高通 CEO 谈即将到来的 AI 手机超级周期

克里斯蒂亚诺，你执掌高通，你们的技术驱动着大量 Android 手机、Meta 的智能眼镜、一批即将到来的可穿戴设备，以及越来越多的汽车。这些我都想聊。但我想先从另一件事说起。你对美国和中国有一个很有意思的看法。高通的骁龙品牌在中国非常强。来到这场峰会之后，我才真正体会到中国的骁龙粉丝有多狂热，我觉得这让你对两个市场有一扇独特的窗口。中国人对 AI 的感受，和美国相比怎么样？你看全球各地对 AI 的民意调查，地域差异非常大。你能不能把两边对照着讲一讲？

So Cristiano, you run Qualcomm, and your technology powers many Android phones, Meta smart glasses, a bunch of coming wearables, and a growing number of cars. I want to get into all of that. But I want to start somewhere else. You have a very interesting view of the United States and China. Qualcomm has a very strong Snapdragon brand in China. Being here at the summit, I have learned just how devoted the Snapdragon fan base is in China, and I think that gives you a unique window into the two markets. How are people in China feeling about AI relative to the United States? When you look at how sentiment on AI polls around the world, it is very different by geography. Could you contrast the two?

如果我做对照，AI 进入设备、进入消费者的大规模采用，在中国发生得非常非常快。在美国，主导对话的是前沿模型的开发，以及企业侧对 AI 的采用。两边谈话的内容，差别就在这里。

If I compare and contrast, the broad adoption of AI into devices and among consumers is happening in China very, very fast. In the United States, what dominates the conversation is the development of frontier models and the enterprise adoption of AI. That is the difference in the conversation between the two.

你跟中国人谈 AI 的时候，会不会觉得他们想象的未来和美国人想的不太一样？美国的民意，你去看调查，负面得厉害。

When you talk to people in China about AI, does it feel as if they are imagining a different future than people in America? The sentiment in America, when you look at the polling, is incredibly negative.

每一种新技术都会带来不确定性。你回头看以前的技术变迁，会看到一些同样的担忧。我对美国的看法是，AI 已经在不少行业里成为重要的生产力工具。高通就是一个例子。今天在另一场对话里有人问我：你们在改造公司，这么多事情并行推进，资本怎么分得过来？你去看，我们在扩张公司的同时，运营费用的规模相对是平的。其中很大一部分，是开发过程中用 AI 换来的生产力。所以你清清楚楚看到 AI 带来生产力，同时也当然会担心这会不会冲击就业。公司也许会靠 AI，不再需要那么多人力。这场对话大概是全球性的，我不认为它是美国独有的。

Every new technology brings uncertainty. If you go back through earlier technology transitions, you see some of the same concerns. My perspective on the United States is that AI has already been an important productivity tool for a number of sectors. Qualcomm is an example. I was asked today, in a different conversation: you are transforming the company and doing a lot of things in parallel, so how can you allocate capital to all of them at once? If you look, we have been expanding the company while keeping operating expenses on a relatively flat scale. A great part of that is productivity gained by using AI through the development process. So you clearly have AI as a source of productivity, and there are obviously concerns about whether that is going to impact jobs. Companies may lean on AI and not need as much manpower. That conversation is probably universal. I do not think it is unique to the United States.

另外还有关于数据中心的对话。数据中心的规模极大。你看着那些部署，对算力的需求一直在涨。我觉得算力需求面前有一堵墙。然后人们会担心：这会不会让我的电费上涨？这些对话在我看来都很自然。电费这种担忧，我在中国没看到。在那边，我更多看到的态度是：这是一项功能，是一次演进。两个国家看这件事的角度有点不一样。比如中国一直非常聚焦物理 AI，以及边缘上的一切。眼下中国在这方面的关注更大。但总体而言，因为 AI 带着不确定性，你会有很多问题。坦率说，我们还在 AI 的早期，非常早。我喜欢一个比喻。这就像互联网早期，如果有人说：我们已经知道结果了。AltaVista 会赢下搜索，MapQuest 会赢下地图，Orkut 会赢下社交。不是的。AI 还有很多事情要发生，其中一些问题会被回答。

You also have the conversation about data centers. Data centers are massive. You look at those deployments and the demand for compute just keeps rising. I think there is a wall of demand for compute. And then you have the concern: is this going to increase my electricity bill? Those conversations feel natural to me. The electricity-bill worry is one I do not see in China. There, I see more of an attitude that this is a feature, an evolution. The two countries are looking at this a little differently. China, for example, has been really focused on physical AI, and on everything at the edge. That focus is bigger in China right now. But in general, because of the uncertainty around AI, you are going to have a lot of questions. To be frank, we are in the early days of AI. Very early. I like this metaphor. It is like the early days of the internet, if someone had said: we already know what is going to happen. AltaVista is going to win search, MapQuest is going to win maps, and Orkut is going to win social media. No. A lot still has to happen in AI, and some of those questions will get answered.

本期节目由 Mercury 赞助。这是包括我在内、超过 30 万名创业者喜爱的 AI 原生银行服务。访问 mercury.com 了解更多。Mercury 是金融科技公司，不是银行。详情见节目说明。节目也由 Atlassian 的 Jira 赞助。团队和智能体在那里获得上下文、协同和控制，把工作往前推。在 jira.com 免费试用。还要感谢 Granola，给一场接一场开会的人用的 AI 记事本。它跟着你的工作走，让你把注意力放在真正重要的事上。到 granola.ai/sources 试用，用优惠码 SOURCES 可免三个月费用。

This episode is brought to you by Mercury, AI-native banking loved by more than 300,000 entrepreneurs, including me. Visit mercury.com to learn more. Mercury is a fintech, not a bank. Check the show notes for details. It is also brought to you by Jira by Atlassian, where teams and agents get the context, coordination, and control to move work forward. Try it free at jira.com. Thanks also to Granola, the AI notepad for people in back-to-back meetings. It works everywhere you do and lets you focus on what matters. Try it at granola.ai/sources and use the code SOURCES for three months off.

眼下在美国，很多人在谈要给前沿 AI 的发展踩节奏。你能看到 Hugging Face 被攻破这类事情，模型能力在变强，网络风险在上升。我很想知道你站在哪一边。你和公司都在华盛顿出现过。你是美国政策讨论里的一个声音。你看这场辩论，现在怎么判断？

Right now in the United States, a lot of people are talking about pacing the frontier of AI development. You see things like the Hugging Face breach, models becoming more capable, and cyber risk going up. I am curious where you stand. You and the company have been in Washington. You are a voice in the US policy conversation. When you look at this debate, what do you make of it?

我免费把意见告诉你，所以它未必很好。而且这只是一个意见。这片地形在变，很难有把握。但我认为答案从来不在两端，答案在中间某个地方。首先，这是一项全新的技术，潜力很大。存在一条发展路线图，持续往这条路线图上投资会很重要。其次，有护栏同样重要。

I will give you my opinion for free, so it may not be very good. And it is just an opinion. This landscape is changing, and it is hard to be certain. But I think the answer is never at the bookends. The answer is somewhere in the middle. First, this is a brand-new technology and it has a lot of potential. There is a development roadmap, and continuing to invest in that roadmap is going to matter. Second, it is also important to have guardrails.

是的，关于 AI 风险的评论很多。我讲几个更实际的。网络安全是一个非常说得通的风险，因为一切都数字化了。AI 之所以存在，是因为我们人类把一切都数字化了。数字信息到处都是，算力也到处都是。但跑在软件上的攻击面非常非常大。如果模型能找到漏洞、攻进系统，你就不再需要一支专门的网络战队伍。任何人拿到一个模型都能做。这是个问题。你需要控制。你需要严肃的公司。你需要恰当的监管。你需要硬的防护带。这和任何同类的重大技术没有什么不同。所以答案不在极端。不是停下来，也不是全速前进。是负责任地做。这没有什么不对。

Yes, there is a lot of comment about the risks of AI. Let me talk about some of the more practical ones. Cybersecurity is a very reasonable risk, because everything is digital. The reason AI exists is that we humans digitized everything. You have all this digital information, and computing power is everywhere. But the surface area that runs on software is very, very large. If models can find vulnerabilities and hack systems, you no longer need a specialized cyber-warfare team. Anybody can get a model and do it. That is a problem. You need control. You need serious companies. You need the right regulation. You need hard guard bands. This is no different from any other major technology of this kind. So the answer is not in the extremes. It is not stop, and it is not full speed. It is doing it responsibly. There is nothing wrong with that.

在我们这个小世界里，我们做芯片。

In our little world, we make chips.

小世界。

Little world.

我们做那些用来算数学的芯片。

We make the chips that do the math.

我不觉得芯片很小，不过没关系。

I do not think chips are little, but that is okay.

跟其他一些大公司比，它是小的。我们做这些计算，同时在做硬件上的钩子、控制和旋钮，让你能安全地存放信息，让你决定个人图谱要不要被存下来，让你手里有控制权。我觉得这很重要。那些公司都需要负责任地做这件事。

It is little compared with some of the other big companies. We do all of this math, and we are building the hardware hooks, controls, and knobs so you can store your information securely, so you can decide whether a personal graph gets stored or not, and so you can have controls. I think that matters. All of those companies need to do this responsibly.

你五月份随特朗普总统去了中国。在那些房间里，你从其他国家的领导人那里听到他们怎么面对 AI 和它的风险？

You traveled to China with President Trump in May. When you are in those rooms, what do you hear from other world leaders about how they are grappling with AI and its risks?

我所在的半导体行业，眼下已经变成一个地缘政治行业。你必须参与这些对话，而 AI 在每一段对话里都很重要。我觉得大家对如何负责任地做这些事有一种共同理解。另一部分，也就是我真正坐在里面的那些对话，是美国企业和中国企业之间有很多强而健康的商业伙伴关系。我所看到的，以及我希望下周还能再看到的，是这些关系在两国之间造出一个真正的稳定点和共同理解。如果这是我们能继续扮演的角色，那我们就会扮演这个角色。

The semiconductor industry I happen to be in has become a geopolitical industry. You have to be part of those conversations, and AI is important in every part of them. I think there is a common understanding about how to do these things responsibly. The other part, the conversations I actually sit in, is that there are a lot of strong, healthy commercial partnerships between American enterprises and Chinese enterprises. What I have seen, and what I hope I will see again next week, is that those relationships create a real point of stability and common understanding between the two countries. If that is a role we can keep playing, that is the role we are going to play.

我们来谈智能手机。这是一个巨大的市场，大概是有史以来最重要的计算市场之一。与此同时，它不再像从前那样增长。我看到一个数字，今年出货量可能收缩大约 14%，明年也许再降一点。这里面有多少是内存和元器件成本飙升、标价吓到人——手机现在已经够好，涨价你就继续用手里这部——又有多少是某种更大的变化？

Let's talk about the smartphone. It is a massive market, one of the most consequential computing markets ever. At the same time it is not growing the way it once did. I saw a figure that units shipped could shrink about 14 percent this year, and maybe a bit more next year. How much of that is the spike in memory and component costs, and sticker shock — the phones are good now, so if the price goes up you just keep the one you have — and how much of it is something bigger?

这个问题里的东西很多。

There is a lot in that question.

是的。

Yes.

我拆开讲，先给事实。人们也许还没意识到：智能手机市场到现在还没有恢复到疫情前的规模。它仍然比 2019 年小，而且眼下是被人为压小的。它一直在慢慢爬回来，原因有好几个。其中一个其实是好事。全世界的人都有智能手机了，所以市场成熟了。它变成一个换机率市场，这完全正常。我大概要暴露年龄了，我是从 1G 开始做的，每一轮周期我都经历过。

Let me break it down, and start with the facts. People may not have realized this yet: the smartphone market has still not recovered to what it was before the pandemic. It is still smaller than it was in 2019, and right now it is artificially smaller. It has been slowly climbing back, and there are several reasons. One of them is actually a good thing. Everybody in the world has a smartphone, so the market matured. It became a replacement-rate market, and that is completely normal. I am going to show my age, but I started working on 1G, so I have been through every cycle.

我听说我们快到 6 了。

I hear we are almost at 6.

对。一轮周期走到后段，市场的一部分会成熟，变成换机市场。即便如此，智能手机仍是一个大市场。每年大约卖出 12 亿部手机。就算你不想要新功能，你用手机的频率也很高，两三年、四五年之后你还是会再买一部。在这个基本盘里，有一件事非常稳定。高端和高档手机占整个市场的比例年复一年在扩大，像钟表一样准。即便市场被压缩，高端档仍在扩张，因为一代又一代，人们用手机做的事越来越多。

Yes. At the end of a cycle, part of the market reaches maturity and becomes a replacement market. Even so, the smartphone is a big market. About 1.2 billion phones are sold every year. Even if you do not want a new feature, you use the phone a lot, and after two, three, four, or five years you buy another one. Inside that base, something has been very consistent. The premium and high tiers keep growing as a share of the total market, year after year, like clockwork. Even in a compressed market, the premium tier keeps expanding, because each generation people do more with the phone.

我相信你还有一台笔记本。笔记本并没有被手机取代。以前有些事你会留给笔记本——一封很长的邮件，你会等到坐下再写。有了 AI，这类工作负载越来越多地挪到手机上。人们想要更好的手机，想要能力更强的手机。就算先不谈向智能体手机的转变，高端档已经在上升。

I am sure you still have a laptop. The laptop was not replaced by the phone. There used to be things you would save for the laptop — a long email, and you would wait until you sat down. With AI, more of those workloads are moving to the phone. People want better phones. People want more capable phones. Separate from the shift to an agentic phone, the premium tier is already rising.

下一层是，市场现在被内存的情况人为压住了。内存价格涨了 5 到 6 倍。高端和高档更有韧性。是的，手机更贵了，但消费者看得见价值，他们还是想买。中端和低端敏感得多。所以今年市场下降 20%，是因为内存价格高。这不是需求问题。现在有被积压的需求。一线希望在时机上。这件事发生的时候，AI 智能手机刚刚开始出现，对那些现在就想买一部 AI 手机的人，这可以是一个机会。

The next layer is that the market is now artificially suppressed by what happened to memory. Memory prices went up five to six times. The premium and high tiers have been more resilient. Yes, the phone is more expensive, but consumers see the value and they still want to buy it. The mid tier and the low tier have been much more sensitive. So the market is down 20 percent this year because of high memory prices. It is not a demand problem. There is now pent-up demand. The silver lining is the timing. This is happening just as AI smartphones are starting to arrive, and that can be an opportunity for people who want to buy an AI phone now.

我们见过这个市场按周期走。过去几年，正如你说的，手机没什么让人兴奋的。手机市场不增长。在 Android 市场，我们仍然以两位数的复合增速在增长，靠的就是高端档扩张、份额提升和更多内容，因为人们想要更好的手机。我的预测是，现在没人想谈手机市场。所有人都在谈数据中心和其他市场，直到 AI 超级周期到来。那一天到来，再加上 6G 到来，关于手机的谈话会重新变多。手机不会消失。手机是人类创造过的最大开发平台。人不会跑到一座数据中心，按门铃说：给我一点 AI。你是通过移动设备去用 AI 的，这些设备会在这幅图景里占一个位置。

We have seen this market move in cycles. For the past few years, as you said, there has not been much excitement about phones. The phone market does not grow. In the Android market we have still been growing at a double-digit compound rate, just from premium-tier expansion, share gains, and more content, because people want a better phone. My prediction is that nobody wants to talk about the phone market. Everyone is talking about data centers and other markets, until the AI supercycle. When that happens, and when 6G happens, there will be a lot of conversation about phones again. Phones are not going anywhere. The phone is the largest development platform ever created. Humans are not going to travel to a data center, ring the bell, and say, give me some AI. You are going to reach AI through mobile devices, and those devices are going to be part of the picture.

我想知道你说的 AI 智能手机到底是什么。但先说一件事：你们在主题演讲的屏幕上放了 Muse。它正在席卷硅谷的早期采用者，另外还有 Instinct、Grok Bot 这类产品。我想压力测试你说的“工作负载在往手机上搬”，因为这些产品是云里的浏览器虚拟机。Muse 在某处云端跑着一个浏览器，不在你的手机上。大部分 token 消耗发生在那边，答案再被送回来。如果这些早期的个人智能体至少眼下是这样工作的，那不就让手机变得没那么关键了吗？

I want to know what you mean by an AI smartphone. But first: you showed Muse on the screen during the keynote. It is taking the early-adopter world in Silicon Valley by storm, and there are products like Instinct, Grok Bot, and others. I want to stress-test the claim that workloads are moving to the phone, because these products are browser virtual machines in the cloud. Muse is running a browser somewhere in a cloud. It is not on your phone. Most of the token use happens there, and the answer gets beamed back. If that is how these early personal agents work, at least right now, does that not make the phone less crucial?

我得小心，不能点名。但眼下有一些 AI 公司，一些站在 AI 最前沿的公司，正在对高通说：到 2028 年，我需要有能力在手机上至少跑一个 1000 亿参数的模型，而且这东西得差不多一直开着。

I have to be careful not to name names. But some AI companies right now, some of the companies at the forefront of AI, are telling Qualcomm: by 2028, I need to be able to run at least a 100-billion-parameter model on a phone, and I need that thing running kind of all the time.

所以你是说，有 AI 公司在谈手机。

So you are saying AI companies are talking about phones.

AI 公司开始做手机了。就拿我们刚才说的中国的例子。豆包手机是字节跳动的手机，由努比亚制造，而且后面还有。大家开始明白，手机对智能体非常重要。会有一场智能体之争：你想让哪个智能体做什么。现在还早，但已经清楚，有些事必须发生在手机上，包括手机为其他设备提供支持的那些事。

AI companies are starting to make phones. Take the example from China we were just talking about. The Doubao phone is a ByteDance phone, made by Nubia, and there is more coming. There is an understanding that the phone matters a lot for agents. There is going to be a battle of the agents: which agent you want to do what. These are early days, but it is already clear that some things have to happen on the phone, including things the phone does in support of other devices.

我觉得把云和边缘说成互相取代，这场争论是徒劳的。越来越多的事会发生在云里。你会找到只能在云里做的复杂工作，然后就在云里做。也有必须在设备上做的事，因为它们只能在设备上做。一边竞争、一边取代另一边，这个想法是错的。拿出一部手机，文件夹里有 300 个应用，试着一个一个说哪个跑在手机上、哪个跑在云里。几乎不重要。可一旦手机有了大量处理能力——我知道每一代芯片都更强——你把它调成飞行模式，它就没用了。你会把它留在口袋里。我觉得 AI 也会这样。有些事会发生在手机上，因为它们必须发生在那里，正如有些事会发生在云里。

I think it is futile to argue cloud versus edge as if one replaces the other. More and more things will happen in the cloud. You will find sophisticated work that can only be done in the cloud, and you will do it there. And there will be things you have to do on the device, because they have to be done on the device. The idea that one competes with and replaces the other is incorrect. Pull up a phone with 300 apps in folders and try to say, one by one, what runs on the phone and what runs in the cloud. It almost does not matter. But the moment the phone has a lot of processing power — and I know each new chip has more — if you put it in airplane mode, it becomes useless. You leave it in your pocket. I think AI will be like that. Some things will happen on the phone because they have to, just as some things will happen in the cloud.

我反复回到的那个点是摩擦。如果你掏出手机，智能体替你做一件事要太久，你就会自己做。另一些事，你会交给智能体。而手机是你个人独有的。我在主题演讲里说过这一点。角落里也许有一个小应用，用来开车库门，或者查一件事。那个应用在你的手机上，智能体得过去操作它。你会同时拥有两边。把它想成一套集成的计算系统，就像今天的智能手机已经是的那样。

The point I keep coming back to is friction. If you pull up the phone and the agent takes too long to do something for you, you will just do it yourself. For other things, you will hand them to the agent. And the phone is unique to you. I said this in the keynote. You may have a little app in the corner that opens the garage door, or checks something. That app is on your phone, and the agent has to go there and operate it. You are going to have a combination. Think of it as one integrated computing system, the way the smartphone already is.

所以你不接受这样一个世界：Muse 和这些个人智能体把计算从设备上抽走了，大部分工作负载、大部分 token 都住在那些云端虚拟机里？

So you do not buy a world in which Muse and these personal agents have abstracted the computing off the device, and most of the workload, most of the tokens, live in those cloud virtual machines?

不接受。两个原因。第一，旧的和新的按定义就是共存的，共存意味着系统必须同时用两个世界。这又要暴露我的年龄了，我记得当年所有人都在谈 Palm 的 webOS，所有应用都要住在云里，手机只是一个壳。事情不是那样发生的。

No. Two reasons. The first is that the old and the new coexist by definition, and that coexistence means the systems have to use both worlds. This is going to show my age, but I remember when everyone talked about Palm's webOS, and all the applications were going to live in the cloud, and the phone was just going to be a shell. It did not happen that way.

我们忘得真快。

How quickly we forget.

你看到应用商店发展起来了。而现在，尤其是开放权重模型出现之后，你看到本地部署在增长。这很有意思。如果我没记错——数字我可能记错，我也记不清是已公布的那个季度还是前瞻指引——英伟达最近一次财报谈到，大约 40% 的收入来自 AI 公司和超大规模云厂商，另外 40% 来自本地部署、新云和主权 AI。一旦你相信本地部署，你就已经接受了混合 AI。有些事在本地做，有些事在云里做。接下来的问题是：本地的边界停在哪里？停在你的 PC？停在你的手机？我觉得事情会朝这个方向走。

You saw app stores develop. And now, especially with open-weight models, you see on-prem growing. It is interesting. If I have this right — I may get the numbers wrong, and I do not remember whether it was the reported quarter or the forward guide — Nvidia's last earnings talked about something like 40 percent of revenue from AI companies and hyperscalers, and another 40 percent from on-prem, neoclouds, and sovereign AI. The moment you believe in on-prem, you have accepted hybrid AI. Some things will be done on-prem, some in the cloud. Then the question is: where does the premises end? Does it end at your PC? Does it end at your phone? I think that is the direction this goes.

我还是没想明白什么是 AI 智能手机。你说过好几次，包括今天早上。你可以说这些手机已经是 AI 手机了。Siri 现在还行。Google Gemini 非常好。每个聊天机器人都已经在手机上。AI 智能手机是什么，真正的差别在哪？

I am still struggling with what an AI smartphone is. You said it a couple of times, including this morning. You could argue these phones are already AI phones. Siri is decent now. Google Gemini is very good. Every chatbot is already on the phone. What is an AI smartphone, and what is actually different?

这个愿景是在 OpenClaw 发布时变清楚的。你可以告诉一个智能体，或者给它发消息：这是我要你替我做的事。你收到一个问题，你说可以，然后智能体去干活。它跑在那台计算机上。它负责规划。它负责把哪些事放在本地做的路由。它会去你的记忆，也就是你上下文里的个人记忆图谱。你的上下文只存在于手机上。

The vision became clear when OpenClaw launched, and you could tell an agent, or text an agent, here is what I want you to do for me. You get a question, you say yes, and the agent does the work. It runs on that computer. It handles the planning. It handles the routing of what should happen locally. It goes to your memory, the personal memory graph inside your context. Your context exists only on the phone.

对。

Right.

上下文不在云里。它不在。它在手机上，传感器数据也来自手机。智能体看这些东西，把它们变得跟你有关，然后采取行动。这和你自己打开浏览器去做，非常不一样。它远远超出浏览器。我觉得这才是 AI 智能手机。我们才刚刚开始。早期，手机制造商和操作系统会放出一组功能，然后称之为 AI。人们也许会用那些功能，但那不是 AI 智能手机。一旦有了编排器和智能体，这就变成第三方的问题，不是第一方的问题。会有不同的智能体、不同的技能，跨公司集成。我们用的例子是支付。应用变成横向的，出现在一切里面。操作系统和应用之间的边界不再重要，体验跨设备发生。这是根本的差别。

Context does not exist in the cloud. It does not. It exists on the phone, and the sensor data comes from the phone. The agent looks at those things, makes them relevant to you, and then takes action. That is very different from opening a browser and doing it yourself. It is way beyond a browser. I think that is the AI smartphone. We are only at the beginning. In the early days, phone makers and operating systems ship a set of features and call that AI. People may use those, but that is not the AI smartphone. Once you have an orchestrator and agents, this becomes a third-party problem, not a first-party problem. There will be different agents, different skills, integrated across companies. Payments are the example we use. The application becomes horizontal and shows up inside everything. The boundary between the OS and the apps stops mattering, and the experience runs across devices. That is the fundamental difference.

我记得手机的早期，Android 发布，iPhone 发布。当时的在位者是诺基亚这样的公司，还有黑莓。他们看着智能手机带出来的那些功能。以 iPhone 为例。你有 iTunes。你有那个可以拨弄鼓和吉他的小音乐应用。操作系统也许自带十五个应用，应用商店里东西不多。于是黑莓和诺基亚说，这些功能我们也能做。但那不是智能手机。智能手机是由第三方应用定义的。当时没有人知道接下来会发明什么，人们会用它做什么。我对 AI 智能手机的感觉一样。一旦 harness 和能力就位，你会看到它怎么长出来，而我认为那会改变手机的体验。

I remember the early days of the phone, the Android launch and the iPhone launch. The incumbents were companies like Nokia, and there was BlackBerry. They looked at the features that shipped with the smartphone. Take the iPhone. You had iTunes. You had that little music app where you could move the drums and the guitar around. You had maybe fifteen apps that came with the OS, and not much in the App Store. So BlackBerry and Nokia said, we can ship those features too. But that was not the smartphone. The smartphone was defined by the third-party applications. Nobody knew then what would be invented, or what people would do with it. I feel the same way about the AI smartphone. Once the harness and the capability are there, you will see how it develops, and I think that changes the experience of the phone.

运营 Sources 最头疼的事之一是账本。我不是会计，市面上的软件又笨，又跟交易真正发生的地方脱节，直到 Mercury 推出了 Mercury Books。这是 AI 驱动的记账，能接上你的 Mercury 银行和信用卡交易，也接外部卡和薪酬系统。你可以选收付实现制或权责发生制，并生成现金流、损益表和资产负债表。设计很好上手，分类大部分由 AI 完成，Mercury 的指挥智能体可以把任务从头做到尾。你可以不额外花钱把会计师请进来，需要的话 Mercury 也会帮你找一位人类会计师。访问 mercury.com/books。Mercury 是金融科技公司，不是 FDIC 承保的银行。银行服务由 Choice Financial Group 和 Column N.A. 提供，二者均为 FDIC 成员。

One of the biggest pains of running Sources has been the books. I am not an accountant, and the software out there was clunky and detached from where the transactions actually live, until Mercury launched Mercury Books. It is AI-powered accounting that works with your Mercury banking and card transactions, plus external cards and payroll. You can choose cash or accrual, and generate cash-flow, profit-and-loss, and balance-sheet reports. The design is approachable, the AI does most of the categorizing, and Mercury's command agent can take tasks end to end. You can invite your accountant in at no extra cost, and Mercury will help you find a human accountant if you need one. Visit mercury.com/books. Mercury is a fintech, not an FDIC-insured bank. Banking services are provided through Choice Financial Group and Column N.A., members FDIC.

AI 的用处取决于它拿到的上下文。上下文散落在工具、讨论串和私信里的时候，你的团队和智能体就是在盲飞。Atlassian 的 Jira 说它解决的就是这个问题。Atlassian 的协作图谱把 Jira、Confluence、GitHub、Slack 等地方真正重要的碎片拉到一起。他们给出的数字是结果准确率提高 44%，token 用量减少 48%。你可以把这份工作上下文分享给已经在用的智能体，比如 Claude、Cursor 和 GitHub Copilot，直接给他们派活，或者通过 MCP 把工具接上。详见 jira.com。

AI is only as useful as the context it has, and when that context is scattered across tools, threads, and DMs, your team and your agents are flying blind. That is the problem Jira by Atlassian says it solves. Atlassian's teamwork graph pulls together what matters from Jira, Confluence, GitHub, Slack, and more. They cite 44 percent more accurate results with 48 percent less token use. You can share that work context with agents you already use, such as Claude, Cursor, and GitHub Copilot, assign them work directly, or connect tools through MCP. Learn more at jira.com.

我一整天都在会议之间切换，常常上一场还没消化，下一场就开始了。Granola 一直在后台跑。它是会议用的 AI 记事本，视频会议、电话、当面开会都能用。我用它回想说过什么、做摘要，并根据邮件起草后续跟进。到 granola.ai/sources 试用，优惠码 SOURCES 可免三个月。Framer 是 Sources 网站 podcast.sources.news 背后的 AI 建站工具。它把智能体放进设计和发布网站的同一块画布。从 framer.com/sources 开始，Framer Pro 年付计划可享 30% 折扣。可能适用规则与限制。

I context-switch between meetings all day, often with no time to process one before the next starts. Granola runs in the background the whole time. It is an AI notepad for meetings that works on video calls, phone calls, and in person. I use it to recall what was said, to make summaries, and to draft follow-ups from my email. Try it at granola.ai/sources with the code SOURCES for three months off. Framer is the AI website builder behind the Sources site at podcast.sources.news. It puts agents on the same canvas where the site is designed and published. Start at framer.com/sources for 30 percent off a Framer Pro annual plan. Rules and restrictions may apply.

转到可穿戴。我的年纪足够记得，当年可穿戴就是 VR，后来加上一点 AR。现在它意味着手表、胸针，以及你和 OpenAI 在做、但不能说的那些东西。

Turning to wearables. I am old enough to remember when wearable meant VR, and then a little AR. Now it means watches, pins, and whatever you are doing with OpenAI that you cannot talk about.

那个我不能谈，是的。

I cannot talk about that, yes.

这些设备都在来的路上。在谈 AI 可穿戴之前，我想先听 VR 和 AR 的现状。高通在这个行业里很久了。它一直很难获得真正的商业牵引力，虽然 Meta 开始靠眼镜做出一些成绩，谷歌要和 Warby Parker 一起做眼镜，我觉得苹果也会来。所有人都在来。你怎么看今天的可穿戴市场？VR 还是你相信的东西，还是我们已经过了那个阶段？

All of these devices are coming. Before the AI wearables, I want the state of VR and AR. Qualcomm has been in that industry for a long time. It has struggled for real commercial traction, although Meta is starting to get it with glasses, Google is coming with Warby Parker, and I think Apple will come too. Everyone is coming. What do you make of the wearable market today, and is VR still something you believe in, or are we past it?

好问题，里面塞了很多东西。我从 VR 说起。它是一种主机型市场。把它想成游戏市场。

Good question, and there is a lot packed into it. I will start with VR. It is a console-type market. Think of it as a gaming market.

故事以前不是这样。人们曾觉得它也许会像移动设备。

That was not always the story. People thought it might be like mobile.

我知道。人们把它说成未来，说成元宇宙。技术发展到体验其实已经相当好，但它是游戏主机式的体验。大量用途是游戏，甚至不是社交，因为它不是你每天方便戴着走来走去的设备。你戴上它，就完全沉浸进去。然后有一条中间道路，混合现实，增强现实加上透视。智能眼镜不一样。重点不再是可视化技术。把眼镜想成一个平台。拐点来自生成式 AI。

I know. People talked about it as the future, the metaverse. The technology got good enough that the experience is actually pretty good, but it is a gaming-console experience. A lot of the use was gaming, not even social, because it is not a device you conveniently walk around in every day. You put it on and you are fully immersed. Then there was the middle of the road, mixed reality, augmented reality plus see-through. Smart glasses are different. The point stopped being the visualization technology. Think of glasses as a platform. The inflection came from generative AI.

这三类里，我相信会非常成功、而且已经成功的，是智能眼镜。我觉得今天的市场是 5000 万到 6000 万部的出货。我们大约有 40 款不同设计，来自不同公司，而且我认为它会急剧增长。最早的智能眼镜是为了延伸相机。拍张照片，录一段视频，发到 Instagram 当限时动态。然后生成式 AI 来了，你开始问不一样的问题。这是什么？把这个翻译一下。带我去那里。这个人是谁？那些智能体式的工作负载开始有牵引力，设备变得更有用。

Of those three, the one I believe will be very successful — and it is already successful — is smart glasses. I think the market today is 50 to 60 million units shipped. We have about 40 different designs, from different companies, and I think it grows dramatically. The first smart glasses were about extending the camera. Take a photo, record a video, post it to Instagram as a story. Then generative AI arrived, and you started asking different things. What is this? Translate this. Take me there. Who is this person? Those agentic workloads got traction, and the device became more useful.

如果你相信我们说了好几年的那句话，AI 是新的用户界面，那这里就是黄金地段。它贴近你的眼睛、耳朵和嘴。这很自然。人类已经知道怎么戴眼镜。有一些规则，而这些规则就是要解决的问题。理想情况下，大约 35 克或更轻。电池大约能用 6 小时、8 小时或 10 小时。而且它必须是一件时尚单品。这样才会起量。在所有可穿戴品类里，我最看好、也会成为最大的，是眼镜：看见我所见，读到我所读，以及其余的事。我们的研究团队做了一件非常简单的事。眼镜里放一个小模型，载入公司通讯录，你走在人群里问这个人是谁。那是 Alex。这是他的工号。这样的应用有很多。

If you believe what we have been saying for a couple of years, that AI is the new UI, then this is prime real estate. It sits close to your eyes, your ears, and your mouth. It is natural. Humans already know how to wear glasses. There are rules, and those rules are the problems to solve. Ideally, about 35 grams or less. About six, eight, or ten hours of battery life. And it has to be a fashion item. That is what gets traction. Of every wearable category, the one I am most bullish on, and the one that becomes the biggest, is glasses: see what I see, read what I read, and the rest. Our research team did something very simple. A small model in the glasses, your company directory loaded, and you walk around and ask who this person is. That is Alex. Here is the employee number. There are many applications like that.

手表会有自己的角色。胸针、吊坠和首饰也会有角色。我们和各家公司有不少设计，现在什么都看得到。人们在试验带摄像头的耳机、首饰、吊坠、徽章——包括我们和微软一起发布的那一款。还早，但有希望。这里面最大的，是智能眼镜。

Watches will have a role. Pins, pendants, and jewelry will have a role. We have a number of designs with companies, and we see all of it now. People are experimenting with earbuds that have cameras, with jewelry, with pendants, with badges — including the one we announced with Microsoft. It is early, and it is promising. Of all of them, the biggest is smart glasses.

先把手表放一边，因为手表已经有健康和运动的用途。如果眼镜是一个品类，智能手机显然继续有用，你会不会在这些可穿戴里押注第三种形态？你提到有几十种，大公司也即将推出我们还没见过的东西。或者有没有第三种形态——一个听你说话的小圆盘、一件首饰，随便什么——让你愿意说，会有成千上万、几百万人用它？

Set the watch aside for a moment, because a watch already has health and fitness utility. If glasses are one category, and the smartphone obviously stays relevant, would you bet on a third form factor among all these wearables? You have mentioned dozens, and big companies are about to ship things we have not seen. Or is there a third form — a puck that listens, a piece of jewelry, whatever — where you would say millions and millions of people will use it?

会。这个问题难答的地方是：那个形态到底是什么？人们在试验，我们会看到结果。感觉像一个 iPod 时刻。

Yes. The hard part of the question is: what is that form factor? People are experimenting, and we will see. It feels like an iPod moment.

我也在想随身听，它是怎么变成一个文化时刻的。

I have been thinking about the Walkman too, and how it became a cultural moment.

就是那种感觉。在随身听之前，或者在 iPod 之前，如果你问每个人会不会随身带着那样一个设备，答案会是：我不知道。然后它就发生了。这又回到我刚才说的。我信不信任它，它有没有价值，摩擦低不低？如果有人把这三件事做通，它就会成功。

It feels like that. Before the Walkman, or before the iPod, if you had asked whether everyone would carry a device like that, the answer would have been, I don't know. Then it happens. It comes back to what I said before. Do I trust it, does it have value, and is the friction low? If someone cracks that, it will be successful.

我们谈谈数据中心。我还在试图理解，高通为什么要做数据中心。

Let's talk about data centers. I am still trying to understand why Qualcomm is doing data centers.

为什么不呢？

Why not?

为什么不。世界缺算力。但你们的机会是什么？你们是一家大公司，面对的是超大规模云厂商。你们的策略，和亚马逊、谷歌云、微软以及其他家，差别在哪里？

Why not. The world is starved of compute. But what is the opportunity for you? You are a large company, and you are dealing with the hyperscalers. What is different about your strategy versus Amazon, Google Cloud, Microsoft, and the rest?

我们有移动的基因。我们从移动起步，也是最专注于移动的公司。我们懂这一点，也会继续做。多元化走了好几步。我们当时认为自己的技术在其他市场也能重要，也能造成颠覆，所以开始建计算资产。我们把计算机做大，而且你不能跳步。我们进了 PC，进了汽车。在汽车里，跑自动驾驶的一些芯片——2++ 级、3 级、4 级——硅片是我们的。那就是一台服务器。它是服务器级的处理器。我们必须把这个能力做出来，下一步，数据中心，就成了公司自然的扩张。

We have a mobile DNA. We started in mobile, and we have been the company most concentrated on mobile. We understand that, and we will keep doing it. Diversifying took several steps. We thought the technology we had could matter, and disrupt, in other markets, so we started building computing assets. We made the computer bigger, and you cannot skip a step. We went into PCs and into cars. In the car, some of the chips that run autonomy — level 2++, level 3, level 4 — that silicon is ours. That is a server. It is a server-class processor. We had to develop that capability, and the next step, the data center, became a natural expansion of the company.

第二部分是，数据中心在以极快的速度变化，人们曾经相信的东西变得很快。一开始全是英伟达 GPU。即便有人想要专门做推理的新硬件，答案也是：你在英伟达 GPU 上训练，不训练的时候就在英伟达 GPU 上推理。只有英伟达。然后推理开始上规模，人们说做推理有别的办法，更高效，性能更好，总拥有成本更低。如果这不是真的，英伟达就不会和 Groq 有任何关系。再往下，这句话变成：算力需求非常高，能源稀缺，所以我需要专门的方案。推理拆成了预填充和解码，连解码自己也在进一步拆开。

The second part is that the data center is moving at breakneck speed, and things people believed changed very fast. At the start it was all about the Nvidia GPU. Even when people wanted new hardware for inference, the answer was: you train on an Nvidia GPU, and when you are not training you infer on an Nvidia GPU. Only Nvidia. Then inference started to scale, and people said there are other ways to do inference, more efficient, better performance, lower total cost. If that were not true, Nvidia would have had nothing to do with Groq. Keep going and the line becomes: compute demand is very high, energy is scarce, so I need a specialized solution. Inference broke into prefill and decode, and even decode is breaking down further.

我们看着这个局面，会说节奏很快。内存技术会变。计算技术会变。在我们的生意里，你什么时候进场并不重要，重要的是你带来的东西有没有创新、有没有颠覆性。有空间。所以我们说：如果我们能做出独特的东西，就会有市场。去专注于做独特的东西。第三层是我们的基因。我们做芯片的时候，假设另一头有一块电池。我看一座数据中心，看到的是一定量的算力，和一定量可用的能源。我们认为自己能在那里做点什么。我们聚焦高能效、高性能的 CPU，以及高性能、高能效的加速器。那是我们的地盘。这是高通的自然演进。

We look at that and say the pace is fast. Memory technology will change. Compute technology will change. In our business it does not matter when you enter, if what you bring is innovative and disruptive. There is room. So we said: if we can do something unique, we will have a market. Go focus on doing something unique. The third layer is our DNA. We build chips assuming there is a battery on the other side. When I look at a data center, I see an amount of compute and an amount of available energy. We think we can do something there. We are focused on energy-efficient, high-performance CPUs, and on high-performance, energy-efficient accelerators. That is our space. It is a natural evolution of Qualcomm.

所以为电池而设计，是你们做这件事背后的一个关键洞察。

So designing for the battery was a key insight behind doing this.

我们一直在谈，是什么让一家公司与众不同。英特尔进数据中心的时候，英伟达还没在做 CPU，但英伟达在做大量并行处理，为图形渲染三角形。那种并行处理后来对 AI 很重要，也给了他们机会。我们对算力与能源这道等式的感觉一样。这道等式给我们创造了机会。

We talk all the time about what makes a company unique. Intel was in the data center before Nvidia was making CPUs, but Nvidia was doing a huge amount of parallel processing, rendering triangles for graphics. That parallel processing turned out to matter for AI, and it created their opportunity. We feel the same way about the equation of compute versus energy. That equation creates the opportunity for us.

在你们收购 Chris Lattner 的公司 Modular 之前，我是最后一个采访他的人。

I was the last person to interview Chris Lattner before you acquired his company, Modular.

他是个很棒的人。

He is a great guy.

他是个很棒的人，也是个传奇。他发明过多种编程语言，包括 Swift，而你们最近买下了 Modular。不了解的人可以这么理解：Modular 是在试图撬开那道软件护城河，那道河把 AI 产业拴在英伟达和 CUDA 上。几个月前他告诉我，不同芯片的浪潮要来了——CPU，以及所有用来跑 AI 的硬件——开发者需要一种办法把它们都用上。他在 Modular 的那套策略，现在有多少变成了高通的数据中心策略？

He is a great guy, and a legend. He has invented multiple programming languages, including Swift, and you bought Modular recently. For people who do not know it, Modular was a bid to crack the software moat that keeps the AI industry tethered to Nvidia and CUDA. He told me, months ago, that a wave of different chips is coming — CPUs and the rest of the hardware used for AI — and developers need a way to use all of them. How much of that Modular strategy is now Qualcomm's data-center strategy?

我见过 Chris，我们很幸运能让他和他的团队留在高通。我肯定他有去非常不同的公司的选择。Modular 和高通合拍，是因为我们对这个行业的看法一样。高通发明过世界上一些最重要的技术，然后授权给所有人。我们相信，技术被广泛部署、有更多公司参与创新的时候，你会得到更多创新和更大的规模。今天软件侧有一个控制点，而行业需要一个跑 AI 的开放替代方案，因为 AI 不会只属于数据中心。它会去到所有地方。

I met Chris, and we are fortunate to have him and his team at Qualcomm. I am sure he had options to go to very different companies. What made Modular and Qualcomm a fit is that we share the same view of the industry. When Qualcomm invented some of the most important technology in the world, we licensed it to everybody. We believe you get more innovation and more scale when the technology is widely deployed and more companies contribute. Today there is a control point on the software side, and the industry needs an open alternative for running AI, because AI will not stay unique to the data center. It is going everywhere.

人们已经在谈数据中心里种类繁多的硬件。现在再加上汽车、PC、手机和可穿戴。开发者必须把这些算力都当作目标，既跑大模型也跑小模型，而且他们需要一个共同平台。在 AI 里，这个平台还不存在。传统数据中心把 Linux 成熟成了操作系统。云有 Kubernetes。在手机上，横向平台是 Android，而且它是开放的。这些东西在 AI 里都还没有。Chris 和 Modular 团队的说法是：开发者需要这个，行业需要这个，我们来做。它是现代的，性能更高，而且跑在任何硬件上。我说，那我们就在高通做，并且把它提供给所有人。

People already talk about a multitude of hardware in the data center. Now add cars, PCs, phones, and wearables. Developers have to target all of that compute, for models big and small, and they need a common platform. In AI, that platform does not exist. Traditional data centers matured Linux as an operating system. The cloud has Kubernetes. In phones, the horizontal platform is Android, and it is open. None of those things exist for AI yet. Chris and the Modular team were saying: developers need this, the industry needs this, we will build it. It is modern, it is higher performance, and it runs on any hardware. I said, let's do it at Qualcomm, and let's offer it to everybody.

高通会为每一次 Modular 部署庆祝，不管它跑在英伟达、AMD、博通、英特尔，还是任何其他芯片上，因为我们相信行业会受益于一个标准、共同、开放的平台。我们说它会用 Apache 2.0。它会开源。对高通自己的芯片来说，平台一旦开放，我们就可以在硬件上竞争。这是我们一辈子在做的事。这就是移动业务。我们不怕。

We at Qualcomm will celebrate every single Modular deployment on Nvidia, on AMD, on Broadcom, on Intel, on any other chip, because we believe the industry benefits from one standard, common, open platform. We said it will be Apache 2.0. It will be open source. For Qualcomm's own chips, once the platform is open we can compete on hardware. That is what we have done our whole lives. That is the mobile business. We are not afraid of it.

在硬件本身的优劣上竞争，而不是靠软件锁定。

Compete on the merits of the hardware, rather than on software lock-in.

我们在硬件上竞争，而这正是硬件创新得以发生的原因。在一个封闭系统里，我也许有更好的芯片，但没有人用得上。我们在数据中心是新来的。我们正在进入，而且我们高兴，因为开放平台创造了在硬件上竞争的机会。等式的另一半，是我们刚才关于 AI 智能手机、PC 和汽车的那场谈话。一旦一切都在跑 AI，边缘就是同一条计算连续体的一部分，你面对的是同一个问题。平台必须能扩展。我们会尽一切努力，让每一家硬件公司都能用它。Modular 会保持硬件中立。它会作为高通旗下的公司来运营，我们会支持很多芯片，同时努力做出我们自己出色的硬件。

We compete on the hardware, and that is what lets hardware innovation happen. In a closed system I might have a better chip, and nobody would ever be able to use it. We are new to the data center. We are entering, and we are happy, because an open platform creates the chance to compete on hardware. The other half of the equation is the conversation we already had about AI smartphones, PCs, and cars. The edge is part of the same compute continuum, once everything is running AI, and you have the same problem. The platform has to scale. We will do everything we can so that every hardware company can use it. Modular will be hardware-agnostic. It will be run as a Qualcomm company, and we will support many chips while we try to build great hardware of our own.

我们用 6G 收尾，这也是我还在想明白的事。我报道过好几代 G。你刚才提到过 1G。6G 要来了，我觉得大多数人很难说清 5G 到底改变了自己的什么。运营商把它吹得很凶。每一代 G 都是一列炒作列车。实际上，6G 会解锁今天做不到的什么？

Let's end on 6G, which I am also trying to figure out. I have covered several Gs. You mentioned 1G earlier. 6G is coming, and I think most people would struggle to explain what 5G changed for them. The carriers hyped it. Every G is a hype train. What, practically, will 6G unlock that is not possible today?

因为你这么问，我觉得应该先回答 5G。我不会说人们只是把它当成理所当然。从消费者的角度看，5G 做了两件事。第一，如果没有 5G，没有人能在智能手机上用无限流量。在 4G 上，这在经济上不可行。

Because of the way you asked, I feel I should answer 5G first. I am not going to say people are simply taking it for granted. From a consumer's point of view, 5G did two things. First, nobody would have unlimited data on a smartphone if it were not for 5G. It is not economically viable on 4G.

这说得通。

That is fair.

不可行。所以如果有一件事改变了日常生活，那就是人们有了无限流量。这不是唯一的一件。4G 是智能手机的开始。宽带来了，手机变成计算机，也就是在那个时点，CD 消失了。所有人开始流式听音乐。带宽够听音乐了。CD 没了，但 DVD 还在。到了 5G，DVD 也走了，因为你可以流式传输高清视频。5G 给消费者的第二件事，是高清视频下载。谈 6G 之前，我得先把这个说清楚。

Not viable. So if one thing changed daily life, it is that people have unlimited data. That is not the only thing. 4G was the beginning of the smartphone. Broadband arrived, the phone became a computer, and at that point CDs disappeared. Everyone started streaming music. There was enough bandwidth for music. CDs were gone, but DVDs were not. With 5G, DVDs went away, because you could stream high-definition video. The second thing 5G gave consumers was high-definition video download. I need to say that before I talk about 6G.

每一个偶数代的 G 都非常成功。2G 是。4G 是。按这个传统，6G 也会非常成功。这是有原因的。一代带来一次重大颠覆，下一代回去把事情补上。手机从 1G 开始，模拟信号，对讲机式的电话，直到人们发现电话可以随身带，也发现频谱不够所有人用。数字蜂窝，也就是 2G，问的是：地球上每个人都能有一部电话吗？它极其成功。移动通信爆发，后付费和预付费都有。3G 是增量。你连上了互联网。你看到黑莓，以及那些早期的 Windows 手机。4G 是宽带，是一次真正的断裂。宽带的另一头是一台计算机。手机变成计算机，改变了世界。5G 补上了一些事：无限流量，高清视频。模式就是这样。

Every even-numbered G has been a huge success. 2G was. 4G was. On that tradition, 6G will be a huge success. There is a reason. A generation brings a major disruption, and the next one goes back and fixes things. Phones started on 1G, analog, walkie-talkie phones, until people realized you could carry a phone, and also realized there was not enough spectrum for everyone. Digital cellular, 2G, asked: can everybody on Earth have a telephone? It was an enormous success. Mobile exploded, postpaid and prepaid. 3G was incremental. You connected to the internet. You saw BlackBerrys, and those early Windows phones. 4G was broadband, a real break. Broadband on one side means a computer on the other. The phone became a computer and changed the world. 5G fixed things: unlimited data, high-definition video. That is the pattern.

那么 6G 是什么？它是为了我们开头谈的那个 AI 未来而设计的。给消费者的简单功能，是无论你在覆盖区的哪里，都能上行高清视频。想想眼镜。看见我所见。6G 会把我们所有人变成这个世界上行走的摄像机。只要我们愿意，每个人都可以一直上行高清视频，而“看见我所见”会变成模型的重要上下文。

So what is 6G? It has been designed for the AI future we were talking about at the start. The simple consumer feature is high-definition video upstream, wherever you are in the coverage area. Think about glasses. See what I see. 6G would turn all of us into walking cameras on this world. Each of us could upstream high-definition video all the time, if we wanted to, and see-what-I-see becomes important context for models.

或者让模型以一种顾及隐私的方式，持续地替你观察。

Or for models, in a privacy-conscious way, to be observing constantly on your behalf.

持续地。我们给每一帧图像做标注。而当你想到世界模型和机器人，后面还有一整串事情。我现在只给你功能视角。更大的变化是，6G 首先是一张数据中心网络，也是一张物理 AI 网络。你去看无线电信号，看基站和手机之间的电磁波，看这个信号怎么表现，你得到的是物理 AI。每一个无线电信号里面——而这样的连接每时每刻有几十亿个——都有载荷，也就是正在传输的数据。信号本身，我可以把它当成雷达，就像汽车为自动驾驶装的雷达。现在我有几十亿个这样的雷达。有了它们，我可以实时绘制世界。网络知道正在发生什么。

Constantly. We annotate every image. And when you think about world models and robots, there is a whole set of things beyond that. I am only giving you the feature view. The bigger change is that 6G is primarily a data-center network and a physical-AI network. You look at the radio signal, the electromagnetic wave between the base station and the phone, and at how that signal behaves, and what you get back is physical AI. Inside each radio signal — and billions of these connections happen all the time — there is a payload, the data being transferred. The signal itself I can treat as a radar, the way cars use radar for autonomous driving. Now I have billions of those radars. With that I can map the world in real time. The network knows what is happening.

无人机探测已经成了一个很常见的例子。一切飞行的东西，整个空中经济的管理，都会由 6G 网络驱动。它们会感知周围。一切在移动的东西：每一辆车、每一个行人、每一辆公交。你可以追踪。对模型来说，这是海量的数据和上下文，尤其是对物理 AI 和机器人。6G 是颠覆性的。它改变基础设施本身是什么。它也让计算可以分布，所以 6G 网络可以运送 token。我们建过一张曾经是铜线、只传声音的网络。然后它变成了兆比特。到了 6G，中间会有 token。更容易的理解方式是：6G 是让新一类智能体设备成为可能的技术，也是给 AI 超大规模云厂商和模型提供实时感知的技术。

Drone detection has become a popular example. Everything that flies, the management of the whole aerial economy, will be driven by 6G networks. They will sense what is around them. Everything that moves: every car, every pedestrian, every bus. You can track things. That is a massive amount of data and context for models, especially for physical AI and for robots. 6G is disruptive. It changes what the infrastructure is. It also lets compute be distributed, so the 6G network can carry tokens. We built a network that used to be copper and carried a voice. Then it became megabits. With 6G there will be tokens in the middle of it. The easier way to think about it: 6G is the technology that enables the new class of agentic devices, and that gives AI hyperscalers and models real-time sensing.

人们什么时候能在日常生活里感受到这些？一年、十八个月、两年，还是三年？

When do people feel this in ordinary life? One year, eighteen months, two years, three?

我们会在 2028 年的洛杉矶奥运会上做演示。第一批硅片，你应该想成 2028 或 2029，二者之一。我不想给一个精确日期。我们会展示原型和硅片，2029 年和 2030 年有早期部署，规模化要到 2030 年之后。

We will do demonstrations at the Los Angeles Olympics in 2028. For the first silicon, you should be thinking about 2028 or 2029, one of the two. I do not want to give an exact date. We will be showing prototypes and silicon, with early deployments in 2029 and 2030, and it reaches scale after 2030.

好。克里斯蒂亚诺，非常感谢。

All right. Cristiano, thank you so much.

谢谢。我很感激。

Thank you. I appreciate it.

银行服务应该有现代软件的感觉。访问 mercury.com 了解更多，几分钟内在线申请。Mercury 是金融科技公司，不是银行。Granola 是 AI 记事本，视频会议、电话、当面，以及 Apple Watch 上都能用。到 granola.ai/sources 试用，优惠码 SOURCES 可免三个月。Atlassian 的 Jira 是你的团队和智能体共享同一份上下文的地方。在 jira.com 免费试用。Framer 是 AI 原生的网站构建器，让你建得更快，同时不交出控制权。访问 framer.com/sources 可享 30% 折扣。可能适用规则与限制。

Banking should feel like modern software. Visit mercury.com to learn more and apply online in minutes. Mercury is a fintech, not a bank. Granola is the AI notepad that works on a video call, a phone call, in person, or on an Apple Watch. Try it at granola.ai/sources and use the code SOURCES for three months off. Jira by Atlassian is where your team and your agents work from the same context. Try it free at jira.com. Framer is the AI-native website builder that lets you build faster without giving up control. Visit framer.com/sources for 30 percent off. Rules and restrictions may apply.