Advancing AI 2026 | Rahul Tikoo, AMD

Clips
More from Advancing AI 2026

Rahul Tikoo

Senior Vice President & General Manager Client

AMD

play_circle_outline From Cloud to Desktop: Ryzen AI Halo Enables Local LLMs — Tokenomics and CIO Strategies to Reduce API Costs

play_circle_outline Lemonade Semantic Routing, Open Ecosystems, and One‑Step Agents for Edge‑Cloud Workloads

play_circle_outline Desktop servers concept: always-on desk PCs hosting agent swarms and services.

play_circle_outline Product roadmap: Strix Halo, Gorgon Halo, upcoming higher-memory agent PCs.

Info

Rahul Tikoo, AMD

Rahul Tikoo

Senior Vice President & General Manager Client AMD

search

John Furrier

>> Welcome back to theCUBE live in San Francisco, California for AMD's Advancing AI event. It's where all the AMD leaders are gathering, the industry leaders as well as practitioners and technologists. It's a free event, sold out, 14,000 people, packed house. I'm John Furrier with Dave Vellante, my co-host. Rahul Tikoo is here, Senior Vice President and General Manager of the Client Business at AMD. I never thought I'd say this early, but as we predicted, AI is going to the client side, which is your devices. But also they got super servers, Helios powering everything in between. Rahul, thanks for coming on theCUBE.

Rahul Tikoo

>> Thank you. Thank you for having me. Good to see you guys.

John Furrier

>> So I love this idea of having my own AI PC that can do stuff. We've done interviews where autonomous vehicles are now becoming personal. So the personal computer client devices are connecting to intelligence.

Rahul Tikoo

>> Yeah.

John Furrier

>> And developers and end users will get value out of that. That's a big driver. Give us an update on current situation with why this is happening and then what's going on with the product.

Rahul Tikoo

>> Yeah, so I think if you look at the last 3 years, we're sort of in this age of chatbots or AI assistants. And so everyone that was using a PC was using some sort of an AI assistant and that could get you 10%, 20%, 30% more productive. What's happened, and a lot of people would ask me, hey, what's the killer app? I don't know what the killer app for AI is. And I think what's happened is the killer app now is agentic AI. And so when you think about agentic, what's happening is people are going from assistants to fully reasoning autonomous agents that work on your behalf, right? And so that productivity of 10%, 20%, 30% because of the assistance is now becoming a force multiplier of 10x, 100x. And so now you can imagine a software developer that has an army of agents working for them and creating code, right? That's the world we're entering into and PCs are becoming a very important part of that world.

John Furrier

>> Yeah, we had a comment on one of my posts on LinkedIn, we brought it up on Dell today, that the human empowerment has gone supernova with AI and things like the graphical user interface WIMP Windows, Icons, Menus, Pointer, is changing to natural language.

Rahul Tikoo

>> Yeah, yeah.

John Furrier

>> So that changes what used to be load Excel, type in data, create a macro, build a graph, import it to PowerPoint. those go away.

Dave Vellante

>> Yeah, absolutely.

John Furrier

>> Now we're in, hey, just run the numbers for the quarter and give me a graph.

Rahul Tikoo

>> Yeah, absolutely. That's AI. AI agents are so smart now that with natural English language, anyone can be a programmer, right? You could be a one-person company. I'll give you a quick example, working with small businesses, right? And there are small businesses that are doctors' offices, that are marketing firms, that are law firms. They don't need to learn AI. They just need to learn how to use these agents to be able to run their back offices. And they're very quickly able to do that and stack their businesses.

John Furrier

>> I want to give— I want to dig into the— now the use case for what you're going to be doing for business. When glue was getting going, when the whole thing went viral. The guy wrote a blog post. How did he do it? Oh, I got a Mac mini and then I just connected all this. It went wild and did all those cool things that everyone made a meme out of and exploded a major viral revolution of developers. And when asked, it's I just said a Mac because that was the cheapest thing I could get. But the point was that was the device that enabled it. Okay, so connect the dots now, that's actually a market.

Rahul Tikoo

>> Yes.

John Furrier

>> So what does that box, that device look like? Obviously small form factor for personal use, but what's in it? What's the requirements? What makes it now acceptable and sustainable, durable to provide value? Because this is what people are doing. They want to get those tokens, they want to play, they want to tinker around, experiment, create value.

Rahul Tikoo

>> So I'll give you a couple of examples on the dialogue and then we can talk about the product. when I meet with the CIOs, everyone is sort of looking at the cost of Agentic AI, especially from a cloud use case perspective. And they're realizing it's a very big P&L item. And so from that perspective, tokenomics has really become an important consideration for C-suites on how to deploy agentic AI. And so when you think about Agentic PCs and the stuff that we're working on, that's become an important strategic node for these CIOs in a hybrid AI infrastructure. So of course they're going to still have cloud, of course they're going to still have private infrastructure, but they're now seriously looking at agentic PCs as an extension of that such that they can save on the token costs while keeping security, privacy, latency, all of those considerations paramount. And so what sort of enables that? So we launched this product called the Ryzen AI Halo. And this Ryzen AI Halo is driven by the Ryzen AI Max+ processor, which is 16 cores, which is 128 gigs of unified GPU, CPU memory. And so now if you look at large language models, things that used to run in the cloud now can run on the endpoints, 9 billion parameter models or 24 billion parameter models are becoming as good a quality as some of these frontier models. So that is enabling us to now be able to run these models on the Ryzen AI Halo box with 128 gigs of memory and be able to have an agent harness running on it and do whatever the workflow you have using agent.

John Furrier

>> So this to me sounds like a developer opportunity because what you just laid out to me, what I heard was, hey, don't go to the cloud and burn tokens and blow the budget. You want to build and build something. You don't— you're playing with it. Here's your sandbox, go play. And by the way, the CIO loves it because let some— let chaos reign and then, okay, I'll fund it. So this is an operational thing. So do you— is that how you're seeing it? am I getting this right? Because I see multiple things— cost, creativity, but also budget. Yes, here's your sandbox, show me what you got. Here's your paintbrush.

Rahul Tikoo

>> Yeah, absolutely. you could start your work on the endpoint or you could deploy your work at the endpoint. when you think about deploying the work at the endpoint, endpoint is probably going to be the best place in terms of inferencing because the data resides at the endpoint, right? So when you think about security, privacy, especially in regulated markets and regulated industries, that's paramount. When you think about latency, how quickly you are able to get responses back, there are certain things that you want to be speedy. And so that happens at the endpoint. And then of course we talked about costs and that is an important consideration.

John Furrier

>> So it's beyond the developer. It's basically what you're saying is if I'm an end user, I can have an agent running on my, with a harness, preprocessing and doing things on my behalf. And then I could interface that from wherever.

Rahul Tikoo

>> Absolutely.

Dave Vellante

>> Well, big picture here. So on average, organizations globally spend about 4% of their revenue on IT. That number we've predicted is going to explode. It's going to increase to, let's call it, 10% by mid-next decade. And that's going to be through spending CapEx to become token generators or go to the cloud APIs. Either way, they're going to be scaling with less labor than they would have historically. The other point is the last sustained productivity boom that we had was in the '90s. Which was a bottoms-up PC productivity boom. You were there, you were selling those things. You guys both remember that. The AI wave started with the CEO saying, thou shalt do AI. And now, I wrote a piece a couple months ago called Personal Agents Light the Fuse. And that's what's happening now. It's just bottoms up. And ultimately it's all, hopefully it's all going to come together. And that's when you're going to start to see these individual silos of productivity come together for corporate productivity. So how do you see that playing out, Rahul, with the CIOs that you talk to? Are they thinking about that? Yeah, because you don't want silos of intelligence. You want that to be a shared corporate capability.

Rahul Tikoo

>> Yeah, I think one of the big things that we are hearing from the CIOs is, hey, I get it, I get the concepts, I understand tokenomics, I'm struggling with it. How can AMD help me with this? And so I think if you think about AMD as an end-to-end silicon provider going from cloud to data center, to the endpoint, to the edge, we provide that entire stack. Now what we're trying to work on is making the software pieces that stitch it all together much simpler. So what I mean by that is if you look at what we're announcing at Advancing AI, we've now built a semantic router. And what that means is this router can look at your agentic workflow and decide in the moment whether things need to go to a cloud LLM or they need to go to a local PC. So an end customer or an IT shop doesn't have to worry about it. It's just taken care of. The other thing is you talked about glue and how glue came about, but again, it had some security challenges, right?

John Furrier

>> Nightmare.

Rahul Tikoo

>> And so what we have done is now create a secure sandbox in which your agentic harness can reside and be controlled with policies. And so when you think about doing this in a 1-person or a 5-person small business versus now scaling it to thousands of seats in IT, it's very different. So what you have to do there is you have to have the IT infrastructure around being able to observe what the agents are doing, being able to govern them with policies. Having the right security infrastructure, making sure that the agents are sticking to your corporation's security policies, being able to manage the tokenomics of, hey, this agent has gone rogue and they're spending a lot of money on things that we don't think need to go to the cloud and managing that locally. All of those are considerations that a CIO or an IT decision maker has, and we can help them in that area.

John Furrier

>> This is a double bottom line opportunity because this end user wins because imagine their future. I got agents working on my behalf while I'm playing golf, checking my email, pinging me when they need me. But the operational side of product efficiency, distributed computing, remember the old grid computing, P2P days? using resources locally, that's an opportunity for the infrastructure as it gets smarter. You guys think about that? Is that part of this, spread the love around in terms of resource loading or resource management?

Rahul Tikoo

>> If I take a step back, an end user wants to be able to use these tools because these tools get them infinite productivity, right? 10x, 100x. We talked about that. They don't need to worry about the security, the policy. That's where the IT infrastructure comes in. So we got to make this so simple that you and I can use it on a day-to-day basis and not have to think about what's happening in the background. So that's where the routing, and all of the

John Furrier

>> infrastructure—Identity, governance.

Rahul Tikoo

>> —our investments, partnerships with Microsoft, partnerships with other ISVs comes in, right? So we're trying to solve that for the CIOs while we're also working with ISVs and we're working with the end customers in developing use cases for this technology.

Dave Vellante

>> This is what's fascinating to me is in the internet we saw the Amazons and the eBays come about. whereas in the cloud, it was SaaS companies like Workday and ServiceNow, and I'd say Salesforce even though they kind of started the whole SaaS thing. And what's going to be fascinating to see is what happens here, because this wave is going to be driven by the client, because you're going to put this capability in individuals' hands, and they're going to work, like you said, on weekends, when they're on the golf course, they're going to be saying, okay, go, approve, approve, approve. You see people walking around San Francisco like this, they don't want to close their laptop because it's running an Agentic job, right? So this is the ultimate innovator's dilemma. Will the large corporations who are now facing that dilemma, the CIO saying, okay, we're going hard after agents, and the CFO says, wow, this is expensive, let's dial that down a little bit. No more open token bar, let's hand out drink tickets. Meanwhile, you got these AI-native startups that are going to come out of the woodwork, and it's going to be fascinating to see. But my prediction is it's happening from the bottom up, and it's going to happen on the client. Give it this

John Furrier

>> time.But the idea of an agent, the idea of an agent computer though, to what Rahul's getting at, I would want to explore is when you nail the governance and identity, I could actually collaborate with each of you with my agent, my machine. I can set up delegation authority that could be managed by the agent. But this brings us to

Dave Vellante

>> the—There's a security nightmare waiting to happen, but that's okay. We'll get over

Rahul Tikoo

>> that.No, no, no. That's the observability concept, right?

John Furrier

>> Absolutely.If you nail the observability and the mechanics around governance, that's collaboration. So my question, Rahul, because if that predicted future happens, which I think it will, it's plausible, quickly, you got to have an ecosystem. So the question for you is, as you lead the group, ecosystem relationships aren't just partner deals or API deals. What are you looking at from a partnership standpoint? Because that flourishes with an ecosystem.

Rahul Tikoo

>> Yes, absolutely.

John Furrier

>> The more people in make it smarter, you can have intelligence in the network on the double bottom line. The users win, of course, but the IT and network compute optimization intelligent layer could be totally optimized. Yeah, you got to do that with the ecosystem.

Rahul Tikoo

>> Yeah, absolutely. And so I think there's two answers to that question. There's the Windows ecosystem and we're partnered very closely with Microsoft in enabling this and through all of the news that they have around Agentic OS and the work that they're doing, we're going to enable that this year with them. So that's good. And then on the open ecosystem side, we're working closely on ROCm and creating open tools through Lemonade. So the semantic routing that I talked about, that's a part of Lemonade. Lemonade is also the server that we use for local LLM serving. So you can get all of the optimized local large or small language models for use on Lemonade. We've also enabled sandboxing with this product that we just announced called Access, and that's your sandbox for the agents. We also announced Skills. So Skills is literally whether you go to any agentic harness, glue, Cursor, whatever it is, you can tell it that I want you to set up an agent for me to do X and it's going to set it up for you. You don't need to go through the exercise of figuring out 5 pages of how to set various things up. Because again, in the end, we want to make it easy.

Dave Vellante

>> This is huge, right? What you're talking about, because I don't want to have to hire a bunch of forward-deployed engineers to come into my organization to make it happen. It's got to be simple. AI is complex to a lot, but as simple as it is to actually bring it through the enterprise is— so the things that you're talking about now, Rahul, a real progress to simplify and democratize AI and adoption.

Rahul Tikoo

>> That's rightAnd I think that's what I hear the most from CIOs is we are bought in, we understand the token cost. There's lots of news cycle out there that you can listen to what people are struggling with this. And so help me simplify it such that I can apply the right PC in my infrastructure. And that actually makes two interesting things happen in the PC world. One is the PC economics change, right? No longer is a PC a cost item. It's a value generation for the organization, just like data centers are, right? Yeah. So now that value equation has changed for the PC. We've become a very important part of that infrastructure. The second thing that happens is it drives a much faster refresh of a 2 billion device PC market. Yeah, right. So when you think about how many PCs are out there that are 5 years old, There's a lot of them out there. Yeah. And they need to be updated.

Dave Vellante

>> You guys remember back in the PC era, every time a new generation processor came out, you had to have it. You wanted to buy it right away. Oh, and you were— you'd happily pay $3,000, $4,000 for a new PC because it was driving your productivity.

John Furrier

>> Well, if it nails the user value, that's one thing. But I'm getting at— which is great. But my point on the earlier was that if you look at the most viral thing in the '90s, early 2000s, was P2P music stealing, music sharing. Yeah, yeah. That was peer-to-peer. That was leveraging people's computers. I just tap into his computer, I get his music. I would opt in. Now that's a little bit random, but my point is, when you get that kind of connection, you can put intelligence— say, oh, Dave's not at his desk. He's got an agent saying, you can use my compute. Yeah.

Rahul Tikoo

>> And I think that's also happening, right? Not only are you going to have agents on your PC that you carry, or you gave the example of people walking around San Francisco. You can have a desk-side— take the Ryzen AI Halo machine, it could be running 24/7 on your desk, running tens of agents on your behalf, and you're just querying into it.

Dave Vellante

>> And you're generating tokens essentially for the electricity that it takes to run it, but you're not paying through the cloud API.

John Furrier

>> You're using your phone.

Dave Vellante

>> Yeah, you don't have to keep talking to

John Furrier

>> WhatsApp.By the way, I was actually at the airport with my phone doing some replies in Paris, waiting for my flight, and I had actually had it open.

Dave Vellante

>> Yeah, well, you have

John Furrier

>> to—I was on the Wi-Fi. To your point, this is where I could say to close

Dave Vellante

>> it.You don't want

John Furrier

>> to.No, it's totally true. But what we're getting at is that my PC can be the key server. It's a server. Yeah, basically. And you'll gladly pay for that on a

Rahul Tikoo

>> PC.That's where it becomes a value play, not a cost play.

Dave Vellante

>> So it's not just that it's— first of all, the refresh cycle was elongating, as we know, and it's, I can live with this. Now it's going to be, I need to have the latest and greatest because it's driving productivity and revenue.

John Furrier

>> It's funny, the servers are now the PC. So you build the hyperscale Helios, now you got desktop servers.

Rahul Tikoo

>> And by the way, the other interesting thing that I see in this market is, so I live in a neighborhood, there's a lot of small businesses, right? It's a grassroots movement. You were saying this, right? Is people are wanting the technology. And so my neighborhood conversations aren't, hey, whose cat or whose dog or, who pooped in my yard? It's all around, hey, this new Agent Harness came up. Does someone know how to use it? And they're doing webinars teaching each other how to do it.

John Furrier

>> Yes, yes. There's a lot of, adult educated people and young kids are all okay, I don't really know the ins and outs of a data center, but I love this AI stuff because they get value. This is a user interface change. And again, that's going to drive you guys to make servers for your desktop, basically call it a PC. Yeah, it's a personal computer.

Rahul Tikoo

>> It's an exciting time to be in the PC industry.

John Furrier

>> What's your focus nowWhat's on your to-do list? Obviously, it's a great opportunity for you and AMD and all that compute coming down to your area for empowering people. What's your focus now? What are you optimizing for? What are your goals?

Rahul Tikoo

>> I think there's a couple of things. Number one agentic PCs are super critical. We're going to continue investing in that roadmap. We announced Strix Halo last year, beginning of last year. We have like 35+ products out in the market. We announced Gorgon Halo, which is the next generation, 192 gigs of local memory, even bigger models or even bigger agent swarms that can run on it. That got announced in May. That'll come to market in the September timeframe. We already have about 22 different products lined up to launch with that. We're already working on the next generation of this, right? So this is a product line that we're going to continue investing in. It's a category we created. We'll continue investing in it. Second most important thing for us is getting the software right. Yeah, it's going to be around orchestration it's going to be around security, it's going to be around observability, and not just doing the software but also getting the ISV partnerships that come along with the key leaders in this space. Microsoft is one example of that. You'll hear some more examples tomorrow in our keynote with key partnerships that we're bringing. Developers are a big part of the equation. Yeah. Because you got to go enable the developers. We are working with a lot of developers here. And so how do we very quickly get to millions of developers developing on our platform such that all of these experiences can get lit up for the end user?

John Furrier

>> I'll give one more final word here to close this out, because I want you to take a moment to explain this. I was talking to someone last week, they're AMD, I'm going to the AMD event. AMD, they just make PCs. A lot of people don't understand the unfair advantage AMD has with its ecosystem. with Microsoft, with all the suppliers. And how does that ecosystem experience translate to today's execution for you?

Rahul Tikoo

>> I think there's a couple of things that people don't understand. AMD is truly one of the very few companies that can go end to end in the infrastructure, whether it is CPUs, GPUs, NPUs, whether it is cloud data center, endpoint, edge, robotics, every industry. We're stitching it all together. And the beauty of it is it's an open ecosystem which is very rich, very vibrant. A lot of development happening in that open ecosystem and people are looking for choice and a player like that to partner with.

John Furrier

>> Would you agree that people are building systems now, not just point products?

Rahul Tikoo

>> People are building solutions now that go end to end.

John Furrier

>> Got it. Yeah, right.

Rahul Tikoo

>> I don't just think about the PC anymore. I think about How does my PC fit in this infrastructure and how do I make it easy for the CIO to be able to run from cloud to endpoint?

John Furrier

>> Well, the first layup for you to win the market is just go to the CIO and say, I could take the token tax and budget off the table, buy these machines. Of course they're going to buy it. Very well. Thank you so much for coming on.

Rahul Tikoo

>> Thank you.

Dave Vellante

>> Appreciate it.

John Furrier

>> Thank you. Again, the desktop, the personal computer has changed the definition of what that means as AI creates more agency and empowers users. That's going to force everything to change. And if that's the action behind the AI infrastructure buildout, AI is now infrastructure. We're doing our part here in theCUBE. I'm John Furrier with Dave Vellante. Thanks for watching.

Rahul Tikoo, AMD

search

John Furrier

>> Welcome back to theCUBE live in San Francisco, California for AMD's Advancing AI event. It's where all the AMD leaders are gathering, the industry leaders as well as practitioners and technologists. It's a free event, sold out, 14,000 people, packed house. I'm John Furrier with Dave Vellante, my co-host. Rahul Tikoo is here, Senior Vice President and General Manager of the Client Business at AMD. I never thought I'd say this early, but as we predicted, AI is going to the client side, which is your devices. But also they got super servers, Helios powering everything in between. Rahul, thanks for coming on theCUBE.

Rahul Tikoo

>> Thank you. Thank you for having me. Good to see you guys.

John Furrier

>> So I love this idea of having my own AI PC that can do stuff. We've done interviews where autonomous vehicles are now becoming personal. So the personal computer client devices are connecting to intelligence.

Rahul Tikoo

>> Yeah.

John Furrier

>> And developers and end users will get value out of that. That's a big driver. Give us an update on current situation with why this is happening and then what's going on with the product.

Rahul Tikoo

>> Yeah, so I think if you look at the last 3 years, we're sort of in this age of chatbots or AI assistants. And so everyone that was using a PC was using some sort of an AI assistant and that could get you 10%, 20%, 30% more productive. What's happened, and a lot of people would ask me, hey, what's the killer app? I don't know what the killer app for AI is. And I think what's happened is the killer app now is agentic AI. And so when you think about agentic, what's happening is people are going from assistants to fully reasoning autonomous agents that work on your behalf, right? And so that productivity of 10%, 20%, 30% because of the assistance is now becoming a force multiplier of 10x, 100x. And so now you can imagine a software developer that has an army of agents working for them and creating code, right? That's the world we're entering into and PCs are becoming a very important part of that world.

John Furrier

>> Yeah, we had a comment on one of my posts on LinkedIn, we brought it up on Dell today, that the human empowerment has gone supernova with AI and things like the graphical user interface WIMP Windows, Icons, Menus, Pointer, is changing to natural language.

Rahul Tikoo

>> Yeah, yeah.

John Furrier

>> So that changes what used to be load Excel, type in data, create a macro, build a graph, import it to PowerPoint. those go away.

Dave Vellante

>> Yeah, absolutely.

John Furrier

>> Now we're in, hey, just run the numbers for the quarter and give me a graph.

Rahul Tikoo

>> Yeah, absolutely. That's AI. AI agents are so smart now that with natural English language, anyone can be a programmer, right? You could be a one-person company. I'll give you a quick example, working with small businesses, right? And there are small businesses that are doctors' offices, that are marketing firms, that are law firms. They don't need to learn AI. They just need to learn how to use these agents to be able to run their back offices. And they're very quickly able to do that and stack their businesses.

John Furrier

>> I want to give— I want to dig into the— now the use case for what you're going to be doing for business. When glue was getting going, when the whole thing went viral. The guy wrote a blog post. How did he do it? Oh, I got a Mac mini and then I just connected all this. It went wild and did all those cool things that everyone made a meme out of and exploded a major viral revolution of developers. And when asked, it's I just said a Mac because that was the cheapest thing I could get. But the point was that was the device that enabled it. Okay, so connect the dots now, that's actually a market.

Rahul Tikoo

>> Yes.

John Furrier

>> So what does that box, that device look like? Obviously small form factor for personal use, but what's in it? What's the requirements? What makes it now acceptable and sustainable, durable to provide value? Because this is what people are doing. They want to get those tokens, they want to play, they want to tinker around, experiment, create value.

Rahul Tikoo

>> So I'll give you a couple of examples on the dialogue and then we can talk about the product. when I meet with the CIOs, everyone is sort of looking at the cost of Agentic AI, especially from a cloud use case perspective. And they're realizing it's a very big P&L item. And so from that perspective, tokenomics has really become an important consideration for C-suites on how to deploy agentic AI. And so when you think about Agentic PCs and the stuff that we're working on, that's become an important strategic node for these CIOs in a hybrid AI infrastructure. So of course they're going to still have cloud, of course they're going to still have private infrastructure, but they're now seriously looking at agentic PCs as an extension of that such that they can save on the token costs while keeping security, privacy, latency, all of those considerations paramount. And so what sort of enables that? So we launched this product called the Ryzen AI Halo. And this Ryzen AI Halo is driven by the Ryzen AI Max+ processor, which is 16 cores, which is 128 gigs of unified GPU, CPU memory. And so now if you look at large language models, things that used to run in the cloud now can run on the endpoints, 9 billion parameter models or 24 billion parameter models are becoming as good a quality as some of these frontier models. So that is enabling us to now be able to run these models on the Ryzen AI Halo box with 128 gigs of memory and be able to have an agent harness running on it and do whatever the workflow you have using agent.

John Furrier

>> So this to me sounds like a developer opportunity because what you just laid out to me, what I heard was, hey, don't go to the cloud and burn tokens and blow the budget. You want to build and build something. You don't— you're playing with it. Here's your sandbox, go play. And by the way, the CIO loves it because let some— let chaos reign and then, okay, I'll fund it. So this is an operational thing. So do you— is that how you're seeing it? am I getting this right? Because I see multiple things— cost, creativity, but also budget. Yes, here's your sandbox, show me what you got. Here's your paintbrush.

Rahul Tikoo

>> Yeah, absolutely. you could start your work on the endpoint or you could deploy your work at the endpoint. when you think about deploying the work at the endpoint, endpoint is probably going to be the best place in terms of inferencing because the data resides at the endpoint, right? So when you think about security, privacy, especially in regulated markets and regulated industries, that's paramount. When you think about latency, how quickly you are able to get responses back, there are certain things that you want to be speedy. And so that happens at the endpoint. And then of course we talked about costs and that is an important consideration.

John Furrier

>> So it's beyond the developer. It's basically what you're saying is if I'm an end user, I can have an agent running on my, with a harness, preprocessing and doing things on my behalf. And then I could interface that from wherever.

Rahul Tikoo

>> Absolutely.

Dave Vellante

>> Well, big picture here. So on average, organizations globally spend about 4% of their revenue on IT. That number we've predicted is going to explode. It's going to increase to, let's call it, 10% by mid-next decade. And that's going to be through spending CapEx to become token generators or go to the cloud APIs. Either way, they're going to be scaling with less labor than they would have historically. The other point is the last sustained productivity boom that we had was in the '90s. Which was a bottoms-up PC productivity boom. You were there, you were selling those things. You guys both remember that. The AI wave started with the CEO saying, thou shalt do AI. And now, I wrote a piece a couple months ago called Personal Agents Light the Fuse. And that's what's happening now. It's just bottoms up. And ultimately it's all, hopefully it's all going to come together. And that's when you're going to start to see these individual silos of productivity come together for corporate productivity. So how do you see that playing out, Rahul, with the CIOs that you talk to? Are they thinking about that? Yeah, because you don't want silos of intelligence. You want that to be a shared corporate capability.

Rahul Tikoo

>> Yeah, I think one of the big things that we are hearing from the CIOs is, hey, I get it, I get the concepts, I understand tokenomics, I'm struggling with it. How can AMD help me with this? And so I think if you think about AMD as an end-to-end silicon provider going from cloud to data center, to the endpoint, to the edge, we provide that entire stack. Now what we're trying to work on is making the software pieces that stitch it all together much simpler. So what I mean by that is if you look at what we're announcing at Advancing AI, we've now built a semantic router. And what that means is this router can look at your agentic workflow and decide in the moment whether things need to go to a cloud LLM or they need to go to a local PC. So an end customer or an IT shop doesn't have to worry about it. It's just taken care of. The other thing is you talked about glue and how glue came about, but again, it had some security challenges, right?

John Furrier

>> Nightmare.

Rahul Tikoo

>> And so what we have done is now create a secure sandbox in which your agentic harness can reside and be controlled with policies. And so when you think about doing this in a 1-person or a 5-person small business versus now scaling it to thousands of seats in IT, it's very different. So what you have to do there is you have to have the IT infrastructure around being able to observe what the agents are doing, being able to govern them with policies. Having the right security infrastructure, making sure that the agents are sticking to your corporation's security policies, being able to manage the tokenomics of, hey, this agent has gone rogue and they're spending a lot of money on things that we don't think need to go to the cloud and managing that locally. All of those are considerations that a CIO or an IT decision maker has, and we can help them in that area.

John Furrier

>> This is a double bottom line opportunity because this end user wins because imagine their future. I got agents working on my behalf while I'm playing golf, checking my email, pinging me when they need me. But the operational side of product efficiency, distributed computing, remember the old grid computing, P2P days? using resources locally, that's an opportunity for the infrastructure as it gets smarter. You guys think about that? Is that part of this, spread the love around in terms of resource loading or resource management?

Rahul Tikoo

>> If I take a step back, an end user wants to be able to use these tools because these tools get them infinite productivity, right? 10x, 100x. We talked about that. They don't need to worry about the security, the policy. That's where the IT infrastructure comes in. So we got to make this so simple that you and I can use it on a day-to-day basis and not have to think about what's happening in the background. So that's where the routing, and all of the

John Furrier

>> infrastructure—Identity, governance.

Rahul Tikoo

>> —our investments, partnerships with Microsoft, partnerships with other ISVs comes in, right? So we're trying to solve that for the CIOs while we're also working with ISVs and we're working with the end customers in developing use cases for this technology.

Dave Vellante

>> This is what's fascinating to me is in the internet we saw the Amazons and the eBays come about. whereas in the cloud, it was SaaS companies like Workday and ServiceNow, and I'd say Salesforce even though they kind of started the whole SaaS thing. And what's going to be fascinating to see is what happens here, because this wave is going to be driven by the client, because you're going to put this capability in individuals' hands, and they're going to work, like you said, on weekends, when they're on the golf course, they're going to be saying, okay, go, approve, approve, approve. You see people walking around San Francisco like this, they don't want to close their laptop because it's running an Agentic job, right? So this is the ultimate innovator's dilemma. Will the large corporations who are now facing that dilemma, the CIO saying, okay, we're going hard after agents, and the CFO says, wow, this is expensive, let's dial that down a little bit. No more open token bar, let's hand out drink tickets. Meanwhile, you got these AI-native startups that are going to come out of the woodwork, and it's going to be fascinating to see. But my prediction is it's happening from the bottom up, and it's going to happen on the client. Give it this

John Furrier

>> time.But the idea of an agent, the idea of an agent computer though, to what Rahul's getting at, I would want to explore is when you nail the governance and identity, I could actually collaborate with each of you with my agent, my machine. I can set up delegation authority that could be managed by the agent. But this brings us to

Dave Vellante

>> the—There's a security nightmare waiting to happen, but that's okay. We'll get over

Rahul Tikoo

>> that.No, no, no. That's the observability concept, right?

John Furrier

>> Absolutely.If you nail the observability and the mechanics around governance, that's collaboration. So my question, Rahul, because if that predicted future happens, which I think it will, it's plausible, quickly, you got to have an ecosystem. So the question for you is, as you lead the group, ecosystem relationships aren't just partner deals or API deals. What are you looking at from a partnership standpoint? Because that flourishes with an ecosystem.

Rahul Tikoo

>> Yes, absolutely.

John Furrier

>> The more people in make it smarter, you can have intelligence in the network on the double bottom line. The users win, of course, but the IT and network compute optimization intelligent layer could be totally optimized. Yeah, you got to do that with the ecosystem.

Rahul Tikoo

>> Yeah, absolutely. And so I think there's two answers to that question. There's the Windows ecosystem and we're partnered very closely with Microsoft in enabling this and through all of the news that they have around Agentic OS and the work that they're doing, we're going to enable that this year with them. So that's good. And then on the open ecosystem side, we're working closely on ROCm and creating open tools through Lemonade. So the semantic routing that I talked about, that's a part of Lemonade. Lemonade is also the server that we use for local LLM serving. So you can get all of the optimized local large or small language models for use on Lemonade. We've also enabled sandboxing with this product that we just announced called Access, and that's your sandbox for the agents. We also announced Skills. So Skills is literally whether you go to any agentic harness, glue, Cursor, whatever it is, you can tell it that I want you to set up an agent for me to do X and it's going to set it up for you. You don't need to go through the exercise of figuring out 5 pages of how to set various things up. Because again, in the end, we want to make it easy.

Dave Vellante

>> This is huge, right? What you're talking about, because I don't want to have to hire a bunch of forward-deployed engineers to come into my organization to make it happen. It's got to be simple. AI is complex to a lot, but as simple as it is to actually bring it through the enterprise is— so the things that you're talking about now, Rahul, a real progress to simplify and democratize AI and adoption.

Rahul Tikoo

>> That's rightAnd I think that's what I hear the most from CIOs is we are bought in, we understand the token cost. There's lots of news cycle out there that you can listen to what people are struggling with this. And so help me simplify it such that I can apply the right PC in my infrastructure. And that actually makes two interesting things happen in the PC world. One is the PC economics change, right? No longer is a PC a cost item. It's a value generation for the organization, just like data centers are, right? Yeah. So now that value equation has changed for the PC. We've become a very important part of that infrastructure. The second thing that happens is it drives a much faster refresh of a 2 billion device PC market. Yeah, right. So when you think about how many PCs are out there that are 5 years old, There's a lot of them out there. Yeah. And they need to be updated.

Dave Vellante

>> You guys remember back in the PC era, every time a new generation processor came out, you had to have it. You wanted to buy it right away. Oh, and you were— you'd happily pay $3,000, $4,000 for a new PC because it was driving your productivity.

John Furrier

>> Well, if it nails the user value, that's one thing. But I'm getting at— which is great. But my point on the earlier was that if you look at the most viral thing in the '90s, early 2000s, was P2P music stealing, music sharing. Yeah, yeah. That was peer-to-peer. That was leveraging people's computers. I just tap into his computer, I get his music. I would opt in. Now that's a little bit random, but my point is, when you get that kind of connection, you can put intelligence— say, oh, Dave's not at his desk. He's got an agent saying, you can use my compute. Yeah.

Rahul Tikoo

>> And I think that's also happening, right? Not only are you going to have agents on your PC that you carry, or you gave the example of people walking around San Francisco. You can have a desk-side— take the Ryzen AI Halo machine, it could be running 24/7 on your desk, running tens of agents on your behalf, and you're just querying into it.

Dave Vellante

>> And you're generating tokens essentially for the electricity that it takes to run it, but you're not paying through the cloud API.

John Furrier

>> You're using your phone.

Dave Vellante

>> Yeah, you don't have to keep talking to

John Furrier

>> WhatsApp.By the way, I was actually at the airport with my phone doing some replies in Paris, waiting for my flight, and I had actually had it open.

Dave Vellante

>> Yeah, well, you have

John Furrier

>> to—I was on the Wi-Fi. To your point, this is where I could say to close

Dave Vellante

>> it.You don't want

John Furrier

>> to.No, it's totally true. But what we're getting at is that my PC can be the key server. It's a server. Yeah, basically. And you'll gladly pay for that on a

Rahul Tikoo

>> PC.That's where it becomes a value play, not a cost play.

Dave Vellante

>> So it's not just that it's— first of all, the refresh cycle was elongating, as we know, and it's, I can live with this. Now it's going to be, I need to have the latest and greatest because it's driving productivity and revenue.

John Furrier

>> It's funny, the servers are now the PC. So you build the hyperscale Helios, now you got desktop servers.

Rahul Tikoo

>> And by the way, the other interesting thing that I see in this market is, so I live in a neighborhood, there's a lot of small businesses, right? It's a grassroots movement. You were saying this, right? Is people are wanting the technology. And so my neighborhood conversations aren't, hey, whose cat or whose dog or, who pooped in my yard? It's all around, hey, this new Agent Harness came up. Does someone know how to use it? And they're doing webinars teaching each other how to do it.

John Furrier

>> Yes, yes. There's a lot of, adult educated people and young kids are all okay, I don't really know the ins and outs of a data center, but I love this AI stuff because they get value. This is a user interface change. And again, that's going to drive you guys to make servers for your desktop, basically call it a PC. Yeah, it's a personal computer.

Rahul Tikoo

>> It's an exciting time to be in the PC industry.

John Furrier

>> What's your focus nowWhat's on your to-do list? Obviously, it's a great opportunity for you and AMD and all that compute coming down to your area for empowering people. What's your focus now? What are you optimizing for? What are your goals?

Rahul Tikoo

>> I think there's a couple of things. Number one agentic PCs are super critical. We're going to continue investing in that roadmap. We announced Strix Halo last year, beginning of last year. We have like 35+ products out in the market. We announced Gorgon Halo, which is the next generation, 192 gigs of local memory, even bigger models or even bigger agent swarms that can run on it. That got announced in May. That'll come to market in the September timeframe. We already have about 22 different products lined up to launch with that. We're already working on the next generation of this, right? So this is a product line that we're going to continue investing in. It's a category we created. We'll continue investing in it. Second most important thing for us is getting the software right. Yeah, it's going to be around orchestration it's going to be around security, it's going to be around observability, and not just doing the software but also getting the ISV partnerships that come along with the key leaders in this space. Microsoft is one example of that. You'll hear some more examples tomorrow in our keynote with key partnerships that we're bringing. Developers are a big part of the equation. Yeah. Because you got to go enable the developers. We are working with a lot of developers here. And so how do we very quickly get to millions of developers developing on our platform such that all of these experiences can get lit up for the end user?

John Furrier

>> I'll give one more final word here to close this out, because I want you to take a moment to explain this. I was talking to someone last week, they're AMD, I'm going to the AMD event. AMD, they just make PCs. A lot of people don't understand the unfair advantage AMD has with its ecosystem. with Microsoft, with all the suppliers. And how does that ecosystem experience translate to today's execution for you?

Rahul Tikoo

>> I think there's a couple of things that people don't understand. AMD is truly one of the very few companies that can go end to end in the infrastructure, whether it is CPUs, GPUs, NPUs, whether it is cloud data center, endpoint, edge, robotics, every industry. We're stitching it all together. And the beauty of it is it's an open ecosystem which is very rich, very vibrant. A lot of development happening in that open ecosystem and people are looking for choice and a player like that to partner with.

John Furrier

>> Would you agree that people are building systems now, not just point products?

Rahul Tikoo

>> People are building solutions now that go end to end.

John Furrier

>> Got it. Yeah, right.

Rahul Tikoo

>> I don't just think about the PC anymore. I think about How does my PC fit in this infrastructure and how do I make it easy for the CIO to be able to run from cloud to endpoint?

John Furrier

>> Well, the first layup for you to win the market is just go to the CIO and say, I could take the token tax and budget off the table, buy these machines. Of course they're going to buy it. Very well. Thank you so much for coming on.

Rahul Tikoo

>> Thank you.

Dave Vellante

>> Appreciate it.

John Furrier

>> Thank you. Again, the desktop, the personal computer has changed the definition of what that means as AI creates more agency and empowers users. That's going to force everything to change. And if that's the action behind the AI infrastructure buildout, AI is now infrastructure. We're doing our part here in theCUBE. I'm John Furrier with Dave Vellante. Thanks for watching.