AI Agents, 6G, and the Future of Computing

Auto-generated from YouTube captions and lightly formatted for readability. It may contain transcription errors. You're going to have now two users of your device, you and your AI and your agent. The agent will understand your intentions and the agent will basically go do things for you on your behalf because now the computers kind of understand our intentions and communicate the way we communicate with each other. I think most people think of 6G and they think, "Oh, cool. I'll get faster internet on my phone." Is that all 6G is or is there more to it? 6G is probably going to be one of the biggest and most different transitions I think we we're going to see about the wireless industry because I think the biggest statement is 6G is actually going to be the wireless technology for the age of AI. Today I'm sitting down with Cristiano Amon, the CEO of Qualcomm, the company behind Snapdragon and much of the technology powering modern phones, cars, computers, and wireless networks. He has a front row seat to where AI is heading next. Not just in the cloud, but inside the devices we use every day. We talked about AI agents, the race for computing power, Qualcomm's move into data centers, and why 6G could be a much bigger deal than simply faster internet. I think you're going to really enjoy this one. So let's get straight to it. So just to give like a general overview and to kick things off, can you give the brief overview of what Qualcomm does and what's the difference between Qualcomm and Snapdragon and just sort of break that down for us? Okay. You know Qualcomm I used to say that Qualcomm is probably uh the biggest company that nobody knows about it. But I think Qualcomm is the home of Snapdragon is the home of Dragon Win is the home now of Dragonfly. So Qualcomm is a semiconductor company. We we started to get to be known for what we have done in the wireless industry. I think that the people that know about Qualcomm and they they kind of know what we're doing in the wireless industry and uh we're the company behind the fundamental inventions that made 3G a reality 4G 5G uh upcoming 6G and u I think every cell phone in the world no exceptions is is based on the technology welcome develop that's how we started we started with our licensing business so every every phone uh that connects to the cellar network whether it's 3G, 4G or 5G um you know take a license for some of the Qualcomm uh technology and our technology part of the standards. That's how we started. But what we really are is a very large semiconductor company. We ship about 40 billion components every year. It's probably more than Starbucks sell cups of coffee I think last time I checked. And uh and our chips are in kind of everything. Our chips are in processor for phones. the Snapdragon. Um, our chips are now in into the digital cockpit of cars or in wearables are entering PCs, uh, but also going to industrial to robot to robots and and soon the data center. So, that's what we do. Amazing. And how many patents again does Qualcomm have? Over 200,000 patents. We were we're we're primarily um an engineering and technology company. I think that's uh the company's deep rooted in technology. I think that's as I said that's how we started right and uh just on 2025 26 is not over yet we're just halfway there but in 2025 we're the number one company in America in invention disclosures uh you know patents applications and uh it's just uh uh kind of reflects I think the innovation and technology engine of the company. Amazing. Amazing. Yeah. I mean you you mentioned a lot of people don't know uh what Qualcomm is in San Diego here. I think everybody knows what Qualcomm is because uh [laughter] you know we used we used to go to Qualcomm Stadium to watch the games. Now we go to Snapdragon Stadium to watch the games. So I feel like in San Diego we're all we're all very familiar. But it's um you know it's exciting to to to see what Qualcomm is doing globally and the amount of patents is just is absolutely insane. But I want to get into talking about agents because I know that's the big sort of story in the world of technology right now. How would how would you describe what an agent even is? Look uh by the way this is a a very exciting I think uh uh topic I think for Qualcomm because I think agents is what really creating a lot of opportunities for us. But the way I would describe it to you is we we kind of spoke about this uh kind of two years ago and I think a little bit before you know it was popular. I think uh when you think about the devices that we use it like in uh we use a phone, we use a PC, you know, you have those constructs that you have an operational system and then you have applications and and but agents exactly how AI is going to get scale. So I'm going to give you I'm going to give you maybe an answer to your question. Thinking about this big milestone that actually happened in the AI industry that was uh I think it was it was a important milestone because it provides clarity how AI is going to get scale which is the the orchestrator called open claw. I think people heard about openclaw and openclaw allow you to create an agent. the agent will understand your intentions and the agent now will basically go do things for you on your behalf and I think that's how we should be thinking about agents we should be thinking about those orchestrators as kind of the future operational systems of devices for oss and we need to think about agents as applications the future of applications um or even some other people think about different skills applications but let me give an example like what agent will do in in your daily life at those things get scale and so I don't make it like an abstract uh conversation. So uh we have uh seen now some some capabilities being developed and uh you it's going to be like that. You're going to be working and then there's your calendar and maybe you have uh you know your outlook calendar and then uh the agent is going to [snorts] tell you it's going to say Matt I just noticed right now you have a 10-minute break. Can I talk to you about your schedule? And you're going to say yes. and they're going to say, "I just saw that it pop up on your Outlook this great interview with Cristiano Roman from Qualcomm uh that is conflicting with your spa appointment. Uh which one's the priority?" And they're going to say, "Of course, the Christian interview is more important." They said, "Can I reschedu your spa appointment for you?" Yes. And we'll actually call right and negotiate a new time and put the cut. So the the agent will do things for you because now what you have with those large language models, large visual models, it changes the main machine interface because now the computers kind of understand uh our intentions and communicate the way we communicate with each other. And once you understand an intention, the agent now or this orchestrator is free to go to the internet do things. You have things like not to get technical but like MCP protocols that it can communicate with the M. you go to your computer and the agents is how AI started to get scale and it started to get um very useful for you. The agent happens because you now have a number of technology intersections that are coming to this point. You used to have one model that does everything. Now you have mix of experts different models. You have a physics engine. You have a model for math. You have a you have large uh visual models, visual action models. And then you have an orchestrator that can orchestrate this and you can you have agents you can create agents and agents can talk to agents and uh and that started to um you know change and I think uh you started to see this happening now but other companies for example Microsoft talk about project solara uh talk about the evolution of copilot to agents you talk about Google's talking about Gemini intelligence so uh it's it's exciting and I Look, I I can probably talk to you about uh the whole day about this, [clears throat] but that kind of changes a lot. Yeah. Yeah. No, I mean, it's really interesting. I think we're seeing this this thing too where these different models people are finding they're they're good for different things, right? Like maybe the OpenAI model is good for this and the anthropic model is good for this and the Google model is good for this and you have these agents who can now decide, okay, well, I'll send this kind of task to this model and this kind of task to this model. Um, I'm curious why does these new agents, why does that increase the demand for compute? Okay. So, I'm I'm going to answer this question, but you brought something interesting in what you said that I think I I should highlight. It's interesting. So, you talk about different uh and I'll get I'll get to your to your question, but you think about different um AI companies do different things. For example, Entrop is good uh for very good for code. You look at cloud code and people use them for coding. Now the interesting thing is if you think about what are the total number of uh folks in today they're you know writing software and of course cloud code and those other things will increase that but there are six billion people that uses phones for example or use and uh probably two billion people with uh or computers uh code is just one application you go I think you're going to see this thing getting democratized and diffused so fast because there going to be other applications and a lot of people going to be using now to that why those things create a lot of demand for tokens and and I'm going to try to give you an example especially how we think about the devices that we use every day the phone has been designed for you to operate and you go operate at your speed and you go to your OS and you pick a app and you do something with one app and it that's how I've been operating Right. The same thing the computer it's going to be uh you are operating it. Even how we thought about how AI was in the beginning there was a chat box for you and you go to the chat box and you ask chat DBT something uh and it's very transactional and is you but now once you have those agents the agents is operating at no at your speed at their speed. The agent is just doing one thing at a time. The agent is doing multiple things. So how the agent starts to use the computer and hows to use the model use uh in in a much higher intensity than we use and what we saw was this exponential growth of tokens because agents are busy. That's how people for example when open claw and that's why open claw was such an important milestone when people bought open claw put in the computer and ask you start communicating with the agent via WhatsApp or or discord and a message and telling them go do this for me go design a web page go do this and the agent starts to go in to different models to the web go to your to computer that consumption of token is now at agent speed not at your speed and I think that's what created this massive demand and he started to change also the architecture of the whole uh thing AI infrastructure I you know we can talk about that but it's kind of changing it's not just about GPUs now you have CPUs to orchestrate things you you have a lot of inference it's not a GPU you have dedicated and it's also creating this whole concept of a hybrid AI which is a whole different conversation [laughter] amazing um so I think a lot of people think that all of this just needs data centers right we're we're typing our prompt it goes to a data center and comes back you know Why why aren't data centers enough? So, um I am going to make a uh a little bit of a provocative uh statement. I think uh I think that I think it's kind of a I' I've been I've been hearing this uh for the past several years. I kind of find a useless discussion. I'm going to tell the reason is that it's almost like people keep arguing about oh this is the data center is enough or or some people are going no I can do this on the edge let's edge is everything outside the data center like the phones the cars the computers uh even like uh even an on-rem server at an enterprise but people say oh we can do this on the edge and there's debate is it edge or it's or it's cloud edge or cloud and I think it's a useless debate I think because the way computer systems. Uh it's almost like go back all the very early days of of PC when big IBM the mainframe said this will never happen. Everything's going to the main frame. It's not true. The cloud will continue to grow and will grow exponentially but the software the application the agents will find its more efficient compute and it's going to go everywhere. Let me give you an example why I'm I'm I'm making this very clear statement on this because I actually think resistance to the edge is futile. I'll tell the reason why I say that. So think about your phone today. Mhm. Um, if you buy a a new phone every year, and I want you to do it, and hopefully with the latest and greatest Snapdragon, but if you buy a phone every year, you're going to go see the specs, and the new phone has a faster processor, a faster graphics, has a GPU, has a new AI processor, it probably generation to generation increases 20%, 30%. And uh, let's And you have, I don't know, 300 apps in your phone. Mhm. What if we we want to have this conversation said, "Matt, let's just go one by one. Let's go write every one of your 300 apps on your phone and let's go have a conversation about what portion of the app is done in the cloud or is done in it almost like doesn't matter, right? It just works." And and the reason I tell you it it is like that because if you get this new phone with a lot of processing power that you bought with the latest processor and you put on airplane mode, it's useless. It's useless. Like you're not going to be you're going to put it back, right? Um that tells me that it's not about edge, it's about cloud. The whole computing system is designed that way. Almost think if you have Microsoft Outlook, you have a client and you have a server. So what's happening as AI becomes a new form of compute, it's going to happen everywhere. And and I'll give you now some simple examples to to make sure people understand this. Let's say that you have those um orchestrators and agents going to be doing things for you and then you give this orchestrator a task and the orchestrator is going to go uh to the cloud to go to your computer and it's going to do that task. Today what you started to see this is one I'm going to give you one example there's many others but today you were starting to see right now the demand for token is exponential. We just talk about that with agents and those things cost money. So you started to see for example companies like OpenAI saying I am not going to continue with Sora to produce video because I have a better use for my tokens that I can monetize with code with codecs and other things right so then you also have a lot of small models open source models that do different things. your agent, your orchestrator is going to say, "Oh, for this thing, I can do right here. Uh, it's cheaper. Uh, I don't have to pay anything. It you, for example, there was this huge debate about cloud and edge. Microsoft at the latest bill talk about unmeter intelligence. Use your PC." So, and that's how the world is going to evolve. I think that's what I mentioned about this hybrid AI. I think you're going to be able to have agents that are going to do things on the cloud that they have to be done in the cloud. They're going to be things that gonna be on device because they have to be done on the device. The device has perception, knows your context. Maybe the stuff that you authorize it, you authorize, you're going to mess up on my apps. I just want it to be done on I want the data to stay here. And maybe there's privacy consideration. There's cost consideration. And this is going to develop naturally. So what you're going to end up with this fabric of compute u I think a lot of people talk about the compute continuum that you're going to be having done AI running everywhere and it's going to run on the cloud is going to grow at the edge and I think the simple answer is what it has to run on the edge is is going to run at the edge. What it has to run on the cloud is going to run on the cloud and uh and that's the way this is going to go. Yeah. So, I'm I'm curious why. So, in in your most recent announcements, you um you guys talked about how you're getting into data centers and you're starting to build hardware for data centers. Um what what's behind that decision? Why are you guys deciding to get into data centers? Well, first of all, I think it's it's interesting. A lot of people ask me, "Oh, is it late for Quan to get into the data center?" I look, I'm maybe maybe it's time to do another provocative statement. Anybody that will declare who the winner is on the AI era right now on anything is probably speculating because we are in the first innings of this. It's almost like they'll go back in the early days of the internet and I'm going to tell you Matt categorically the winner of the social network is MySpace or it's going to be Orcut. [clears throat] Nobody knows or you can say the winner of map is Map Quest. There's no question about it. No. So I think uh across silicon across models we're just at the very beginning. So Qualcomm saw a couple of things that were happening in the data center that creates the right entry point and that's we've been developing in the past two years uh for for doing this and I think we came up with a comprehensive solution uh it's we have a dragon for every market. So we have dragonfly uh it's uh for the data center we develop a comprehensive solution because of those things that we talk about it the data center is moving from training to inferencing right that so then you're doing inferencing to run those those put those models to work that's one the other one is the data center needs a CPU for those orchestrators uh and the inference is going to be disagregated because you need a lot of performance at the lowest possible cost and the number three thing is you have something that uh we've been dealing with this for decades because of our phone experience which is this is the demand for compute this availability of energy so how you close that gap right and and I'll tell you there's something very interesting about the phone the phone is like this every year I have to increase the computing uh capabilities about 20 30%. But the battery is has to be smaller because you want your phone to be smaller, has to fit in your pocket. I it has to be cool, has to last all day. So we have to come up with very power efficient solutions. So this create an environment for Qualcomm to enter. And what we do in a data center is um we're we're building what we think is going to be the highest performance, lowest power consumption CPU for orchestrators uh that we that we announced. We're doing an an accelerator uh for inference and you see in the accelerator for inference the Qualcomm efficiency and power DNA coming true. We designed a completely different memory solution that doesn't require uh HBM. I think if you see what's happening with prices of memory, you have the memory Armageddon that is going on. Right. Right. I saw something that was uh on the internet and that I thought was hilarious. I think uh somebody put a picture of a DRAM card and he put a picture of a Dodge Ram and he said this this the DRAM that I want. This is the DRAM I can afford. Uh but but but I think it's because there's so much more demand for memory and I think what we did we said let's create a different solution that is more efficient in power and cost. So we kind of create a solution we call high bandwidth compute. They integrate the logic and the memory in a way that you have to move the electrons and and uh consume power generating efficiencies. So all those things point to the reason we enter the data center because I think it's the market is now ripe I think for Qualcomm uh to be able to participate. So with the with what you're building in the data centers, is is that going to help solve some of the the power constraints that are happening right now? I look as a new entrant, I think what we offer is a significant benefit of the total cost of ownership, how to get the maximum amount of performance [snorts] for the power you have available and and hopefully start to make uh um AI more affordable. Yeah. So, you know, along these same lines, you know, you're talking about mobile phones and there was a, you know, another mobile phone company out there that's, you know, a very big name and they're talking about how they're raising all of their prices. Um, you know, how is how is, uh, you know, Snapdragon and Qualcomm, how how are we going to make sure that this this compute is available for everybody on device? Look, we have a short-term issue and I think uh the issue is not it's not fault of their own. And I think what happened is the demand uh for AI and especially on the data center so high that uh all of of the memory went to the data center. If you look at the amount of uh uh of the expenses in the data center I think uh how some of those products are being sold into the data center they're very expensive. So uh it kind of memory vendors decided that's a more attractive market for them. Mhm. So then the rest of consumer electronics is being probably starved of DM and chips are in everything. Chips are in cards and phones and in appliances and uh so the price of memory went up and I think nobody nobody that is in in the phone business in the PC business is immune to this. So I think what we see there's a shortterm um uh I think uh situation that is memories are not available and they're too expensive and I think that is driving to higher prices of things that people buy uh like their phones their their PC even some other consumer electronic devices they all being impacted by this now that's the short term I think we all hopefully that more capacity is going to be built available and uh there's going to more supply and those things going to go back to normal because the f the phone market does not have a demand problem. It has a memory availability problem, right? Uh the second part of the answer is the exciting part. The exciting part is uh those devices, phones and PC, they're all going to become intelligent. They all going to be running uh you know agents and and you're going to start to see the coexistence. So you're going to be operating your phone, but then the agent's going to be operating your phone as well for you. do the same thing with your PC, the same thing in your car when you're behind the wheel and that will create an incredible diffusion of AI will democratize AI and actually will will increase the scale and I think we know once you start to increase the scale and it's every computer that is available in the world started processing AI AI is going to become more affordable and it's going to become available for for every industry for every person for every application. Yeah. So, I mean, maybe all of this extra AI that's going around, all this extra intelligence could solve a lot of these short-term problems that we have. Well, I [laughter] think the short-term problem is I think uh I don't know. I have an opinion about it. I think uh you know, you have a lot of other established industries. I think uh it's important I think for the memory uh you know, suppliers to support existing industry established and uh we need uh hopefully they're going to use I think all of the uh incredible earnings that they have to build more memory factories. Right. Right. Let's let's talk about the the modular acquisition. What does what does that do for Qualcomm and what does that do for you know customers of Qualcomm? Look, I'm incredibly excited about acquisition. I'm going to say I don't expect and and I'm not I'm not going to say this to uh you know and I understand what the challenges ahead. I'm not saying this to be pretentious or anything like that, but I don't think not everybody understand it. Uh but I think the modular acquisition I am incredibly proud of it. I think we're, you know, we can't wait uh uh for that process to be completed because uh this is an incredible team. This is a team that had a very a very simple uh and and mission which is to democratize uh AI to create a platform that is available to everyone that is open for developers to have you know a simple platform that they can target yeah any hardware any in any environment data center or the edge and uh and and make I think AI uh uh available for everyone which is kind of a a mission that really resonate with Qualcomm. I think I have always been a big believer that closed systems uh are not the ones that win at the end. I think uh open horizontal systems are the ones that win at the end. I don't think it's never the job of one single company to innovate. I think it's uh if you actually have an open platform, you allow everybody to innovate and I think you you see the best uh of human ingenuity and I think that's what we're trying uh we share that vision of modular and then we're trying to make sure they give scale is to create an open AI platform uh that's anybody can use it any hardware can use it. it it doesn't have to be on Qualcomm hardware and I think what the industry wanted I think our our ambition is to make modular the Linux of AI and I think that's why this is very significant and is also good for the industry. Gotcha. Cool. Well, we're running out of time here so I'm going to I'm going to jump to some of these uh questions here. You said um the future of devices might have two users um agents and humans. What what does that mean? Well, that mean just think about uh uh you know, just think about your phone on your PC, right? Uh and uh you're going to be using it, but then you're going to be telling your agent to do things for you and he's going to be using it. And I think we're going to started to see uh those things uh they're going to start to coexist. And the reason is easy to see that. I go back to that when I talked to about Open Claw. A lot of people bought a separate computer, put open claw running, right? and and you were operating your computer and openclaw is operating the separate computer and you message to to open claw do this for me do that but the interesting thing when you look at that in your office what happens when you walk outside you don't carry two computers you just carry one phone so which means those things are going to fuse into one device and I think the phone is going to set the direction how humans are going to utilize it and you're going to have those two things coexistent that's why we're kind of designing the next uh processor for this future because I think there will be a big upgrade cycle coming uh for phones, for PCs and and I think that's really what it means. You're going to have now two users of your device, you and your AI and your agent. Gotcha. Can we can we talk about physical AI real quick? What does that mean? And you know, how how are you guys getting involved in physical AI? Yes. So when when I look about the future of Qualcomm and the future of Qualcomm we've kind of presented at our last investor today you should be thinking about three big business of course one is going to be like the the personal AI business that includes you know smartphones and PCs and those new smart wearables that connect to agents how people are going to use it uh the data center we talk about another one is physical AI and that's how we look about um our automotive business our industrial business our robotics business and uh and our networking business, but I think physical AI is about really AI uh dealing with the real world with uh the physical environment. So, so it's interesting um it is also a lot of that is edged. There's a lot of cloud to but a lot of that is edge. For example, when you think about a car that is drive itself is an AI that is driving the car itself. is processing a lot of uh input from sensors from the physical world and in making decisions about what the car is going to do. The other great example that is robot. A robot is kind of trained on the physical world and gets input. It has sensor and operates like humanoid robots. It's everybody's choosing to do humanoid robot because the word is designed for humans. So you just use the tools you use uh everything and that is physical AI and that's require you to also think about the processor a little bit different uh because you have to have uh you have to design processors and integrate them with sensors they need to get they have a lot of perception a lot of sensing being trained on what are those physical inputs mean it's not like trained on texts and language created by humans is trained on that and uh I actually think we have a right to win uh in those markets because uh it's all about edge. It's all about high level integration of computing and sensors and uh and uh that's is the traction that we had in the marketplace. It shows that we're in the right spot. Great. I I've got just a couple more questions. Um I want to ask about 6G really quick because I think most people think of 6G and they think, "Oh, cool. I'll get faster internet on my phone." Is that all 6G is or is there more to it? Okay. So 6G is probably going to be one of the biggest and most uh different transitions I think we we're going to see about the wireless industry. And I think for me to tell you what about uh 6G is I need to tell you about every generation of wireless there was a problem to be solved. And what is the problem that 6G is solving? because I think the biggest statement is 6G is actually going to be the wireless technology for the age of AI. Um so 2G digital seller the mission was can every person on earth have a phone and I think you saw when 2G happened everybody got a phone and it was uh you know that's what the beginning of the mobile revolution 3G there was an incremental change which is connected to the internet and you probably remember your Blackberry and uh and so forth 4G it was a big change broadband which will transform your phone into a computer and you know what the smartphone revolution is um it became the largest development platform ever created by mankind. It was a big deal. 5G was another incremental step which is now allow you to have unlimited data rates, allow you to have high-speed video streaming, mission critical capabilities, enable other industries. So what is 6G? So 6G is created for the age of AI. We spend time talking about agents and how new classes of devices, how uh you have an agent operating your device, understanding communicating with you the way we communicate to each other. We also talk about physical I'm going to put everything together with 6G. So one of those new classes of the mobile devices that uh is going to interact with agents are glasses. Mh. So glasses it's uh a very important real estate because if now your human uh machine interface is how we interact with each other because now AI understand uh what we see what we read what we say. So then I need to design a wireless interface that see what I see. So so if we have glasses with cameras and and everything I see the AI see so I can talk about it. I'm talking and I said uh and asked my agent, who's this person that I talk to? And I say, well, that's Matt. I just check him on LinkedIn. It's Matt. Uh so I need to have see what I see. So I need to create a wireless uh technology that will be able to send highdefinition video to the cloud all the time. So on the connectivity side, uh 6G will transform all of us into walking cameras in this world. But that's only one portion of 6G. The other portion of 6G is now we talk about physical AI. So the radio environment is is a physical uh uh you know uh uh input for models that can be trained on on how that radio environment works. So now you can create a a real time sensing network. Think about radars everywhere like a radar at scale. So you can detect everything that moves, everything that flies. Uh these days when I say drone detection, everybody gets very interested. Uh but you can you'll be able to detect drones, manage the low altitude economy, but also see uh everything that moves in the roads, every bus, every cars, every person, every bicycle. Those things are real important context for models. They're important for robots. They're important for autonomous system. And I think that's what 6G is all about. It's about probably the the next big architecture of the of the wireless network for the age of AI that will enable those new classes of devices that see what we see and create a lot of sensing capabilities uh for for robots, for agents, for autonomous systems, and uh and all of those new things are going to be powered by AI. Awesome. That sounds exciting. So this is my final question. um what are you most excited about that you can do with AI right now? Uh it's hard to answer that question because there's so many things happening at Qualcomm right now from uh from the new uh you know personal AI devices, the new class of devices to robots to the change of industries to to self-driving cars to data center. So I guess I'm I'm excited about everything that Qualcomm is doing right now. Amazing. Awesome. That's it. Thank you so much. I really [laughter] appreciate it. Great talking to you. Yeah, you too.