저자: LisaSu · 외국인
요약 보기
AMD가 AI 가속기 MI355를 공개하며 토큰당 비용 효율 40% 개선, 이전 세대 대비 3배 성능을 주장. 2026년 연간 가속기 로드맵(MI400 시리즈, 10배 개선 예고)과 랙 스케일 시스템 확대, ZT Systems 인수 완료 및 설계 엔지니어 1,000명 추가를 발표. 7대 AI 모델 빌더가 AMD 제품 사용(Meta, Oracle, OpenAI, xAI, Tesla 등). AI 시장이 연 60% 이상 성장해 2028년 5,000억 달러 규모가 될 것으로 전망. 오픈 생태계 강조.
다듬은 글
I'm here at AMD's Advancing AI event in San Jose, California, where AMD just unveiled a slate of new AI hardware, software, and systems designed to compete with Nvidia and others. Uh, shares are flat for the year, underperforming its major peers, but have rallied 17% in the past 3 months. Joining me here in San Jose, an exclusive interview is Lisa Sue, AMD's CEO and chair.
Um, thanks for having me here. Thanks. Thanks for being here, John. So, talk to me about um these systems. You got a chip sitting there next to you that's I guess fresh out of the oven. Uh what is it that you're announcing here? What uh what level of competition does that raise you to uh with the others out there?
Yeah, absolutely. So, first of all, um it's been a big day for AMD. Really exciting. I love talking about our new products. You know that. So, we just recently uh announced our MI 355 chip. is our newest uh AI chip and it is fantastic. It's the best inference accelerator out there in the market and you know if you look at what's happening in AI today um we're just seeing so many more use cases John like people are using AI everywhere and so one of the most important things for our customers is to get more efficiency um from their chips.
So when you think about um you know tokens per dollar that's kind of the metric that is used and this chip will give you 40% more tokens per dollar which just gives you the opportunity to offer AI at a much more affordable price um across the world. So yes very exciting. Uh we also previewed our 2026 lineup.
We're on an annual cadence now of of new accelerators and and we're not just doing chips but we're doing you know big rack scale systems. And so all of that was part of our announcements today and there uh investors are are puzzling over how the innovations will translate into share gains versus the likes of Nvidia.
So why why is it beyond chips now and why is it about racks? That was a big part of the ZT systems acquisition. Put that into perspective with the the go to market and the design component to satisfy the customer need. Yeah, absolutely. So look, we're really excited about the momentum that we're seeing um in the AI market these days.
You know, if you look today, you know, seven out of the top 10, you know, model builders and AI companies are using um AMD products. You know, that includes Meta, Oracle, OpenAI, Sam Alman was here with us today. Meta and Oracle were here. Um you know, XAI, Tesla to name a few. And when you think about, you know, what they need, what they really want is the latest generation hardware that gets into the data center and then gets into serving production workloads as fast as possible.
And so to do that, you require, you know, of course, we require chips. Uh, but we're also uh putting a significant amount of investment into the software layer to make it just super easy to use. And we're also uh building uh rack scale solutions. So we just completed our ZT acquisition. We've added a thousand design engineers.
And the whole purpose is so that we can help our customers get to the market with our AI solutions as fast as physically possible. You got competition with Nvidia at the high end certainly. Then the hyperscalers, the big cloud providers who are also now big AI providers, AWS, Azure, uh Google, they've got their homegrown AI chips that they're trying to use as much as possible to keep their costs lower, their margins higher.
How do you play to both ends of that and win? How does openness fit into your strategy? Yeah. Well, you know, John, when you think about it, you know, I view we're still in the very early innings of AI. And frankly, um, as much as we've seen a ramp up in AI, you know, we have, um, sort of said that the market is going to grow, um, over 60% annually for the next number of years to over 500 billion dollars in 2028.
Um that's a huge market and you know in that market what people want is the most flexible and the most programmable um you know architectures out there and and that's really what we're building and you know the key point in this is again we believe in deep co-development with our partners. So hardware software systems coming together making sure that we have the right compute for the right application and you know that's what we've shown.
If you look at just how much improvement, you know, I can just give you some stats, right? If you look at um the solution that we're just talking about, the MI355, you know, it's Hold it up high. Oh, hold it up high. Yes. I want to make sure everyone gets to see my new baby here. We're very happy that um it is now out in production.
But um you know, this is like three times the performance of just what you saw last year. So, you know, that's just how much value you're getting. and uh we previewed our next generation our MI400 series and there we're going to see like 10x you know types of improvements. So that's how much we can get uh from the hardware and working closely with our partners.
And uh you asked about open um open is the foundation of what we do at AMD. You know I'm a real believer in the fact that an open ecosystem is how you get the best innovation into the market. So we have you know thousands of developers here today. They're playing with our hardware. They're learning our software and we want all of them developing on our stuff so that uh you know we get the benefit of all of that innovation and they get the benefit of bringing their best ideas
Comments 0