it sounds really simple, but if I hold up a phone number, I can say, look and call this number. And it's like these small simple things, but they add, like, little bits of value throughout your day. I think I think you're starting to see the potential.

让我问你这个问题。目前阻碍你实现这款产品宏伟愿景的最大技术瓶颈是什么？是电池问题？重量问题？现阶段你最希望突破的最大技术挑战是什么？

Let me ask you this question. What is the the biggest sort of technical holdup right now from you achieving sort of your grandest visions for this product? I mean, is it battery? Is it weight? What is the biggest technical challenge for you right now that you wish you could just really break through it?

哦，这就是工作流程开发的特性——它不是单一问题，而是所有这些因素的平衡。

Oh, that's the thing about working on workflows. It's not one thing. It's the balance of all these

给我们列个清单吧。比如你现在白板上写的，哪些是真正需要突破的限速步骤？

Tell mean, us a list. You know, on your whiteboard right now, what are the the the rate limiting steps that you just really need to break through?

我认为我们已经展示了研发范围的两个标杆。首先是Ray Ban Meta智能眼镜，它们已在市场销售，昨天公布的升级让产品更出色。在与SLR和Luxottica合作过程中，我们的合作伙伴对'全天候舒适佩戴眼镜'有着严苛标准——包括尺寸、舒适度和重量。

Well, I think we showed the bookends of what we're working on. So we've got the Ray Ban Meta Glasses. They're in the market right now. They just got better with the announcement that we had yesterday. And these, you know, working with SLR and Luxottica, our partners who have a really strict definition of what it means to be an all day wearable, comfortable pair of glasses around dimensions, the comfort, the weight.

因此这些眼镜有着绝不能逾越的硬性限制。我们基本是以外形设计为先，然后在保持外形的前提下尽可能融入技术。这是一个标杆。另一个标杆是去年Connect大会上展示的Orion原型机，那款眼镜内含七块定制硅芯片，波导采用我们精选的最高端材料碳化硅制成。

And so we had real strict limits that we just could not violate in these glasses. And so these were basically, like, form factor first, and then we'll put as much technology in them as we can without sacrificing the form factors. That's one bookend. I'd say the Orion prototype that we showed last year at Connect is the other bookend, where, you know, those glasses had seven pieces of custom silicon in them. The the waveguides were made out of silicon carbide, which we chose is the material with the highest end.

这本质上就是你想要达到的目标。

That's what you wanna you wanna get to there, essentially.

是的。它配备了大量传感器，内含大量硅元件，还有一个70度大视场角的显示屏。对我们来说，路线图上的关键问题是——我们刚发布了一款介于两者之间的设备，就是Beta Ray Ban智能眼镜的显示屏。

Yeah. A bunch of it had a bunch of sensors. It had a bunch of silicon. It had a big 70 degree field of view display. And so the question for us, road map wise, is, you know, and we just announced a device, the, you know, Beta Ray Ban display, which is in between there.

没错。我们正在逐步推进。如果以FormFactor为起点，再加上Ryan这边的进展，我们正在向上攀登。但核心问题在于每个环节都需要权衡取舍。这很正常。

Right. So we're starting to march up. So if you had, you know, FormFactor first and then Ryan over here, we're starting to march up. But the the big question for us is, like, each thing is these trade offs. It's okay.

每个硬件组件都能创造附加值。比如，如果我们在眼镜中加入眼球追踪传感器，AI就能确知你的视线焦点，而不必猜测你面前场景中的关注点。但每项功能都会影响外形设计，这些都需要权衡考量。

For each of these hardware components, there's value you can add. Mhmm. Like, for example, if we added eye tracking sensors into these glasses, the AI would know what you were looking at as opposed to having to guess, you know, in the in the scene that's in front of you. But with each of them, you get form factor trade offs. And so each of these things were kinda weighing.

这本质上是一系列选择题。我们把Orion项目中的所有元素拆解开来，这些就是当前需要权衡的各个维度。当然，电池效率提升了，软件优化了，显示屏在保持尺寸不变的情况下实现了更高能效和亮度——这些都是我们正在攻关的方向。

And so it's, you know, it's a list of things. You basically take everything that was in Orion, and, you know, those are the different trade offs that we're looking at. But, yeah, you know, batteries got more efficient, software got more efficient. If the displays got more efficient, meaning more power efficient, had to be brighter with the same size. Like, these are all the things we're working through.

我想聊聊神经传感头带，但在你提到的权衡清单里，当前最困难的取舍是什么？哪个排第一位？

I wanna talk about the neural band, but on that list of trade offs, what is what is the hardest trade off right now? What's number one?

显示屏是最大的技术难点。绝对是所有组件中最具挑战性的。昨天发布的Meta Ray Ban智能眼镜，其显示屏达到5000尼特亮度，42像素/度，分辨率超过市面上所有主流头显设备。

The display is the hardest thing to do. The display is by far the hardest thing to do. I mean, the glasses that we announced yesterday, the Meta Ray Ban display. Yeah. I mean, they're 5,000 nits, 42 pixels per degree, which is a higher resolution than any headset major headset in the market.

佩戴时完全不会有光线泄漏，旁观者既看不到你正在浏览的内容，也听不到任何提示音。

And when you put them on, there's no light leaks out. So you can't so bystanders can't see what you're looking at or tell the noise.

所以要把所有这些东西都装进去，我我能理解

So fit fitting all that in, I I can see

但这是20度的单目视场。要在70度的双目视场中实现这一点，就像猎户座那样，但以那种外形尺寸来说，我是说，这，你知道，我们最大的投资是在显示技术上。我们不止采用一种技术，而是多种技术，这绝对是行业中最难解决的问题。

But it's a 20 degree monocular field of view. Accomplishing that in a 70 degree binocular, which is what Orion was, but with that form factor, I mean, that's, you know, the the largest investment we have is in display. We're doing not one technology, multiple technologies, and that's definitely the hardest industry problem to solve.

对。我们来谈谈神经腕带，这是你们本周公布的最酷的东西之一。我觉得它真的很令人兴奋。你知道，我立刻开始想象能用这个腕带做的所有事情。所以我有一大堆问题想问，但很快地，你们有没有考虑过在腕带上加装某种屏幕或附加其他东西？

Right. Let's talk about the neuro band, which is one of the coolest things that you unveiled this week. So, look, I thought it was really exciting. You know, I immediately started thinking of all the things you could do with the band. And so I mean, I've got a laundry list of things I want to get to, but I mean, very quickly, could you ever see yourself putting a screen of some kind or attaching something else to the band?

这是第一个问题。我是说，考虑过。我们多年前收购了一家位于纽约的公司Control Labs，他们一直在研究这种表面肌电技术。他们当时开发的腕带——真希望我今天能带来——非常大。看起来像是只能在布鲁克林找到的赛博朋克风格，但那是个研究用原型。

That's the first question. Mean, consider it. So we acquired a company, Control Labs, many years ago based in New York City, and they've been working on this surface electromyography technology. The band they were developing at the time, I wish I had it with me here, is is enormous. It looked like a cyberpunk thing that you find in Brooklyn only, but it was a research band.

它有很多不同的通道，体积庞大，他们当时试图建立机器学习模型来预测你大脑向手腕肌肉发送的信号。当他们取得一定突破后，我们完成了收购。后来我们把它变成了更理想的外形。

So it had a lot of different channels. It was big, and they were trying to figure out if they could build machine learning models to predict the signals that you were sending for your brain to your muscles on your wrist. And they'd reached a certain level of breakthrough that we made that acquisition. We've since turned it into a much more desirable form factor.

是啊。

Yeah.

但我们这次发布的第一代产品，重点在于，你知道，它只专注于一件事。唯一的一件事，那就是输入。是为眼镜服务的输入设备。这就是我们的起点。你知道，我们收到了大量相关问题。

But the generation one that we did here, the whole point was, you know, it's focused on one thing. One thing only, and that is input. And it's input for the glasses. And so that's where we started. There's you know, we get a ton of questions for us.

老兄，我真希望你能试试健身龙。这很有趣，懂吗？问题在于，你能把它变成手表或添加个...你懂的？确实有这种可能。

Like, man, I wish you could do fitness dragon. It's interesting. You know? It's the question, could you turn in a watch or add a you know? There's Yeah.

你真的可以考虑这些点子，要知道，我们正在审视所有可能的形态因素和功能特性。但目前我们全力确保第一代产品能完美运行，专注于输入体验优化——还有很多输入功能我们想添加到Well里。

You really can think of these things, and, you know, there's no form factor we're not looking at or or feature not considering. But right now, we're we're so focused on getting it to work with gen one and having it be great for input, and there's a lot of different input things we wanna add to Well,

你们会像与SR Lexotica合作眼镜那样，比如说与Garmin建立合作开发神经波段产品吗？

would you ever do, like, the the way you're you're partnered with SR Lexotica for the glasses? Would you do a partnership with Garmin, for example, for for the the Neural Band?

巧了，你提到Garmin。我们昨天刚宣布了与Garmin的合作关系。

Well, we did. It's funny now you mentioned Garmin. We announced a partnership with Garmin yesterday.

哦好吧，可能我漏看了这条消息。抱歉。嗯...

Oh, okay. Maybe I I missed that. Sorry. Well, it's

没...没在正式渠道公布。我们宣布的是：如果你佩戴Oakley Meta Vanguard系列...

not not in the it's so we announced it. So with the if you're wearing the Oakley Meta Vanguard's

对。

Right.

而且你戴的是佳明手表，它支持自动拍摄功能。比如昨天我在迪普洛伦俱乐部的跑步过程中——这地方你应该去看看，他就在附近。非常

And you're wearing a Garmin watch, it enables auto capture. So if you're going for so with a run I went on yesterday in the Diploren Club, which you should check out. He's he's around here. It's very

有趣。

fun.

对，对，是设置好的。

Yeah. Yeah. Was set up.

我知道你是跑步爱好者，所以那对你来说肯定特别有意思。

I know I know you're a runner, so that that I know that was a lot of fun for you.

但它的设定是每跑完四分之一英里，眼镜就会拍一段短视频，跑完后对吧？我们的应用会自动把这些片段拼接成集锦视频，并叠加你的佳明数据。佳明手表就是触发设备，你可以设置固定间隔触发——这对马拉松很实用，或者按特定速度触发。比如滑雪下坡达到某个速度时就会自动拍摄。

But it was set up so every quarter mile, the glasses would take a short video, and then at the end of the run Right. We automatically stitched together in our app a reel for you that had those short videos stitched together and it overlaid your Garmin sets on it. And the Garmin watch is what creates the trigger. So it'll you can do it at certain intervals, which be great for marathon, or at certain speeds. So, like, once you hit a certain speed skiing down a mountain, it'll Yeah.

没错，自动生成你的专属视频。

Yeah. The video for you.

等等等等，听我说。我是说，我们有没有可能和佳明达成合作？我觉得这简直太合适了，不是吗？

But but but stick stick with me. I mean I mean, could we could we see a Garmin a a Garmin partnership? I mean, it makes so much sense to me. No?

我是说，你看。我戴Garmin手表已经超过十五年了。我非常喜欢他们做的产品。我们还有其他合作项目，比如让跑步、骑行、滑雪、高尔夫或冲浪更有趣、更社交化。眼镜是防水的。

I mean, look. I've been wearing Garmin watches for over fifteen years. I'm a huge fan of the stuff they're doing. We've got other I mean, the partnership we started is they can do the, you know, fun make the runs more fun, more social, or cycling, or skiing, or golfing, or surfing. Glasses are waterproof.

不过，是的，也许还有更多可以做的，但现在我们真的专注于让输入功能正常运作。

But, yeah, hey, maybe there's more to do, but right now we're just really focused on getting the input to work.

好的。最后一个问题，因为我知道你得走了。我特别想问问你，人们具体会如何使用这款显示眼镜。过去24小时里我一直在思考的一点是，Marge说过，它能让你更活在当下，更专注。你不用一直看手机之类的。

Okay. Last question for you, because I know you gotta go. I really want to just ask you about how people how you see them using the display glasses specifically. And, one of the things that I've been reflecting on a little bit over the last twenty four hours is, Marge has said, it allows you to be sort of in the moment, more present. You're not looking at your phone and stuff like that.

我理解这个观点，但另一方面，我也在想这会不会反而让人更容易分心。我的意思是，你把手机收起来了，不想看它，对吧？但如果短信直接出现在我眼前，就像我说的那个例子，如果我在和你说话时不想因为发短信分心（那样很粗鲁），我可以用眼镜更隐蔽地操作。但这也意味着我没有真正和你在一起，没有全身心投入。

And I take that point, but on the flip side, I've been thinking a lot about how this could just be more distracting for people. What I mean by that is, you know, you put your phone away, I don't want to look at it, okay? And meanwhile, if I have the text messages coming in my eyes, you know, gave that story of, If I'm talking to you and I don't want to get distracted texting somebody because it's rude, right? I could just do it, you know, more subtly with my But, I mean, it means that I'm not there with you. I'm not present.

我反而更分心了。你怎么看待这个问题？

I'm more distracted. So how do you think about that?

所以，我的意思是，你得亲自试试这款眼镜。我们得给你安排个演示。戴上试用后会更清楚——这是单目显示器，意味着它只在单侧显示。20度的视野范围，大概相当于这里智能手表的大小。

So, I mean, you you gotta try the glasses. I mean, we we gotta get you a demo. We gotta I mean, it it this will be more clear and obvious, like, if you put a pair on and and give it a try. But, you know, it's a it's a monocular display, which means it's on one side. It's 20 degree field of view, so it's about the size of a smartwatch right here.

而且我们特意把它设计得稍微偏一点，所以它本质上是用于快速瞥视的显示。不适合长时间浏览内容，我们也没打算那么做。我们刻意没有同步手机所有通知（虽然技术上很容易实现），因为我们希望用户能选择性接收。比如我只接收WhatsApp通知，而我的WhatsApp基本上只联系我妻子。

And it's intentionally a little bit off to the side, so it's really meant to be a glanceable display. It's it's not intended for long form consumption of stuff. It's it's really not tailored for that. And we intentionally are not mirroring the note it's it would be easy to do, but we're not doing it mirroring all the notifications from your phone because we want it to be really selective for you in terms of what comes in there. So for me, I just get WhatsApp notifications, and for me, WhatsApp is very, like, it's my wife.

你知道，这就像一份精选的通知清单。实际上，我之所以带着手机跑步而不是把它留在家里，唯一原因就是如果我妻子打来电话，我总得接听对吧？所以它本应传递高价值信息。这样你就能快速处理，而不必像看手表那样需要抬头点击，或从口袋里掏出手机查看是否重要——它真的只需一次点击，毫不分心，让你更能活在当下。而且我们还有个简易控制中心。

You know, it's it's like a curated list of when that comes in, I actually like, the only reason I run with a cell phone as opposed to leaving that at home is that if I do get a call from my wife, I kinda need to answer it. Right. So it's meant to be high signal. And so you can just really quickly instead of, like, having to, like, look up at a watch and hit it, or grab your phone out of pocket to see if it's important, it's really, like, it's such a one click, non distracting tap that it's like, it really does help you be more in the moment. And then we there's an easy there's a control center.

如果你向左滑动，就能轻松开启勿扰模式。是的。

If you swipe left, it's really easy to hit do not disturb if Yeah.

但是，我的意思是，听我说。我在想，比如家庭晚餐时，你和妻子或孩子们坐在一起。我不知道未来愿景是否人人都戴这种眼镜，也许连你的孩子也戴。但假设在这个增强现实里所有人都会佩戴眼镜——难道你不担心孩子们吃饭时，我们正聊着当天的事，他们却不断收到短信？

But, I mean, I hear me. I'm I'm just thinking at at dinner, you know, your family dinner, you're sitting with your wife or or with your kids, And I don't know if the vision is to have everyone wearing the glasses, maybe even your kids are wearing the glasses. But let's say the idea is that everyone will be wearing glasses in this augmented reality. I mean, would you not worry that, hey. My kids are you know, they're eating dinner, we're talking about our day, and they're getting text messages.

就像...你知道，他们虽然人在场，心思却不在这里。

Like, you know, and they're they're not with me. They're not here.

是的。我认为不同情境会形成不同规范。就像现在家庭晚餐时，大家会把手机扣着放或放在别的房间。同理，眼镜也有关闭开关，如果你想专注当下的话。但设计初衷是，当你需要被重要通知提醒时——比如我在等一个电话...

Yeah. I mean, I think there's, you know, different norms develop in different circumstances. Like, you know, right now, you've got the, If you're having a family dinner, you know, it's like, put the phone face down or in the other room. I mean, similarly, like, glasses have a an off switch, you know, if if you wanna just be present in it. But the idea is that, you know, these are really intended to be great for, you know, when you're when you want when you when you want to be alerted if something kinda like, I'm waiting for a phone call.

我需要知道特定联系人的消息。目前我们认为AI会随时间推移优化这一点，学会区分高价值和低价值信息。就像那些邮件应用的AI扩展功能，能帮你筛选并突出紧急内容。产品设计理念就是低干扰地快速处理高价值信息。当然最终取决于用户使用习惯和逐渐形成的规范，其他设备形态也是类似发展路径。

Like, I need to know if a message from this person comes in. And, like, right now, at settings, we think AI will make that even better over time to learn what's high signal and what's not. You know, similar to how you've got all these AI extensions to email apps that allow you to really curate through and and boost signal for stuff that's urgent and just kinda push away stuff that's not. And so, like, the intention of the product and the way you've designed it is to make it, like, low friction and quick to get high signal stuff and deal with it. But, of course, like, a lot of this is just gonna spend on how people use it and the norms that develop over time and and similar to see with other form factors.

太棒了。Alex，这真是个很酷的产品。非常感谢你参加节目并分享战略思考。我们很珍惜这次交流，期待你未来发布更多产品时再次做客——可穿戴设备领域确实令人着迷。

Great. Well, Alex, it's a very cool product. Thank you so much for coming on the show and telling us more about how you're thinking about the strategy. We appreciate your time, and and I look forward to having you on more as you unveil more and more products. Wearables is a fascinating space.

这位是Alex Himmel，Meta公司可穿戴设备部门的副总裁。

That is Alex Himmel, the VP of wearables at Meta.

好的。非常感谢邀请我参加。

Alright. Thanks so much for having me.

谢谢。

Thank you.

刚才发言的是来自Meta的Alex Himmel。好的。那么，英伟达昨天发布了一些重大消息，宣布将斥资50亿美元入股英特尔，而这还不是其唯一的大手笔投资。该公司还同意支付9亿美元聘请一家名为Enfabrica公司的首席执行官，并吸纳其部分员工。为了让我们不仅了解Enfabrica是什么，还要解读这笔英特尔交易的长期意义，我为大家请来一个强大的专家小组。Anissa Gardesi是The Information网站的云计算领域记者。

That was Alex Himmel from Meta. Okay. Well, NVIDIA made some major news yesterday, announcing it is taking a $5,000,000,000 stake in Intel, and that wasn't the only big investment it is making. The company also agreed to pay $900,000,000 to hire the chief executive at a company called Enfabrica and also bring on some of their employees to tell us not just what Enfabrica is, but also what this Intel deal could mean in the long run, I want to bring on a great panel for you. Anissa Gardesi is our cloud and compute reporter at The Information.

Aaron Ginn是GPU托管公司Hydra的首席执行官，Jay Goldberg是Seaport Research Partners的高级分析师。非常感谢三位今天到场。周五能组成这么强大的专家团实属难得，我已经迫不及待要开始了。Anisa，请简单直接地为我们梳理核心要点——关于这笔英特尔交易，我们需要了解哪些关键信息？

Aaron Ginn is the CEO of a GPU hosting company called Hydra, and Jay Goldberg is a senior analyst at Seaport Research Partners. Thank you so much to the three of you for being here. It's a big panel for a Friday, but I am excited to get to it. Anisa, very quickly, mean, walk me through the brass tacks here. What do we need to know about the Intel deal here?

好的。英伟达正以数十亿美元入股英特尔，两家公司已同意共同设计几条产品线。我认为最引人关注的是，英伟达不再满足于仅提供与自家GPU配套的ARM架构CPU，而是希望与x86架构（即英特尔和AMD生产的那类处理器）实现更深度整合。简而言之，当前重点是英伟达持有英特尔大量股份，双方正在合作开发两款产品。

Right. So NVIDIA is taking a multibillion dollar stake in Intel, and the two companies have agreed to co design a few product lines. I think the thing that people will will be most interested in is that now instead of just offering an ARM based CPU next to the NVIDIA GPU, NVIDIA wants a tighter integration with the x 86 architecture, which is the kind of processor that Intel and and AMD build. So yeah, what you need to know is that there is NVIDIA has a large stake in Intel now, and they're working on two products.

再简单说明一下——考虑到可能有观众不太熟悉——能否解释GPU和CPU之间的关系，以及它们在这个事件中是如何相互关联的？

And very quickly, mean, just, you know, for people who might not be as familiar, just explain to us the relationship between GPUs and CPUs and how they sort of relate to each other in this story here.

所以在AI之前，大多数数据中心都充斥着您提到的CPU，即中央处理器。它们是数据中心传统的核心部件，能处理各种常规计算任务。但过去两年间，大量数据中心资源已转向GPU——这种由英伟达制造的图形处理器芯片。不过GPU离不开CPU运作，这正是英特尔这桩交易的意义所在。

So before AI, most data centers were filled with what you mentioned is called a CPU, essential processing unit. They're the traditional workhorse of the data center and can do all kinds of typical computing tasks. But over the past two years, a lot of data center capacity has shifted to GPUs, which is the chip and video makes. And so, but you can't have a GPU work without a CPU, and so that's sort of where this Intel deal comes into play.

明白了。这正是我希望厘清的细节。那么亚伦，我想听听你的看法——你认为这对英特尔来说是笔好交易吗？英伟达才是赢家？

Got it. Okay, that was the detail that I was hoping to get clarity on. Okay, so Aaron, I want to come to you. Do you think this is a good deal for Intel? Is NVIDIA the winner?

你怎么看？

What what do you think?

我认为现阶段对英特尔来说，任何交易可能都是好交易。英伟达此举的意图在于持续扩大对其生产线的控制权——你看他们最近的收购步骤，始终在设法保护其周边生态系统，提升客户体验的可靠性。因为在之前的CPU基础设施阶段，终端客户其实失去了对半导体之外环节的控制权。黄仁勋本质上想成为硅晶界的苹果公司，通过掌控更多客户体验环节来确保最佳用户体验。

I think at this point, probably, any deal is a good deal for Intel. Right? So so I I well well, I think what NVIDIA is trying to do here is they're trying to figure out ways of continuing to grow their control of their production line as well as, like, they if you look at the steps even of the most recent acquisition, they're always trying to find ways to protect their kinda surrounding ecosystem and finding ways of increasing customer experience, reliability, be because even under the the previous CPU stage of, you could say, infrastructure, the end customer kinda lost control outside of the semiconductor. So And what he's trying to do is control more of that customer experience, basically, to be the apple of silicon. So he's constantly trying to think of ways of assuring that the customer gets, like, the best experience possible.

选择英特尔部分原因是这家挣扎求存的企业拥有美国本土合法的晶圆代工业务。另一个选择AMD则存在风险——AMD的GPU产品线相当强劲。如果英特尔倒闭，英伟达就不得不从竞争对手那里采购支撑GPU业务的CPU。

And so some of the Intel stuff is to, you know, that the fact of, like, they're a struggling company. They're the one that has a legitimate foundry business in America. But the other one is that the the alternative would be AMD. And AMD has a a quite robust GPU, you know, business line. And so if Intel goes under or disappears, then they have to buy the CPU that could fund the GPU that is the alternative of their, you know, from their competitor.

所以我认为这更多是防御性布局。正如Anisi所说，他们的HDX系列仍需CPU支持，且ARM架构在数据中心层面仍存在技术限制，要说服客户从英特尔x86架构转向ARM仍需时日。虽然ARM终将成为英伟达旗舰平台，但未来多年他们仍需要x86平台来维持业务增长。

So so I think it's it's more structured around that where they're kinda playing its own defense here, but as well as, like, these still as as Anisi was saying that they need CPUs to run-in their HDX series and that there still is some technical limitations of ARM in the data center layer that they have to work through to get customers to move from x 86, which is the Intel CPUs, into ARM. The ARM, I suspect, will continue to be the flagship of NVIDIA, but there's still gonna be many years in which they need an x86 platform to continue to grow their business.

懂了。杰，亚伦提出的这些观点很有见地。你怎么看？你同意吗？

Got it. Okay. Well, Jay, that that that was those are some good points that Aaron brought up. What do you think? Do you agree?

我我并不是完全同意。我认为首先，让我明确一点，英特尔必须生存下去。这对行业很重要，对美国也很重要。所以他们能获得资金注入，我认为这是有帮助的。

I I don't I don't totally agree. I think I think the like, first, let me just say, Intel needs to survive. It's important for the industry. It's important for America. So them getting cash in the door, I think that that's help that helps.

他们需要资金来维持运营。

That's they they need they need money to keep this going.

没错。

Right.

但就这笔交易而言，仅此而已。我不认为英特尔能从中获得太多。这对英伟达来说是个好交易，让他们得以进入英特尔的核心云CPU领域。从绝对金额来看，这里的收入机会相当有限，因为如今大多数使用GPU的AI系统，正如所说，每个GPU集群都需要配备一定数量的CPU，比例大约是一比四或一比八。目前英伟达在其Grace系列CPU中提供这种CPU，并计划明年推出Vera进行更新。

But that that's about it as far as this deal goes. I don't think there's a lot in here for Intel. It's it's a good deal for NVIDIA, and then it gets them into Intel's sort of core cloud CPU base. In terms of absolute, you know, dollars, it's a pretty small revenue opportunity here, because in most AI systems today that use GPUs, like, this is says, every every GPU cluster needs some number of CPUs attached to it, sort of one to four, one to eight kind of ratio. And today, NVIDIA has that CPU in their Grace line of CPUs, and they're gonna update that next year with Vera.

所以他们在这一块已经准备得非常充分。这笔交易的作用是为那些不想使用英伟达CPU、不想完全依赖英伟达的客户提供AMD之外的替代选择。而英特尔就是这个替代选项。

And so they're in really good good shape around that. What this does is it it does provide an alternative to AMD for those customers who who want to use somebody else's CPU. They don't wanna use an NVIDIA CPU. They don't wanna be all NVIDIA, so they have they want some alternatives. And, you know, it's gonna be Intel.

但同样重要的是，这笔交易对英特尔的最新股东——美国政府很有利。对吧？黄仁勋昨天让唐纳德·特朗普赚了30亿美元，能做到这一点的人可不多。我认为这...

But I think most important too is this is a good deal for the Intel's new newest shareholder, the United States government. Right? Jensen Jensen Huang made Donald Trump $3,000,000,000 yesterday, so not many people can say that. And I think that is

是考量的部分因素。

part the calculus.

你认为这就是你认为这是这项投资中的主要考量因素吗？

You think that's you think that's the the the major calculus here in this investment?

不确定是否主要，但我认为这绝对是考量的一部分。

Don't know about major, but I think that is absolutely part of the calculus.

好的。亚伦，你怎么看？

Okay. Aaron, what do you think of of that?

我是说，我觉得我们其实表达了相同的观点，但从x86架构转向ARM存在相当显著的结构性差异。我们公司曾是Grace系列最大的采购方之一。

I mean, I I think we actually said the same thing, but but but there is, like, a there is a a structural difference of adopting ARM to x 86 that is quite significant. And so, you know, like, my firm is one of the well, we were one of the largest buyers of the Grace series.

再提醒一下，x86是指CPU架构。

And And just remind x 86 is the is the CPU.

对。对。它是一套指令集。没错。

Yeah. Yeah. Yeah. It's a it's a instruction set. Yep.

是的。所以你看，基带架构的特性就是ARM。这是最符合逻辑的迭代方向。你的手机用的是ARM。任何密集型计算框架通常都基于ARM的高效特性。

Yeah. So so and so, you know, like, the the the feature of baseband architecture is is ARM. It's the most logical kinda x and x iteration. Your phone is ARM. And any type of, like, dense computing framework, it generally means an ARM based on efficiency.

同样，这也延续了黄仁勋希望继续整合区域生态系统的趋势，使之达到类似苹果的水平。所以不应忽视这一点——他确实用了这个比喻。要知道，苹果使用的是ARM架构。而且别忘了，他们曾试图收购ARM对吧？这清晰展现了所有企业都向往的发展轨迹。

And as well as they it continues the same trend of which that Jensen is wants to continue the integrated zone ecosystem together to where he is like Apple. So it shouldn't be lost I mean, he's literally used that metaphor. So it shouldn't be lost on you that, you know, Apple uses ARM. So, and again, to try to buy ARM, Right? So, that is a clear trajectory where everyone wants to go as a company.

但x86架构问题在争取企业客户时是个重大障碍。正如Shreya所说，考虑到英伟达60%-70%的利润率（视季度和业务而定），这笔钱根本不算什么。相比每年数千亿美元的规模，这对英伟达来说微不足道。但代工计划对美国主权和黄仁勋本人而言都至关重要。

But the x 86 problem is a significant problem if you're trying to win over enterprise customers. Right. And know Shreya was saying that they I mean, this money is like nothing compared to when you have a 60 to 70%, you know, profit margin, depending on what quarter, what you're doing, you know, it's not like hundreds of billions of dollars a year. Like, this is a nothing on the on the spectrum for NVIDIA. But the but the foundry program is is is super important to, like, America's sovereignty and and to Jensen himself.

不过黄仁勋已押注台积电。但正如Jay所言，这里存在明显的政治导向——忽视这点就太愚蠢了。英特尔确实需要支持，这是摆在眼前的事实。

But, you know, Jensen has put his bag with TSMC. But but as Jay was saying that there is clearly political orientation here. Like, to to look past that would be silly. It is it's very it's staring right in front of your Right. The intel needs to be supported.

确实如此。在当前政府构建的某种特殊框架下，这某种程度上已半国有化。作为最重要的半导体企业，英伟达必须找到整合这两者的方法。

That's true. And it's semi nationalized in some, you know, weird construct that we kind of embedded this this administration. Right. And and so Jensen is the most important semiconductor company in the So so you you have to find ways to which they integrate these two things.

Anisa，你昨晚参加了媒体电话会议。有什么关于代工业务的特别内容让你印象深刻吗？

Anisa, you were on the the the press call last night. What what did you get from that? Was there anything about the the foundry business that that they said that stood out to you?

是的。现场有很多关于这笔初始交易是否可能促使英伟达未来成为英特尔代工客户的提问。两位高管都未正面回应：黄仁勋称英伟达始终在评估各种选项，而Lip Bu则完全回避了这个话题。正如Erin和Jay所说，美国政府看重英特尔的正是其代工能力。

Yes. There were tons of questions to Jensen and Lip Bu about whether this initial deal could potentially lead to NVIDIA becoming a customer of the Intel Foundry down the line. And both executives did did not directly answer that question. Jensen said something about how NVIDIA is always evaluating different options, but Lip Bu didn't really talk about this angle. And to what both Erin and Jay are saying, the the reason that that the US government cares so much about Intel is for its foundry.

而这次交易丝毫未能让人看到英特尔与英伟达在这方面关系发展的乐观前景。

And this deal kind of had no optimism around, you know, that aspect of an Intel Nvidia relationship.

听起来他们甚至没有讨论过这个问题，对吧。

It sounds like they didn't they didn't even talk about Right.

他们讨论过了

They did

没有讨论这个。

not talk about it.

Jay，我是说我们抱有希望。我们想要这样。嗯，我是说，Jay，就像，我们确实讨论过。Jay，我是说，你当时对这种情况感到非常震惊。

Jay, I mean We're hopeful. We want this. Well, I mean, Jay, like like, I mean, we were talking yes. And Jay, I mean, you you were you were astounded that that this was the case.

我是说，这对我来说真的很不可思议，他们花了这么多时间——昨天会议上至少一半的问题都是关于这个的，但他们拒绝承诺使用英特尔作为代工厂。真正让我印象深刻的是，英特尔CEO Lip Bu Tan 对自己的代工工艺连一句好话都说不出来。对吧？反而是NVIDIA的Jensen能够称赞说，哦，英特尔有很棒的封装技术。但就连Lip Bu Tan自己也没法说，我们会评估所有选项，但我们真的很喜欢自己的内部工艺。

I mean, it was it was really remarkable to me that they spent all this time at least half the calls half half the questions at the conference yesterday were about this, and they refused to commit to using Intel as a foundry. And what really stood out for me was Lip Bu Tan, the CEO of Intel, could not say a single good thing about his own foundry process. Right? It was it was Jensen from NVIDIA who actually was able to talk up oh, you know, Intel has this great packaging. But, like, even Liplu Liplu because he couldn't say, like, oh, we're gonna evaluate everything, but we really like our internal process.

他甚至连这个都说不出口。在我看来，这已经很能说明问题了。我...我觉得他...我非常担心这件事，因为就像Aaron说的，我们需要英特尔的代工业务成功。是的，为了整个行业，为了国家，我非常非常担心它可能没有未来。

He couldn't even say that. And, I mean, that that to me spoke volumes. I I think he's I I think I'm very concerned about it because I I like like Aaron was saying, like, we need we need Intel Foundry to work Yeah. For the industry, for the country, and I'm very, very concerned that's that's not doesn't have a future.

我...Aaron，我给你30秒时间，最后再说一下你的想法，然后我们想继续讨论Fabrica，请说说你最后的想法，之后我们就进入下一个话题。

I I I Aaron, I'm I'm gonna give you thirty seconds here just to just to give us one last thought, and then I wanna go on to and Fabrica, just get your last thought here, then we'll move on.

我认为，早期Aaron的电话，也就是英特尔CEO最近那次通话，他基本上是和Asher谈了关于他的Foundry业务，并表示他们只会承接那些客户订单门槛极高的项目。但任何在Foundry领域工作的人都知道，这需要你深度融入客户，与他们合作多年才能最终推出新产品。从技术角度看，英特尔实际上是一家非常优秀的工程公司，在芯片创新和进步方面有着悠久的历史记录。所以，他们能做到这一点。

Well, I think that the early Aaron's call, the last one that the the Intel CEO did, he did, like, basically kinda talk to Asher about his Foundry and said that, like, we're only gonna do things at which have this, like, really high bar of, like, customer orders. But as anyone who works in Foundry knows that it gets much more of you have to integrate yourself to the customer and work with them over a number of years to finally launch a new product. And so the Intel's actually, from a technical perspective, very good engineering company. It has a long track record of having great innovations, great advancements on the chip side. So, they can do this.

只是在结构上，他们陷入了困境，不知道如何为公司外部的人提供服务。我认为Pat犯了一个错误，他说英特尔必须成为自己的客户。这可能是错误的决定。但Pat的大方向是对的，那就是这项业务必须继续下去。

It's just structurally, they got stuck into this, they don't know how to service people outside of their own company. And I think it was a mistake under Pat to just say, Intel has to be its own customer. I think that that was probably the mistake. But, you know, Pat had the general right orientation, which is like this business has to continue. Right.

我们必须找到加速这一进程的方法。但我认为他们确实在努力思考：我们能否真正做到Gentle希望我们做的一切？

We have find a way of accelerating this. But I think they really struggle with trying to the idea of like, can we actually do everything Gentle wants us to do?

对，对。

Right. Right.

Adisa，我想听听你的看法。关于Enfabrica的这笔交易。Enfabrica是我们Top 50榜单上的公司，你还记得我们怎么称呼它吗？就是信息行业50家最有前途的初创企业。我们在新闻编辑室称之为TI50。

Adisa, I wanna go to you. So this deal with Enfabrica. Enfabrica is a company that you actually know from our top 50 what call it? It's the information 50 most promising Yeah. Startups We call it TI50 here in the newsroom.

但Fabrica，如果我没记错的话，是你整理的计算机类别中的首选。我想知道为什么它在你的榜单上排名如此之高，这可能告诉我们为什么NVIDIA如此渴望获得这些人才。

But Fabrica was, I think, if I recall, the top pick in our compute category that you put together. I wanna know why they were so high on your list because that maybe might tell us something about why NVIDIA was so eager to go after the talent.

我将Enfabrica排在计算类初创企业榜单的前列，是因为他们正在解决企业在组装NVIDIA GPU集群时面临的主要问题。这些芯片集群越来越大，而芯片之间以及机架系统之间的网络连接技术却没有以同样的速度发展。因此，Enfabrica的创始人希望创造一种更好的网络结构——你大概能猜到他们公司名字的由来——用于连接大型GPU集群。

I I ranked Enfabrica high on the list of computing startups last year because they were tackling one of the major problems that companies face when they put together NVIDIA GPUs. These clusters of chips are getting larger and larger. And the networking between each chip and between each rack system, is not advancing at the same rate. And so, the, the founders of end fabric wanted to create a better network fabric. I think you can sort of get how they chose their name, a better network fabric to connect large clusters of GPUs.

是的。他们当时正在开发一项技术，这项技术显然注定有可能在英伟达集群中发挥作用。英伟达非常重视网络技术，他们收购了迈络思就是明证。这就是为什么他们去年在负载方面处于领先地位。

Yeah. And so they were, they were building a technology that was clearly destined to work potentially, you know, in Nvidia clusters. And NVIDIA cares a lot about networking. They obviously bought Mellanox. So that's why they they were highest

去年在负载方面。

on the load last year.

杰伊，这种真实感让我觉得这与我们上次节目中的谈话有关，你提到英伟达想向客户销售整套解决方案，而你实际上认为这可能会阻碍业务发展。

Jay, this this kind of real I feel like this relates to a conversation that we had last time on the show, which is you talking about NVIDIA wanting to sell customers the entire package, and you actually saw that as something that could hold that that could hold the business back.

好吧，我这么说吧。现在围绕AI计算的争夺战很大程度上正在演变为网络技术的较量。如果你私下与英伟达交流，他们会表示相较于竞争对手，其GPU的一大性能优势在于拥有这套网络技术栈。而另一方面，他们的主要竞争对手博通则具备深厚的网络技术专长。因此这实质上是两家公司之间的对决。

Well, I'll I'll say this. The the fight over AI compute is now very much shaping up to be a fight around networking. If you if you talk to NVIDIA, off off the record, they'll say one of the big performance gains advantages they have versus the competition in GPU is the fact that they have this networking stack. And then, but in contrast, you have their their rival in this space is shipping up to be Broadcom, which had to come with deep, deep networking expertise. And so this is really the the the battle between these two companies.

今年或明年即将展开的最大战线将围绕网络技术展开。因此苹果这次对Enfabrica的招聘很重要，因为这实际上是在这场即将到来的战斗中进一步加码。

The the biggest front that's gonna open up this year or next year is gonna be around networking. And so this Apple hire or at Enfabrica is important because it really is sort of one more big raising the stakes even further in this upcoming fight.

没错。

Right.

艾琳，如果我没记错的话，Enfabrica的CEO曾在博通工作，确实是业界知名人士。但我要补充的是，昨晚我与一些了解交易内情的人交谈过，他们提到英伟达不仅聘用了这位CEO，还招募了该公司相当大部分的工程团队及其他成员。这不像某些AI初创公司那样只重点挖角某个核心人物，我感觉该公司很多人都会加入英伟达。

Erin, if I if I remember correctly, the CEO of Enfabrica was at Broadcom. So he's definitely like a well known person, but I will mention, I I talked to some people familiar with the deal last night and they mentioned, yes, this is we're hiring, you know, NVIDIA's hiring the CEO, but they're also hiring a very large part of the engineering team and others at the firm. So it's sort of not like one of those AI startups where you could have one person who's, who's the main deal that you want to hire. I get the sense that a lot of people at the company are going to go to Nvidia.

亚伦，你刚才点头了。你对此有什么想法？你觉得这个举动是否有些出人意料？

Aaron, you you were nodding. What what what were your thoughts here? Do you think it's a it was a surprising move at all?

不。我是说，他们非常青睐收购方。英伟达的大部分收购对象都是收购方。再看看他们对Leptin的收购，如果你观察那个团队的成员，很多人都是ARM专家，他们在构建ARM框架方面有着长期的成功记录。所以他们喜欢招募经验丰富的高手加入团队，对他们而言回报巨大，而对英伟达来说成本却很小。

No. I mean, they they they love Acquirers. The that's a large portion of NVIDIA's acquisitions are Acquirers. And as well as if you look at their their Leptin acquisition, if you look at the talent of that team, many of them are experts in ARM, and they have, like, a long track record of, like, building out ARM frameworks. So so they they like hiring very good, highly experienced people to, like, join the team and as well as, like, you know, for them, the the payout is immense and as well as, like, to NVIDIA, small.

正如杰伊所说，在这种规模下，竞争本质上是网络技术的较量。如果你查看数据中心设备的折旧周期——毕竟我运营过数据中心——网络设备的折旧周期是计算节点的两倍。这意味着网络设备会比计算设备多服役一倍时间。

And so when you're at this such scale, as Jay was saying, that it is a battle over networking. And and if you actually look at the depreciation schedules of because, you know, I operate data centers. You look at depreciation schedules for the node versus the networking. Networking is about twice as long as the nodes. So it's gonna stay there twice as long as the actual, like, compute equipment itself.

因此网络设备在数据中心里具有更强的永久性，而且网络部署比其他技术栈要困难得多。设计数据中心时，计算节点可以随时更换维护，但网络架构通常固定不变。这就是为什么你会看到那些复杂的线缆工程——需要耗费大量人力。所以掌控网络技术对维持市场地位至关重要。

And so, it's a much more permanent set into the actual data center as well as implementing networking is significantly harder to do than the rest of the actual stack. Because you can slot in you're designing when you're implementing data center, you can slot the nodes out because you gotta, well, work on them occasionally. But the networking generally just stays there. And that's why all the crazy cables you see, right, the cable work, is like that's a lot of main man hours to do. So it's an important part of maintaining his position in the market is maintaining control of networking.

没错。我觉得这个话题值得深入探讨，我们应该邀请更多网络初创公司上节目——这是我从这次讨论中得到的心得。阿妮萨，或许可以请你联系去年TI50榜单上的公司，看看能否邀请几位CEO来节目？毕竟这些初创公司表现相当出色。非常感谢三位今天的分享，我们非常感激。

Right. Well, look, I think we could talk a lot longer about this, and, know, we should have more networking startups on the show, is my takeaway from this. So Anisa, maybe we'll get you to go back to those companies that you picked last year on TI50, we'll see if a couple of their CEOs will come on because it's I mean, look, we're batting a pretty high average here with those startups doing great things. Thank you so much to the three of you for coming on. We really appreciate it.

这三位是阿妮萨、杰伊和亚伦。好的，现在说说埃隆·马斯克——他有多位得力助手协助管理商业帝国。今天我的同事西奥·韦特发表了一篇人物特写，关于这位帮助马斯克实现数据中心野心的关键人物：布伦特·梅奥。他从全球石油钻井平台的工作经历，到现在主导xAI超级计算机项目，同时还是孟菲斯市一场污染纠纷中的重要角色。

That is Anisa, Jay, and Aaron. Okay. Well, Elon Musk has a number of lieutenants helping him run his empire. And today, my colleague, Theo Waite, published a profile on one of the most interesting people responsible for helping Musk achieve all of his data center ambitions. Brent Mayo has gone from working on oil rigs all over the world to now being key to x AI supercomputer efforts, and he's also an important character in a pollution battle that is shaking out in the city of Memphis.

现在有请我们的埃隆·马斯克领域专职记者西奥·韦特，为我们详细讲述这个故事。欢迎来到节目，西奥。很高兴再次见到你。

Joining me now is Theo Waite, our reporter on all things Elon Musk, to tell us more about this story. Welcome to the show, Theo. Great to have you back.

感谢邀请我参加。

Thanks for having me on.

好的。在我们深入讨论布伦特·梅奥之前，我想先铺垫一下背景。我们知道数据中心是人工智能发展的重要组成部分，但有一点我们常常忽略，那就是建设这些数据中心需要多大的投入，以及它们需要多么迅速地投入使用。你能详细讲讲这方面的情况吗？

Okay. So now before we get into Brent Mayo, I mean, I just want to set the stage a bit. We know that data centers are a huge part of the AI story, but I think one thing that we forget is just how big of a lift constructing these data centers are and how quickly they need to come up. Can you tell us more about that?

是的。过去建造一个数据中心需要多年时间。你得先建好大楼，等待公用事业公司和政府部门的审批。通常来说，一个设施从开始建设到投入使用可能需要三到五年。

Yeah. Building a data center used to be something that took many years. You know, you'd have to build a building. You'd have to wait for approvals from utilities and governments. And, you know, oftentimes it could take three, four, five years to get a facility online.

人工智能的蓬勃发展使得企业不惜一切代价，采取极端措施以最快速度让数据中心投入使用。在这方面，XAI表现得最为激进。他们在不到两年时间里，从零开始建成了两座超大规模设施——位于孟菲斯的'巨像1号'和'巨像2号'。如地图所示，这些设施位于城市人口密集区，这在当地引发了很大争议。我们为此做了分析，发现在第一个数据中心五英里半径内约有2万人居住，第二个则达到15万人。问题在于，XAI使用天然气涡轮机为这些设施供电，而非等待接入电网，这源于埃隆设定的极其激进的计划——要求以最快速度让计算设备上线。

The AI boom has, you know, made it so that companies are doing pretty extreme things to get data centers online as quickly as possible at all costs, and x AI is pretty much the most aggressive in in that way. They've gone from zero data centers to two pretty massive super sized facilities in Memphis in less than two years called Colossus 1 and Colossus 2. You know, as you can see on the map there, they're in pretty dense areas of a city, which which makes them pretty controversial locally. You know, this we did an analysis for this story and found that, you know, within a five mile radius of these, data centers, there's about 20,000 people, within a five mile radius of the first 150,000 within a five mile radius of the second one. And the reason that matters is that XAI is using natural gas turbines to power these facilities rather than waiting for a connection to the grid, which is because of, you know, the super aggressive pace that Elon has set for them to get compute online as as quickly as possible.

你提到了污染问题。这些天然气涡轮机与之有何关联？它们如何加剧了污染？

And so you you talked about the pollution issue. What do natural gas turbines here have to have to do with that? How how do they make it worse?

XAI使用的涡轮机原先常用于石油天然气行业的水力压裂作业，通常位于偏远地区。通过卡车运输大量天然气到现场，就能快速获得电力，远比等待电网接入要快。但作为天然气涡轮机，它们同时也在排放污染物——这在荒无人烟的压裂作业区问题不大，但在城市里就成了一大隐患。

So they you know, the the turbines that XAI is using are turbines that were previously, you know, used often in, like, fracking for the oil and gas industry where they're sitting in the middle of nowhere, and you can truck in a bunch of natural gas, put it in there, and get a bunch of power online way quicker than you could waiting for a hookup to the grid. But they are, you know, natural gas turbines, therefore, they're emitting pollution at the same time, which is less of an issue when you're in the middle of nowhere building a, you know, fracking setup, but a much bigger issue when you're in a city.

我想了解你报道的布伦特·梅奥，但我们是否知道他们为何选择这个特定区域？为何要建在人口更密集的城市和城镇附近？今天不便多说。

And I want to get to this person you wrote about, Brent Mayo, but do we know why they picked this particular area, and why they're so close to these, I guess, more dense, or urbanly dense cities and towns. Can't speak today.

当然不是。因为那里已经有现成的建筑等着他们。第一个数据中心就设在一家闲置的电器工厂里，那里原本就有一些电网接口可供后续扩建。第二个数据中心则直接选在了仓库区，周围全是同类建筑。

Sure. No. Because there there were buildings ready for them. The first data center is in a former appliance factory that was sitting empty and had some, you know, connections to the grid that they could eventually build on. The second one was just a a warehouse in an area with a bunch of other warehouses.

所以他们基本上可以直接入驻，开始往里面堆放设备。

So they could just, you know, show up and start putting stuff in.

说说布伦特·梅奥吧，他在这个故事里扮演什么角色？

So tell me about Brent Mayo. How does he fit into the story?

他算是那种只可能存在于埃隆·马斯克宇宙里的人，因为他的职位描述极其灵活。在加入马斯克阵营前，他长期在墨西哥湾、中东等全球各地的石油钻井平台和油气工程领域工作。2022年加入SpaceX后，去年又随着xAI筹建孟菲斯首个数据中心时转岗过来。最初是个幕后工程型人才，但随着xAI成为焦点，他也越来越多走到台前——不断会见政要、商界领袖和社区代表，为xAI在当地争取支持。

So he's he's kind of a person that could only exist in Elon Musk's universe because his job description is incredibly fluid. He spent, you know, a lot of his career pre Elon working in oil rigs and oil and gas engineering in the Gulf Of Mexico, The Middle East, all around the world. Then he joined SpaceX in 2022 and moved to x AI last year when the company was starting to build its first data center in Memphis. He was initially kind of a behind the scenes engineering type of guy, but he's gotten a lot more public as xAI has become a big story. You know, he's constantly meeting politicians, business leaders, community leaders to try to win support for x AI in the city.

马斯克究竟特别欣赏他哪一点？梅奥有什么特质吸引了马斯克？

And why why does Musk like him specifically? What is it about Mayo that Musk has taken a liking to?

简单说就是典型的'马斯克生态适应者'——精力超群，随便丢到什么难题面前都能24小时连轴转的那种人。他完全践行马斯克的硬核文化，随时准备接听

Well, he's the, you know, kind of person that thrives in Musk world, which is someone that's extremely energetic that you can just kind of throw at any problem, and he'll work twenty four seven toward it. You know, he's all about the Musk hardcore culture where you could get a

老板的深夜来电

call from the boss in

无论是深夜还是白天任何时刻，他都会对你要求做的事充满狂热。同时他也是个相当有魅力的人。我在孟菲斯与许多政客和活动家交谈过，他们可能并不完全赞同他，但都说他虽然强势，却有种让人难以抗拒的魅力，即便你不同意他的观点。

the middle of the night or, you know, at any hour of the day and and, you know, be super gung ho about whatever he wants you to do. And he's also, you know, a pretty personable guy. I I talked to plenty of people in Memphis, you know, politicians and activists who might not necessarily agree with him, who said that, you know, he's intense, but he's, you know, somewhat charming even if you disagree with him.

如果我们把视角拉远一点，随着所有AI公司都在加速建设数据中心，想到马斯克有这样一位得力助手挺有意思。或许Sam Altman也有自己的心腹。这类工业界的关键人物在这些亿万富翁的故事中，尤其是那些急于推进项目的人眼里，想必正变得越来越重要。

And tell me about if we just zoom out a little bit, I mean, I imagine as all of these AI companies seek to ramp up their data center efforts, It's kind of interesting to think that, I mean, Musk has this guy. I don't know, maybe Sam Altman has his own person. You know, like, feel like these sort of industrial characters must be getting more and more, I guess They must be becoming more and more important to this story for all of these billionaires, especially that are trying to wrap this stuff up.

确实。XAI是个极端案例，他们在数据中心建设上表现得尤为激进迅速。但Meta、OpenAI等公司现在也开始使用天然气涡轮机——这曾经是它们绝对避讳的。当这些企业进入城市时，像AWS、谷歌这样拥有庞大公关部门和政府关系团队的公司，会派西装革履的专业人士处理社区和政界事务。梅约显然不是这种风格。

Yeah. I mean, XAI is an extreme example. They are, you know, being especially aggressive and and fast with building data centers, but Meta and OpenAI and, you know, other companies are also using natural gas turbines now, which which used to be a huge no no for basically all of them. But when they go into a city, you know, AWS, Google, like, these are companies that have huge public relations departments and government affairs people that, you know, are slick and they wear suits and they you know, that that's fully their job is dealing with communities and politicians. Mayo is obviously not that.

据参会者描述，他在与孟菲斯市长及当地要人的会议上，穿着连帽衫出现，满口粗话。AWS的人绝不会这样——他们必定西装革履，保持高度专业。

Like, I talked to people who were in this meeting with him and the mayor of Memphis and several other prominent locals where he, you know, shows up wearing a hoodie and he's, like, dropping f bombs and stuff. That's not what someone from AWS is gonna do. Like, they're they're gonna be in a suit and they're gonna be very professional.

他接受过你的采访吗？

And did he talk to you for the story at all?

很遗憾，没有。

No. Unfortunately.

如果有机会采访，你会问他什么问题？

If you did have an interview with him, what what would you ask him?

根据他谈论为埃隆工作的方式，以及他前女友动态里的帖子，我想问，你每晚睡多久，怎么能保持如此充沛的精力？

Based on the way he talks about working for Elon and, you know, the post on his ex feed, I would say, how much sleep do you get every night, and how do you have so much energy?

但我的意思是，跟我一起即兴聊聊。从战略角度看，我有点好奇他会如何看待XCAI正在构建的超级计算机策略。你知道吗？我甚至在想，他是不是觉得‘更快、更快、更快’就是竞争优势所在？

But I I I mean, riff with me here for a second. I mean, you know, strategically, I I I kinda wonder, you know, how he might think about the supercomputer strategy that XCAI is building out. You know? I almost wonder, you know, how he's thinking about it. Like, is it just faster, faster, faster that is, you know, the competitive advantage?

或者，如果他在我面前，我可能会问，你如何让这个数据中心与众不同？我不知道。我好奇这些话题你是否和HasOrbit的其他人讨论过。

Or, I don't know, if I've had him in front of me, I might ask, you know, how are you building this data center to be different? I don't know. Those I wonder if those are questions that you that you talked about with the other people in HasOrbit at all.

不，我觉得这些问题很合理。XAI的长期战略相当不明确。他们投入大量资金和资源建设这些数据中心，但我不确定他们是否完全清楚长远如何供电——比如除了现在正在安装的涡轮机之外。关于长期战略，我有太多问题想问他。

No. I I I think those are fair questions. I mean, the long term strategy for XAI is pretty unclear. Like, they're throwing a bunch of money and a bunch of resources at these data centers, but I don't think they fully know how they're gonna power them in the long term, like, beyond, you know, the turbines that they're they're setting up right now. I I I think there are tons of questions about the long term strategy that I'd love to ask him.

没错。太棒了。西奥，非常感谢你参加节目，也感谢你和我们的图形团队一起制作了那张地图。这对不熟悉美国那片区域的观众非常有帮助。这位是The Information的埃隆·马斯克专题记者西奥·韦特。

Right. Great. Well, Theo, thank you so much for coming on the show, and thank you for putting together that map with our graphics team too. That was very helpful for those of us who are not familiar with that part of America. That is Theo Waite, our Elon Musk reporter at The Information.

好的。OpenAI已明确其硬件野心，而我们今天在The Information发布的新报道显示，乔尼·艾维并非该公司唯一想挖角的苹果前员工。OpenAI还从苹果挖来了其他关键硬件人才，甚至接触了部分相同供应商。我想请参与该报道的记者亚伦·蒂利来详细聊聊他的发现。亚伦，欢迎回到节目。

Okay. Well, OpenAI has made its hardware ambitions clear, and now a new story we published today at The Information reveals that Joni Ive is not the only Apple alumni that the company is going after hiring. OpenAI has been poaching other key hardware talent from Apple, and it is even talking to some of the same suppliers. I want to bring on Aaron Tilly, one of the reporters on that story, to talk more about what he's found. Aaron, welcome back to the show.

很高兴你能来。

It's great to have you.

嘿，谢谢邀请我来。

Hey, thanks for having me.

那么，告诉我你在报道中发现了什么。

So tell me about what you found in your reporting.

是的，我们发现OpenAI确实在加速推进这些硬件项目。显然，他们计划在未来一年左右发布一款设备。为此，他们充分利用了与苹果的人脉网络。他们聘请了苹果的员工来启动这个项目，并且正在不断扩大这个团队。这些人脉全部来自苹果的硬件团队。

Yeah, so we found that OpenAI has really been accelerating these hardware efforts. Obviously, it's on track to release a device in the next year or so. And to do that, they've really engaged their Apple network. They hired Apple people to start this, and they're continuing to build out that team. And the network is all from Apple hardware team.

所以他们非常积极地招揽这些人，而且进展相当顺利。这些人才对加入OpenAI感到非常兴奋，这是一个有趣的机会。对他们来说，挖角这些人才并不困难。

So they're they're really aggressively going after these folks and they're doing and it's it's a pretty it's a pretty easy job to kind of find these folks. They're very excited to move to OpenAI. It's a it's an interesting opportunity. And, they're it's they're finding an easy time poaching these folks.

所以说服他们跳槽过来很容易？

So so it's been it's been an easy sell to to get them to come over?

是的。根据我与相关人员的交谈来看，因为

Yeah. It's been I I I think from what I've spoken to folks because

苹果是一家很棒的——我是说，苹果是一家非常值得为之工作的公司。

Apple's a great I mean, Apple's a great company to work for, I think.

对吧？哦，是的。它很稳定。他们不像科技界的其他公司那样裁员。我认为这是个很棒的地方。

Right? Oh, yeah. It's it's stable. They don't lay off people like the rest of the tech world. I think it's a great place.

但和硬件领域的人交流时，我觉得苹果硬件工程部门的很多人对那种例行工作感到有些缺乏灵感和厌倦。就是那种对iPhone、Mac和Apple Watch进行渐进式更新的工作。所以我认为这个网站相当令人兴奋。

But speaking with folks in sort of that hardware universe, I think a lot of people in hardware engineering at Apple have become a little uninspired and bored with sort of the rote work. Doing kind of the incremental sort of updates to the iPhones and Macs and Apple Watches. So I think this is a quite exciting site And

我是说，不知道你的报道是否透露了OpenAI试图构建哪些类型的硬件产品？因为有很多想法被抛出来。你在这方面有什么发现吗？

I mean, you know, I wonder if your reporting told you anything about what kinds of hardware products OpenAI is trying to build? Because I mean, there's been a lot of ideas thrown around. Sort Did of get any insight there?

是的，我们听到了一些消息。主要看来，他们在法庭文件中承认了一个关于发布一系列设备的计划。有多种形态因素在考虑中，包括没有显示屏的智能音箱，你可以想象成类似亚马逊Alexa的设备。

Yeah, we heard a few things. I mean, I think primarily it appears and they've and in court documents, they have acknowledged sort of a plan around a family of devices that they want to release. And there's a range of sort of form factors. They're looking at a smart speaker without a display. So obviously something like you could think like, you know, an Amazon Alexa like device.

智能眼镜，我们听说也在讨论形态因素，还有数字录音机。

Smart glasses, we've heard as talk about form factor, digital recorders.

他们在招募有构建这类产品经验的人才吗？他们是在考虑制造方面的问题吗？我是说，我不是在问你的消息来源是谁，而是想知道我们是怎么知道这些的？

And are going after talent that have experience building these types of products? Are they looking at it in terms of manufacturing? How are they I guess I'm not asking you what who your sources are, but I'm asking how do we sort of know this?

哦，我采访过实验室里的人。我们的供应链消息人士透露，苹果供应商戈尔科技和立讯精密正在参与这些设备的制造。就像苹果的生态一样，这里也有一个庞大复杂的供应链网络，现在有部分人开始与OpenAI合作。

Oh, I mean, I've spoken to people who've been in the lab. We we have people in supply chain, GoreTech in Luxshare, which are Apple suppliers, are working on these devices. So similar to like in the Apple world, there's a very, very large, complicated supply chain with a lot of folks, a lot of people in that chain. There's people now entering into and doing work with OpenAI here.

没错。我想了解乔尼·艾夫在这些讨论中的重要性。感觉这可能更容易被接受，因为人们会说，看，我们想和一位有远见的人合作。他在这些讨论中占很大比重吗？

Right. You know, I want to get some sense from how important Joni Ive is to these conversations. It feels like maybe this is an easier sell because people are saying, Well, look, I mean, we want to go work with a visionary. Is he a pretty big part of these discussions?

哦，当然。他在招揽和挖角方面参与得非常深入。乔尼·艾夫的吸引力毋庸置疑。他是九十年代末与史蒂夫·乔布斯一起将苹果从悬崖边缘拉回来的标志性人物。

Oh, for sure. Yeah. Involved very deeply in sort of the outreach and poaching, I think. And just, yeah, the draw of Johnny Ive. He is he is sort of the icon that brought Apple back from the brink with Steve Jobs in the late nineties.

这两人之间的深厚合作造就了今天的苹果公司。因此，将他视为塑造苹果的关键人物——那些曾让苹果伟大的人眼中的关键部分——他确实是其中的核心。

The deep partnership between those two made Apple into the companies today. And so, you know, just seeing him as that key piece of what made Apple, what used to make, know, some of these people think what Apple used to be, the great people of Apple, he is a key part of that. Right.

人才是一方面。我想知道你是否能谈谈这有多困难，因为很多公司尝试过做硬件。当然，很多公司没有乔尼和艾夫这样的导师在身边指导，但我不认为只要招到几个优秀的硬件工程师就能轻松成功。对OpenAI来说，这甚至可能是一场长达数年乃至数十年的旅程。

Talent is one side of the story. I wonder if you can talk a little bit about just how difficult it is, because a lot of companies have tried to do hardware. Of course, a lot of companies have not had Jodi and I have by their side guiding them, but it doesn't strike me that this is a home run just if you can get a couple of great hardware engineers on board. I mean, this is I mean, we're talking a multiyear, multi decade journey even for OpenAI.

完全同意。硅谷遍地都是试图创业的前苹果硬件员工。将苹果这样拥有庞大资源、每个产品环节都有团队支持的环境，转换到更小、更艰苦的创业环境，对许多人来说极其困难。硬件在每个层面都需要团队协作，在苹果之外执行这一点极具挑战性。因此，很多苹果硬件人才有好点子，却难以实现。

Totally, for sure. I mean, I think, you know, Silicon Valley is littered with former Apple hardware employees trying to do startups. I think it's like the trying to translate being at a big company like Apple with tons of resources surrounded by massive teams working on every element of a product to a smaller, more scrappy kind of environment, that's really difficult for a lot of folks because there's just so much that goes into hardware at every sort of layer that, you know, you just need a team that can execute is really challenging outside of Apple. So you just there's been a lot of examples out there of sort of, yeah, Apple hardware folks with a good idea, but they just can't, they just can't bring it to life.

最后一个问题。伟大的故事往往引发更多疑问。关于OpenAI的硬件策略，这让你产生了哪些问题？你会关注什么？

Last question for you. You know, every great story comes with more questions than answers sometimes. What questions did this raise for you with respect to OpenAI's hardware strategy, and what are you watching for?

是的，我主要思考的是：世界是否需要另一个设备，尤其是AI设备？自从……

Yeah, I mean, primarily the thing I was thinking about a lot in this is, does the world need another device, especially an AI device? Like, since, you know, the

我们拍摄了

We shot

这周我们拍摄了眼镜。我是说，你懂的吧？

them we shot the the glasses this week. I mean, you know?

是啊是啊。我是说，自从ChatGPT崛起后，出现了很多AI设备，比如那个Humane AI Pin。但我觉得这些产品都没能向消费者展示任何实际价值或用途，对评测者和市场需求来说完全是失败之作。

Yeah. Yeah. I mean, since since the rise of ChatGPT, there's been a lot of, you know, AI devices, the humane AI pen. Yeah. There's just like none of these have shown, I think, any sort of value or purpose for consumers and have been complete flops for reviewers and demand.

所以我在想，强尼·艾维能向我们证明这条路走得通吗？世界真的需要AI设备吗？因为我持怀疑态度。智能手机甚至是Mac，都完全能满足与AI交互的需求。大多数人不过就是在Safari浏览器里输入ChatGPT的提示词而已。

So I think is Johnny Ive, is he going to show us that there is a path forward? Does the world need an AI device? Because I'm not convinced. I think the smartphone is a perfectly fine kind of device to interact with AI, even your Mac. I mean, most people are just typing into a Safari browser, their chat GPT prompts.

他们根本不需要专门设备来做这些事。所以这就是我在思考的——这东西到底有多必要。

They don't need a device to do this stuff. So I think that's the thing I'm thinking about, like how necessary is this.

没错。而且我们几分钟前才在节目里采访了Alex Kimmel，其中有个问题...

Yeah. And, you know, we just had Alex Kimmel on the show just a few minutes ago, and one of

我问他的问题之一

the questions I asked him

刚才说到，嘿，戴着这些眼镜，我是说，这难道不会真的很分心吗？你提出的论点是眼镜能让你更专注，而我对他的观点是，实际上我认为这会增加干扰，所以我其实是支持你的。和你一样，所以我很好奇他们会推出什么产品。亚伦，非常感谢你参加我们的节目。这位是亚伦·蒂利，《The Information》的苹果产品记者。

was about, Hey, with these glasses, I mean, isn't it distracting, really? You make the argument that glasses allow you to be more present, and my point to him was, I actually think it adds distractions, and so I'm actually with you. With you, and so I'm curious to see what they come out with. Aaron, thank you so much for coming on the show. That is Aaron Tilly, our Apple reporter here at The Information.

好了，今天的节目就到这里。提醒一下，我们每周一到周五太平洋时间上午10点、东部时间下午1点在这个频道进行直播。我要感谢亚马逊网络服务对本节目的赞助支持，也感谢各位的收看。我们非常珍视每一位观众。我已经开始期待下周一的新节目了。

Well, that does it for today's show. A reminder that we are live on this stream Monday through Friday at 10AM Pacific, 1PM Eastern. I want to thank Amazon Web Services who is our presenting sponsor for this production, and I want to thank you for tuning in. We really do appreciate your viewership. I am already excited for our next show on Monday.

我们为你们准备了一周满满的精彩内容，到时见。祝大家周末愉快，暂时先告别啦。

We got a busy week planned for you, and so we will see you then. Until then, have a great weekend. Bye bye for now.