Chip market is moving 'very, very quickly', says AMD CEO Lisa Su

Lisa Su
2 followers
6 Video Views·Jan 7, 2025
Lisa Su, AMD CEO, joins 'Closing Bell Overtime' and CNBC's Kristina Partsinevelos to talk its new AI chip, pricing, chip manufacturing and more.
AMD shares closing in the red today after climbing in anticipation of the company's
advancing AI event today. That's where CEO Lisa Sue unveiled a new AI chip, the MI325X. And
joining us now, fresh off the stage in San Francisco, is Lisa Sue, AMD chair and CEO,
along with our own Christina Partzenevos. Christina. Thank you, John. I have to start
actually by saying congratulations. 10 years as CEO running this company. It's no small feat,
that's for sure. But I want to get straight into cost. Often within the research world,
you have a lot of, I guess, analyst reports that say AMD is a cheaper alternative. And I know on
stage you talked about the total cost of ownership being lowered with this latest AI chip. But what
does that actually mean? Does that mean that your chips are cheaper or they're cheaper if you buy
a bunch of different chips from AMD? How does it work? Yeah. So first of all, Christina,
thank you for being here, John. Great to be on the program today. This was a big day for us.
We launched a ton of new products across our fifth generation EPYC, our new AI chip, MI325.
We talked about our roadmap going forward, which we're accelerating. We talked about some new
networking solutions. We talked about AI PCs, and we had a ton of partners on stage. So it's
been a fantastic day. Look, when you look at what's happening in data centers today, Christina,
we want to help our customers and the industry really take advantage of new technology.
And the beauty of new technology is you can actually take what was, let's call it a legacy
data center, and do that same work for much lower total cost of ownership. So back to your question
about costs and what it is. The new technology is much more power efficient, has more performance,
gives you just a tremendous amount of capability. And one of the metrics that we talked about today
is for our newest generation of server processors, our turn generation, we actually
can take seven legacy servers and replace them with one new fifth generation EPYC server. So
that's just tremendous total cost of ownership. Now, what do we want to do at AMD? What we want
to do is provide value to our customers. So it is about more performance. It is about better
performance per dollar. It's about reducing their footprint and giving them more computing for their
applications. Which is a perfect segue to talk about the MI325X, which you're comparing to the
H200 for NVIDIA, which has been on the market for three years. So is it, are you saying that there's
just a such a large market for that type of chip? Or is it this constant game of catch up with
competitors? Well, the best way to say it is, first of all, the market is moving very, very
quickly. You know, if I just say, you know, our MI300, which has been in the market now just
a little bit less than a year, was competing against the H100, which was the previous generation.
It's done phenomenally well, and we've seen tremendous customer adoption. We're very excited
to have, you know, Microsoft and Meta and Oracle on stage today. We had a number of AI developers
talking about just how much performance they're getting on MI300. We're now launching MI325,
which compares very well to H200. Actually, if you think about some of the most important
inferencing workloads, we're going to see, you know, 20 to 40% more performance on our new chip.
And it's not just, you know, it's not just what we're doing today, but we're really excited also
about our next generation, our MI350 series and what's going on after that.
That's a few years out.
No, MI350 series is next year. It's in the second half of 2025 and going to MI400 in 2026. So
we are accelerating the pace of innovation. It feels really good to kind of see all of this stuff,
you know, come together. And the most important thing is developers love developing
on our accelerators, and that's what we're really focused on.
