高通 CEO 克里斯蒂亚诺·安蒙谈 AI 设备的未来

未来的 AI 设备长什么样？让我们来问问那位打造为其提供动力的芯片的 CEO。接下来就是 Cristiano Amon。欢迎来到 Big Technology Podcast，一个对科技世界及其之外进行冷静而细致对话的节目。我们在达沃斯的 Qualcomm 空间，今天为你准备了一期精彩的节目。我们将讨论 AI 设备的未来。我们将讨论什么是 AI PC，以及是否有人会想要它。我们将讨论数据中心建设、机器人和工业 AI。来和我们一起做这件事的是完美的嘉宾，Qualcomm CEO Cristiano Amon。Cristiano，很高兴见到你。

What does the AI device of the future look like? Let's ask the CEO building the chips that will power it. That's coming up with Cristiano Amon right after this. Welcome to Big Technology Podcast, a show for coolheaded and nuanced conversation of the tech world and beyond. We are here at Davos at the Qualcomm space and we have a great show for you today. We're going to talk about the future of the AI device. We're going to talk about what an AI PC is and whether anybody's going to want it. We're going to talk about the data center buildout, robotics, and industrial AI. And here to do it with us is the perfect guest, Qualcomm CEO, Cristiano Amon. Cristiano, great to see you.

也很高兴见到你。非常开心能和你进行这次对话。

Great to see you, too. Very happy having this conversation with you.

确实，现在是我们进行这次对话的完美时机，因为关于 AI 设备的讨论正从理论走向具体，而 Qualcomm 可能正处于中心。所以，让我为我们的听众，如果你对 Qualcomm 不熟悉，简单介绍一下这家公司。一家市值 1700 亿美元的公司。所以它非常大。它是 Snapdragon 芯片的设计者，该芯片用于移动手机，尤其是高端 Android 手机，也用于 PC、汽车，并越来越多地用于可穿戴设备。还有 Dragon Wing 芯片，我们将会谈到，它用于机器人等工业用例，而且你们刚刚进入 AI 数据中心，为 AI 推理构建服务器。所以，一家芯片设计公司真正处于 AI 故事的中心，无论是可穿戴设备还是数据中心。

Definitely, it's a perfect time for us to have this conversation because talk of an AI device is going from theoretical to concrete and Qualcomm might be at the center of it. So, let me give for our audience, if you're new to Qualcomm, a little bit of an introduction to the company. $170 billion company. So, it's very big. It's the designer of the Snapdragon chip, which is in mobile phones, notably high-end Androids, also PCs, autos, and increasingly wearables. There's also the Dragon Wing chip, which we're going to talk about, which is in industrial use cases like robotics, and you just got into AI data center, building servers for AI inference. So a chip designer really at the center of the AI story whether it comes to wearables or in the data center.

我喜欢这个。非常好。我认为这是对 Qualcomm 的很好的介绍。也许我只补充一点。我认为你知道 Qualcomm 是一家非常独特的半导体公司。我认为尤其是在当今环境下，连接很重要，计算很重要，AI 处理很重要，我们是少数几家将所有这些集于一身的公司之一。而且我们可能是少数几家从你耳机中的 5 瓦到数据中心的 500 瓦的半导体公司之一。这对公司来说是一个激动人心的时刻，对技术来说也是一个激动人心的时刻，因为 AI 正在进入一切。

I like that. Very good. I think that's a great introduction of Qualcomm. Maybe I'll just add one thing to it. I think you know Qualcomm is a very unique semiconductor company. I think especially in today's environment when connectivity is important, computing is important, AI processing is important, one of the few companies that had all of it under a single roof. And we're probably one of the few semiconductor companies that go from five watts to your earbud now to 500 watts when you think about a data center. And it's an exciting time for the company also exciting time for technology since AI is going into everything.

为智能手机设计芯片让你处于一个非常有趣的位置，因为当我们都开始想象 AI 设备会是什么样子时，显然在 AI 方面，底层的算力非常重要，而你有能力做到。最近你谈到你的信念是，AI 设备的市场机会——我们将讨论外形会是什么样——但市场机会是 100 亿台设备，这将使其比智能手机市场更大。你是怎么得出这个数字的？

And designing the chip for the smartphone has put you in a very interesting position because as we all start to imagine what an AI device is going to look like, obviously when it comes to AI the compute underneath is really important and you're in position to do it. And recently you've talked about how your belief is that the market opportunity for an AI device—and we're going to get into what the form factor is going to look like—but the market opportunity is 10 billion devices, which would make it bigger than the smartphone market. How do you get to that number?

所以，这很有趣，我认为要得出那个数字，实际上重要的是看看智能手机如何在不同代际中演变，我认为有几个方面。有手机的演变，有计算的演变，然后 AI 如何改变未来的发展，也许我会带我们稍微进入那段旅程来谈谈。最大的变化之一——我不会一直追溯到 2G——但手机行业发生的最大变化之一是我们将宽带引入蜂窝网络，我们说我们可以有宽带速度。我们意识到在宽带的另一端你需要一台计算机。所以你的手机需要变成一台计算机，你需要开发一台能放在手掌中的计算机。那就是智能手机。那就是永远改变了计算机的智能手机。我想它是我们不可分割的设备。我们一直随身携带，它一直是我们的数字生活的中心。现在随着你不断进步，我认为在智能手机方面，我们现在每年购买 12 亿部手机。它是第一大消费电子产品，每个人都有一部。但当你开始思考 AI 正在发生什么，尤其是计算机使用 AI 现在理解我们时，那么你开始不仅进入你携带的计算机，还进入你穿戴的计算机，特别是因为如果智能体对你有用，它们将一直和你在一起。然后你开始从携带手机到也拥有眼镜或戒指或手镯、手表以及所有这些东西。但它们改变了可穿戴设备过去的性质。可穿戴设备——当你谈论可穿戴设备和技术时——被设计为只是扩展你的手机功能。例如，是的，你有智能手表会告诉你时间，但也现在把你的传感器数据传回手机，并把通知从手机给你。但这一切都将改变。这一切都关乎连接到一个模型，连接到一个智能体。随着这些变化，我们都将开始穿戴这些东西，然后你开始思考大数字。你知道，如果每个人最终都得到一个手表、一个戒指或一个眼镜，那不是连接到智能体，那么你谈论的是和手机一样大的数量级，我认为这很令人兴奋，这就是我们思考移动行业未来的方式。

So, it's interesting and I think for you to get at that number, it's actually important to see how the smartphone had evolved over different generations and I think you have a couple things. You have the evolution of phones, you have the evolution of compute and then how AI changes that going forward and maybe I'll take us a little bit into that journey just to talk about it. One of the biggest changes—I won't go all the way back to 2G—but one of the biggest changes that happened in the phone industry when we developed broadband into cellular and we said we can have broadband speeds. We realized that on the other side of that broadband you need a computer. So your phone needs to become a computer and you need to develop a computer that will fit in the palm of your hand. And that's the smartphone. That's the smartphone that changed computer forever. I guess it's our inseparable device. We carry with us all the time and it is doing—it's been at the center of our digital life. Now as you keep advancing, I think in smartphones right now we are in the billion every single year is 1.2 billion phones are purchased. It's the number one consumer electronics and everybody has one. But when you start thinking about what's happening with AI and especially as computers using AI now understand us, then you're starting to go into not only the computer that you carry, but also the computer that you wear, especially because if agents are going to be useful for you, they're going to be with you all the time. And then you start to go from carrying a phone to also having a glass or a ring or a bracelet, a watch and all those things. But they changed the nature of what wearable used to be. Wearable was—when you talk about wearables and technology—was designed to just extend your phone functionality. Like for example, yes, you have a smartwatch will tell you the time but also give you now your sensors back to the phone and give notifications from the phone to you. But that's all going to change. It's all about connecting to a model, connecting to an agent. As those things change and we all are going to start wearing those things, then you start to think about big numbers. You know if everybody ends up getting a watch, a ring or a glass that's not connected to an agent then you're talking about order of magnitude as big as the phone and I think that's exciting, that's how we think about the future of the mobile industry.

但问题是：为什么它需要是可穿戴的？我在年底前和 Sam Altman 谈过。现在 OpenAI 将构建一系列设备，但传闻一直是它将是一个智能手机大小的设备，没有屏幕，它只是听你说话，然后向你推送关于你生活的通知。我就想，好吧，为什么它不能只是手机上的一个应用？为什么它必须是可穿戴的？

But here's the question: why does it need to be a wearable? I was speaking with Sam Altman right before the end of the year. And now OpenAI is going to build a family of devices, but the rumor had been that it's going to be a smartphone-sized device, no screen, and it just listens to you and then it will push you notifications about your life. And I was like, well, why can't it just be an app on the phone? Why does it have to be a wearable?

好吧，它不一定非要如此。看，我们正在和他们合作。不幸的是，我不能告诉你它是什么。你会看到的，它会很令人兴奋。但这里——我们需要以稍微不同的方式思考这个问题，对吧，可穿戴设备是其中将更重要的事情之一。我先回答你的问题：这整个个人 AI 设备类别——人类很久以前就已经决定了他们要穿什么，对吧？所以我不认为你我会戴一个大头盔。我认为我们可以戴眼镜。我们可以戴珠宝。

Okay, it doesn't have to. Look, we're working with them. Unfortunately, I cannot tell you what it is. You will see and it's going to be exciting. But here's—we need to be thinking about this a little bit different, right, and wearable is one of the things it's going to be more so. I'll start first answering your question: this whole category of personal AI devices—humans already decided what they're going to wear a long time ago, right? So I don't think you and I are going to be wearing like a big helmet. I think we can wear glasses. We can wear jewelry.

我们人类决定自己穿什么，而我们的工作是让电子设备高度集成，把大量算力塞进小尺寸里。这来自我们手机的基因。你可以把电子设备放进所有这些里，再加上连接性。连接到智能体会非常有用。但你也可以在桌上放一个东西，或者在床边放一个东西，然后连接到不同设备中的智能体。我认为我们会看到一切都会以某种方式变得智能。最根本的一点是，现在计算机能理解我们看到的、我们说的、我们写的，这稍微改变了人机界面，也随之改变了计算机的整个定义。所以可穿戴设备对我们来说是最合乎逻辑的，因为我们在考虑移动性和你会随身携带的东西，但你也可以在桌上放东西。思考这个问题的方式是想想那些在技术转型中被困住的设备。比如，你现在面前有一台笔记本电脑，对吧？我现在可以打赌——我看到它是高通芯片的——我可以打赌这台笔记本有触摸屏功能，但你大概不常触摸。你用键盘。那是它被设计的方式。用户界面就是为此设计的。你触摸你的手机。当你从口袋里掏出手机时，你会去触摸应用。你像这样指着手机试图拍照，并不太自然。眼镜戴在头上，摄像头随你移动。也许你可以对手机说话。也许手机在这里，你拿起它之前就对它说话。因此，会有其他东西放在你桌上，你会对它们说话。所以我们不知道这些会如何发展。但回到你的问题，可穿戴设备是合乎逻辑的——可穿戴设备将成为我们穿戴和携带的东西。

We humans decide what we're going to wear, and it's our job to make electronics very dense and pack a lot of computing power into small form factors. That comes from our phone DNA. You can put electronics in all of this, plus connectivity. Connecting to an agent is going to be very useful. But you could have something on your desk, or next to your bed, and you can connect to agents in different devices. I think what we'll see is everything will become smart in one way. The biggest fundamental thing is that computers now understand what we see, what we say, what we write, and that changes the human-computer interface a little bit, and with that changes the whole definition of what a computer is. So wearables are the most logical thing to us because we're thinking about mobility and things you're going to carry with you, but you could have things on your desk. The way to think about this is to think about devices that get caught in the transition of technology. For example, you have a laptop right in front of you, right? I can bet you right now—and I see it's Qualcomm-powered—I can bet you that the laptop has the ability for you to touch the screen, but you probably don't touch it often. You use the keyboard. That's what it was designed for. The user interface was designed for this. You touch your phone. When you pull your phone out of your pocket, you're going to be touching apps. It's not very natural for you to point at the phone like this to try to record images. Glasses are on your head; the camera moves with you. Maybe you can talk to the phone. Maybe the phone is here and you talk to it before you pick it up. Therefore, there are going to be other things on your desk that you're going to talk to. So we don't know how those things are going to pan out. But going back to your question, wearables are logical—wearables are going to be things that we wear and carry around.

但请帮我们稍微展开一下，这种体验会是什么样的。

But help us flesh out a little bit about what this experience will be like.

好的。

Yes.

我的意思是，显然我们还没到那一步。我们经历了许多起起落落。Google Glass 就是一个例子。很久以前人们就把计算设备戴在头上。现在技术似乎真的快要达到可能有用的程度，也许它能理解我们的上下文。所以，Cristiano，当你想到，好吧，我要把芯片放进眼镜，也许还有其他不同形式，人们会使用它们并获得某种体验。那是什么体验？

I mean, obviously we're not there yet. And we've had many stops and starts. Google Glass was an example. People were wearing computing on their heads a long time ago. Now it seems like the technology is actually getting there to the point where maybe it will be useful, maybe it can make sense of our context. So, Cristiano, when you think about, all right, I'm going to put chips in glasses and maybe some other different formats and people will use them and have x experience. What is that experience?

好的，我们来谈谈体验。现在我们要把这段对话拆开。我会谈体验，然后谈背后的技术。想想今天的眼镜能做什么。比如 Meta Ray-Ban 眼镜。我认为今年 Google 生态内还会有其他眼镜推出。今天的眼镜能做什么？有摄像头，所以你能看到你所看到的，能理解图像。它可以标注图像，还有麦克风。它有扬声器，可能有也可能没有显示屏。即使没有显示屏也有使用场景，比如 Meta Ray-Ban 眼镜。体验是什么样的？首先，这些东西要规模化，必须摩擦很低，体验必须有用。否则就是噱头，你不会用。所以体验会是这样：我正在和你说话，然后假设我看到观众席上某个人，我就说‘这是谁？’，眼镜会告诉我‘我不知道，让我查一下’。我在网上查——这个人在这里吗，这个人叫什么名字？我说‘哦，好的，你以前见过她，有一封从这个人发给你的邮件。’它必须像你一直有个朋友在身边。你走在街上，我问‘这是什么？’这就是什么。甚至像这样：你开始一天的生活，你的智能体会来找你说‘你知道吗，我注意到你现在似乎有空。我能谈谈你的日程吗？有一个冲突我们需要解决。’这些就是体验会如何的例子。将是这个智能体，它有能力理解你的上下文，理解你周围的环境、你看到的、你说的，并实时反应。有趣的是我们还没到那一步，但你看到了变化的开端。我喜欢做类比。所以我要告诉你与智能手机的类比。当第一部智能手机出现时，比如你看到 iPhone，看到 Android——也许这个数字我会说错——但也许有 10 个应用，你说‘好吧，这就是 10 个新应用。’当时你无法想象会有几十万个应用。如果你现在看你的手机，你有大量应用。所以你的手机随时间变得更好，因为突然应用商店里有了新应用。我认为那些智能体也会这样。最终智能体会与其他服务集成，你开始看到它。例如，我们在印度有一个做智能眼镜的客户。他们集成了数字支付系统。所以现在你可以看着二维码说‘支付这个’，它就会支付。所以从‘翻译这个’、‘给我解释这个’，到‘支付这个’。你可以收到账单，你说‘我收到了这个账单。请支付这个账单。从我的支票账户扣款，完成后通知我。’你可能拍张照片并电邮给我，因为我想留个副本。这些就是你将如何与那些计算机交互，体验就会是那样。

Yes, let's talk about the experience. And now we're going to break this conversation. I'll talk about the experience and I'm going to talk about the technology that goes behind it. So think about how glasses are performing today. You have, for example, the Meta Ray-Ban glasses. I think there's going to be other glasses coming within the Google ecosystem this year. And what are the glasses doing today? You have cameras so you see what you see, can understand the image. It can annotate the image and you have a microphone. It has a speaker, may or may not have a display. You have use cases even without the display, like the Meta Ray-Ban glasses. What does the experience look like? First of all, for those things to get scale, they have to have very low friction and the experience has to be useful. Otherwise, it's like a gimmick. You're not going to use it. So the experience is going to be like this: I'm talking to you and then let's say I see somebody in the audience and I just say, 'Who is this person?' and the glasses will tell me, 'I don't know, let me check.' I check on the web—is this person here, what is this person's name? I say, 'Oh, okay, yeah, you met her before, there was an email that was sent to you from this person.' It has to be something like you have your friend with you all the time. You're walking on the street and I say, 'What is this?' This is what it is. Or even something like you go into your day and your agent is going to come to you and say, 'You know, I noticed that right now you seem to be free. Can I talk about your agenda? There's a conflict we need to resolve.' Those are examples of how this experience is going to be. It's going to be this agent that has the ability to understand your context, understand what is around you, what you see, what you say, and react in real time. And what is interesting is we're not there yet, but you see the beginnings of the change. And I like to draw parallels. So I'm going to tell you the parallel with the smartphone. When the first smartphone arrived, like when you saw the iPhone, you saw the Android—maybe I may get this number wrong—but maybe there were 10 apps and you say, 'Okay, those are the 10 new apps.' You couldn't at the time imagine that you're going to have probably hundreds of thousands of apps. And if you look at your phone right now, you have a ton of apps. So your phone got better over time because all of a sudden a new app became available in the app store. And I think that's how it's going to be with those agents. Eventually the agent gets integrated with some other service and you start to see it. For example, we have a customer in India that is doing smart glasses. They integrated with the digital payment system. So now you can look at a QR code and say 'pay this' and it will pay. And so you go from 'translate this', 'explain this to me', to 'pay this'. You can get a bill and you say, 'I got this bill. Please pay this bill. Get it out of my checking account, notify me when it's done.' And you may take a picture and email it to me because I want to keep a copy of it. Those are going to be how you interact with those computers and that's what the experience is going to look like.

有没有一个世界，我们与计算机走得太近？想想有时候空闲时间真的很美好，而现在智能体会像‘啊哈，他有一刻空闲，我要去帮他解决一个冲突，或者帮他了解这个人是谁’，而不是他自己上前问‘我们以前见过吗？’最终会不会有一个点，人类和计算机走得太近？

Is there a world where we get too close to computers? Think about sometimes that free time is really nice, and now the agents being like 'Aha, he has a moment, I'm going to go and help him resolve a conflict or I'll help him understand who this person is' as opposed to him going up and asking 'Have we met before?' Does there eventually come a point where humanity and computers come too close together?

这是个好问题。我不知道答案，但我认为像一切事物一样，这将由你决定。你看，我们中有些人，不是所有人，有时候你只需放下手机，就会是这样。

That's a good question. I don't know the answer to the question, but I think like everything, it's going to be for you to decide. Look, some of us, not all of us, sometimes you just put your phone down and it's going to be like that.

你只能自己决定什么时候该断开连接。但我觉得情况会有点不一样，因为现在与计算机协作会更容易，计算机也会更容易与我们协作。我想借你问我的这个问题讲件趣事。我当时没在 CES，我正在和高通的一位客户聊的正是这件事——关于智能眼镜和摄像头，以及现在摄像头能看到你所看到的、还能对图像做标注。然后有人说，你知道，万一有时候有些事是你想忘掉的呢？结果答案是：你或许可以忘，但 AI 不会忘。

You're just going to have to decide when it's time to disconnect. But I feel it's going to be a little bit different, because now it's going to be easier to work with computers, and the computers are going to be easier to work with us. I'm going to use this question that you asked me to tell something funny. I wasn't at CES, and I was having a conversation with a customer of Qualcomm about exactly this thing — about the smart glasses and the camera, and the fact that now the camera sees what you see and can annotate the image. And then somebody said, you know, what if sometimes there are things that you want to forget? And then the answer was, well, you may, but the AI won't forget.

但这些都会是很有意思的事——就像技术本身一样，人类会怎么用它、这些东西会怎么发展，我们拭目以待。

But those are going to be interesting things — like with technology in general, I think how humans are going to use it and how those are going to be developed, we're going to see.

这个话题的自然延伸是：随着 AI 越来越强大、人类与 AI 越来越接近，总会有人想说，干脆让我们融为一体吧。Elon Musk 谈过，他创办脑机接口公司 Neuralink 的理由是——他说——最终 AI 会比人类更强大，我们最好与它们融合，否则它们会毁灭我们。所以我就想问你：你愿意与 AI 融合吗？

The natural extension of this conversation is that as AI becomes more powerful and humanity comes closer to AI, there are going to be people who want to say, let's just bring us together. Elon Musk has talked about how the reason for building Neuralink, his brain-computer interface company, is — he said — eventually AI is going to get more powerful than humans, and we better merge with them or they're going to destroy us. So I just want to ask you: would you merge with AI?

不愿意。但你看，在我们刚才的对话里，你说技术太多的时候，我们谈的非常以消费者为中心。但当你从消费者转向企业，就更容易理解了。如果你真的想一想：假如你有能力实时学习一切——就像我们现在已经看到的一些用例，尤其是在工业领域——当你有一个操作设备、操作炼油厂之类的人，突然身边有了这个智能体。你走到某台设备前，你说：“这个我怎么操作？”它就会说：“你要这样操作。你做这个，做那个。”所以，能够实时获取知识——我认为这是一个极其巨大的机会，真正让知识和学习民主化。

No. But look, in the conversation we just had, we were talking very consumer-centric when you said too much technology. But it's easier to also understand when you move from the consumer to the enterprise. If you actually think about the fact that if you have the ability to learn everything in real time — like we're actually seeing some use cases right now, especially for industrial — when you have somebody who is an operator of an equipment or of a refinery or everything, and then all of a sudden you have this agent with you. You get to a particular piece of equipment and you say, "How do I operate this?" And it will say, "Here's how you're going to operate it. You do this, you do that." So the ability for you to have access to knowledge in real time — I think there's an incredibly incredible opportunity to actually democratize knowledge and learning.

我也这么认为。这是 AI 与增强人类能力之间联系的另一点。我们可以这么说，因为我们在手机上看到了这一点，对吧？

I think so. That's another thing about the connection between AI and in terms of augmenting human capabilities. We can say that because we saw that with phones, right?

手机让多少国家接入了互联网，通过手机获得了数字化接入。你知道，不是通过电脑。我认为让人们连接起来、拥有上网的能力，是极其赋能的。我想那些个人 AI 设备也许也会是一样的情况。

With phones, it's how many nations got access to the internet and became — you know — got access to digital through the phone. You know, it wasn't through a computer. And I think it was incredibly empowering to have people connected and have the ability to access the internet. I think maybe that's going to be the same thing with those personal AI devices.

好，我马上换个话题，但我问你愿不愿意与 AI 融合时，你非常快地说了不。

Okay, I'm going to move off this in a second, but I asked if you'd merge with AI, and you said no very quickly.

是的。

Yes.

为什么是这种条件反射式的不？

Why the reflexive no?

哦，因为你看，这是不一样的。我觉得这挺有意思。我觉得人们喜欢这类科幻故事。我有一个非常清晰的信念。我认为有人类，有人性。AI 是我们的创造物，是用我们做的事情训练出来的。我觉得如果你看看很多那些模型，它真的就是一个旨在增强我们的工具，但它不会夺走我们的人性。

Oh, because look, it's different. I think it's fun. I think people like to have those stories about science fiction. I have a very clear belief. I think there's humans, there's humanity. AI is our creation, trained on the stuff that we do. I think if you look at a lot of those models, it's really a tool designed to augment, but it won't take away our humanity.

好，快速聊一下形态：你多次提到眼镜，但没提耳机。你知道，想想这场竞争正在如何成形，不同公司在不同形态上下不同的注——尤其是看看科技巨头，就像我们《Big Technology Podcast》喜欢报道的那样。Meta 在 AI 眼镜上下了大注。Google，如你所说，我觉得我们会看到他们下非常大的注——Google Glass 第二部，虽然也许他们会换个新名字。Apple，可能要等到 2027 年才能看到他们的一款眼镜。也许他们的大注会是 AirPods，以及 AI 已经如何通过翻译这类功能在 AirPods 中交付——我是说里面的 Siri，但还有不少工作要做，也许他们会通过和 Google 的合作来做。你为什么认为眼镜胜过耳机？

Okay, very quickly on form factor: you've mentioned glasses a number of times, you didn't mention earbuds. And you know, when you think about the way this competition is shaping up, you have different companies making different bets on different form factors — especially when you look at the tech giants, big technology, as we like to cover here on the Big Technology Podcast. You have Meta making a big bet on AI-powered glasses. Google, as you mentioned, I think we're going to see a very big bet from them — Google Glass part two, although maybe they'll have a new name. Apple, it might be 2027 until we see a pair of glasses for them. Maybe their big bet is going to be the AirPods and how AI already is delivered in the AirPods with things like translate — and I mean Siri inside there, but still has some work to do, and maybe they'll do it with their Google partnership. Why do you think glasses over earbuds?

你看，我不会说一个胜过另一个。我们有幸——我猜——与大多数真正在打造个人 AI 设备的公司合作。我觉得我们有幸与他们共事。所以我们有相当广阔的视野。我举个例子。现在有些公司在设计带摄像头的耳机。

Look, I won't say one over the other. We have the benefit — I would assume — of being with the majority of the companies that are actually building personal AI devices. I think we have the benefit of working with them. So we have pretty broad visibility. Let me give an example. There are some companies right now designing an earbud with a camera.

带摄像头的耳机？

An earbud with a camera?

带摄像头，因为如果你把它放进耳朵里、又有个摄像头，它就能看到你前方。所以除了扬声器和麦克风之外，它还能提供一些上下文。我觉得这又回到“怎么做”的对话。人类会戴、而且大部分时间都戴着的东西是什么？眼镜。我坚信眼镜是最自然的。也许因为我从 13 岁起就戴眼镜，所以我习惯了。但当你转头时，摄像头跟着你走，它离你的眼睛很近——你该想想这一点。这——你问可穿戴设备那个问题时我本该想到这一点，因为那是回答那个问题最简单的方式。如果 AI 理解我们所看到的、我们所说的、我们所在的位置，它就会更贴近我们的感官，而眼镜能捕捉一切。它离你的嘴更近，离你的耳朵更近。但耳机也一样，只是缺了视觉，这就是为什么有些人在耳机上装摄像头。但如果你只是有个连到 IP 地址的耳机，你就能连上一个智能体，和智能体对话。

With a camera, because if you put it in your ear and you have a camera, it can see in front of you. So it can provide some context in addition to just having a speaker and a microphone. I think it goes back to the how-to's conversation. What are the things that humans are going to wear and wear most of the time? Glasses. I am a believer that glasses is the most natural. And maybe because I wear glasses since I was 13, so I'm used to them. But when you turn your head, your camera goes with you, it's close to your eyes — you should think about this. This is — I should have thought about that when you asked the question about wearables, because that's the most simple way to answer that question. If the AI understands what we see, what we say, where we're here, it's going to be closer to our senses, and glasses captures everything. It's closer to your mouth, closer to your ear. But earbud, it's the same thing, it's just missing the vision, and that's why some people are putting a camera on an earbud. But if you just have an earbud connected to an IP address, you can connect to an agent and you can have a conversation with the agent.

那别针呢？

What about a pin?

一样。那是装摄像头的另一种方式。有吊坠，有首饰。所以拭目以待。但我觉得你会看到人们在形态上不断尝试。我认为眼镜很可能成为这类设备的主要构建方式。

The same thing. It's another way to put a camera on it. There's pendants, there's jewelry. So we'll see. But I think you're going to see people experimenting with form factors. I think glasses is likely going to be the primary way that those devices are going to be built.

那么，假设眼镜是赢家。你觉得款式重要吗？我给你一个二选一。我有一副更时尚的眼镜，但助手最差；或者一副不那么时尚的眼镜，但助手非常棒。哪个赢？

So, let's say glasses is the winner. Do you think that style matters? Let me give you a binary here. I have the more stylish glasses with the worst assistant, or the less stylish glasses with an amazing assistant. Which wins?

这是个好问题。

This is a great question.

这是个好问题，因为我们将看到行业里发生的另一件事：当你开始考虑可穿戴设备时，你会看到时尚与科技的融合。实际上，我想在这里做个预测。我不想冒犯任何其他公司，但我认为这正是横向模式会战胜纵向模式的地方。我这么说的原因是，地球上不太可能每个人都使用完全相同的眼镜。人们想要不同的外形。他们想要不同的颜色。尤其是你穿戴的东西，更是如此。因此，我认为会出现不同的品牌。这会是一个有点有趣的动态，因为你戴的是雷朋吗？还是 Meta 的？如果是由消费电子公司制造的雷朋，那它是消费电子品牌还是雷朋？我们拭目以待。但我认为你会看到时尚与科技的结合，会有选择——不同品牌面向不同人群、不同年龄段等等。所以我认为我们会看到很多多样性，这与手机领域非常不同，在手机领域大多数人都会携带相似的手机。我认为情况会有所不同。

This is a great question because we're going to see another thing happening in the industry: when you start thinking about wearables, you're going to have the mix of fashion and technology. I actually think I'm going to make a prediction here. I don't want to be offensive to any other company, but I think that's where the horizontal model is going to win versus the vertical model. The reason I'm saying that is because it's very unlikely that everybody on earth is going to use the same exact glasses. People want different form factors. They want different colors. It's the same especially for things that you wear. As a result, I think you're going to have different brands. It will be a little bit of an interesting dynamic because is that a Ray-Ban that you're wearing, or is it a Meta? If it is a Ray-Ban made by a consumer electronics company, is it the consumer electronics brand or is it Ray-Ban? We'll see. But I think you're going to have the combination of fashion and technology, and there's going to be choices—different brands for different people from different age groups, etc. So I think we're going to see a lot of diversity, very unlike the phone space where most people will carry a similar phone. I think that's going to be different.

我要回答我自己的问题。我宁愿要更好的助手和难看的眼镜，也不要漂亮的眼镜和糟糕的助手。

I'm going to answer my own question. I'll take the better assistant and the ugly glasses over the nice glasses and the bad assistant.

是的。最好的情况也许是……好看。

Yeah. The best thing is maybe the most... be nice.

也许最成功的眼镜会与最好的助手配对。

Maybe the most successful glasses is going to pair with the best assistant.

最终会这样。你会认为我们会达到那一步。

Eventually. You would think we get there.

是的，我想会的。

Yeah, I think so.

嗯，请为我们分析一下 AI 设备竞赛。有很多公司都在朝这个方向努力。Meta 多年来押注元宇宙，这实际上已经转变为押注智能眼镜。谷歌，所有迹象都表明——如果你看他们最近的思考游戏纪录片，他们只是把手机对准物体说：“这是什么？”就像在说：“你需要眼镜。”嗯，还有 OpenAI。你正在与 OpenAI 合作这个项目。一系列设备将出现在不同地方。而苹果显然也必须被视为这里的有力竞争者。谁会赢？

Um handicap the AI device race for us. We have many companies that are running at this. We have Meta that's been making this multi-year metaverse bet which has really transformed into the smart glasses bet. We have Google, which all indications are—I mean, if you look at their recent thinking game documentary, they're just like pointing their phone at things and saying, "What is this?" It's like, "You need glasses." Um you have OpenAI. You're working with OpenAI on this project. Family of devices that are going to be in a bunch of different places. And Apple obviously has to be considered a power player here as well. Who wins?

看，我会通过回顾互联网的起点来回答这个问题。Orkut 并不是最终胜出的社交媒体；最终是 Facebook，然后是 Instagram。MapQuest 并不是主要的地图；最终是谷歌地图。所以现在下结论还为时过早。我认为你看到所有这些公司；它们有庞大的生态系统，并且正在投资于它们的生态系统。我们拭目以待。不过，我会尝试给你一个答案。我有这样一个观点——这也许是我们有时间进行的更长对话——但我认为最终，边缘的赢家将是 AI 竞赛的赢家。我这么说的原因是，尤其是对于所有个人化的东西，边缘具有真正的上下文。

Look, I will answer this question by going into the beginning of the internet. Orkut wasn't the social media that won; it ended up being Facebook and then later Instagram. MapQuest wasn't the main map; eventually it was Google Maps. So it's early to call. I think you see all those companies; they have big ecosystems and they're investing in their ecosystems. We'll see what happens. However, I'm going to try to give you a little bit of an answer. I have this view—and this is maybe a longer conversation that we're going to have time for—but I think at the end of the day, the winner of the edge is going to be the winner of the AI race. And the reason I say that is because, especially for everything that is personal, the edge has real context.

意思是你的手机、你的设备、你使用的设备。

Meaning your phone, your device, the devices that you use.

在人类所在的地方。人类不会敲开数据中心的门说：“给我一些 AI。”他们是通过那里的其他设备来体验的。实际情况是，如果你看看模型是如何训练的，模型是在互联网上可用的信息上训练的。但当你快进到一个模型——当你加入物理 AI 的理解，理解我们的世界，理解你的上下文，理解你——那对你来说会比一个在互联网可用数据上训练的通用模型有用得多。所以谁能访问那些数据，谁就处于非常非常有利的位置。因此，是那些已经在所有这些不同设备中都有存在的公司。我认为它们有优势。我不会赌它们输。

Where the humans are. Humans don't knock on the data center and say, "Give me some AI." They're experiencing it through some other devices over there. And what happened is, if you look at how models got trained, models got trained on the information available on the internet. But when you fast forward to a model that is—when you add physical AI understanding, understanding our world, understanding your context, understanding you—that's going to be a lot more useful for you than a generic model that got trained on data available on the internet. So whoever has access to that data is in a very, very strong position. So it's companies that have presence in all of those different devices already. I think they have an advantage. I will not bet against them.

好的。但让我和你深入探讨一下，因为我们已经看到那些公司。我就点名吧。亚马逊、苹果、谷歌、Meta。它们都试图构建这种情境感知的个人助手。我们听过关于 Alexa Plus、Apple Intelligence 以及 Meta 各种伙伴的演示。谷歌显然有 Gemini，但尽管它们拥有所有这些数据，我们仍然没有真正拥有一个能够做到它们所承诺的事情的助手。我的意思是，苹果可能是最引人注目的，承诺这种情境感知助手能帮你弄清楚航班时间并告诉你：“好了，该去机场了。”他们还没有做到。是什么阻碍了这些公司？是硬件问题，还是 AI 问题？瓶颈在哪里？

All right. But then let me take this a level deeper then with you because we have seen those companies. I'll just name them. Amazon, Apple, Google, Meta. They've all tried to build this contextually aware personal assistant. We've heard presentations about Alexa Plus and Apple Intelligence and the Meta all the different buddies you can have in the meta properties. Google obviously with Gemini but even though they have all this data we still don't really have an assistant that's capable of doing what they've promised. I mean, Apple might be the most notable and promising this contextually aware assistant that will help you figure out when your flight is and tell you, "All right, time to get to the airport." They haven't done that yet. What is holding these companies back? Is it a hardware problem, an AI problem? Where is the bottleneck?

这是多种因素的结合，但我比你描述的更乐观。我认为我们开始看到一些真实体验的开端。你必须达到成熟——首先，AI 模型需要变得更成熟。它们需要变得更有能力。即使在 AI 内部也发生了很多变化。你开始看到混合专家模型。你开始看到思维链推理。所以你有不同的东西专注于特定任务。我认为我们才刚刚开始物理 AI，这对于获得上下文非常重要。所以我认为这将会发生。另一部分是算力。你需要大量的高性能计算，这就是我们介入的地方，因为由于延迟，你不能在云端做所有事情。如果——我回到你让我描述体验的时候。如果你和我在街上一起走，我会说：“嘿，这个人是谁？”你说：“这是某某。”答案不能是：“等等，让我想想。我们继续走。我一直在想，”然后那个人走过去了——你就错过了重点。我认为你需要在设备上做某些事情。它需要快速。就像现在所有公司一样，语音转文字——他们开始本地处理，因为你不会容忍任何延迟。所以，我们会达到那一步。

It's a combination of things, but I am more optimistic than you describe. I think we're starting to see the beginning of some real experiences. You have to get the maturity—first of all, the AI models need to get more mature. They need to get more capable. You had a lot of changes even within AI. You started to see mixture of experts. You started to see chain of thought reasoning. So you have different things specialized in specific tasks. I think we're just at the beginning of physical AI, which is really important for you to have context. So I think this is going to happen. The other part of it is compute. You need to have a lot of high-performance computing, and this is where we come into the picture because you cannot do everything on the cloud because of latency. It is not going to be useful for you if—I go back to when you ask me to describe the experience. If you and I are walking together in the street and I'm going to say, "Hey, who's this person?" and you say, "This is so and so," the answer can't be, "Hold on, let me think about it. Let's keep walking. I've been thinking about it," and then the person went by—you missed the point. I think you're going to have to have certain things you need to do on the device. It needs to be fast. Like all companies right now, voice-to-text—they're starting to do locally because you won't tolerate any delay. So, and we're going to get there.

是的，我们刚才在房间里谈到，可能是在滑雪场上，让眼镜为你指向适合你技能水平的坡道。但如果你必须等上两分钟，你知道，你可能是个初学者坡道的滑雪者，却指向黑钻道。所以，你真的希望它能够快速工作。

Yeah, we were just talking earlier in the room here about potentially being on the ski hill and having the glasses point you down the hill that suits your skill set. But if you have to wait like 2 minutes, you know, you might be a bunny hill skier and down the black diamond. So, you really want to be able to work fast when you do that.

眼镜。

Glasses.

是的。

Yeah.

你的眼镜会是第一个牺牲品。

Your glasses will be the first casualty.

好的。我们请到了高通 CEO 克里斯蒂亚诺·安蒙。这里是达沃斯的高通空间，我们这周会做四场对话，非常高兴能在这里。休息之后，我们会聊 AI PC、AI 数据中心、AI 建设面临的制约，以及机器人。如果时间允许，我们马上回来。

All right. We're here with Cristiano Amon, the CEO of Qualcomm. Here at Qualcomm Space at Davos, we're going to be doing four conversations through the week here and thrilled to be here. On the other side of this break, we're going to talk about AI PCs, AI data centers, the constraints on the AI buildout and robotics. If we have time, we'll be back right after this.

我们回到达沃斯 Big Technology Podcast 特别节目，周一录制，周二在我们的各个频道上线。我们继续聊 AI 将如何改变设备。AI PC 一直是我很感兴趣的话题。有很多声音说，如果电脑内置 AI，你就能更高效，它能真正改变你的工作方式。那是营销。现实中，这个承诺兑现得很慢。这是戴尔产品负责人对 The Verge 说的话。他说，过去一年我们学到的是，尤其从消费者的角度看，他们并不是因为 AI 才买的。事实上，我认为 AI 更多是让他们困惑，而不是帮他们理解某个具体结果。显然，高通在 AI PC 的成功上有利益。现在正在发生什么，未来会走向哪里？

And we're back here on Big Technology Podcast special edition here at Davos and we're here broadcasting, talking together on a Monday going live across our channels on Tuesday. And let's keep going here about how AI will transform devices. The AI PC is a subject that has been interesting to me. A lot of noise about how if you have AI baked into your computer, then you'll be able to be more productive and it can really transform the way that you work. That's the marketing. In reality, that rollout, that promise has been slow to meet the reality. This is from the head of product at Dell speaking to The Verge. He says, what we've learned over the course of the year, especially from a consumer's perspective, is they're not buying based on AI. And in fact, I think AI probably confuses them more than it helps them understand a specific outcome. Obviously, Qualcomm has a stake in the success of AI PCs. What is happening today and where is it going?

这是个很好的话题。首先，在我们进入 PC 领域时，我认为推动骁龙 PC 销售的一个重要原因，是我们提供了多天续航、强劲性能，以及非常令人兴奋的轻薄形态。所以我们只是造了一台更好的 PC。在消费者这边，我同意你的说法，你还没看到很多智能体，我知道人们希望马上看到。我也希望马上就能看到。在消费端我并不一定反对这个说法，因为微软刚刚为 Windows 推出了智能体，才刚推出。所以我认为，随着人们开始依赖智能体，会用得越来越多。你会看到一些东西在你的设备上运行。但我认为那不是 AI PC 的故事。故事有点不一样。我们看到 AI PC 正在发生的事，以及我们确实有能力在笔记本上运行大规模高性能推理，我们看到的是另一回事。我们看到的是，现在你的 PC 上有很多很多应用和服务在做大量云端计算，如果你能依赖 PC 上可用的算力，不仅会更快，而且经济性完全不同。我举个例子。

It's a great topic of conversation. Look, first of all, as we enter the PC space, I would argue that a lot of what's driving the sale of Snapdragon-powered PCs is the fact that we deliver multi-day battery life, a lot of performance in a very exciting thin and light form factor. So we just build a better PC. On the consumer side, I would agree with that, that you don't see yet a lot of agents, and I know people want to see this right away. I wish it was seen right away. I don't necessarily disagree with that on the consumer front because Microsoft just launched agents for Windows. It just launched. So I think people are going to use it more and more as you start to rely on agents. And I think you're going to see things that are going to be running on your device. But I think that's not the story for AI PC. The story is a little bit different. What we're seeing happening with AI PC and the fact that we actually have the ability to run significant high-performance inference on a laptop, we're seeing something else. What we're seeing is right now you have many, many applications and services on your PC that are doing a lot of cloud computation, and if you could rely on the computing that is available on the PC, not only is it going to be faster but it has a completely different economics. I'll give an example.

如果你是 SaaS 公司，现在所有 SaaS 公司都受到 AI 的威胁。如果你是 SaaS 公司，你说我要在应用里放一个智能体，每次我有这些数据就要跑一遍，而你在为云端算力付费，如果你真正用上设备里的算力，你的经济性会彻底改变。我举个实际的例子。现在有很多东西，你只要按一个按钮。你在 Microsoft Copilot 上看到，在很多不同应用上看到。总结这个。你有一堆数据，有几页文档，总结这个。你可以一路走到云端，承担云端算力成本来跑模型，或者你可以在电脑上跑那个总结文本的模型，那是免费的，因为那是你已经拥有的电脑。所以我们开始看到很多企业甚至应用有兴趣，把应用的一部分放到设备上的 AI 引擎上运行，这正在开始。

If you're a SaaS company, and all the SaaS companies right now are being threatened by AI. If you're a SaaS company and you say I'm going to have an agent within my application, and every time I have this data I'm going to run it, and you're paying for compute in the cloud, your economics change dramatically if you actually use the compute in the device. I'll give you a practical example. There's many things now. You just have a button. You see that on Microsoft Copilot. You see that across a number of different applications. Summarize this. You have a bunch of data. You have several pages of a document. Summarize this. You can go all the way to the cloud and have a cost of cloud compute to run the model, or you can run that model that summarizes your text on the computer, that's free, because it's the computer that you already have. So we're starting to see a lot of interest from enterprises or even applications to start running a portion of the application on the AI engine on the device, and that's starting right now.

所以购买 AI PC 硬件、而不是让云端代码接管你的电脑，主要原因基本上是成本？

So the reason to buy AI PC hardware as opposed to, let's say, letting Claude Code take over your computer, is mostly its cost?

我认为，你看，我刚给你举了一个例子。还有更多，比如游戏。现在很多游戏引擎都在考虑在 PC 上使用 AI。比如在 RPG 游戏里，你可以和一个角色像模型一样对话，你对话，玩法就变了。有成本的例子，有新用例的例子，有智能体的例子。我认为你问题的答案是，首先，为什么要买骁龙 PC？因为从定义上说，即使你不用 AI，它也会更快、多天续航，而且感觉像你的手机。你可以整天用笔记本，你可以去很多地方，不用带充电器。第二部分，作为消费者为什么要买 AI PC？作为消费者，我认为随着时间推移，你会看到越来越多应用有 AI 前端，它们会利用 PC 上的能力，但对你来说是透明的。在企业端，我认为经济性会改变，因为很多 ISV 和 SaaS 应用会需要板载算力，我认为那会带来不同。

I think you see, well, I just gave you one example. There's more, like gaming for example. A lot of the gaming engines right now are thinking about using AI on the PC. For example, you can have on an RPG game a dialogue with a character like a model, you have a dialogue, the gameplay changes. There's examples of cost, examples of new use cases, examples of agents. I think the answer to your question is, first of all, why should you buy a Snapdragon-powered PC? Because by definition, even if you're not using AI, it's going to be faster, multi-day battery life, and it's going to feel like your phone. You can use your laptop all day without, you can go places, don't take the charger with you. The second part of it, why should you buy an AI PC as a consumer? As a consumer, I think over time you're going to see more and more apps having an AI front end and they're going to leverage the capabilities on the PC, but it's going to be transparent to you. On the enterprise, I think the economics are going to change because a lot of the ISVs and SaaS applications are going to require the onboard computing, and I think that's going to make a difference.

很有意思。所以那会成为软件公司的要求。高通也进入了数据中心领域，你们在建设数据中心。显然你们有设备里的芯片，就像我们聊的，但现在你们在为 AI 推理建设数据中心。我们聊聊，其实你先给我们讲讲，为什么这是高通要做的动作。

Very interesting. So that'll be a requirement from software companies. So Qualcomm has also gotten into the data center world and you're building data centers. So obviously you have the chips in the devices like we talked about, but now you're working on building data centers for AI inference. So let's talk a little bit about, well actually why don't you first give us a little bit about why this is a move that Qualcomm is making.

是的。你看，我们一直相信 AI 在数据中心会发生的事，你开始看到所有为训练而建的基础设施，但最终，现在这已经很好理解了，当我们开始开发我们的方案时，我们就是这么想的：最终推理会超过训练。因为你想一下，如果你是一家公司，花数十亿美元建一个训练数据中心，你期望获得投资回报。所以当你把 AI 投入生产时，你就是在做推理。我们一直相信，当你进入推理时，不同 AI 玩家之间会有大量竞争。所以我认为总拥有成本很重要，你消耗多少电力很重要，架构也很重要。

Yes. Look, we always believe that what's going to happen with AI in a data center, you started to see all this buildout for training, but eventually, and now it's well understood, when we started developing our solutions, that's what we thought: eventually inference is going to take over training. Because just think about that for a second. If you're a company spending billions of dollars building a data center for training, you expect to get a return on that investment. So when you start putting AI into production, you're doing inference. And we always believe that when you go to inference, there's going to be a lot of competition between the different AI players. So then I think the total cost of ownership matters, how much power you consume matters, and the architecture matters.

所以首先回答你的问题，我们意识到当数据中心开始转向推理时，我们有机会利用我们的资产，为数据中心构建一个非常节能的推理解决方案，将我们为边缘设备开发的技术进行 Scaling（规模扩张）。

So first answer to your question is we realize when the data centers start to transition to inference we have an opportunity to leverage our assets to build a very power efficient inference solution for the data center, scaling the technology that we develop for the edge.

因为手机能效很高，而电力是 AI 的瓶颈，你可以利用这一优势并将其应用到数据中心。这就是逻辑。

Because the power is efficient in the phone and power is such a bottleneck in AI, you can use that advantage and put it in a data center. That's the logic.

如果你看看今天，AI 的增长非常迅猛，但能源的增长却没有跟上。你已经看到可用能源和 AI 之间存在差距。所以我认为能源也将成为运营推理数据中心的资源。这是运营支出中最大的项目之一。然后我认为人们想要不同的架构，这是我答案的第二部分。

If you just look at today you have this very aggressive ramp of growth of AI and you don't have the same ramp on energy. You know already there's a gap between available energy and AI. So I think energy is going to be a resource also to operate an inference data center. That's one of the biggest items in operating expenses. And then I think people wanted to have a different architecture which is the second part of my answer.

我答案的第二部分是我们相信数据中心正在经历另一个解聚过程。让我解释一下我的意思。移动行业发生的关键事情之一，如果你看看今天的智能手机，从半导体的角度来看，智能手机是一个非常困难的工程挑战，因为我必须在你的智能手机中集成大量计算。它必须能放进口袋。它不能发热。你会触摸屏幕、你的脸。它不能发热。我不能有风扇。我不能在智能手机上做液冷。而且你的电池必须能用一整天。否则，它就没用。

The second part of my answer is we believe that the data center is going through another process of disaggregation. Let me explain what I mean by that. One of the key things that happened in the mobile industry, if you look at your smartphone today, your smartphone is a very difficult engineering challenge from a semiconductor standpoint because I have to pack a lot of computing in your smartphone. It has to fit in your pocket. It cannot get hot. You're going to touch your screen, your face. It cannot get hot. I cannot have fans. I cannot do liquid cooling on the smartphone. And your battery has to last all day. Otherwise, it's not useful.

那就没价值了。

It's worthless.

是的。所以，为了做到这一点，我们必须完善计算的解聚，没有更好的描述方式。我举个例子。在 PC 中，一切都是围绕 CPU 的。所以如果你要解码音乐或解码视频，你会加载 CPU。在手机上你不能这样做。它消耗太多电力。所以你创建了专用于音乐解码的硬件，专用于拍照时 JPEG 编码的硬件。专用于视频解码的硬件，一切都被聚合。我认为你这样做是因为你想最大化利用电池中的可用能量。

Yes. So, in order to do that, we had to perfect the disaggregation of the compute, for lack of a better way to describe it. I'll give you an example. In the PC everything was CPU-centric. So if you're going to do a decode of music or you do decode of a video, you go and load up the CPU. You can't do that on the phone. It burns too much power. So you create a dedicated hardware just for music decode, a dedicated hardware just to JPEG encode when you take a picture. A dedicated hardware for you to do video decode and everything is aggregated. And I think you do that because you wanted to maximize the use of the available energy in the battery for you.

而这一切都存在于手机中。

And this all exists in the phone.

存在于手机中。这就是我们所说的异构计算。如果你看看今天的骁龙，它有多个引擎用于不同的事情。我们不会在 CPU 甚至 GPU 上运行所有东西。数据中心也进入这个领域，我们开始看到解聚。他们有一种用于预填充的架构。他们有一种用于解码的架构。所以当开始做推理并且需要专用引擎时，我们正在构建我们认为的后 GPU 时代。我们正在构建那个。我实际上相信英伟达收购 Groq 验证了你需要不同的引擎来做不同的事情，我认为这就是我们正在做的，我认为这就是我们在数据中心上的重点。

Exists in the phone. It's what we call heterogeneous compute. If you look at a Snapdragon today, it has several engines for different things. We don't run everything on the CPU or even on the GPU. Data centers go into that and we're starting to see disaggregation. There's an architecture that they use for prefill. There's an architecture they use for decode. So we're building what we believe is post-GPU when you start to do inference and you need dedicated engines. We're building that. I actually believe that the Nvidia acquisition of Groq validates that you need different engines for different things and I think that's what we're doing and I think that's our focus on data center.

好的，我们来谈谈机器人。你相信人形机器人的炒作吗？

Okay, let's talk about robotics. Are you buying the hype on humanoid robots?

我喜欢和你的整个对话。我一直在做比较，我将与汽车行业做个比较来概述我们的战略。但让我先给你答案。我买人形机器人的机会。然而，机会将会有不同的表现方式，其中一些事情需要时间。例如，直接回答你的问题，一个机器人将在你家里，做你要求它做的一切。训练它需要时间。这非常困难。

I like this whole conversation with you. I've been doing comparisons and I'm going to do a comparison with automotive to kind of outline our strategy. But let me give you the answer first. I buy the opportunity to humanoid robot. However, the opportunity is going to manifest itself differently and some of those things are going to take time. For example, to get straight to your question, a robot that is going to be with you in your house and it's going to do everything you ask the robot to do. It's going to take time to train that. It's very difficult.

远程操作员？

Teleoperators?

这很困难。每个家庭不会一样。每个任务不会一样。需要大量的训练。话虽如此，一个机器人能完成某些任务并反复执行，这实际上不是一个难以解决的问题。因此，我将给你我的比较隐喻。当我们开始做汽车时，当我们开始为汽车构建平台，我们现在为我们的汽车业务感到非常自豪。我们也进入了自动驾驶的技术栈。当你想到自动驾驶，当你想到 Robotaxi，比如 L5 级别，没有方向盘，你坐到后座打个盹，那需要大量的训练。因为你可以达到 0 到 95%，但要达到 99.999% 的极端情况，你必须进行大量训练。然而，如果你做辅助驾驶，人类仍然负责接管方向盘，如果发生什么，那么你有能力将其应用到每辆车上，从 L2 到 L2+ 到 L3，一直到 L4。所以，这是一个巨大的市场机会，这就是我们现在正在做的。你可以为每一款车型带来某种形式的辅助驾驶。我对机器人也有同样的感觉。如果你做一个类人机器人或类人手臂，或者你做任何可以利用为我们设计的世界的事情，并且你训练机器人执行特定任务，我认为我们非常——这已经在发生了，我相信从商业角度来看机会是巨大的。这就是为什么我们真正专注于工业机器人。因为你可以训练一个机器人，例如，你的任务是晚上去超市把东西放回货架。这是一个自包含的问题。你不是在训练机器人做所有事情。我认为能做好一切的机器人还需要一点时间才能实现。

It's difficult. Every house is not going to be the same. Every task is not going to be the same. It's going to be a lot of training required. Having said that, a robot that can do certain tasks and do that task over and over, that's actually not a hard problem to solve. So with that, I'm going to give you my comparison metaphor. When we started in auto, when we started building platforms for automotive and we're very proud of our automotive business right now. We also got into a stack for autonomous driving. When you think about autonomous driving, when you think about robotaxi, like a level five, no steering wheel, you go to the back seat and you take a nap, that requires a lot of training. Because you can get to 0 to 95%, but for you to get to 99.999% of the corner cases, you have to do a lot of training. However, if you do assisted driving with the human still responsible to pick up the steering wheel and something happens, then you have the ability to put this in every car from level two to plus two plus to level three and all the way to level four. So, that's a massive market opportunity and that's what we're doing right now. You can bring some form of assisted driving to every single model. I feel the same way about robotics. If you do a humanoid robot or humanoid arm or you do anything that it can leverage the world that's been designed for us and you train the robot on a particular task, I think we're very—it's already happening and I believe the opportunity from a business standpoint is massive. That's why we're really focused on industrial robots. Because you can train a robot, for example, your task is going to go to the supermarket at night and put the stuff back on the shelf. That's a self-contained problem. You're not training a robot to do everything. I think the robot that will do everything is going to take a little bit of time until we get there.

在中国有一场人形机器人半程马拉松，精彩片段看起来真的很有趣。机器人在起跑线上拉自己的脸，机器人带着整个团队，抓着绳子，把它们甩到赛道边。人们跑得又快又远，机器人冲出赛道时带着的力量。但有些机器人完成得相当快。我不会说它们超过了我半程马拉松的时间，但最终会的。但它们的完赛成绩值得尊重，而且包括了换电池的时间。一直有观点认为，在中国，中国离生产过程如此之近。想想他们的汽车，对吧？他们有电动汽车的繁荣，因为他们长期以来一直在制造电池和电子产品。谷歌 DeepMind 的 CEO 德米斯·哈萨比斯最近表示，中国只落后西方最先进模型几个月。但在机器人方面，他们似乎领先。你同意这个观点吗？

There was a half marathon in China of humanoid robots and the highlights looked really funny. Robots pulling on their face at the starting line and robots taking their whole team, holding on to ropes and like flinging them into the side of the course. And people went pretty fast and pretty far with that the power the robot had as it sort of crashed out of the course. But some of those robots finished pretty fast. I won't say they beat my half marathon time, which eventually they will. But they were respectable in their finish and that included time for battery changes. And the argument has been that in China, China is so close to the production process. Think about their cars, right? They have this electric car boom because they've been building things with batteries and electronics for so long. Demis Hassabis, the CEO of Google DeepMind, recently said that China is only a couple months behind the state-of-the-art western models. But it seems like they're ahead on robotics. Do you agree with that argument?

看，我认为中国有很多事情——我认为他们所做的非常了不起。

Look, there are many things I think that China—it's remarkable, I think what they're doing.

我觉得大家都在谈论中国速度。我们从在中国有许多不同的合作伙伴使用我们的技术中了解到这一点，不仅是在汽车领域，现在还有手机、机器人和工业领域。我认为有一种观点有一定道理：你离一个非常庞大的工业基础更近，你可以快速制作原型，快速制造，快速失败。我认为这些对技术开发有帮助。但机器人技术所需的技术非常非常广泛。从先进半导体开始——我认为这是中国公司与高通等公司合作的一个领域。你会有很多生态系统，这对于训练大量软件很重要。但是的，这很迷人。每个人都在竞赛，事情进展得很快。

I think everybody talks about the China speed. We know that from having a number of different partners in China using our technology, from not only cars but also phones now robots and industrial. I think there is some merit in the argument that you're closer to a very large industrial base and you can prototype fast, you can build things fast, you can fail fast. And I think those things are helpful in developing the technology. But the technology required for robotics is very, very broad. You go from advanced semiconductors—I think that's one area that the China companies are partnering with companies like Qualcomm and others. You're going to have a lot of ecosystem that is going to be important for training a lot of software. But yes, this is fascinating. Everybody is in a race and things are moving fast.

最后，我想谈谈工业 AI，我认为在 AI 对话中，这可能是最少被提及的，但却是当今发生的一些最有趣的事情。我的意思是，即使在这个空间里，我们有一个机器人，我们正在看它，它是在短短几周内用 50 美元的高通芯片建造的，而且它移动得相当好。请稍微谈谈 AI 在工业领域的应用，也许你认为为什么人们没有太关注它。只是因为它不够吸引眼球，不足以成为头条新闻吗？

Lastly, I want to talk about industrial AI, which is something that I think as far as the AI conversation gets probably the least ink, but is some of the most interesting stuff that's happening today. I mean, even here at the space, we have a robot that we're looking at that was built in just a couple of weeks with a $50 Qualcomm chip and it's moving pretty well. Talk a little bit about the applications of AI in the industrial space and maybe why you think people aren't paying so much attention to it. It's just not sexy enough for the headlines?

你知道，那可能是——我会说现在对数据中心的关注太多了，它可能占据了对话中的所有空气，数据中心。我甚至会产生共鸣，因为我说我们正在为数据中心建造一些东西，它得到了很多关注。但现实是，AI 的工业机会是巨大的。它是巨大的，因为你可以把 AI 处理放在几乎任何东西上，你会发现每个行业、每个垂直领域都有大量的用例。零售业如此，仓储业如此，医疗保健业如此，制造业和能源业也是如此。我们实际上看到了令人难以置信的需求，特别是因为如果你有能力实时处理来自物理 AI 的东西——电机、机器，所有你可以放置传感器的东西。但只是给你一个例子，如果我们不对不同的机器过于花哨，仅计算机视觉方面：一个摄像头——你可以把摄像头放在生产线上，你训练模型只是看看传送带上来的东西是否符合模板，是否符合你的预期。你用摄像头做质量控制。例如，你把摄像头放在超市的货架前。你现在有能力实时检查库存。你实际上可以在线销售商店里的东西，并实时管理库存。你可以把同样的摄像头放在智慧城市中，读取车牌。我认为这是一个巨大的、巨大的机会。我在达沃斯这里举行的许多会议中，有一些是与工业公司进行的。他们对工业 AI 非常感兴趣，我认为这实际上正在发生。

You know, that's probably—I'll say there's so much attention on data center right now that it probably takes all of the air in the conversation, data center. I'll even resonate with the fact that we said we're building something for the data center and it got a lot of attention. But the reality is the industrial opportunity for AI is massive. It's massive because you can put AI processing on pretty much everything and you find that every single industry, every single vertical has a massive number of use cases. It's true in retail. It's true in warehousing. It's true in healthcare. It's true in manufacturing, in energy. And we're actually seeing an incredible amount of demand, especially because if you have the ability to process in real time things that come from physical AI—motors, machines, all of those things you can put sensors on. But just to give you an example, if we don't get too fancy with different machines, just in computer vision alone: a camera—you can put a camera on a manufacturing line and you train the model just to see if what's coming in on the conveyor belt against the template is what you expected. You do quality control with just a camera. You put the camera, for example, looking at a shelf of a supermarket. You now can have the ability to check inventory in real time. You can actually sell online what's in the store with real-time inventory management. You can put the same camera on a smart city and you're reading license plates. And I think it's a massive, massive opportunity. Some of the many meetings I'm having here at Davos are with industrial companies. They're super interested in industrial AI and I think that's actually happening right now.

好的，还剩五分钟。有两个问题要问你。我很高兴与你交谈的原因之一是，在某种意义上，你能看到未来，对吧？因为当某样东西要大规模生产时，你是第一个被打电话的人，比如说，有人正在打造 AI 可穿戴设备。你与 Meta 密切合作，所以你很了解他们预期的需求，因为他们需要你的芯片来构建产品。考虑到 AI 的构建，也许还有 AI 设备的构建，并看看你拥有的关于未来面貌的水晶球：事情会继续以这个速度发展吗？它们可能保持像以前一样快吗？

Okay, five minutes left. Two questions for you. One of the reasons why I'm so happy to be speaking with you is because in a sense you can see the future, right? Because when something is going to be mass-produced, you're the first call that's being made from, let's say, someone building an AI wearable. You're working closely with Meta, so you have a pretty good understanding of the demand that they're anticipating because they need your chips to be able to build things out. Thinking about the AI buildout and maybe also the AI device buildout and looking into the crystal ball that you have of what the future looks like: Are things going to continue at pace? Can they possibly keep moving as fast as they have been?

看，我觉得——我认为那个问题真正针对的是数据中心可能正在发生的事情，因为在个人设备方面，我们才刚刚开始。我认为我们看到了一个很大的轨迹，比如眼镜继续逐季增长。但我认为更广泛的问题是数据中心的速度。这是我的回答：如果我们回到 2000 年互联网泡沫破灭的时候，对吧，你有互联网泡沫的调整。回到 2000 年，想想我们当时认为互联网会是什么样子。我会告诉你，今天，25 年后，现在 26 年了，它实际上比人们想象的要大得多。所以无论他们在 2000 年认为互联网会是什么样，现在都要大得多。

Look, I feel that—and I think that question is really directed at what's probably happening on the data center because on the personal, on the device side we're just at the beginning. I think we're seeing a big trajectory, like for example glasses continue to increase quarter over quarter. But I think the broader question is to the speed on the data center. And here's my answer: if we go back to the year 2000 when the dot-com crash happened, right, you have that correction on the dot-com. Go back to year 2000 and think about what we thought back then what the internet would be. I will tell you that today, 25 years later, 26 years now, it is actually way bigger than people thought it would be. So whatever they thought in 2000 the internet would be, it's way bigger right now.

而且你仍然可以在这个上面买宠物食品。

And you can still buy pet food on this one.

是的。然而，它并不是在 2000 年全部发生的。它发生了。所以我认为现在 AI 从长远来看会比人们想象的更大。从长远来看，它可能被低估了。现在，它部署的速度有多快，有多普及，我们拭目以待。我们能继续以这个速度建设吗？有可能。会不会出现放缓？也有可能。嗯，我们对此感到兴奋。我认为终于——这更多是针对高通——终于，人们刚刚意识到边缘机会是巨大的，所有关于数据中心的空气，其中一些开始关注边缘了，我认为我们只是在那条曲线的起点。

Yes. However, it didn't happen all in 2000. It happened. So I think what's going to happen is AI right now in the long run is going to be bigger than people think. It's probably under-hyped for the long run. Now, how fast this is going to get deployed and how pervasive, we'll see. Could we continue to build at this pace? It's possible. Could a slowdown also happen? It's possible. Well, we're excited about it. I think it's finally—and this is more for Qualcomm—finally, people just woke up that the edge opportunity is massive and all of this air that was all about data center is some of it started going to paying attention to the edge right now and I think we're just at the beginning of that curve.

好的。最后，我得问你一个达沃斯问题。我们在达沃斯。我们身后有雪坡。这是真的。对于那些想知道的人，公司已经经历了这段非常有趣的旅程。曾经有段时间，他们进入了所谓的利益相关者资本主义，他们考虑股东之外的人群。我认为我们现在正处于一个更赤裸裸追求底线的时刻。我不是在说高通。我只是广泛地说，似乎公司更加——他们似乎抛弃了这种幻觉，即他们关心底线之外的其他东西。我想知道——你知道，我们就在世界经济论坛外面。在这个活动中将有 48 场关于 AI 的对话。

Okay. Finally, I got to ask you a Davos question. We're here at Davos. We have the slopes behind us. This is real. For those wondering, corporations have been through this really interesting journey. There's been moments where they've been into what's called stakeholder capitalism where they think about the group of people beyond the shareholder. And I think we're kind of in a moment now where there's more of a naked pursuit of the bottom line. I'm not speaking about Qualcomm. I'm just saying broadly it seems like corporations are much more—they've sort of put away this illusion that they care about much else than the bottom line. And I wonder if—you know, we're here right outside the World Economic Forum. There's 48 conversations that will happen in this event that will be about AI.

人们会谈论 AI 如何能够治愈癌症，或者为我们提供治愈癌症的最佳机会，并赋能弱势群体。所以我很好奇，从你的角度来看，你认为 AI 会成为新的利他主义，还是新的企业利他主义？这是好事还是坏事？

People will be talking about how AI will be able to cure cancer or get our best chance at curing cancer and empower the disempowered. So I'm curious, from your perspective, do you think AI is going to be the new altruism or the new corporate altruism, and is that a good or a bad thing?

这是个复杂的问题。你看，我认为技术是一种工具。我认为它会像计算机那样，并且会继续如此。我认为它会帮助加速许多事情。例如，它会帮助加速药物研发。它会帮助许多事情，提高生产力，正如我之前所说。它可能会使教育民主化。它会改变我们对教育的看法。这是不断变化的事情。它将是一种工具。我不认为它会成为那种改变社会的东西。我给你一个非常个人的回答。当我大学毕业时——这会很糟糕，因为这会暴露我的年龄——但当我大学毕业时，互联网才刚刚开始。我记得去我的第一份工作，那里有一台传真机，你得去传真机那里取你 overnight 收到的传真，然后把其他传真放进去，还有人仍在打办公室备忘录，就像我们现在不再谈论这个了。我认为当互联网和电子邮件到来时，那是一场革命。我认为 AI 将会是那种革命。几乎像计算机一样，但会是我们用计算机做事情，只是更多。这就是我对它的感觉。

It's a complicated question. Look, I think technology is a tool. I think it's going to be like computers did and will continue to do. I think it will help accelerate many things. It will help, for example, accelerate drug discovery as an example. It will help many things, will increase productivity as I said before. It's probably going to democratize education. It's going to change how we think about education. This is something that keeps changing. It's going to be a tool. I don't think it's going to be this change-society kind of thing. I'll give you a very personal answer. When I got out of college—and this is going to be terrible because it's going to show my age—but when I got out of college, it was just the beginning of the internet. I remember going to my first job and there was a fax machine, and you had to go to the fax machine and get the faxes that you got overnight and put the other faxes in there, and you had somebody still typing interoffice memos like we don't talk about this anymore. I think when the internet arrived and email arrived, it was a revolution. And I think AI is going to be that kind of revolution. Almost like computers, but it's going to be like us doing things with computers, just more. That's how I feel about it.

好的。嗯，关注这个领域真是太棒了，因为每次我以为自己跟上了，就会有新东西出现。而且我认为你将会处于这一切的中心，随着所有即将推出的设备。你知道，也许当 OpenAI 真的发布这一系列设备时，我们可以再次谈论竞争状态。顺便说一句，我们有一个很棒的现场观众。大家，制造点噪音，让人们能听到。致 Cristiano 和高通团队，感谢你们邀请我来到你们在达沃斯的空间。非常兴奋能参与一系列关于 AI 现状的精彩对话。我相信我们的观众在结束时会对事情的发展有很好的理解，而这是一个很好的开端。所以，Cristiano，非常感谢你来参加节目。

All right. Well, it's been amazing following this space because every time I think I'm caught up, there's something new. And I think that you're going to be right at the center of it with all the devices that are going to come out. And you know, maybe when OpenAI does release this family of devices, we can talk again about the state of the competition. By the way, we have a great live audience with us. Guys, make some noise so people can hear that. To Cristiano and the Qualcomm team, thank you for having me here at your space at Davos. And very excited to be engaging in a number of really great conversations about the state of AI. I'm sure that our audience by the end of them will have a really good understanding of where things are going, and this was a great way to kick it off. So, Cristiano, thank you so much for coming on the show.

不，谢谢你。谢谢。和你进行这次对话我真的很有趣。谢谢。

No, thank you. Thank you. I really had fun having this conversation with you. Thank you.

我也是。好了，各位。感谢收听，我们下次在 Big Technology Podcast 再见。谢谢。

You too. All right, everybody. Thanks for listening and we'll see you next time on Big Technology Podcast. Thank you.

非常感谢。

Thank you very much.