It has been a huge week for Nvidia rival AMD at its advancing AI conference in San Francisco. The unveiling of new chips, a new $5 billion investment in Anthropic, and a reiteration that demand for AI remains incredibly strong. I sat down for a conversation with AMD chair and CEO Dr. Lisa Su. Su has been the driving force behind AMD's ascent the past decade.
First of all, for the industry, we're calling the industry market growth to be a $2 trillion market opportunity for us as we think about from now through 2030. So it's just an incredibly large market. We're seeing this point in AI, we've called the fact that AI is at an inflection point where more and more people are doing useful work, doing meaningful work, and that has made inference the largest growth driver overall. And I am super excited to say that we are in full production for Helios, our MI455 rack scale architecture, our new Venice CPUs, and a tremendous amount of new content. So it's a pretty exciting time for us. I think it's an exciting time for industry, and more importantly for all of us in the AI ecosystem. This is the time to really accelerate what we can bring to the market.
Amongst all the important points you made at the event, Lisa, this one also stood out to me. AI is evolving. You talked about just now too. AI is evolving from training to what does that mean for the AI revolution and what does it mean for a company like AMD?
Well, what it really means, Brian, is all of us are now starting to experience the real power of AI. We have great customers that are training and doing foundational models. But the key is the model is only as good as the output that you get from it. And so with inference, that is all of us asking questions. That's agents working. And that is growing at a really accelerated rate and pace. And that's why we're saying that just the accelerator market alone is going to reach $1.4 trillion by the time we get to 2030. And inference is really the way we turn AI from a technology to something that really changes the way we do business, the way we do research, the way we do healthcare, all of those aspects of it. So I'm really excited to see the rate and pace of adoption of AI.
I think I've told you this before, but it bears repeating. I live vicariously through your X feed. I see you traveling the country. I've seen you at the White House before. Lots of cool things you've been doing this year as you've traveled the world and you've done your important work at AMD. What has surprised you about this AI movement?
I think what has surprised me, Brian, in a very positive way is the rate and pace of adoption is actually faster than any of us thought. Frankly, we usually think about what's going to happen in a market, we think about updating it on an annual basis. And with AI and with everything that we've seen around the world with our customers, things are changing on a monthly and a quarterly basis. So just over the last five or six months, agents have become really productive. The idea that you're not just asking an AI model a question, you're actually asking an AI model to help you solve a problem, help you code a new software module, or help you plan your next trip. Those are things that just a year ago we wouldn't have thought were possible, and all of that requires more and more compute infrastructure. So that is probably the most interesting, exciting, surprising thing: for those of us in the industry, I've never seen a technology adoption curve like what we're seeing with AI right now, and agents are just taking it to a whole new level.
You really increased your total addressable market, or TAM, for your data center business, your accelerator market, pretty significantly as you look out to 2030. On the topic of AI agents in 2030, what are these agents doing that they aren't doing today?
Well, let me start with the very important thing to remember about AI: AI is a tool. So AI is a tool to make every one of us more productive, to make every one of our companies more productive. And so the idea of agents is think about the productivity you can get from 10 employees. You have really smart people, but if each of those 10 employees had 10 agents or a hundred agents or even a thousand agents, they can be much more productive. And so the whole idea of Agentic AI is to create a closed loop situation where you can ask your AI to help you solve more complicated problems so that you as an employee can be much more efficient. At the end, the human is the one that decides the right or wrong answer, but AI can help you look at a whole set of options much faster and more efficiently than if you had to do it by hand. That's the power of Agentic AI.
Within the technology that you unveiled this event, Helios for example, you just mentioned it. What does something like that do that prior iterations of your products don't do?
One of the most important things when we think about AI is just how many questions can you answer and how much does it cost. So you hear things like AI tokens or how many AI tokens can you get per dollar. What Helios does is it's an incredible feat of engineering that allows you to scale much broader in terms of the size of the models, the types of questions, the number of users that can use a certain system. The way to think about it is we're seeing more than a 30 times improvement in performance when we look at the new Helios systems. From a customer standpoint, that just means you can get a lot more done for your money. And the way I think about it is it's a very positive feedback loop: the more AI can do, the more compute you need; the more compute you have, the more intelligence you have; the more problems you can solve. And all of that relates to the importance and the essential foundation is the computing layer, which is what we do at AMD.
You've mentioned this to me in the past, Lisa, but you brought it up again in your keynote. It's one of your philosophies or one of your beliefs that one chip player will not win it all during this AI revolution. Why is that still the case?
Well, I think the most important thing to think about, Brian, is the world is a very heterogeneous place. We're all different. Our companies are different. Our problem sets are different. What we're trying to solve is different. As a result, there's not just one chip that can do it all or one company that can do it all. What you really need is an ecosystem. You need an end-to-end set of engines. That's what I like to say at AMD. We have an end-to-end set of compute engines from CPUs to GPUs to FPGAs to A6, the ability to put all that together. And we have a true belief in an open ecosystem. The idea is the more developers we have on AMD, the faster we can move the ecosystem, the more progress we can make. That's just good for the overall industry. So I'm a big believer in there's no one-size-fits-all, but there's incredible power in bringing the ecosystem together and allowing us to innovate simultaneously.
