# Why AMD CEO Lisa Su is so confident about the AI boom | press Q&A > Source: > Published: 2026-07-26 15:30:00+00:00 *Become a member of GB MAX to gain exclusive access to the industry and to the most influential global B2B leadership community in the business of gaming, entertainment, and tech.* [Join now](https://go.gamesbeat.com/gb-max/)and also get a VIP ticket to GamesBeat Next (Nov 2-3, SF).Lisa Su, CEO of Advanced Micro Devices, unleashed a broadside of competitive announcements last week about AI processors and graphics processing units and other major moves in the AMD ecosystem. She spoke briefly with the press about her views on AI and how the company’s portfolio of new products is highly competitive with rivals such as Nvidia and Intel. The questions here came from a wide variety of press. She also made bold predictions about how big the total available market for AI products will grow 45% a year through 2030 to $1.4 trillion — compared to her previous estimate of $500 billion made just half a year ago. “No matter what, it’s up and to the right,” Su said. She said AMD’s approach isn’t just about designing and making one AI chip. It’s about a whole portfolio of chips. On stage with Su were Vamsi Boppana, senior vice president of AI at AMD; and Dan McNamara, senior vice president of compute & enterprise AI at AMD; The session was moderated by Caroline Guss, chief communications officer at AMD. AMD is in the quiet period now and so the questions weren’t focused on financials. Here’s an edited transcript of the press Q&A. **Lisa Su: **I’ve already done a lot of talking, so I’m not going to do a lot more right now. It’s been a great pleasure to be able to talk about our entire product portfolio. You can see and feel the excitement that we have around AI, the AI adoption curve, how we’re working together with our top customers to get the full capability of our product portfolio, the full spectrum of giving AI developers–this event is actually very focused on AI developers. Seeing so many developers who are, many of them, learning about AMD for the first time is what we’re excited about. I’m happy to talk about anything you guys would like to talk about. Let’s turn it over for questions. **Question: Given the comments you’ve made about the growing TAM that you’re seeing, it suggests that even the past few months it’s changed a lot. What makes you confident it’s going to grow at that rate, even though we’re seeing reports like enterprises having to watch their token budgets and other types of pressures? Why do you think it will keep growing like that?** **Su: **Let me start with just taking a step back to say–predicting a TAM is one of the hardest things you can do, because you’re probably not going to be right. But what we can say is, we spend a lot of time with our customers, really understanding what’s happening fundamentally at the workload level. We see, right now, tremendous demand for compute. It’s true on the GPU and accelerator side, but it’s been even more aggressive in terms of the rate of change of the TAM on the CPU side. To your question about what gives us the confidence that’s going to continue–we’re in a place–you heard it from our largest customers – OpenAI, Meta, Anthropic – that we have to plan multiple years in advance for us to be ready. Ready with the chips, with the power, with the data centers, with the entire supply chain. We’re all working much more closely together. We see that it’s the classic case of the more useful AI gets, the more you want to use it. Of course we’re going to optimize. That was some of the commentary that Dan had working with our enterprise customers. But no matter what, it’s up and to the right with AI adoption. We see it make a larger and larger impact on–let’s call it product development. Those key areas that enterprises need it for, for value creation. It’s much more than a cost reduction, which is what it originally started as, and much more now about real enterprise value creation. For that reason, we feel good about the demand curve. We also all now recognize the importance of–it’s not just about one chip. It truly is about the entire family or portfolio of chips and hardware. That’s one thing we have uniquely always believed in over the last few years. **Question: I’ve been in and around the ecosystem for about 30 years now, aging myself a bit. Getting to Helios was a challenge, but also a milestone. Can you talk about the complexity of what it takes to get there? And also, second half deployments. How does AMD define deployment? Can you talk about your confidence in getting there?** **Su: **Let me start first with a very clear acknowledgement. Hopefully you saw it from the hardware we had on stage. Helios is incredible. It’s an incredible feat of engineering capability that has incredible compute, cooling, integration, density, all of those things. Now, I think the important thing for people to remember is we have been planning for this moment for the last several years, in terms of both what we have built internally, Vamsi and his entire team on the platform and the software capability for Helios, Forrest and his entire team on the networking capability, our acquisition of Pensando, our acquisition of ZT Systems, our integration of all those capabilities. Perhaps the secret sauce that I hope you can get a feel for from today is, we are in lockstep with our partners. In other words, we are co-developing. It’s not like AMD is developing and then we hand it over to our partners. OpenAI is co-developing with us. Meta is co-developing with us. You saw some of the comments from Tom Brown at Anthropic. We’re in a place where–I think we are highly confident in the ramp. Declaring production means we are ready to ship. We’re going to start shipments at the end of the third quarter. They’re going to continue to ramp into the fourth quarter and the first half of next year. We have planned the exact data centers that the first shipments will go into. We’re working very closely with our OEM partners, who are together ensuring–I think the difference with Helios is, every part of the supply chain needs to be aligned. That has also been one of our key vectors. We feel really good about it. **Question: You announced that you’re launching a data center rack, the partnership with Cerebras. It looks like you’re still following Nvidia, depending on the timeline. I guess I wonder is, what can you tell me about your plan to gain technical or marketing leadership over the company? Especially given that they can outspend you in a number of different ways. Or is your strategy more around what kind of market share you can take, and being second place is fine in the world of expanding AI because the market is going to be so big it doesn’t matter?** **Su: **Our view is, we are absolutely paving the road for what we think AI compute needs. That is, GPUs are very important. I think we feel great about the road map with GPUs. As I said, a little bit of a preview as we go from MI450 to MI500, we’re not doing a small step. We’re actually doing another major leap forward. We believe we will have leadership in the scale-up compute domain. That’s a big statement. But that’s what we see. I think we have chosen a different path. Our path has been leaning into chiplet architectures and portfolio architectures for the last, I would say, practically the last 10 years, but especially the last five years. When you look at our Epyc portfolio–I appreciate that everyone thinks CPUs are interesting these days. We’ve been thinking about this moment for years. The way these chips have been optimized–Mark Papermaster and his team have really thought about, how do I create an ecosystem such that I have the right chip for the right workload? That extends throughout our road map to the work we’re doing on the Instinct side. I think we feel very good about where we are. It’s a huge market. For me to say that we believe the AMD market opportunity is $2 trillion, that’s pretty large. But I think our portion of that market–we fully expect to grow ahead of the market across the segments we’re in. We’re going to bring our take to it. Our take is about leadership technology. It’s about open ecosystems. It’s about consuming the right compute for the right workload. It is with that breadth and depth. **Vamsi Boppana:** Specifically, there’s always going to be a segment that will emerge. We have to stay focused, with disciplined execution, on what needs to be in front of us. Otherwise it’s easy for us to get distracted. But there will come a point in time where we’ll pick a segment and say, this will be undisputed, where AMD will lead. I’m very confident in Lisa’s comments. **Su:** We can say we have undisputed leadership in CPUs. **Question: Some of the big news today is that you announced this partnership with Anthropic. Last year, you announced a similar partnership with OpenAI, for OpenAI to buy your chips, deploy your chips. The deal was structured a bit differently at the time. You granted OpenAI awards that could potentially equal 10% of AMD if that deal were to be executed upon. Why were these two deals with these two frontier AI labs structured differently? What has changed in the market since last year that made you think about the deal with Anthropic differently? What is the status of the OpenAI deployment, which I believe is supposed to start in the second half of 2026?** **Su:** First of all, we are extremely pleased and honored and excited, all those words, to be dealing with the most important frontier AI model companies in the world. OpenAI, Anthropic, Meta, and others. The way I would explain it is, every deal is different. When we think about what we have with OpenAI, it’s a very special relationship. Hopefully you saw some of that with Sasha and Philip. We started early with them, with MI300. We’ve worked on road maps together. We’ve worked on software together. Frankly, they have bet big on AMD with an up to six gigawatt deployment. We said the first gigawatt would start in the second half of 2026 and continue into 2027. I can say that’s exactly on track. Frankly, without being specific on customers, I would say our general demand for MI450 and Helios is above our original expectations. Customers are seeing what they see in the hardware and they’re liking it. We’re increasing production capacity as we go through the next few quarters. As to your question about Anthropic, I’m thrilled to be working with Anthropic. I said it a bit on stage. We have really had our eye on Anthropic for the last several years, when Dario and Tom first started Anthropic. With these things, you always have to find the right time to intersect. It’s a significant engineering effort for anyone to bring on a new chip. From that standpoint, we recognize that for Anthropic to invest their engineering capability, resources, capital, and time in AMD is a big thing. Each deal is different. The Anthropic deal is very specifically around MI450, up to two gigawatts of MI450. The first deployments will start in the first half of 2027, for the first gigawatt. We’re working with them on longer-term plans as well, but they’re just structured a little bit differently. **Question: I was hoping to get more color on the Anthropic partnership. What does this look like? Are they throwing their specs, their needs over the fence to you, and you’re meeting them? Does the collaboration also extend to securing data center space? Leases, backstops, power, that kind of thing.** **Su: **The relationship with Anthropic really starts foundationally at providing a significant amount of capacity for them as they scale up. You should expect that they’re going to consume AMD GPUs and CPUs through a number of different sources, including clouds, neoclouds, as well as perhaps some of their own data centers. We’re working closely with them to enable their deployments. I think the scope of the relationship is beyond just, let’s call it, supplying chips, but it is really working together on the system bring-up. It’s also working together on Claude for AMD. Ensuring that from a software standpoint–it is really the idea that, if we look at both Claude and Codex, these are two of the most powerful and popular platforms out there. We want them to be able to have developers use them to develop on AMD at a very efficient scale. Those are the things that are included. From a data center standpoint, we’re continuing to be active in ensuring that there is enough capacity out there. Certainly that’s one of the areas we’re focused on. But that’s not unique to Anthropic. That’s really overall across the ecosystem. **Question: One of the big conversations, broadly, around AI over the last couple of weeks in policy circles has been this question about open source AI and how we should deal with it. Is distillation the standard practice, or is it a bad thing? Should the U.S. have access to Chinese open weights models? I know you have a lot of discussions with folks. What is some of the advice you give on the role–again, we saw this week with the Hugging Face thing. They needed Chinese open source to handle the intrusion they detected because they couldn’t use a frontier model. You always have a fairly nuanced take on all this. What’s some of the advice you would give?** **Su: **This is an area that’s very fast-evolving. I don’t want to say that I have all of the answers. But I certainly have some opinions. The opinions are really around the fact that–I think open source is a great thing. It is, because it gives people a level of transparency and control that enables you to do a lot. Now, that being that case, I think we are looking at, as we’re an ecosystem that has many different types of models – foundational models all the way to open models, to open weight models – the key for us in that framework is–I think we’re all going to use what works the best for us. This active conversation about restricting open models is an area where we all believe that they have a significant place in the ecosystem. We just have to make sure that we manage all the pieces of that. **Boppana: **There are some things that can be accomplished through regulation. Constitutions about how models are built and distributed and so on. But over time, it’s our belief that the open source will also evolve to ensure that such regulation becomes part of the open communities, which will regulate themselves. We’re seeing a lot of interesting startups that are doing this. For example, models that are being trained with open constitutions. That would be a future that would be a lot more scalable. Of course you always have some regulatory environments you operate in, but there has to be some other environmental things that emerge in the ecosystem itself. **Question: The Cerebras announcement today wasn’t necessarily surprising, but given the $40 billion your competitors spent for much the same reason–given the resources and experience that AMD has with chip design, should we expect the partnership with Cerebras to be a long-term one? Or is this something like a stopgap while you build something similar?** **Su: **Let me say, I wouldn’t view anything as–we don’t start anything with the notion that it’s going to be a stopgap. We start with the notion that–I think Cerebras has very interesting technology. There are lots of ways to get workload-specific acceleration done. I think Cerebras has a very interesting technology that works very well with Helios. The idea of our open ecosystem is that we will work with a number of different companies that may have technology that can be useful. In terms of what we see in the long term, the long term view is–we’re going to see more workload disaggregation as we go forward. It’s to be expected. The rate and pace will be determined a bit by the cost structure of these price points. But from our perspective, we have a very strong road map with our baseline MI400, MI500, MI600 road map. We also have the ability to mix and match different compute technologies. We’re excited about the relationship with Cerebras. You can expect that we’re going to do more workload disaggregation going forward. **Boppana: **Just to add a quick point, you actually don’t have to look too far. If you look at our current portfolio, we have little AI engines that do super low latency things, super low power things, real time responses. If you look at our embedded – cars, automobiles, medical instruments. The central point we believe in is that AI is going to be pervasively infused into all forms of computing with different characteristics. We will have solutions that are ready to solve them.