Hi Cristiano, great to see you. How are you doing? So, how's building about this year's event and what are you most looking forward to?
Yeah, well, thank you for being here with us at Snapdragon Summit. I think the main theme of this event is really the AI smartphone. I think the AI smartphone is just happening, has happened with new agent experiences. And of course, besides announcing the latest Snapdragon 8 Elite Gen 5 family of products, I think we're kind of showing how the smartphone and the mobile market will transition to an AI smartphone.
Yeah. And congratulations on the launch of the new Snapdragon platform. The first question is, these chips are designed to support more powerful device AI. So my question is, how do you see AI agents evolving across devices over the next 3 to 5 years?
Yes. So it's interesting. I think we're going to start seeing these two dimensions of how you're going to use your phone. We always going to look at the phone as a user and we want a better camera. We want a fast UI response. You wanted just better graphics. You wanted the ability to have higher connectivity and speed for you to operate. But at the same time now agents are going to be operating your phone for you even when you're not operating the phone. And the agents want to have ability to process millions of tokens, wanted to do this at a very high speed and low latency. So when we think about the next 5 years, the evolution of agents and how much more we're going to start relying on agents to do things, we have to design Snapdragon for those two worlds.
Yeah. In previous years, we often talk about generative AI, but compared with generative AI, does agentic AI require different design?
Well, it does. I think what we've seen is when you have agents, you have like orchestrators that are going to understand the intent, plan, they will reason, they're going to operate different models, some large language models, large visual models, going to go to the cloud, use protocols to go to different services. So for all of this, you need much faster CPU because the CPU needs to be running the orchestrator. You need a much faster NPU to be able to run a lot of very capable models on the device, and you needed to have also a very efficient way to do this to preserve all day battery life because now even when you're not using your phone, the agent is using your phone.
Yeah. So what new capabilities need to be built into the SoC?
First, one of the new innovations we talked about in my presentation, and we're going to have a lot more details for that coming including at the future Mobile World Congress, is you're going to have an AI core processor alongside the AOC. We talk about HPC, high bandwidth computing, because you're going to have a lot of memory bandwidth intensive processing for those models and we need a brand new architecture.
Many AI agents still rely on the cloud. We know that, and what are the biggest challenges to running them on device?
What is really happening now is as AI is evolving, every computer is going to run some form of inference because AI is everywhere. It's no different than your smartphone today. Your smartphone today, when you're using an app, some portions of the app run on your device, some portions of the app and functionality runs on the cloud. And I think it's going to be the same thing with AI and you're going to have this hybrid AI that you have models running on your device working alongside models that are running on the cloud and you need those orchestrators to do it. So some of the capabilities you need on device to do this is ability to run much bigger models because some users or some functions through privacy and security they may want to run that on a device. Also we needed to have a lot of processing power for the orchestrator to basically route functions between the models on the device and the models in the cloud.
Yeah. So when do you expect device agents to become mainstream?
Oh, it's already happening right now. We're super excited about the approval of the DAO phone which has just launched. I think we've seen a number of other devices being built with different agents from Xiaomi, from some of the Chromebooks from Alibaba and some of those other devices. This is also true with the Meta Quest, Google Intelligence. So, this is already happening right now.
Yeah, that's right. Anyways, the new Snapdragon 8 Elite Extreme MG6 and AI is becoming central to SoC design. So, how will this change the capabilities of flagship chips?
It's a very interesting question because in one hand, I think we continue to see the flagship increasing in performance year over year. You want, you know, faster CPU, you want a better GPU, you want a better camera and you want to have more battery life. And I think like we have been building those flagship phones but now we also see a leap in the amount of compute. Like we talked about it today when we launched Snapdragon 8 Elite, you know, Gen 6 Extreme, we talked about supporting 30 billion parameter models. I see that in a very short period of time the requirement for mobile phones is going to exceed 100 billion parameter models and I think that is the kind of leap in performance we're going to see happening in the flagship.
For a long time, Qualcomm was best known for mobile phone chipsets and wireless connectivity. So how do you define Qualcomm today?
