Lisa Su sees physical AI chips as a five-year-plus market

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 19:51

   And how do you see the market evolving in these next five or six years? Is it, there's a standard set of chips for training, a standard set for inference, or do you just see an explosion? like a Cambrian explosion of different ASICs, different designs, different use cases? Yeah, I like that question because I am a believer in there will be diversity of chips. And the reason is there's so many use cases, right?

2. 20:12

   If you think about use cases from, you know, whether you're talking about science or manufacturing or design or backend or, you know, frankly, personal AI, I think we're going to see AI in everything that we do, you know, certainly in your phones and your PCs. And so you have all these pieces. You're going to have different types of chips that do that. You know, certainly for the largest systems, we tend to believe that, you know, you need the most compute you can get.

3. 20:37

   And so, you know, GPUs are there, but lots of ASICs are in the process. And, you know, we'll see a variety of different chips. You opened up a really interesting line of questioning there. When mainframes were so expensive and then eventually wound up having PCs that were more expensive on their desktop, you alluded to AI being run locally. Yes. When would we have a local computer, a laptop, a desktop computer that would have the power we're seeing to run some of these LLM models in your mind?

4. 21:07

   And do you see that as a specific market to go after? Look, I definitely see the idea that AI will be at every part of our ecosystem is a real thing. I think that's one of the advantages. If you think about the power of AI, you want it everywhere and you want it across all different applications. And I think when you think about PCs today, we're putting a significant amount of AI in them to run local models.

5. 21:33

   And why would you want that? It's like, well, maybe I don't want all my personal data all over the place. On that point, can you make a prediction on when the market for physical AI chips is greater than the market for chips and data centers? That's a great question. I'm a big believer in physical AI. I still think it's, let's call it five years. You think five years? Is that fast? At least five years.

6. 21:57

   So you're saying five plus? Five plus, yes. Okay. But that is ultimately the biggest end market, do you think? Do you think physical AI becomes the biggest end market? I think it becomes a significant end market. I think you look at chips in data centers and you look at chips at the edge, they're also significant markets. When you look at the most cutting-edge techniques today, EV lithography, all of this whole stuff to make chips, one of the things that's observable is we're only as good as what humans have been able to invent.

7. 22:30

   And I often ask the recursive question, what happens when the AI is able to invent its own method of manufacturing? Different materials, different material sciences, different approaches that we may not necessarily understand. Is any of that R&D happening, whether in AMD or in other places? Like, how are we trying to get beyond the physical limits of electrons shunting across a Junction. I think this idea that the AI can be extremely smart and extremely capable, like we think about how AI can design the future chips and it will design pieces of it, but there's still a creativity of bringing it all together that I think humans are still absolutely at the center of that. So I don't necessarily see the AI designing our next generation GPU, but I do see at helping us design the next generation GPU much faster and more reliably.