As generative AI evolves, smart devices are undergoing a paradigm shift from 'application-centric' to 'AI agent-centric,' which will trigger a comprehensive upgrade cycle for global computing infrastructure and underlying semiconductor architectures.
Earlier this week, Qualcomm CEO Cristiano Amon sat down for an exclusive interview with CNBC’s 'The Tech Download' at the network’s new London studio. In a conversation with veteran technology journalist Arjun Kharpal, Amon outlined his vision for the evolution of personal electronics in the AI era.
Addressing market concerns about weak end-demand and future growth drivers, Amon offered a highly bullish outlook: 2026 will be the 'Year of the AI Agent,' as current smartphone, PC, and automotive computing capabilities will fall far short of requirements, ushering in a new global cycle of device replacement and chip upgrades.
He emphasized that AI is not merely adding new features but fundamentally reshaping the entire computing platform. Current device compute capacity is vastly insufficient to support the forthcoming AI agent ecosystem.
“I believe none of the devices we have today are ready for this future,” Amon stated candidly during the interview. “You will have to upgrade your phone, you will have to upgrade your PC, and your car will also need an upgrade. This represents a completely new form of computing and software—and it requires a completely new silicon architecture.”
To accommodate this transformation, Qualcomm’s entire product roadmap is undergoing a comprehensive overhaul. Amon revealed that running increasingly complex models on the device side means the silicon area dedicated specifically to AI model execution must expand significantly. Not only are GPUs and neural processing units (NPUs) becoming critical, but central processing units (CPUs) are also experiencing a resurgence. Future AI agents will need to perform highly sophisticated tasks—such as inspecting data, invoking models, and connecting to the internet—and “these orchestrators reside on the CPU.” Qualcomm is set to unveil its latest data center CPU on June 24.
2026 will be the 'Year of the AI Agent,' and smartphones will gain a 'dual persona.'
The market has long sought a true 'killer app' capable of stimulating consumer device upgrades. Amon noted that conventional AI chatbots are just the beginning; the real inflection point for the industry lies in the widespread adoption of 'agentic AI'—intelligent agents capable of autonomously executing tasks on behalf of users.
“I am absolutely certain that 2026 will be the ‘Year of the AI Agent,’ and we are already witnessing this trend begin to scale,” Amon predicted.
Regarding the future of smartphones, Amon affirmed unequivocally that phones will not disappear—but how we use them will undergo a fundamental transformation. He offered a vivid analogy: “Your phone will have ‘dual personas.’ When you pull it out of your pocket, you can still open apps and do things just as before; but simultaneously, your phone will also host orchestrators and agents that act on your behalf, automatically invoking your applications.”
This shift will completely disrupt the existing app ecosystem. Amon cited banking as an example: in the future, users won’t need to open a bank app and adapt to an interface designed by developers. Instead, they’ll simply issue instructions directly to an AI agent. The agent, equipped with the user’s credentials, will automatically perform tasks such as checking balances, generating charts, and initiating transfers, rendering the results on-screen in the user’s preferred colors and format. ‘Apps aren’t dead—they’ll evolve into agents that are more powerful, highly personalized, and seamlessly integrated into everything.’
Smart glasses will become the next core endpoint, as AI giants enter the hardware race to secure 'data endpoints.'
When discussing the potential of future devices, Amon expressed strong confidence in the commercial prospects of smart glasses.
‘One hundred percent [they will become one of the most important devices]! The numbers speak for themselves,’ Amon stated clearly regarding market expectations. ‘Annual shipments of smart glasses have already reached tens of millions of units. We’re seeing a clear trajectory—within a few years, this figure could grow to hundreds of millions, or even rival the scale of smartphones. For context, annual smartphone sales are around 1.2 billion units.’
Amon explained the inherent advantages of glasses as a form factor: humans already have the habit of wearing eyewear, and glasses sit closest to the eyes, ears, and mouth—making them ideal for integrating cameras and microphones so AI models can ‘see what you see and hear what you hear,’ enabling natural multimodal interaction. Beyond glasses, entirely new device forms—such as pins, jewelry, and pendants—will also flourish.
Notably, Amon pointed out that the current landscape of consumer electronics players is undergoing dramatic change. In the past, smartphone manufacturers dominated vertical ecosystems ranging from smartwatches to earbuds; but in the AI era, ‘agents’ have become the center of digital life, creating opportunities for non-traditional hardware companies. Pure-play AI software giants like OpenAI are now venturing into hardware design, driven primarily by their hunger for future data.
Amon cut straight to the heart of the business logic: ‘When people wear smart glasses and move around, they generate enormous volumes of data—exponentially greater than the data we’ve historically used to train models. These AI companies need to win these agents’ “endpoints” because access to this personalized consumer data is critical for training future models.’
The semiconductor industry is not merely experiencing cyclical fluctuations; a capacity inflection point may arrive in the second half of 2027.
With explosive growth in computing demand, the entire semiconductor supply chain—from high-bandwidth memory to wafer foundry services—is facing severe capacity constraints.
When asked how to address challenges such as memory shortages, Amon quipped, ‘First, if you have memory, I’ll buy it from you.’
Addressing market debates over whether the semiconductor industry is merely experiencing traditional cyclical fluctuations, Amon offered a dissenting view. He emphasized that this is not simply an industry cycle but a fundamental structural transformation. 'I think this time it’s a bit different. The entire global computing infrastructure has already been upgraded. AI is essentially a new way of running software, and you need different—and more powerful—computers. Tokens are the new currency in the AI world, and existing computing infrastructure simply lacks the capacity to handle them; an upgrade process is inevitable.'
Regarding when the capacity bottleneck might ease, Amon provided a relatively optimistic timeline. Contrary to the pessimistic market sentiment that 'it won’t be resolved until 2028,' Amon stated: 'I’m relatively more optimistic. I believe at some point in 2027—or at least in the second half of 2027—we should start seeing significantly more capacity coming online.'
He believes that entirely new chip architectures that utilize memory more efficiently will create substantial market opportunities for companies like Qualcomm, which excel in low-power and high-efficiency design.
Below is the full transcript of the interview, translated with AI assistance.
Kharpal:
Welcome to The Tech Download.
I’m Arjun Kharpal, Senior Technology Correspondent at CNBC, and we’re kicking off with an exceptionally interesting guest: Cristiano Amon, CEO of Qualcomm.
In case you’re unfamiliar with Qualcomm, here’s a quick introduction: they are one of the world’s largest semiconductor companies. They design what are known as 'system-on-chips'—we’ll dive deeper into SoCs, a term you may have heard before—under the Snapdragon brand.
These chips power smartphones, integrating numerous components onto a single piece of silicon, and they drive a significant portion of the world’s smartphones. They also power other types of devices, such as laptops. As such, their CEO offers valuable insights into consumer electronics and where the industry is headed in the AI era. We’ll discuss everything—from the upcoming wave of new devices to some unreleased products he shared with us.
He’ll talk about how AI is transforming our relationship with core devices—particularly smartphones—and what this shift means for how Qualcomm must rethink chip design for future devices.
One term you’ll hear frequently in this conversation and in future episodes is “Agentic AI.”
The idea here is that we started with chatbots of this kind—where we input a query and receive a response—but AI companies are now aiming to go further. They want these AI chatbots to effectively act as assistants, serving as agents that represent us and sometimes autonomously carry out tasks on our behalf.
You’ll also hear about other innovations mentioned by Qualcomm CEO that enable AI-powered devices. One of these is MCP, or Model Context Protocol.
This is actually an open-source system that allows AI assistants to connect and communicate with other data sources and applications. This is important—if you want your agent to perform tasks for you within another application, for instance.
The final innovation you’ll hear about is what’s known as cloud-and-edge processing.
So, often when we use these AI chatbots, your data is sent to the cloud, processed there, and then returned to your device—where you actually receive your answer.
But many device manufacturers are now exploring how to run those AI models and AI processing tasks directly on the device itself.
This could mean faster performance and enhanced security—and it’s precisely the direction Qualcomm is pursuing with its chips.
I’m very pleased to bring you this conversation with Qualcomm CEO Cristiano Amon. Welcome to The Tech Download.
Cristiano, thank you so much for joining The Tech Download.
Amon:
It’s a great pleasure to be here, and I’m really looking forward to this conversation, Arjun.
Kharpal:
Cristiano, you see, today we’re discussing AI devices—how we’ll interact with personal electronic devices in the future, as AI is becoming increasingly pervasive within these devices.
But first, I think it’s important for our audience to understand what we actually mean when we talk about AI devices, AI smartphones, or AI laptops.
Well, that’s an excellent point for discussion—especially since we’ve now reached a pivotal moment in our AI transformation journey, where we have a clear understanding of what those devices will be and what you’ll be able to do with them.
Amon:
I’d answer your question this way: you’ll still have your phone—it won’t disappear—but your phone will become better. It will have agents acting on your behalf to operate it, and there will also be entirely new categories of devices that we’ll wear.
To address your question directly—I’d say, what do these things actually mean? We’ve been talking about so-called personal AI devices. When large language models and large vision models understand the world the way we do—and understand how we communicate—the natural way to interact with them is through devices that are closer to our senses: closer to our eyes, our mouths, and our ears.
And then there will be things we wear. Glasses are a very natural form factor, but there will also be jewelry, pendants, and pins—because the type of user interface will be fundamentally different from what we’ve had on PCs and smartphones. That will represent a new device category, and it will be part of our mobile experience.
Kharpal:
You mentioned something interesting about large language models—what has changed to make today’s devices far more personalized?
Right, this involves multiple technologies—I’ll break it down for you. I think the first point is that those large language models have become significantly better, and they’ve also become highly specialized.
Amon:
It’s not just language models—it’s also vision models. You’re starting to hear terms like 'multimodal,' meaning models can accept different types of input: audio, language, and vision.
You’re hearing about smaller models and mixture-of-experts models that are highly specialized for certain functions. You’re also hearing about model distillation—where a large model is trained first, and then a smaller version is derived from it—and about these models being connected to various applications.
You’re hearing about technologies like the MCP protocol, which allows models to connect to the internet, other applications, and cloud services. And then there’s the final piece of this technological stack—I believe you’ve heard of OpenClaw.
These are technologies called orchestrators—agents that operate your computer or device on your behalf, access your data, interact with models, and complete tasks for you.
When you combine all of these elements, they create the conditions for transforming our mobile experience—you’ll start interacting with mobile devices through agents in a fundamentally different way.
No keyboard needed—you won’t need to touch anything; you’ll just look and speak. I believe this is laying the groundwork for the AI-driven mobile revolution we’re about to witness.
Kharpal:
So, Cristiano, let’s address some of these issues. You mentioned models, as we’ve been talking extensively about large language models—these massive, enormous models that are trained and then power the chatbots used by so many people.
But there’s also discussion about how more of these models are becoming smaller so they can run on devices rather than in the cloud. That’s a significant part of your work at Qualcomm. How important is model development right now in enabling the future agent-based AI experiences you’ve been describing?
Amon:
Look, it’s truly important—these models will continue to be developed. But I think we’re now reaching a point of maturity in the industry where we’re beginning to understand how these models and computers will work together.
Let me be slightly provocative here—because I recall that about a year ago, maybe two years ago, there was a lot of discussion around whether a model should run in the cloud or at the edge. Could I run the exact same model on an edge device as in the cloud? But I think that was the wrong conversation to have.
For example, Arjun—if I asked you how many apps you have on your phone, you’d probably tell me you have over 200, organized into folders. And if I asked you to go through each one and identify which parts run in the cloud versus which run on the device, that would likely be a futile exercise. I know they all operate across both, because if you switch your phone to airplane mode, it becomes significantly less useful.
So, I believe the same thing is happening with AI.
AI is simply a new form of computing. Models are now evolving in such a way that specific models run at the edge while others run in the cloud, and they’ll work together as a unified system. In fact, there’s now an industry term for this: the 'computing continuum.' It functions like an integrated system. Let me give you an example of what I mean.
If I’m going to interact with a future personal AI device—say, a pair of smart glasses—and I want to interact just as I am with you right now, I need instantaneous responses. Imagine I’m walking down the street and I say, 'Who is that person?' The model responds, 'That’s Arjun from CNBC. You know him—you might want to say hello.' I can’t wait. I can’t ask the question, keep walking, and then wait for tokens to arrive. Similarly, if I’m reading a menu and ask, 'What does this say?' I want an immediate answer. Or if I scan a QR code and say, 'Please help me pay for this.' What’s happening is that computation naturally partitions between what runs at the edge and what runs in the cloud—and they’ll be integrated. One cannot function without the other. I believe this dynamic will, to some extent, define how we view the future of consumer electronics and devices in the AI era. And I’ll tell you—it’s no different from what’s happening in automobiles and robotics. In fact, those are excellent examples of edge AI—they actually run AI models within robots and vehicles.
So, let’s delve into this term—AI devices—and what it actually means. Well, we obviously have smartphones, but over the past few years, there’s certainly been an explosion of all these different types of chatbots. We’re thinking of OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini—all of these are now available and have become significantly more sophisticated. Now device manufacturers are wondering: how can we integrate all this AI into various kinds of consumer electronics? Ultimately, what does this mean for how we’ll use our devices?
This part is really important because we’re about to discuss different types of upcoming AI devices, and you’ll hear terms like ‘smart glasses.’ If you haven’t seen them yet, Meta has produced some smart glasses equipped with a camera. Samsung, Qualcomm, and Google are all collaborating on smart glasses as well. Many companies are focusing on this space, and many believe it could represent a new frontier for AI devices.
Another thing we should mention here is the AI pin. You may not have encountered this type of device yet, as it only enjoyed a brief surge in popularity in 2024. A company called Humane brought one such product to market. Essentially, it’s a small square that you clip onto your clothing. It has a camera but no screen—you can speak to it, and it responds verbally while also projecting text onto surfaces like your hand or a table, powered by AI.
That product actually failed. It received rather poor reviews, and you can read extensively about why it didn’t catch on at the time. However, it offered a glimpse into how device companies envision the future of hardware in an AI-driven world. We’re now starting to see more and more non-traditional players entering the consumer electronics space.
One example is OpenAI, the maker of ChatGPT, which hired renowned Apple designer Jony Ive to develop a device—so it will be very interesting to see what that product looks like when it launches. But this also signals that new entrants are stepping into the consumer electronics market, and we may witness entirely new form factors that we’ve never seen before.
Now, a major buzzword we’ve already mentioned is the concept of Agentic AI.
When I hear people talk about it, the utopian description is essentially that these AI systems will be able to act autonomously on our behalf—like the smartest personal assistant imaginable.
Kharpal:
When it comes to agentic experiences on consumer devices, what do you envision that looking like?
Amon:
Great—this is actually one of my favorite questions to answer, because, you know, even though we’ve been transforming Qualcomm and becoming a more diversified company, we remain one of the largest providers of mobile technology.
It’s now incredibly fascinating to see what’s happening with agents and on smartphones.
When we talk about agents, I’m fairly confident that 2026 will be the year of the agent—we’re already starting to see this scale.
These things are becoming extremely popular and going viral online. I think it started around the time OpenClaw launched—you now have this software installed on your computer, you set up an agent to do things for you, you create agents that access your data, act on your behalf, operate your machines, and go out onto the internet.
Then what you saw happen—I remember people buying computers, with many purchasing Mac Minis, which sold out entirely. They’d put them in their offices, install OpenClaw, keep it running constantly, burning through tokens like crazy, connecting to their Anthropic and OpenAI accounts and exhausting all their token allowances.
Now, this illustrates a new way we’ll interact with our computers.
We’ll still use our computers ourselves, but agents will also use our computers on our behalf and perform tasks for us.
In a mobile context, however, you can’t carry around a backpack with a car battery, a Mac Mini, and your phone. So these capabilities will need to run directly on the phone itself. Given that billions of people use smartphones, I believe this will define how the mobile experience evolves—and now we clearly understand what constitutes an AI phone versus an AI PC.
Let me break it down for you. What’s going to happen is that your phone will take on a dual role. Your phone will remain your phone—it won’t disappear. You’ll still pull it out, open apps, and do things yourself. But it will also host orchestrators and ‘claws.’ Between now and summer, you’ll see every operating system vendor talking about how their OS will include an orchestrator and a claw. You’ll be able to instruct agents to act on your behalf—they’ll operate your phone, access your apps, and execute your commands.
This shift is profoundly transformative. Even if I momentarily set aside the mobile aspect and focus only on what’s happening in the software industry, the entire SaaS sector is evolving: software companies today build applications not just for human users—with user interfaces designed for people—but also for agents, enabling them to enter and operate within those applications. The way this evolution is unfolding is truly fascinating.
For example, there will be things you do yourself, but also things your agent does for you.
I’ve seen some examples where someone has a meeting, and then another invitation comes in saying, ‘Cristiano, you have this podcast with Arjun from CNBC—should this meeting take priority afterward?’ I’d say, ‘Absolutely.’ Then it says, ‘Okay, can I reschedule your other meeting for you?’ And you say, ‘Yes.’ The agent calls that person, sends an email if it needs to talk to someone, and figures out how to reschedule things based on your calendar.
Gradually, we’ll get used to these new experiences, and over time, your phone will simultaneously be operated both by an agent and by you.
I’m just waiting for an agent that can clear out my inbox for me. That would be the best thing ever.
Kharpal:
Cristiano, when you look ahead at the roadmap, you said 2026 will be the year of agents. It feels like over the past two years, many smartphone companies have been talking about integrating AI into their devices, but none have actually delivered the kind of agent-driven experience you’re describing. Do you really believe that will happen in 2026? To what extent—and as you collaborate with all your device partners, what does that more advanced roadmap look like?
Amon:
That’s interesting. If you recall the transition from feature phones to smartphones, it was a massive shift—the phone became a computer. What happened early on in the smartphone era, very early on, was that the apps you had were those provided by OEMs.
So I’m sure when you bought an iPhone, it came with iTunes and that app where you could move drums and instruments around to play music. But eventually, the App Store emerged, and developers started building all kinds of different applications. I can tell you now that the vast majority of the experience on your phone doesn’t come from the operating system itself or the OEM—it comes from developers. It’s a platform; it’s a computer. In fact, the smartphone is the largest platform humanity has ever created.
Alright, now let’s draw a parallel with AI. AI arrived. The first applications we saw were chatbots. People would go in, ask questions, and get answers. Then they started doing more advanced things, like ‘summarize this for me.’ After that, many OEMs began asking, ‘How can I apply AI to my camera? How can I apply AI to these various features?’
There’s a mix of usability. Some things are extremely useful, while others are still in early stages—but then something happened.
Those orchestrators and agents emerged.
Then everything started to change, because now AI can become genuinely helpful—it can do things for you. I believe we’re at this inflection point, which is why I say 2026 is the year when these developments will start to become clear and these agents will begin to scale.
So much so that between now and summer, you’ll see every operating system vendor discussing how they’ll embed these orchestrators directly into the OS—because agents are the new apps, and those orchestrators are the new app stores. We’ll see old and new coexisting on the same device, because with smartphones, you carry only one device—you don’t carry two.
Then we’ll see other categories of devices that are purely agent-driven, unburdened by legacy constraints—such as smart glasses, personal AI devices, and AI pins.
I think that’s precisely why I like drawing this parallel: it shifts from how OEMs or operating systems integrate first-party AI capabilities to what I see as infinite possibilities, as people start creating their own agents—and those agents begin acting on your behalf to operate devices.
We’ll discuss those new types of devices later. But my first question, following that response, is: does this mean apps are dead?
They’re not dead—but apps will change, absolutely change.
I want to go back to something I said two years ago. We all have banking apps—you have a banking app.
You’ve chosen your bank and installed its app. So, there’s a developer who imagines how this app should appeal to the majority of bank users—what color scheme to use, how it fits your device screen. It comes with instructions: when you launch the app, it says, 'Here’s my checking account, here’s how much money I have, here are my recent transactions, here’s my credit card, and here are my investments.'
The way it will be accomplished is the way most people would prefer.
Alright, now that app on your phone—your banking app—has your credentials. So it can go to the bank, authenticate on your behalf, and bring the information back to you.
Now, you could have an agent to which you give your banking credentials, and it would work through the same application experience. Let me describe it for you.
You’d say, 'How much do I have in my checking account?' It would respond, 'Here’s how much you have. Here are your three most recent transactions. Can you plot this for me?' And if you then said, 'I want this information displayed on my screen, but I don’t like the app’s color scheme—I’d prefer different colors,' it would render it exactly as you requested.
So that’s the distinction. That’s the fundamental difference in how applications will evolve into agents—they’ll become more powerful, more customizable, and integrated into everything. For example, you’re at a restaurant, looking at a QR code on your bill. You could say, 'Pay this for me. Please deduct it from my checking account and notify me once it’s done.' That experience would differ from you opening the app yourself and completing the process manually.
Among the many important points you’ve made so far in this episode, you still believe smartphones remain essential.
Kharpal:
Do you believe it will still be at the core of the agentic AI world, and how do you expect our relationship with smartphones to change?
Amon:
Yes and no. So let me step back a bit and talk about how our digital lives—particularly from a consumer perspective—will function.
We are currently entering a world where the absolute center of our digital lives is our smartphone—the device we are most inseparable from. Everything revolves around the smartphone. It’s as if the smartphone were the sun, with every other planet orbiting around it. This extends even to so-called smart wearable devices, whose purpose is to extend the functionality of your phone onto your body.
Take a simple example: the smartwatch. It tells you the time, sends sensor data from you to your phone, and delivers notifications back to you from your phone. The integration is so deep that, in practice, you see the same smartphone companies also developing many wearables and their associated ecosystems.
Now, there’s a major shift underway.
In the near future—very near, especially as agents begin to scale—the center of your digital experience will be the agent or agents you choose. Everything will revolve around the agent.
Smartphones will orbit around the agent, as will new categories of devices—which we’ll discuss shortly. The agent will be the entity that understands human intent and acts on your behalf.
Thus, the focal point of your digital life is shifting. That will transform our relationship with devices—and with that, I’ll answer your question.
Will smartphones disappear? Absolutely not. Smartphones are extremely useful devices, and their scale is immense. However, we will use our phones in slightly different ways.
Recall when smartphones first emerged, some claimed laptops were dead. Not at all. Laptops never died—they’re still here and remain powerful. But the way we use laptops has changed. Workloads have shifted. I like to cite this example because it’s easy to grasp: we used to conduct e-commerce on laptops. We used to engage in social media on laptops—that’s where you went to access Facebook.
Today, most e-commerce and social media activity happens on mobile devices. So your phone will still be there, but certain workloads will migrate to new categories of devices because interacting with agents—and with those new devices—will simply be better for you.
To summarize: your phone will still be there. Your phone will evolve. It will have even greater computing power because you’ll still do things directly on the phone—but you’ll run agents that operate the phone on your behalf.
Then, some of your experiences on your phone will migrate to new devices that interact with you more naturally through human language and vision.
So, let’s talk about those devices, Cristiano, because you mentioned several in our conversation—perhaps we should start with glasses. After all, as you know, this is a project you’ve been collaborating on with Google and Samsung, and we know Samsung will launch them this year. Do you think glasses will be the most important next device?
Absolutely, 100%. Look, the numbers speak for themselves. We’re already shipping tens of millions of smart glasses annually. We expect that within a few years, this could reach hundreds of millions—and potentially rival the scale of smartphones, which, as you know, number around 1.2 billion units per year.
The reason it’s such a clear winner is, first, that people already wear glasses—some of us do—so there’s already a large-scale eyewear industry in place.
Second, glasses are ideally suited for cameras—when you turn your head, the camera sees exactly what you see: ‘see what I see.’ They’re close to the ears for speakers and near the mouth for microphones. They’ll also be able to read what you’re reading. So glasses become a very natural interface for interacting with models through multimodal inputs—I believe primarily natural language and vision. It’s then easy to see how certain low-friction workloads can seamlessly migrate to glasses as we begin to scale.
Going back to what I said earlier—about the role of wearables in a smartphone-centric world versus their role in an agent-centric world.
It’s very interesting. When the first smart glasses launched, their purpose was to take photos and post them to Instagram or generate Instagram Stories. Essentially, they were just an extension of your smartphone camera, with the added ability to play music—that was it. But now, with agents, you gain new capabilities every week. Each week, you can integrate another cloud service, and its evolution becomes entirely independent of the phone’s development cycle. I think this is precisely what we’re describing—and as you know, I’m extremely bullish on glasses.
Now, the initial wave of smart glasses focused heavily on integrating cameras, but there’s been a lot of discussion about incorporating displays—augmented reality-style or mixed reality-style visuals—and enabling control through hand gestures and other modalities. What major innovations do you foresee for smart glasses in the coming years?
Yes, exactly what you just mentioned. First, there will be different categories and price points. I think the most fundamental capability is the model’s ability to consume and interpret images and audio. But then we’ll start seeing many exciting technologies emerge for displaying information directly in the glasses, and I believe that will become mainstream.
So, I think the first question is: how do you receive information? How do you get text? How do you get directions? How do you overlay information? There are many very interesting display and projection technologies under development for this purpose, and I believe they will become mainstream. Another key aspect is how you’ll leverage technologies—many of which we originally developed for virtual reality headsets—such as eye tracking and tracking objects around your hands, which will create intuitive control points, especially when visual information is being displayed back to you.
Kharpal:
Another device you mentioned in this conversation, Cristiano, is the pin. I recall there was a viral moment for the pin a few years ago, and it failed. Was it simply too early? Or was there no demand for such a device? Does this kind of device have a future?
Amon:
Yes, I think we all know the answer now. It was too early—because that was before GenAI, before orchestrators and agents existed. But now that you have them, you can see its value. That said, aside from glasses, other form factors haven’t been settled yet. I believe we’ll see many different form factor experiments. We’ve been working with every cloud company and AI company. Right now, we have over 40 designs of these kinds of devices, and I can tell you the range of form factors is extremely broad. You’ll see jewelry. You’ll see earbuds with cameras. You’ll see pendants. You’ll see pins. You’ll see watches. You’ll see other objects people carry. We’ll see how this evolves. But the principle is this: it’s something you wear, something that’s always with you, something that can see the world around you—so you have context and can access agents and interact with them. Within the next few years, we will definitely see all these devices hit the market, and we’ll learn what people actually want.
Kharpal:
Cristiano, I know it’s an open secret that you’re collaborating with OpenAI on their device, and I realize you can’t say much about it—but this underscores how non-traditional companies, which previously didn’t venture into consumer devices, are now rethinking consumer hardware for the AI era. You mentioned you have all these different designs underway. Looking ahead, do you expect more companies that typically aren’t in consumer electronics to enter this space because they see the opportunity in AI devices?
Amon:
Absolutely, 100%. Let me explain why this is happening. I’ll touch on a point I made earlier, Arjun. Remember when I told you that when the smartphone was at the center of your digital experience, it favored vertically integrated ecosystems—where one company provided both the phone and all the wearables?
But now, when the agent—or a collection of agents—is at the center of the ecosystem, you might see more horizontal plays. You could have multiple devices because they’re loosely connected—they all link back to the agent. This creates opportunities for many companies to enter the space.
In fact, if you look at the eyewear humans buy, those are fashion accessories. Those fashion brands are becoming tech companies—they’re not necessarily consumer electronics firms. I’d like to circle back to your question about why these companies are getting involved.
As we undergo this transformation, first came model training and AI entering data centers. Then came inference that generates tokens. But now, you have agents driving demand for those tokens.
Those companies realize that with the rise of agents, AI will be used by humans—not by humans going to data centers and ringing the doorbell to say, 'Give me some AI.'
So what’s happening is that all the devices we wear become endpoints for agents, and those AI companies understand they must win those agent endpoints.
And there’s one final piece of data that no one is talking about yet—but we’ll remember this conversation years from now when people start discussing it.
The reason we have AI today is because we’ve digitized all information—our books, everything we’ve written, emails, content on the internet, social media posts—all this information, all web pages, are digital, so now you can train models.
But once we start using agents daily—walking around wearing smart glasses and capturing everything we see—the volume of that data will be enormous, growing exponentially compared to the data we used to train current models. So those companies want access to this data because it’s critical for training future models and providing context tailored to you, as your consumer experience will be highly customized.
All of this explains why those companies are interested in building devices. We’re incredibly excited about all the chips we’re making—from earbuds consuming less than 2 milliwatts to 2000-watt chips—as we move into data centers, all of these become relevant to every one of those companies.
Kharpal:
Right. I’d like to talk a bit about chips—not the edible kind, you know, which means something completely different depending on whether you’re in the UK or the US.
But I’d like to discuss a term called SoC, or system-on-a-chip.
Now, this is actually a single piece of silicon with various types of chips on it. You’ll hear Cristiano Amon mention some of them. So, I’d like to quickly explain what they are. The first is the central processing unit, or CPU. This is essentially the brain of a device. It handles all general-purpose computing tasks, such as running applications and the operating system.
The second is the graphics processing unit, or GPU. It renders graphics for games or the device itself. Another is called the neural processing unit, or NPU, which handles AI workloads. All of these are integrated onto a single piece of silicon and then packed into your device—and that’s precisely what Qualcomm designs.
However, that is manufactured by other companies. If you’ve listened to previous episodes of the podcast, we’ve discussed extensively the challenges in the chip supply chain.
But right now, there are significant bottlenecks throughout the semiconductor supply chain. One of them is memory. All your devices—laptops, smartphones—now include some form of memory.
What’s happening is that only a handful of companies in the world can produce these memory chips. Currently, all of this capacity is being directed toward AI data centers and AI chips. As a result, there’s a severe shortage of memory for consumer electronics. This is a major topic of concern across the entire electronics industry right now—and it’s exactly what Cristiano Amon and I discussed.
At the heart of all this are the chips that power these devices, and Qualcomm is certainly one of the leaders in this field. With the surge in new types of AI-enabled devices, what does this mean for how you think about chip architecture? There’s a strong push to integrate more functionality onto a single chip—to make them more powerful and more energy-efficient. So, when you consider all these new devices, what does this imply for the future design of chips?
This is an incredibly exciting moment for semiconductor companies—especially for a company like Qualcomm.
Amon:
We’re honored—truly honored—to have developed our capabilities in the mobile industry, because it’s an extremely demanding sector. With each generation, users want more powerful phones that can do more—play games, feature large GPUs and CPUs—but it still has to fit in your pocket, last all day on a single charge, and not overheat.
I can’t install liquid cooling or heatsinks inside a smartphone. So this has instilled a kind of DNA in our semiconductor engineering: you must pack more computing power into a smaller space in an energy-efficient way. And I think that’s a fantastic challenge for a company like Qualcomm.
The same applies to smart glasses—how do I pack substantial computing power into them while maintaining highly efficient battery life? So I believe this fundamentally changes the architecture. We are significantly enhancing the capabilities of neural processing units (NPUs), which will run smaller models in coordination with large cloud-based models.
This means a significant increase in silicon area dedicated specifically to running these models—both on GPUs and neural processing units (NPUs).
We need faster and more energy-efficient CPUs because agents drive CPU usage. We also need intelligent management of connectivity and battery life.
Our entire roadmap is now undergoing an upgrade—our entire roadmap. Because I believe none of the devices we have today are ready for this future. You’ll have to upgrade your phone. You’ll upgrade your PC. You’ll adopt new categories of devices, such as smart glasses and pendants. Your car will be upgraded, enabling you to interact with agents inside it. I believe this transformation will occur across all the things we interact with. It’s a new form of computing, a new form of software, requiring new silicon architectures.
Kharpal:
How important will CPUs be? This has been a major point of debate—people are talking about a CPU renaissance, not only in data centers but also as we consider these consumer devices.
Amon:
Extremely important—100% correct. Going back to the example I gave you earlier, when we start discussing agents, you have massively parallel processing from GPUs and even NPUs, but the orchestrators handle very complex tasks. They’ll check your data, inspect models, access the internet—that orchestration is all done by the CPU. Those orchestrators run on the CPU. So right now, all boats are rising. At Qualcomm, our CPUs have become critically important, and we already have industry-leading CPUs in smartphones and PCs. Very soon, on June 24, we’ll announce our data center CPU. CPUs will be extremely important.
NPUs optimized for inference—and especially for energy-efficient inference—will be very important, and GPUs will remain crucial as well. That’s why I believe this is a great time for the semiconductor industry.
Kharpal:
The semiconductor industry is experiencing a massive surge in innovation and demand. Naturally, this has led to numerous bottlenecks across the supply chain—both in memory availability and in actual chip manufacturing. From the perspective of foundries, there is also a major debate underway about whether the cyclical patterns historically observed in the semiconductor industry are changing. How do you navigate all these different challenges, particularly those related to memory and capacity constraints?
Amon:
Well, first, before I answer your question—if you have memory, I’ll buy it from you. Okay.
So, I think the industry is unfortunately dealing with a situation similar to what we experienced during the pandemic, where demand far outstrips supply, and memory has become a bottleneck.
I don’t believe our current capacity was built for this AI reality, but I’m optimistic that the industry has already found ways to address these issues, and new capacity will come online.
Regarding the question about cyclicality, I feel this time is somewhat different. I believe the world’s entire computing infrastructure is undergoing an upgrade—across the board. AI is essentially a fundamentally different way of running software. You need different and more powerful computers. What you’re generating are tokens—the new currency of the AI world—as I’ve mentioned in previous conversations. The existing computing infrastructure simply doesn’t meet requirements and must go through an upgrade cycle. We’re already seeing this demand emerge in data centers, and as AI agents gain traction and begin to scale, we’ll see similar demand arise on other devices as well.
I wish we had more memory. We’ve been working closely with the industry to secure additional capacity. The same applies to semiconductor manufacturing—it’s a great question, you know.
I’m quite optimistic. Some people say it won’t be resolved until 2028, but I’m more bullish. I think at some point in 2027—or at least in the second half—we should start seeing more available capacity. New architectures will also enable more efficient memory utilization, creating opportunities for companies like Qualcomm, which has always been highly efficient in power consumption and memory usage. We’re building edge devices capable of disrupting data center paradigms—that’s exactly what we aim to showcase when we unveil our roadmap at the end of June.
With all the innovation and interest in semiconductors, there’s now fascinating competition emerging—particularly in the data center space. How do you see competition evolving in consumer electronics? After all, we haven’t seen many new entrants in that segment yet. Apple, for instance, has long focused on designing its own chips for its products. Google uses its Tensor chips in Pixel devices. Samsung, of course, is your partner but also uses its own Exynos chips. Do you anticipate more competition and new players entering the market as new devices and form factors emerge?
That’s a fantastic question. Look, I can’t predict the future, but there’s one thing I can tell you—I can predict this: the shifts we’ve witnessed in the mobile industry, from 2G to 3G to 4G to 5G, were transformative. Players changed, devices changed. And although we haven’t mentioned it in this conversation, we’re already working on 6G—the next generation of wireless technology designed specifically for the AI era and those new device categories.
So, what we are now beginning to see is that when we talk about personal AI devices—and we’ve reached this point—we’re witnessing a shift among the players interested in the mobile industry. Now, all AI companies are focusing on mobile devices as endpoints, and they are building hardware.
So, one thing I’m starting to see now is the convergence between smartphone OEMs and AI companies—between smartphone OEMs and models. I actually believe this landscape will look very different because workloads on phones are increasingly being driven by agents and models, and the players involved will be very different. Therefore, I think this industry will undergo transformation. I can’t predict the winners and losers, but I can foresee change, and I believe it’s highly likely to reshape the relationship among smartphone OEMs, operating systems, and agents—that will change.
On the silicon front, look—we’ve always had competition.
We’ve always had competition, and I believe competition has always been great and incredible for these industries—which is why I think data centers will also see competition.
The rule we follow—I think what’s unique about Qualcomm is that while players come and go, and we’ve seen many companies in the mobile industry go from heroes to zeros, we remain. We stay because if you have the right technology—if you have leading-edge technology—there’s always a place for you.
Today’s success does not guarantee tomorrow’s success. So, we must continue to reinvent silicon and innovate, and I believe that’s exactly what we’re facing amid the transformation of the mobile industry and AI-powered consumer devices.
Kharpal:
Cristiano, I always learn so much from speaking with you. This has given me a lot to think about, and I’m excited to see some of the things you hinted at in this episode unfold over the next few months and roughly a year. Thank you so much for joining me.
Amon:
It’s my pleasure. It’s always a joy speaking with you, Arjun, and I look forward to seeing you again soon.
Editor/Deng