高通 CEO 克里斯蒂亚诺·安蒙：下一场移动革命

AI 最特别的地方其实在于人。正如你所说，我们开始看到可穿戴设备出现这种趋势。我认为这将是下一场移动革命。就像计算机现在能理解我们看到什么、听到什么一样，网络也将能够做到这一点。大家好，我叫 Ismael Herbert，欢迎回到 Humane 的 AI 播客《The End of Limits》。今天的嘉宾是 Qualcomm 的 CEO Cristiano Amon。非常感谢你今天来到这里，Cristiano。

The most especial thing about AI is actually the human. We're starting to see that happening, as you said, with wearables. And I think that's going to be the next mobile revolution. In the same way that computers now understand what we see, what we hear, the network is also going to be able to do that. Hi everyone. My name is Ismael Herbert, and welcome back to The End of Limits, the AI podcast by Humane. My guest today is Cristiano Amon, the CEO of Qualcomm. Thank you so much for being here today, Cristiano.

哦，非常高兴来到这里。谢谢。

Oh, super happy to be here. Thank you.

Qualcomm 是全球移动通信和计算领域的领先公司之一。你可能没听说过他们，但如果你用手机、开车，很可能用到他们的一些产品或至少是他们的知识产权。Cristiano 今天来这里谈谈他们如何推动下一代技术。也许这是个好的开始，因为你今天在 Humane 办公室。告诉我们你为什么在这里。

So, Qualcomm is one of the leading companies in the world in the mobile communications and computation space. You might not have heard of them, but if you use a phone, if you use a car, chances are you probably use some of their products or at least their IP. And Cristiano is here today to talk about how they are powering the next generation of technology. And maybe, you know, that's a good place to begin because you're in the Humane offices today. Tell us why you're here.

是的，今天是个非常特别的日子。我想我们正在发布一些非常有趣的公告，但你看，我们在这里是因为我们喜欢与 Humane 的合作。我认为 Humane 已经理解了 AI 的未来走向。很早就认识到 AI 既要在数据中心又要在边缘的重要性，并愿意做新的事情。我们对正在做的事情感到非常兴奋。从这种改变未来的个人计算开始。我们如何做计算机视觉，我认为今天的公告是我们将在 AI 加速方面做的事情。这非常令人兴奋。

Yes, today is a very special day. I think we're making some very interesting announcement, but look, we're here because we love the partnership of Humane. I think Humane has understood where the future of AI is going. Understood early on the importance of having AI both on the data center, on the edge, and is willing to do new things. And I think we're very excited about what we're doing. Starting with this future-changing personal computing. How we're doing the computer vision, and I think the announcement today is what we're going to do with AI acceleration. It's super exciting.

也许为我们的听众稍微展开一下，当我们谈论边缘 AI 与云端和混合 AI 时，我们指的是什么。

And maybe for our audience, we can expand a little bit on what we mean by when we're talking about AI on the edge versus the cloud and hybrid.

是的。所以，有趣的是。我们看到很多关于 AI 在数据中心的讨论，然后你应该在边缘做 AI。但我想稍微拉远一点看。自个人计算开始以来，计算的未来一直是边缘和数据中心两者。从一开始就是这样。实际上，如果你看看今天，我们的智能手机，我们不可分割的设备，是世界上连接云端最多的设备。实际上，如果你把它调到飞行模式并尝试使用手机，会非常沮丧。但与此同时，它是连接云端的。它有很多计算能力，设备上有很多事情要做。在 AI 上也不会不同。我认为 AI 正在改变一般的计算。它正在改变计算架构。它正在改变操作系统。它正在改变应用程序。它将在数据中心和边缘同时发生。我认为混合就是当这两者一起工作的时候。你有 AI 在边缘运行，在数据中心运行。现在，对我来说最简单的解释方式就是想想 AI 在与我们所有人互动中在做什么。现在，有了那些大型语言模型、那些大型视觉模型，AI 理解人类语言，理解我们说什么、写什么、看到什么。当然，如果那是计算机的用户界面，那它就在我们所在的地方。那就是在边缘。我认为这是一个很好的例子，说明这在边缘是如何发生的。但还有更多事情。隐私，你想要非常快速响应、低延迟的事情。然后甚至如何卸载计算，让边缘做一些计算，然后发送到云端进行更复杂的计算。

Yes. So, it's interesting. We see a lot of discussions about AI on the data center, and then you should do AI on the edge. But I like to kind of zoom out a little bit. The future of computing, since the beginning of personal computing, has been both the edge and the data center. Since the beginning. Actually, if you just look at today, our smartphone, our inseparable device, is the most cloud-connected device in the world. Actually, if you put it in airplane mode and try to use your phone, it's going to be very frustrated. But at the same time, it's cloud-connected. It has a lot of computing capabilities and a lot of things to do on the device. And it's not going to be different on AI. I think AI is changing computing in general. It's changing computing architectures. It's changing the OS. It's changing the application. It's going to happen both on the data center as well as on the edge. And I think hybrid is when you have those two things working together. You have AI running on the edge, running on the data center. Now, the easiest way for me to explain this, it's just thinking about what AI is doing in the interaction with all of us. Now, with those large language models, those large visual models, the AI understands human language, understands what we say, what we write, what we see. And of course, if that's the user interface of the computer, that stays where we are. That's on the edge. I think that's one great example of understanding how this is happening on the edge. But then there's more things. Privacy, things that you wanted to have a very fast response, low latency. And then even how you offload computing, having the edge doing some of the computation and then send it to the cloud for more complex computations.

对。我认为本地数据中心是这种交换中非常关键的一部分，让事情保持低延迟，你知道，相对低的延迟，但仍然在云端，这样你就不会占用计算机太多空间。所以，告诉我们一些关于 Qualcomm 扩展到数据中心的事情。

Right. And I think local data centers are a very critical part of that sort of exchange between having things be low, you know, relatively low latency, but still on the cloud so that you're not taking up too much space on the computer. So, tell us a little bit about Qualcomm's expansion into data centers.

是的。所以我们看到，首先，基于这个边缘和混合，我们来到数据中心，从手机领域、汽车领域和 PC 领域来到数据中心。我们决定去数据中心的原因，这对我们来说是一种自然的扩展，是因为我们相信我们现在正进入 AI 的下一个阶段。在 AI 的下一个阶段，我们将从 AI 的创建，也就是所有为训练而建立的数据中心容量，转向实际将 AI 投入生产。我认为有很多预测表明，在数据中心容量方面，推理将超过训练。一旦发生这种情况，公司将在 AI 计算业务上竞争，你有非常创新的公司，这个领域的新进入者，比如 Humane。我们相信，在功耗和总拥有成本方面更高效的新架构将发挥作用。因为我们的移动 DNA，移动设备从工程角度来看是一个非常困难的设备，因为你必须打包大量的计算能力，但你的电池必须持续一整天，它必须能放进口袋，而且不能发热。我们一直在思考，我们实际上可以将其扩展到数据中心，并在效率方面拥有非常颠覆性的技术。我认为这与 Humane 想要做的事情完美匹配，特别是为数据中心带来颠覆性的商业模式。所以，这是一种自然的结合。

Yes. So, we saw that there was a First of all, building on this edge and hybrid, we're coming to the data center, coming from the phone space and the car space and the PC space, going to the data center. The reason we decided to go to the data center, and it was a kind of natural expansion for us, is because we believe that we're now going to the next phase of AI. In the next phase of AI, we're going to move from creation of AI, and that's all the data center capacity that has been built out for training. We're going to go from creation of AI to actually putting AI into production. And I think there's a lot of projections that inference is going to overtake training in terms of data center capacity. Once that happens, and companies are going to compete for the AI computing business, you have companies that are very innovative, new entrants in the space, like Humane. We believe that new architectures that are going to be much more efficient in power consumption and total cost of ownership will make a difference. And because of our mobile DNA, the mobile device is a very difficult device from an engineering perspective because you have to pack a lot of computing power, but your battery has to last all day, it has to fit in your pocket, and it cannot be hot. And we've been thinking we can actually scale that to the data center and actually have a very disruptive technology in terms of efficiency. And I think that it's a perfect match to what Humane wants to do, especially as coming with a disruptive business model for the data center. So, it was kind of a natural combination.

所以我理解为什么电源效率在手机上如此重要，因为你列出的所有这些原因。但即使当你扩展到像数据中心这样巨大的东西时，它似乎仍然是你们最关心的事情。稍微展开一下为什么。

So, I understand why power efficiency is so important on the phone for all these reasons you laid out. But even when you expand to something huge like a data center, it still seems to be something that's top of mind for you guys. Expand a little bit on why that is.

是的，我认为有两个原因。一是如何提高可用电力的效率。如果你看看现在发生的事情，数据中心消耗大量电力。然后如果你用你拥有的电力获得最大量的计算或 token，如果你考虑推理，我认为这将是极具竞争力的数据中心和竞争力较低的数据中心之间的区别。这是一个原因。我认为另一个原因不仅是能源消耗本身，还有提供该能力的总体成本。正如我之前所说，随着 AI 规模扩大，将是 AI 上的竞争。实际上能够以最低可能成本提供那组推理 token 的公司将更有利可图，也将更具竞争力。所以，在我看来这是不可避免的。我们必须构建更高效的 AI。

Yes, I think for two reasons. One is how do you increase efficiency of the power that's available. If you look at what's happening right now, there's a lot The data centers are consuming a lot of power. And then if you get the most amount of computation or tokens, if you think about inference, with the power you have, I think that's going to be the difference between a very competitive data center and a less competitive data center. That's one reason. I think the other reason is it's not only the energy consumption itself, but the overall cost of providing that capability. And as I said before, as we get into AI getting a scale, it's going to be competition on AI. The companies that actually can provide that set of tokens for inference at the lowest possible cost are going to be more profitable and it's going to be more competitive. So, it seems to me that it's inevitable. We have to build AI that is going to be more efficient.

是的，我的意思是，每一个达到规模的 AI 创始人，我都会问他们，你知道，是什么让你夜不能寐，你最担心的问题是什么？他们总是说推理和成本，你知道，我们怎么才能让那行得通？所以，这确实引起了共鸣。但我们也回到混合这个点上，你认为有哪些基本的架构原则能够实现未来这种高效的混合 AI 使用？

Yeah, I mean, every AI founder who reaches scale, and I ask them, you know, what keeps you up at night, what's the number one problem you're worried about? And they always say the inference and cost, you know, how are we going to make that work? So, that definitely resonates. But we're also to circle back to the hybrid point, what do you think are some of the fundamental architectural principles that will enable this sort of efficient hybrid AI usage in the future?

是的。你知道，有趣的是，尽可能多地在靠近传感器数据的地方进行计算的能力。举个例子，就是我们在计算机视觉方面与 Humane 合作做的事情。你知道，让摄像头能够看到、标注图像，然后具备处理能力来学习大型视觉模型、处理数据、理解数据的含义，然后将其发送到云端。我认为这不仅效率极高。想想看，如果你在沙特阿拉伯王国各地部署摄像头，所有高清数据流同时发送到云端。从存储、从网络消耗的角度来看，那将极其低效。这是一个完美的例子，让你看到混合实际上是如何工作的。你部署模型。模型对图像有非常具体的任务，会做出决策，理解事件是什么，然后当需要采取行动时，会将其传达给云端。所以，你节省了带宽和网络传输数据以及存储，然后你在响应时间上获得了收益。

Yes. You know, the interesting thing is the ability to do as much as the computation that you can, you know, closer with the sensor data is. Like, if I can give an example, is what we're doing with Humane on computer vision. You know, the ability to have cameras that are able to, you know, see, annotate the image, and then have the processing capabilities to learn a large visual model, process the data, understand what the data means, and then send that to the cloud. I think not only it has an incredible efficiency. Think about it, if you deploy cameras all over the Kingdom of Saudi Arabia, and all of the data stream, you know, in high definition is being sending to the cloud at the same time. From a storage, from a network consumption point of view, it would be incredibly inefficient. And that is a perfect example for you to see how hybrid actually works. You deploy the model. The model has a very specific task on the image, will make a decision, understand what the event is, and then when it's needed to take action, will communicate that to the cloud. So, you save on bandwidth and off the network to transfer data as well in the storage, and then you gain in response time.

而现在，假设我们受到能源以及这些数据中心可用空间总量的瓶颈限制。那么，你认为半导体设计上会有突破，使这些设备在未来十年内变得更强大吗？

And right now, let's say we're bottlenecked by energy and just the sheer amount of space we have for these data centers. So, do you think there is a breakthrough in semiconductor design that will enable these devices to be more capable, let's say, over the next decade?

我认为是的。甚至不用等到下一个十年。我认为就在未来几年内。我们一直在做的一些事情，实际上现在是我们与 Humane 合作的一部分，就是如何改变数据中心的架构，进一步分解任务，然后为那个任务创建完美的计算。在某种程度上，我想稍微与手机做个类比，只是为了提供一个例子。你知道，当你开始设计手机时，早期的一些计算范式。你知道，在 PC 中，例如，你有一个 CPU 做所有事情。所以，你需要的每个计算都发送给 CPU。但是，CPU 对某些事情效率不高。所以，在手机中，出于必要，你开始分解计算。如果你想处理视频，你有一个视频引擎。你要处理图像，你有一个图像信号处理器。如果你要处理信号，你有一个信号处理器。你开始分解成我们所谓的异构计算，因为这是你能以最有效的方式完成这件事的方法。我们在数据中心看到了这种情况。你开始分解，你开始有专门的集群做不同的事情。我认为这就是我们正在做的事情。进一步分解，从移动端吸取一些经验，改变内存架构。所以，这将是效率的突破。我认为这本质上就是我们在扩展数据中心时正在做的一些事情。

I think so. Not even I don't have to go to the next decade. I think just within the next few years. Some of the things we have been doing and actually is now part of the partnership that we have with Humane is how can we change the architecture of the data center, do further disaggregation of tasks, and then create the perfect computing for that task. To some extent, I want to do a parallel a little bit with phones just to provide an example. You know, when you start designing a phone, early on some of the paradigms of computing. You know, things in a PC, for example, you have a CPU that does everything. So, every computation you need that you send it to the CPU. But, the CPU is not efficient for certain things. So, what you've started to do in a phone, out of necessity, you've started to disaggregate the computation. If you want to process video, you have a video engine. You're going to process image, you have an image signal processor. If you're going to process signals, you have a signals processor. And you started to break down in what we call heterogeneous computing because it's way that you can get this to done in the most efficient way. And we saw that happen in the data center. You started to disaggregate and you started to have special cluster doing different things. And I think that's kind of what we're doing. Further disaggregating, taking some of the learning from mobile, changing memory architectures. So, that's going to be a breakthrough in efficiency. And I think that's in the essence some of the things we're doing in our expansion into the data center.

有趣。当我们更普遍地看待设备时，你知道，显然半导体构成了设备，但设备本身，你认为会随着时间改变吗？所以，不仅仅是使用手机，你知道，可穿戴技术今天被越来越多地使用。

Interesting. And when we look at devices in more in general, so, you know, there's obviously the semiconductors build up the devices, but the devices themselves, do you believe will change over time? So, it's not just going to be using phones, you know, wearable technology is becoming a lot more used today.

这是个好问题。实际上，我喜欢这个问题，因为我们经常被问到移动的未来是什么？比如，手机会像从功能手机到智能手机那样改变吗？我认为设备的变化与用户界面有很大关系。也许我要稍微回忆一下，对吧？我认为当个人电脑起步时，比如你的第一台 PC，我们用键盘与计算机交互。我们今天仍然使用它。然后操作系统和应用程序都是为此设计的。然后发生了一些事情，鼠标出现了，突然计算机改变了，显示器改变了，操作系统改变了，应用程序改变了。我们现在有了图形用户界面。然后 PC，我们使用 PC，但 PC 现在可以放在手掌中，你可以随身携带。那就是智能手机，它是你可以触摸的屏幕。一切都围绕它改变了。所以，应用程序是为触摸设计的，我认为这就是我们现在所处的位置。所以，现在我们进入用户界面的下一个重大变化。那些 AI 模型，它们能看到我们所看到的，能听到我们所听到的。所以，现在什么类型的设备变得更自然？我们现在开始看到的是，如果你有 AI 智能体陪伴在你身边，实时对你所说的话、你听到的话、你看到的东西做出反应。它需要更接近我们的感官。一个例子是眼镜。眼镜靠近我们的眼睛，我们看到东西的地方，摄像头朝同一方向，靠近我们的嘴巴、耳朵。然后我们开始看到这种情况发生，正如你所说，可穿戴设备。可穿戴设备正从仅仅扩展手机功能的设备转变为与 AI 智能体连接的方式。我认为这将是下一次移动革命。在这一点上，我们现在确信我们将继续使用手机，就像我们继续使用笔记本电脑一样，但我们现在将拥有一个可穿戴设备，随着新用例的开发而变得越来越好。我认为那将是我们的下一个移动平台。

That's a great question. Actually, I love this question because we often get asked about what is the future of mobile? Like, is the phone going to change like it changed, for example, from the feature phone to the smartphone? And I think changes in devices have a lot to do with the user interface. And maybe I'm just going to go a little bit into memory lane, right? I think when personal computer started like your first PC, we interact with the computer with the ask to keyboard. We still use that today. And then the OS and applications were all designed for that. Then something happened, the mouse happened, and all of sudden the computer changed, display changed, the OS changed, applications changed. We have now a graphical user interface. Then the PC we're using the PC, but the PC now could fit in the palm of your hand, and you could carry around with you. That was the smartphone, and it was the screen that you can touch. And everything changed around that. So, the applications was designed for touch, and I think that's where we are right now. So, now we go to the next big change of the user interface. Those AI models, they can see what we see, they can hear what we hear. So, now what is the type of device that becomes more natural for that? And what we're starting to see right now is if you have AI agents that is going to be alongside you kind of reacting in real time to the things that you say, the things that you hear, the things that you see. It needs to be closer to our senses. So, one example of that is glasses. Glasses is close to our eyes where we see thing, the camera goes in the same directions, close it to our mouth, to our ears. And then we're starting to see that happening, as you said, with wearables. Wearables are moving from devices that were just extending the functionality of the phone into ways that we connect with the AI agent. And I think that's going to be the next mobile revolution. I at this point now we have conviction that we're going to continue to use our phones like we're continuing to use our laptops, but we're now going to have a wearable device that gets better and better as new use cases are started to be developed. And I think that's going to be our next mobile platform.

有趣。所以我们与技术交互的方式将减少触摸和点击，更多是看到和与之交谈。

Interesting. So, the way we interact with technology will be less touching and clicking and more seeing and talking to it.

绝对。就像我现在与你互动的方式一样。

Absolutely. In the same way that I'm interacting with you right now.

是的。是的。那会很好。

Yeah. Yeah. That'll be good.

我的意思是，未来我们可能也会变成 AI 采访者，不过继续聊未来这个话题——高通显然以 3G、4G、5G 技术闻名。当你思考这项技术的未来时，有没有什么比“稍微快一点的 5G”更激进的东西让你感到兴奋？

I mean, in the future we'll probably be an AI interviewer as well, but continuing this discussion about the future — Qualcomm obviously is known for 3G, 4G, 5G technology. When you think about the future of that, is there something more radical than just a slightly faster version of 5G that gets you excited?

哦，当然。我们现在就在研发 6G。而且我认为 6G 和我们刚才聊的一切都密切相关。每一代无线技术都有一个要解决的问题。2G 是：我们能不能创造一种技术，让地球上的每个人都能拥有一部随身携带的电话？那就是 2G。3G 是把那部数字无线电话连接到互联网。4G 是带来宽带，把它变成一台计算机。而 5G 是让任何人在手机里都能拥有无限的数据速率、无限的数据服务，就像我们在 Wi-Fi 上那样，还有关键任务型服务。那么 6G 的角色是什么？6G 的角色是：为 AI 的世界提供蜂窝连接。所以除了提供高带宽、更高速度、更低时延的连接之外，网络还将成为一个感知网络。网络将感知我们周围的一切，因为就像计算机现在能理解我们所见、所听一样，网络也将能够做到这一点。而这些传感器数据将被用来训练关于我们物理世界的模型，并持续演进 AI。

Oh, yeah. We're working on 6G right now. And I think 6G has a lot to do with everything we just talked about. Every generation of wireless had a problem to be solved. So 2G was about: can we make sure we create a technology that allows every person on Earth to have a telephone that you can carry with you? That was 2G. 3G was about connecting that digital wireless phone to the internet. 4G was about bringing broadband and making that a computer. And 5G was the ability to provide anybody the ability in a cell phone to have unlimited data rate, unlimited data service like we had it on Wi-Fi, and mission-critical services. So what is the role of 6G? The role of 6G is: what is the cellular connectivity for the world of AI? So in addition to providing high bandwidth, higher speed, lower latency connectivity, the network is going to be a sensing network. The network will sense everything that is around us, because the same way that computers now understand what we see, what we hear, the network is also going to be able to do that. And that sensor data is going to be used to train models on our physical world and continue to evolve AI.

那这在实际中意味着什么？你现在谈的有点像是多模态模型，而不只是大语言模型。这会如何改变我们的日常生活？

And what does that mean practically? So you're talking a little bit about multimodal models now as opposed to just large language models. How will that change everyday life for us?

让我给你举一个 6G 产品的例子。我们刚才在聊，比如智能眼镜。一款 6G 产品——6G 智能眼镜——会是一副首先通过蜂窝网络连接的眼镜。你将连接到蜂窝网络。你周围的上下文将被理解。网络将拥有你周围的所有上下文，当智能体与你互动、处理你的日常事务时，它就能理解这些上下文。我认为这将是巨大的区别。它不只是关于连接，而是关于你周围发生的一切、你的环境、你身边的人等等。不只是摄像头能看到的东西，网络将看到一切。

Let me give you an example of what will be a 6G product. We're just talking about, for example, smart glasses. A 6G product — a 6G smart glasses — will be a glass that is first connected through cellular. You're going to be connected to cellular. And the context around you is going to be understood. The network will have all the context around you, and it's going to be understood by the agent as it interacts with you on your everyday things. And I think that's going to be the big difference. It's not going to be just about connectivity, but everything that is happening around you, your surroundings, the people around you, and so on. Not only what the camera will see, but the network will see everything.

那么，在技术路线图之外，高通想用 AI 解锁怎样的人类体验？

And so, beyond the technical roadmaps, what is the human experience Qualcomm is trying to unlock with AI?

这是个有趣的问题。我想第一个是——从高通的角度来看，尤其是因为我们是一家身处这么多不同行业的公司，而这些行业正是人所在的地方，对吧？所以你可以说，有时候数据中心是去往能源和基础设施所在的地方。但我们用 PC、用手机、用汽车所做的大部分事情，都是在人所在的地方。我认为 AI 的一个作用是：我们能否让人的潜能自动化？比如智能手机——在高通，我们非常自豪的一件事是它让互联网民主化了。如果你看很多国家，比如巴西、印度，互联网实际上是以移动互联网的形式发展起来的。它真正弥合了数字鸿沟，让人们同时获得了计算和互联网的接入。我认为 AI 在这些个人 AI 设备和智能体式体验中也有潜力，在信息和知识上做同样的事。它能够真正让知识民主化，让做新事情的能力民主化，而且它很可能会增强人的潜能。

It's an interesting question. I think the first one is — from a Qualcomm perspective, and especially because we're a company that is in so many different industries where the humans are, right? So you can say sometimes the data center is going where the energy and the infrastructure is. But most of what we do with PCs, with phones, with cars is where the human is. And I think one role of AI is: can we automate the human potential? For example, the smartphone — one of the things we're very proud of at Qualcomm is that it democratized the internet. If you look at many countries, if you look at countries like Brazil, like India, the internet actually developed as a mobile internet. It was the ability to actually bridge the digital divide and get people access to both computing and the internet. And I think there's a potential for AI in those personal AI devices and those agentic experiences to do the same thing with information and knowledge. It's the ability to actually democratize knowledge, democratize the ability to do new things, and it will probably augment human potential.

但不止于此——那是消费端。在企业端，我相信企业随着用 AI 进行数字化转型，可以发生根本性的改变，尤其是在边缘侧，因为过去几十年企业把大量数据搬到了云端，但现在它们所有的资产都可以变得智能、可以互联，我认为这很可能就是企业数字化转型的下一波浪潮。

But it's not only that — that's on the consumer side. On the enterprise side, I believe that companies can fundamentally change as they digitally transform with AI, especially on the edge, because companies have spent the past few decades moving a lot of the data to the cloud, but now all of their assets can be intelligent, can be connected, and I think that will probably be the next wave of digital transformation of enterprises.

对。所以一切互联、人人互联，你就不再会遇到现在用 AI 时的这个问题——你得先给它大量上下文，才能让它做某件事，而这多少浪费了时间。

Right. So everything is connected, everyone is connected, and you don't have this issue that you have right now with AI where you have to give it a ton of context before you can tell it to do something, which sort of takes away the time of it.

这个说法很好——我觉得比我自己的描述更好——说明了 5G 设备和 6G 设备的区别。它将拥有上下文，而不需要你去提供。

And that's a great way — better than I described myself, I think — of saying what's going to be the difference between a 5G and a 6G device. It's going to have the context without you having to provide it.

好。那么当我们思考 AI 如何改变的不只是技术、而可能是人类的轨迹时，你希望高通扮演什么角色，你希望公司留下什么遗产？

Okay. So when we think about how AI is shifting the trajectory of not just technology, but potentially humanity, what role do you hope Qualcomm will have, and what legacy do you want the company to have?

这是个很深的问题，但我想大概会是我们做无线技术时留下的同一种遗产。我有幸在高通工作了 30 年。我 1995 年加入高通，那时第一代 CDMA 还没出现，我想我们可以说，我们通过让每个人都能连接，在真正改变这个社会上出了一小份力，靠的就是我们在无线技术上所做的一切。我们希望能在 AI 上做同样的事，并且正如我所说，继续赋能于人。

That's a very deep question, but I think probably the same legacy that we had with everything we've done in wireless. I've had the privilege of being at Qualcomm for 30 years. I joined Qualcomm back in 1995, before the first CDMA, and I think we can say that we had a small part in actually changing this society by enabling everyone to be connected, with what we have done with wireless technology. We hope to be able to do the same with AI and, as I said, continue to empower people.

当然，从 1995 年以来，你们也变成了一家全球化程度高得多的公司。那么你如何排定优先级、如何推进，既保持战略一致，又进入所有这些新市场，并和沙特阿拉伯这样的国家、以及亚洲许多其他国家建立合作？

And of course, since '95 you guys have become a much more global company as well. So how do you prioritize and work through keeping aligned on strategy, as well as entering all of these new markets and partnerships with countries like Saudi Arabia, but also so many others in Asia?

我想这是目前为止最难的问题，因为公司确实正在经历一个转型过程。我们从一家以移动为中心的公司，扩展到汽车，扩展到个人计算、工业、个人 AI、机器人，再到数据中心——而且我们是同时在做所有这些事。我觉得我们有点着急。我们看到了一个绝佳的机会去扩展我们的技术，在这么多行业中都占据一席之地。而排优先级很难，因为我们对所有这些机会都同样兴奋。所以我们就继续往前推进。

That's the most difficult question I think so far, because the company is really going through a process of transformation. We went from a mobile-centric company into being automotive, into personal computing, and industrial, and personal AI, and robotics, into the data center — and we're doing all of it at the same time. I think we are a bit in a hurry. I think we see an incredible opportunity to expand our technology and be relevant in so many industries. And it's hard to prioritize because we're equally excited about all those opportunities. So we're just going to keep going.

我觉得这说得通。我想现在很多人在 AI 的应用层会讨论：当软件——现在任何人都能开发软件——你该如何为自己建立护城河。

I think that makes sense. I think a lot of people now, for AI, the app layer, they talk about how do you build a moat for yourself when software — anyone can build software now.

高通有那么多优势——它全部的历史、所有投入其中的研究和知识产权——以至于无论你转向哪个方向，你们天然就已经有了一定的先发优势，尽可能多元化是合理的。

And Qualcomm has so much of an advantage with all of its history and all of the research and IPs that have gone into the work they've done so far that any corner you pivot into naturally, you guys already have sort of a head start, and it makes sense to go diversify as much as possible.

是的，我同意。当我们和一些半导体同行比较时，我认为高通的独特之处在于，我们同时在做的芯片有的不到 5 瓦，有些情况下只有 1 瓦——用在眼镜或耳机、智能耳机里的芯片——一直到用在数据中心里的 500 瓦芯片。所以我认为覆盖范围非常广。另外一点是，我们大概有能力出现在这么多行业里，而且我们理解所有这些不同行业的痛点。所以我觉得，继续同时做所有这些事情是非常有诱惑力的。

No, I agree. I feel when we compare ourselves with some of our semiconductor peers, I think what makes Qualcomm unique is that at the same time we're doing chips that are less than 5 W in some cases, like 1 W, the chips that go into a glass or an earbud or a smart earbud, all the way to a 500 W chip that goes into the data center. So I think it's a very wide range. And the other thing is we have probably the ability to be present in so many industries, and we understand the pain points of all those different industries. And then so I think it's very tempting to continue to do all those things at the same time.

你如何思考带领一家公司完成这样的转型？你当 CEO 已经 5 年了？

How do you think about leading a company through such a transformation? So you've been CEO for 5 years now?

是的，从 2021 年开始。

Yes, since 2021.

这场转型进展如何？

How has this transformation been?

这么说吧，我很有偏向性。我认为我们正在一段旅程上。我觉得我们只是迈出了旅程的最初几步。这段旅程远未完成——把高通转型成一家不仅因我们在无线领域的成就而被认可，也因我们能在计算和 AI 领域做到的事而被认可的公司。而且正如我所说，要与我们相信技术能帮助转型的众多行业保持相关性。这非常紧张，因为如你所说，我们在并行做很多事情，但同时也令人兴奋。我们挺高兴的是，从技术角度看，即便是那些非常新、我们还在建立规模的行业，我们实际上也建立起了领先的技术地位。所以看起来我们走在正确的道路上，只需要继续走下去。

Look, I'm very partial. I think we're on a journey. I think we just took initial steps on a journey. The journey is far from being completed of transforming Qualcomm into a company that was recognized for what we have done in wireless, but also recognized for what we can do in computing and AI. And as I said, be relevant to many industries that we believe our technology can help transform. And it's very intense, I think, because as you said, we're doing a lot of things in parallel, but it's exciting at the same time. And we're kind of happy that from a technology standpoint, even industries that were very new and where we're still building scale, we actually were able to build a leading technology position. So it seems like we're on the right path and we just need to keep going.

你觉得为什么会这样？你在高通的成功公式是什么？

Why do you think that is? So what is your formula for success at Qualcomm?

我认为这个公式是高通独有的一点：我们是一家非常、非常以技术和工程为中心的公司。

I think the formula is one thing that is unique to Qualcomm. We're a very, very technology- and engineering-centric company.

对，你是工程师出身。

Right, you're an engineer by training.

是的，我是。举个例子，看看美国——2025 年还没过完，但在 2024 年，在所有美国公司中，所有规模的公司，甚至比高通更大的公司，我们在专利申请数量上是第一名。我认为这种对创新的热情、对技术和工程真理的追求，大概正是帮助高通发展出新的核心能力、新的本领，并成功进入新行业的原因。如果你问我，作为这家公司的领导者，最重要的事情是什么，那就是确保我们永远不丢掉这种文化，继续在让我们迄今取得成功的这种文化上建设。

Yes, I am. For example, just look in the United States — we're not done with 2025 yet, but in 2024, of all the American companies, all of them, all sizes, even bigger than Qualcomm, we've been the number one company in patent applications. I think this passion for innovation, this for technology and engineering truth is probably what has helped Qualcomm develop new core competencies, new capabilities, and enter new industries successfully. And I think if you ask me the question, what's the most important thing for you as you lead this company, it's make sure we never lose this culture, that we continue to build on this culture that made us successful so far.

你如何看待硬件和软件的互动？对你来说，理想的协作是什么样的？

And how do you think about the interaction of hardware and software? So what is that ideal collaboration like for you?

这是个好问题。尽管我们是一家芯片公司，我们生产硬件——但今天我们超过一半的工程工作是软件。我认为真正把东西作为一个系统来开发的能力极其重要。有时我们甚至对自己说，我们不是芯片公司，我们是系统公司。我认为同时在硬件和软件上演进的能力极其重要。而当我们进入新行业时，软件一直是开发一些新能力的重要领域。举个例子，作为汽车领域的硬件公司，我们上个月刚在慕尼黑的 IAA 车展上发布了我们的辅助驾驶和自动驾驶软件栈。这是高通的一个纯软件产品，运行在我们的辅助驾驶硬件上，首发于宝马 iX3。我在公司内部说，看，我们应该非常自豪，我们有了一个纯软件产品，建立在这些能力之上，这样我们就能在硬件创新的同时做大量软件创新。

That's a great question. Even though we're a chip company, we produce hardware — more than half of our engineering is software today. I think the ability to actually develop things as a system is incredibly important. Sometimes we even say to ourselves that we're not a chip company, we're a system company. And I think that ability to actually evolve on those hardware and software at the same time is incredibly important. And as we enter into new industries, software has been a very important area of actually developing some new capabilities. If I give you an example, as a hardware company in automotive, we just launched last month at the IAA in Munich our assisted driving and autonomous stack. It's just a software product from Qualcomm that runs on our hardware for assisted driving, and it was launched in the BMW iX3. And internally in the company I said, look, we should be very proud we have a software-only product that we have built on top of the capabilities, so that we can do a lot of innovation in software alongside innovation in hardware.

我觉得这真的很令人兴奋，而且我认为 AI 实际上也真正促成了硬件和软件之间的这种协作，因为硬件的迭代——现在很多都可以通过模型来完成，而不必自己动手去试。所以迭代时间快得多。我也说不好，我只是觉得硬件侧会有很多令人兴奋的创新正以这种方式到来。

I think that's really exciting, and I think AI actually also really helps enable this collaboration between hardware and software, because the iteration on hardware — so much of it can now be done through models instead of having to work and try things out yourself. So the iteration time is so much faster. And I don't know, I just think there's a lot of exciting innovation on the hardware side coming that way.

完全同意。而且你看，已经有好几年了，我们很多芯片或硬件——我们一直在用 AI 设计这些芯片。不仅在芯片设计上，也在芯片如何运行上。比如你可能想不到，我们连接智能手机和其他联网设备与无线电波的 5G 调制解调器，上面有一个 AI 引擎。因为有了这个 AI 引擎，我们实际上改善了设备的覆盖范围，因为即使有时信号很弱，AI 也能帮助推理，某种程度上恢复并预测信号到底是什么。即使在弱信号下，你实际上也能有很好的接收效果，能取回信息。

Absolutely. And look, it's been now a number of years that a lot of our chips or hardware — we've been using AI in designing those chips. And not only in the design of our chips, but also in how the chips operate. For example, you would not think, but our 5G modem that connects smartphones and other connected devices to the radio waves, it has an AI engine on it. And because of the AI engine, we actually improve the coverage area of devices, because even though sometimes you get a very weak signal, the AI helps with inference, kind of recover and predict what the signal really is. And even in a weak signal, you actually have a very good reception and can retrieve the information.

这真是个好问题，我有点无意中问出来的——你们是一家为 AI 设备制造硬件的公司，但你们自己又是怎么用 AI 来帮助制造这些硬件的？这有点像循环引用，但我觉得很高兴看到它在两端都有益。

That's a really good question that I kind of asked without meaning to ask — like, you are a company that makes hardware for AI devices, but then how do you yourselves use AI to help you make those? So it's sort of like a circular reference, but I think it's great to see that it's being beneficial on both ends.

完全同意。你知道，我喜欢讲一个故事——大多数人不会知道我要说的这件事。这大概会暴露我的年纪，但在高通早期，那是互联网的早期、电子邮件的早期，我们需要一种技术来帮助我们处理内部沟通。所以我们最终开发出了——它曾是公司的一部分，我们把它作为产品提供——Eudora，它在互联网早期某个时候曾是世界上最流行的电子邮件客户端。我们开发它是因为我们有内部需求。就像今天，我们自己开发了一些叫 Q Genie 的 AI 工具，实际上帮助我们的工程师用 AI 开发产品。我们今天在一些团队中已经看到 AI 带来的 10%、20% 的效率提升。这是我们吃自己的狗粮的一个绝佳例子。

Absolutely. You know, I like to tell the story — most people will not know this that I'm going to say. And it kind of is probably going to show my age, but in the early days of Qualcomm, it was the early days of the internet, the early days of email, and we needed to come up with a technology that helped us deal with internal communications. So we ended up actually developing — and it was part of the company and we offered as a product — Eudora, which was at some point the most popular email client in the world in the early days of the internet. And it was something that we developed because we had an internal necessity. Like today, we have developed a number of AI tools ourselves called Q Genie that actually help our engineers develop our products with AI. And we have seen gains today in some groups of 10, 20% efficiency enabled by AI. And this is a great example of us eating our own dog food.

是的，太棒了。

Yeah, that's great.

尤其是因为今天有太多争论围绕着一个问题：好吧，人们落地 AI 的成果在哪里？你已经在内部有了这些成果，也证明了这件事。嗯，最后我想问几个快速的快问快答。其中一个问题是，在这个 AI 时代，你认为是什么让人类独一无二？还有什么对我们来说依然特别？

I mean, especially because there's so much debate today around, okay, where are the results from people implementing AI? You already have them in house and it proves the story. Um, and to wrap up, I think I'd love to ask some quick rapid-fire questions. One of them being, you know, in this age of AI, what do you think makes humans unique? What remains special to us?

你看，我个人有这样的观点：AI 最特别的地方其实在于人。是帮助创造它的人，是提供训练数据的人。所以人现在是、将来也会继续处在中心。我们其实不把 AI 看作某种会比人做得更好的东西，而是它帮助人更好地去做人最擅长的事。我认为我属于那种相信很难取代人类的人，而且实际上正是人会让 AI 变得有用、让 AI 大放异彩。

Look, I personally have this view: the most especial thing about AI is actually the human. It's the human that helped create it, the human that provided the data for training. So the human is and will continue to be at the center. And we actually think about AI not as something that is going to do something better than humans, but it actually helps humans do what they do best in a better way. And I think I am one of those people that believe it's going to be pretty difficult to replace humans, and actually humans are the ones that are going to make AI useful and make AI shine.

太棒了。反过来，有什么问题是你还在思考如何攻克的？什么让你夜不能寐？

Amazing. And what conversely is a problem that you're still thinking about how to crack? What keeps you up at night?

有很多事情，但我觉得一个问题是我们如何继续提升算力，并且能以非常高效的方式做到。因为我很清楚，对计算的需求会在各处持续增长。你看半导体，我们现在在建 2 纳米芯片，之后还要走向 2 纳米以下。我们如何把更多性能塞进硅里，并确保这些硅能用最少的功耗？因为我认为 AI 正在带来所需算力数量级的跃迁。比如我们现在就在想，如何把多个数百亿参数的模型跑在一部小小的手机上，然后我们还要试着把它做到一副很小的眼镜框里。我们必须解决这些问题，而且我们多少觉得，如果我们不解决这些问题，没有别的公司会去解决，所以这是我们要解决的问题。

There are many things, but I think one problem is how we continue to increase compute in a way that we can do it very efficiently, because I think it's very clear to us that the need for computing is going to continue to grow everywhere. I think if you look in semis, we're building chips now at two nanometer, and then we're going to go beyond two nanometer. And how do we pack more performance into silicon and make sure that silicon can use the minimum amount of power? Because I think AI is creating a shift into the order of magnitude that computing is required. Like we're thinking right now how we can run multiple tens of billion parameter models into a tiny phone, and then we're going to try to do that into a very small frame of a glass. And we have to solve those problems, and we kind of feel if we don't solve those problems no other company will, so it's our problem to solve.

那肯定也会让我夜不能寐。最后，作为结尾，你对进入职场的年轻人，或者那些想了解该如何为这个快速变化的市场中的职业生涯做好准备的人，有什么建议？

That would also keep me up at night for sure. Then finally, to end, on advice you have for young people entering the workforce, or people looking to understand how they should best prepare for a career in this rapidly changing market.

你看，我会把注意力放在这个快速变化的收入市场上。我认为现在的潜力和机会大概和我起步时非常不同。我起步时，可以说有幸在无线革命和手机刚刚兴起的时候进入职场——那是一件了不起的事。我认为今天有了 AI，各种可能性都不可思议，每个领域的创新都在以非常快的节奏推进。所以我觉得这是那种你可以说“如果你有一个好想法，AI 会帮你把它落地”的时代之一。我认为这大概是创造力和创新的绝佳时机。所以，有一个好想法，然后去把它做出来。

Look, I would just focus on this rapidly changing market of income. And I think the potential and the opportunities right now are probably very different than when I started. When I started, I had the privilege, I think, of going to the workforce at the time the wireless revolution and cell phones were just starting — it was an incredible thing. I think today with AI the possibilities are incredible, innovation at a very fast pace in every area. So I think this is one of those times that you can say: if you have a good idea, AI will help you put it to work. And I think that's probably an incredible time for creativity and innovation. So have a good idea and go put it into work.

是的。好，太棒了。非常感谢你，Cristiano，感谢你抽出时间，真的很感激。

Yes. All right, amazing. Well, thank you so much Cristiano for your time, really appreciate it.

谢谢。谢谢。我真的很享受这次对话。

Thank you. Thank you. I really enjoyed having this conversation.