He was with me for 55 x AMD's first rack scale systems solution, but inside it AMD first in the world's first two nanometer chip of that type. A lot was made of it when you actually just stood on stage and held it in your hand for the first time. Why is it significant?
Well, first of all, Ed, it's great to be here with you at CBS. I think CBS is always a great way to kick off the year because you get so much perspective. So it was fun giving the keynote last night. Look, Helios is a massive system. You can see it in the background here. MI 355 is just an incredibly powerful chip. And probably the context I would give Ed is, one of the things that we're so clear about is that the demand for AI compute is just continuing to increase. And we have seen that over the last five years. When you think about just how much new capabilities have come on board, we've now seen a real inflection in the number of people who are using AI. So if you think today there are probably more than a billion active users using AI, and we expect that to scale to over 5 billion users over the next five years. So for all of that, you need compute and lots and lots of compute. And from that standpoint, MI 355 is a significant leap forward in terms of technology capability, made up of two and three nanometer chips, 320 billion transistors, just a lot of performance.
And what's the timeline for it to be deployed in the real world? And then who will be the principal first user of it?
You'll see in the second half of 26 and it will ramp from there. And we have very strong partnerships. Greg Brockman was on stage with us last night talking about all the use cases that they see. And we've announced a partnership with Oracle, many others as well. So given that it is in full production now, it's getting ready to ship.
That's at one end of the sort of scale and spectrum. At the other, you have MI 40X, which is for small data centers. I'm trying to simplify it, but it's basically an enterprise product. What was it that you were trying to solve for with that?
Yeah, I think what we're trying to solve for is the world is a very heterogeneous world. You have all kinds of use cases for AI from the very biggest cloud data centers that are doing large scale training and inference to enterprise applications as well as supercomputers. And so we actually have a family of chips. At the highest end is MI 455 for the cloud environment. But we announced last night MI 40X, which is actually using the same basic building blocks but is now really focused on enterprise applications so that you can go into current data centers with the new technology. So we're excited about that as well. Enterprises are starting to increase their adoption of AI. In some cases, they want their own control of their data centers in terms of on-prem environments.
What are they doing with it, though? I mean, we've been so fixated on frontier models with hundreds of billions of parameters and the scale of infrastructure needed for that. With MI 40X, we're talking about something slightly different. I think it's just really interesting if you could explain what the demand is from those enterprises, what they want with the technology.
Well, I think you see many enterprises now using AI all throughout their business processes, whether you're talking about things in their workflow. Even AMD, we're using AI through every part of our development process. We're seeing a lot of applications in financial services, in health care. These are areas, especially in financial services, that people actually don't want everything necessarily in the cloud. They'd like to be able to have their own on-prem deployments or private cloud deployments. And in this case, you don't want to have to build a brand new data center for every new generation of chip. MI 40X allows us to use some of those existing data centers and upgrade with the new capabilities.
Welcome. If you're watching us on Bloomberg Television or you're listening on Bloomberg Radio, we're live in Las Vegas and we're with AMD CEO Lisa Su, and we're talking about the latest generation of accelerators. What makes this generation of AMD accelerators the better option, particularly for on-prem and at the edge over what NVIDIA is offering right now?
Well, the best way to think about it is we're in this place where AI is at an inflection point. We're seeing now in every part of compute. We see it in the largest models. When you're thinking about things like ChatGPT and Gemini and Grok, we're also seeing many use cases in new capabilities, like video production, entertainment, health care, where you're doing drug discovery, all of these various things. Our claim to fame is really outstanding performance at very advantageous total cost of ownership. And the other thing that we believe very strongly in is an open ecosystem and deep partnerships with our overall ecosystem coming together. So when you put those things in perspective, I think we have a great set of applications that will take advantage of these newest generation chips.
You mentioned that Greg Brockman, who's the opening president, was on stage with you last night. And one of the basic points that he made was there are tools and functions they would love to release and put out into the world, but are compute constrained. I often ask you to quantify demand, but is there a way to quantify the severity in the lack of compute, the deficit that's out there right now?
Well, let me just give you some numbers to kind of ground what we think the demand environment is looking like. So if you think, today we have about a billion active users and we're ramping that to 5 billion over the next five years. And we have about 100 zettaflops of compute all around the world. I mean, that's just a generic number that aggregates all of that. We think we have to increase compute by another 100 times as you go over the next four or five years. And I introduced a term last night, the yottaflop. People are like, what is a yottaflop? A yottaflop is actually 10 to the 24th in terms of flops. So that's a one followed by 24 zeros. And to give you just a view of just how much things have really increased, that's another 100 times more compute than we have today. So that gives you an idea. Now you think, what are you going to use all that compute for? The truth is the models that we have today are great. They do amazing things. We talked about a number of use cases. Perhaps one that hits very close to home is writing software. People are using the tools right now to significantly enhance the productivity of software developers. But it's good, but it can get so much better. And I think that's the key point. We like to say that AI is really going to be everywhere and it's really for everyone and it's for each one of us to make our businesses more productive, each one of us more productive going forward. And so we're still in the very early innings of really unlocking the power of AI.
So where we stand is there's a compute deficit. And software is kind of hit the limits of what current generation compute can offer. Help us understand the bottlenecks and barriers to deploying that compute. A lot at the moment about memory chips. What else? Energy, electricity. What's crossing your desk, Lisa, that gives you pause and say this is a problem right now?
Well, our job as a technology industry is to push the bleeding edge. That is our job. And so when we think about MI 455 deploying two nanometer and three nanometer chips, having the latest generation memory, high bandwidth memory that is out there and really deploying these big systems, the important thing is that the entire ecosystem come together and we plan together for this next big inflection in compute. And that's exactly what we're doing right now. I think we're working very closely with the entire supply chain to ensure that we have the resources to expand this compute environment. And yes, some of the things that you mentioned are constrained, but which is most severely? I don't think that's any one thing. I think we're all looking at how do we build faster? Certainly power is one of those areas where you see throughout the world, power is being brought online as fast as possible. Certainly from a silicon standpoint, we're ramping our production capabilities with our partners. From memory standpoint, our partners are ramping as well. So it's not any one thing. I think it's all of these things have to go in tandem. And that's why partnership is just so important in this business.
We started this conversation talking about Helios, first rack scale architecture and infrastructure from AMD. Could you talk about the future and how much of the content you want to own in a server? We started this story with the CPU. Frankly, if you look at what NVIDIA is doing, they want to increasingly own all of what's inside the box. Is that something that AMD is focused on too?
What's most important for us is to ensure that we have turnkey solutions that are very, very easy for our customers to deploy. Because when you think about how do you use all of this AI compute most effectively, you want it to go into the data center and really be up and running on day one. And for that you have to optimize a full system. But from that standpoint, we are very focused on an open ecosystem. So yes, we designed the CPUs and the GPUs and some of the networking elements, but we also work really with the broad ecosystem of partners with industry standards. It's all about ensuring that we get the best of all worlds when we put our solutions together.
Looking ahead to MI 500 in 2027, that has 1000 times the performance of the 300 generation. So your last generation of real world deployed gear, something's coming. It's a thousand times better. How did you make it a thousand times better? It is just incredible engineering at every level.
MI 355 is ten times better than the MI 300 that we just launched six months ago. MI 500 is another 10x on top of that. We are using the most advanced technology out there. We have a very clear focus on hardware, software, system co-design. And it is clearly pushing the bleeding edge of capabilities.
What is the status of AMD's ability to sell products into China right now?
