AMD CEO Lisa Su is undoubtedly a prominent figure in the chip industry, especially as we enter the new era of AI, her reputation has reached its highest point in a decade.
Looking back at her growth history, after obtaining a Ph.D. in Electrical Engineering from the Massachusetts Institute of Technology (studying for eight and a half years and obtaining three degrees in Electrical Engineering), Su began her career at Texas Instruments, where she played an important role in developing silicon transistor technology on insulators. Subsequently, Su worked at IBM for 12 years, leading the development of semiconductor copper interconnects, leading the team that developed the Cell microprocessor used in the PlayStation 3, and serving as the technical assistant to CEO Lou Gerstner.
After serving as the Chief Technology Officer at Freescale Semiconductor for a period of time, Su joined AMD in 2012 and was promoted to CEO in 2014.
In the ten years of leading AMD, Lisa Su has achieved extraordinary success—after decades of lagging behind Intel, AMD has developed the world's best x86 chips and continues to take a significant share of the data center market from Intel. In addition to traditional PC and graphics chip businesses, AMD is also a major player in the gaming console game field. With AMD competing with Nvidia in the data center GPU market, the GPU business is now increasingly in the spotlight.
Advertisement
Earlier, Su was interviewed by foreign media Stratechery, discussing her career path, including the lessons she learned during her promotion process, and then discussed why AMD was able to achieve such great success during her tenure. In the interview, Su also shared her views on "ChatGPT" changing the industry and how AMD is responding.
The following is the interview transcript:
I. Lessons Learned from IBM
Q: I know you don't want to talk too much about yourself, but I need some fact-checking. We just talked about how you were born in Taiwan, immigrated to the United States at a young age, and finally got into the Massachusetts Institute of Technology. It is said that you were hesitant between computer science and electrical engineering at the time, and finally chose electrical engineering, mainly because it was more difficult. Is that true?
Lisa Su: That's true. I have always been dealing with mathematics and science, and my parents always said: "You must do these difficult things." When I went to the Massachusetts Institute of Technology, I was choosing between electrical engineering and computer science at the time. Computer science, you just need to write software programs, and electrical engineering, you have to build things. I wanted to build things.Q: Your Ph.D. focused on silicon and insulator technology, and then you went to IBM. You pioneered the use of copper interconnects on chips. Regarding your experience at IBM and what you might have learned, I have three questions: First, speaking of copper interconnects, you once mentioned in an interview that after you completed this technology development, you wanted to embark on a new journey, but your boss asked you to stay. You felt that the actual learnings you accumulated at that time when you thought you were done were some of the most impactful. What were those learnings?
Lisa Su: I did indeed learn a lot during my time at IBM, which was early in my career. When you go to school and get a Ph.D., you think the most interesting things are the research you do and the papers you write; we all write papers and so on.
When you actually join a company and participate in a project, these projects often take several years to complete. But the exciting part is at the beginning when you come up with new ideas.
What I learned is that one of the first products I was involved in developing was a microprocessor with copper interconnects, and it turns out that the last 5% of what it takes to get a product out is probably the hardest, where most of the secret sauce is. If you learn how to do that, then frankly, it is.
Q: All the software engineers are saying, "Hey, it's the same for us too. Didn't you know?"
Lisa Su: (laughs) That might be true, that might be true. But we all have our own views on the "secret sauce." When problems arise, the secret sauce is yield and reliability. When you're trying to produce millions instead of just five, you learn a lot, at least I did.
Yes, as a young researcher, you would think, "Hey, I'm ready to start my next research," and you realize how gratifying it is to see your product actually shipped and on the shelves, where you can walk into Best Buy and buy it. These are the things I learned.
Q: Even today, do you feel that your time and energy ultimately achieve a balance between the goals you are building and the actual execution and fulfillment of your commitments?
Lisa Su: Of course, today I personally spend a lot of time looking at future roadmaps and technology.
Q: The following question is purely out of curiosity, how deeply do you need to be involved in such matters, not specifically about you, but AMD in general, since you are already a fabless company, how much do you need to be involved when considering the actual last mile? What is the level of interaction with TSMC or your packaging partners or any other companies, and how to truly improve yield?Lisa Su: As a fabless company or design house, that is indeed the case for us. We are actually engaged in end-to-end development, so you can imagine from the very first day of product conception—actually even before that, we are thinking about which technologies will be ready, what the next big thing we should bet on? This goes on throughout the entire process. Sometimes, it may take five years or even longer before the technology is truly realized. We are also in the final stages, ensuring that the product is delivered with high quality, appropriate yield, the right cost structure, and in large quantities.
So, this is actually end-to-end, and the difference is that it is not all done by one company, which is very common in a more traditional integrated manufacturing model, but rather through partnerships. We have found that it is actually very effective because experts from all parties are working together.
Question: During the IBM phase, I am curious about the second question: You were involved in the development of the PlayStation 3's Cell processor. This chip was a technological marvel, but the PlayStation 3 is considered the least successful PlayStation, which in the long run, prompted Sony's strategy to really shift from hardware differentiation to exclusive products. I think this question can be divided into two parts: the first is: What did you learn from that experience? The second question is related to this: How much impact did that experience have on you later? I am more curious about whether all the work you put into the Cell processor and its actual market performance has brought any management insights?
Lisa Su: Yes, it's interesting that you mention that. I have been involved in PlayStation development for a long time, and if you think about it carefully, PlayStation 3, 4, 5, etc...
Question: It's like a thread that runs through your career.
Lisa Su: Yes, across multiple companies. I have to be honest, these decisions are more about architectural decisions. From this perspective, whether it's the PlayStation console or other work we have collaborated on (we are AMD, but the situation at IBM at the time was similar), it is actually a close cooperation that customers or partners are trying to achieve.
At that time, the Cell processor was ambitious, considering the type of parallelism it was trying to achieve. I say again, from a business perspective, it was definitely successful. When you rank things, I think history will tell you that there may be different rankings.
Question: My point is that the era of game consoles has gone through several stages. In the stage of PlayStation 1 and PlayStation 2, they made wise hardware decisions, which made their approach quite different from Nintendo's. But once you move to HD, the cost of creation increases significantly, developers actively support multiple processors, and game engines emerge accordingly. Suddenly, no one is willing to bear the burden of differentiation on the Cell, they just want to run on the Cell.
Lisa Su: Perhaps some people would say that if you look back, programmability is so important.
To achieve real commercial success from day one, we must consider both hardware and software. As we have seen, over the past 10 years, one of the things I am most proud of in the work that AMD has done or has completed is the PlayStation 4 and PlayStation 5, where we have always made new leaps in hardware.And they are compatible with previous generations of products, which is very helpful.
Question: The third question about IBM is, you served as Lou Gerstner's technical assistant for a year. What did you learn from him?
Lisa Su: (laughs) You've done your homework, haven't you? The year with Lou was one of the most educational experiences in my career. IBM is a very good company in terms of talent development, so they identify them early in their career and ask them: "Hey, what kind of experience do you want?"
In my case, they asked me, do you want to go the technical route, or do you want to go more towards the management route, the terms being IBM Fellow or IBM Vice President. Honestly, I didn't think I was smart enough to be an IBM Fellow like Bob Dennard.
There are a lot of great people there, so I thought, "Okay, let me try management and business." They gave me an opportunity to spend a year with Lou, who is truly an amazing person. If you think about someone who has been out of school for five years and has really only done what we call pure engineering, then attending what is essentially the best MBA in the world.
What was most interesting for me was to really understand where he spent his time. He always spent time learning, very focused on external things, understanding market dynamics, understanding customer dynamics. How does this change your strategy, how does this change the way you guide the leadership team?
Question: Another thing I've always admired about Lou Gerstner is, as you said, he didn't just observe the market from the outside, understand what was happening, but really understood IBM, understood the intrinsic capabilities and unique differentiation of IBM. Basically, my point is, IBM is big, what does that actually mean? What kind of impact can you bring in some way? The whole middleware revolution, look, we can solve this internet problem for companies that are older and bigger than us, which will be a differentiated thing. But then, obviously, everything collapsed. IBM should do cloud computing, Lou actually wrote in his book, I don't know how much that is a retrospective. If you succeed him, can you lead IBM to greater heights?
Lisa Su: I don't know if I would go down that path. I was a semiconductor practitioner, and I am still a semiconductor practitioner. If you think about it seriously, IBM was a great career for me, but if I wanted to continue being a semiconductor practitioner, I had to go to a semiconductor company. So I went to Freescale (semiconductor company), taking on more business roles.
Question: Have you personally acknowledged, "Okay, now I'm a businessman"? Or did you choose this path and just move forward in this direction?
Lisa Su: I have always played both technical and business roles. At Freescale, I started as the Chief Technology Officer. I joined as the Chief Technology Officer, and after a few years, I eventually took charge of the networking and multimedia business, which was definitely a choice, and this choice was ultimately that I wanted to drive results, and driving results requires, yes, technology is great, but you need the right business strategy.Q: Does this limit many technologists? Do they underestimate all the outcome-driven factors unrelated to technology?
Lisa Su: I think this is something technologists must learn. By the way, there are many outstanding CTOs who really understand this. My current CTO is Mark Papermaster, who was my partner at IBM, we grew up together, and later we became partners at AMD, he really understands that technology is great, but you also need to drive business outcomes. This is why I like what I do, because yes, I can integrate outstanding technology with a great team, but also have the opportunity to drive very important business outcomes.
II. Entering the AMD Era
Q: Let's talk about AMD. I mentioned the game console strategy earlier, which was a major shift in focus after you joined. Was the idea at the time like "Look, this is an easy win, the sales are high, we can get back into gaming"? What was the idea at the time?
Lisa Su: Well, I would never say anything is an easy win.
First of all, I want to say that when I first joined AMD, our market share in the PC market was probably over 90%, by the way, I really like the PC market. I believe we will discuss this more. But I want to remind everyone of this point. The PC market is cyclical, and the cycles can be very intense.
They can be very striking. Therefore, from a business strategy perspective, at the beginning of AMD's establishment, diversification and formulating a strategy based on high-performance computing principles were very important for us. We are a computing company, good at building computing capabilities, so which markets can really utilize these capabilities now? Gaming is one of them, and we are very fortunate that Sony and Microsoft, the two leading game console manufacturers, chose us.
Q: Who drove the shift of game consoles to x86? How much did Sony learn from the Cell? Did you go to them and say, "Look, this is a feasible approach"? How did the universality of this architecture develop?
Lisa Su: Yes, I think this is a series of choices, so this is a choice between x86 and other architectures, if you are considering the developer ecosystem around x86 only when considering software development, I think this is a very critical part, but I don't know if the architecture itself is enough. I think the incredible graphics capabilities and graphics, especially if you want to customize graphics, few companies can do this, AMD is one of them.
Q: How integrated are the CPUs and GPUs you provide? AMD acquired ATI in 2006. So my question is, before you went to AMD, were there any other companies that could really provide the products you did for game consoles?Lisa Su: I believe we have been able to achieve this for two reasons. First, we have foundational IP, which is the combination of what we call CPU or microprocessor cores and graphics IP capabilities, and we are willing to customize. Frankly, we have a large team dedicated to these projects, to make customizations.
Question: Do you think this is a pattern: initially, everything revolves around cutting-edge technology for the best performance, but as it (I don't want to say it slows down, but as the functionality becomes commoditized) develops, customization becomes more important. For example, you acquired Xilinx.
Lisa Su: I think the best approach is to have a few principles. First, the fact is, the world needs more semiconductors. Semiconductors, chips are now the foundation of a lot of what we do, a lot of what we do, we call standard products suitable for a wide range of use cases. But you will find those high-volume applications, such as gaming consoles, such as some of the work now done in the cloud, such as some AI work that I believe will be customized, in these cases, because the quantity is so large, so customization makes sense. This is something I have always believed in. This is part of our strategy, and it is also part of our deep collaboration strategy. So if you have the right building blocks, then you can work with a wide range of customers to really figure out what they need to achieve their vision.
Question: But is there a situation where, as the process curve continues to decline, the design cost becomes higher and higher, and customization has a bottom line, and only AMD has enough scale to customize, is this somewhat contradictory?
Lisa Su: I think it's important to look at which markets are really suitable for large-scale customization, but it's not everything. Maybe your IoT device you wouldn't want to do this, because the return on investment is not high. But for large computing power, I think it requires a combination of the right IP and the ability to work deeply with partners. By the way, it's not necessarily about hardware customization, we can also do a lot in software, and I think this is one of the important trends for the future.
Question: So I have to ask, you came to AMD, stayed there for a few years, and then took over as CEO. Is this another example of a difficult choice?
Lisa Su: I think so. I can say that when I joined AMD, my real thought was that I have been working on high-performance processors all my life, this is my background, and in the United States, there are few companies that can let you do this kind of work. I have always respected AMD and thought it was an important company, but I thought I could make a difference, so after joining this company, I realized, "Wow, I still have a lot to learn." In the first few years, I did learn a lot about the market dynamics of this world, but it was also a great opportunity to make a difference.
Question: In which areas can you make a breakthrough? We can see the difference - I mean, just look at the stock chart, and we can see the performance of your chips. So, in this case, it may be difficult to go back to your exact mindset 10 years ago, but what was your plan at the time? What did you say, "Look, I can do this, there is a way, there is a path, I see it?" What path did you see?
Lisa Su: I clearly saw that we had the foundation to build an incredible roadmap. We are very different in these foundations.
Question: What are these foundations? Is it intellectual property or customer relationships?Lisa Su: High-performance CPUs and high-performance GPUs are our pillars. If you think about it carefully, you will find that these are all incredibly remarkable cornerstones. Now, what we lack is a very clear strategy, that is, what we want to become when we grow up, and the execution machine that can achieve this goal.
Therefore, from a strategic point of view, I think we have some choices. If you still remember, it was 2014, and at that time, the most exciting thing was mobile phones, such as application processors. So we would discuss, "Should we enter the mobile field?" Our answer was, "No, we shouldn't, because we are not a mobile phone company. Other companies do better in this area, and we are a high-performance computing company, so we must make a roadmap to make full use of our advantages, which requires us to reform our architecture, design, and manufacturing methods." I know how to do this, and it takes time. You can't do it in 12 months. I think it takes five years. It does take five years, but it is clear that we have these elements, and we just need to build an execution engine in a truly orderly manner.
Question: You just mentioned manufacturing. We know that before you took over, AMD had already spun off GlobalFoundries. I want to use professional terms here, how troublesome is the constantly revised wafer agreement between you and GlobalFoundries? Is this something you have to deal with constantly when trying to execute the strategy?
Lisa Su: AMD and GlobalFoundries used to be one company, yes.
That wafer supply agreement was also signed before I took office, but if you consider the several major strategies we must do, if you want to manufacture high-performance processors, you need the best technology partners, the best manufacturing partners, and GlobalFoundries is a great company, they were still a great partner at that time. It's just that you need scale to manufacture at the forefront, and the scale does not exist.
When they realized this and said, "We will not develop the 7nm process," it was a very good decision for both GF and AMD, and from a financial point of view, AMD had to return all the money it originally received.
Although there is business cooperation between the two parties, from a technical point of view, this is absolutely the right choice. As I said, GlobalFoundries is our excellent partner. I have great respect for [GlobalFoundries CEO] Tom Caulfield as a partner, and I think focusing on what each company is good at is beneficial for both companies.
Question: You are the first high-performance chip manufacturer to switch to chiplets, and now everyone is moving in this direction, so you must be leading in this area. Were you forced to do this because of the wafer agreement, so that you could do some batch production with GlobalFoundries and TSMC while still delivering chips?
Lisa Su: Not at all. In fact, I think this is obviously one of the best decisions we have made. Of course, we couldn't have anticipated this at the time.
What we were considering at the time was where Moore's Law was heading, and how could we stand out? Frankly, our idea at the time was that we needed to bring something different to the processor market, so manufacturing these low-yield, expensive giant chips was not the answer.I remember we spent some time with Mark and our architects trying to decide, "Is this the time for us to switch to small chips? Is this the time for us to bet the company on switching to small chips?" We said, "Yes, because we will get higher performance, more cores, and a better cost point," which gave us great flexibility, and we learned a lot in the process.
The first generation Zen 1 chip was decent, but we encountered some programming model issues that needed to be addressed, which were improved in Zen 2 and really made progress in Zen 3.
Q: In 2014, when you took over the company and felt you could make a difference, I saw several significant shifts. For example, you have switched to small chips, and at that time, TSMC was also starting or transitioning to EUV. To what extent did you see the long-term changes in the market and made the "look, I can do something here" decision?
Lisa Su: Yes, we did study the technology roadmap and TSMC's progress at the time, as well as the packaging technology at the time, and we decided that it was time to bet. I would say that the world we live in is that we must bet, sometimes it takes three to five years to achieve.
Q: Yes. I don't mind asking you about the decision in 2014, because today's important decisions are often made at that time.
Lisa Su: Exactly, and there are risks in doing so, such as "Can we really achieve the expected performance by adopting small chips?" But we learned a lot, and I think history will prove that we made the right choice, but at the time, some of our competitors called it glue, they glued the chips together. Just like "We are not gluing chips together."
Q: Now they are doing the same thing. Looking back at the past 10 years, AMD has achieved a real performance lead in the x86 field, between design decisions and TSMC's leading process, who do you think deserves the most credit? What about the return?
Lisa Su: I do believe they are closely related.
TSMC is an outstanding partner in this field. When you take a lot of design risks, you want to know if your technology is reliable, so you know where to spend your time and energy.
Q: This is what TSMC and ASML have done, such as adopting 300mm first, and then adopting EUV, this cooperation has proven to be feasible, and then both parties can cooperate at the same time.Lisa Su: That's right, I think this is a very synergistic partnership.
Q: Before your tenure, the most important moment for AMD was actually, as we've discussed before, their shift from x86 to 64-bit and cornering Intel in this regard, which is a story of both hardware and software. That was before your tenure, but I think one of the ongoing criticisms of AMD is the need for software improvement. Where is the software? You can't just be a hardware cowboy. When you joined, was there a feeling of, "Look, we have this opportunity, and we can continue to build on this over time"? What is AMD's cautious attitude towards software? How have you been working to change this situation?
Lisa Su: Well, let me be clear, there is no reservation at all.
I believe we have always believed in the importance of the combination of hardware and software, and the key to software is that we should make it easy for customers to use all the incredible features we have integrated into these chips, which is completely clear.
I think you will see that we are actually on several arcs of technology development. So, the CPU arc and everything we have done to build the Zen product portfolio. Now, we just previewed Zen 5 in the data center at Computex, and then launched it in client products. That particular arc is an arc.
Now we are in the next arc, which is AI and GPU.
Q: I want to ask you another thing. Regarding this trend, we talked about the chiplet trend, and we talked about EUV. How important is the rise of HPC to your success? Because what I see from it is that they are buying on a large scale, and they are actually doing LTV calculations to say, "Look, yes, these AMD processors are worth it in the long run." Second, if there is a software vulnerability, they will work hard to fill it because they can see the long-term benefits. When you consider what we can actually win here, has this had an impact on you? Is it a driving factor?
Lisa Su: Yes, your observation is very insightful. When you consider high-performance computing and how things have changed, the fact is that HPC is a very important part of the entire market, and we have spent a lot of time there. The point you raised is absolutely correct, that is - you want to think that the product always wins in every market, but that's not necessarily true. In the hyperscale computing market, the best product wins.
We have been able to prove this. Frankly, the key to this market is that one victory is not enough, and temporary victory is not enough. You have to win the roadmap, and that's exactly what we did at that specific point in time.
It turns out that there are indeed some customers who will make purchases according to the roadmap.By the way, they will ask you to prove this point. In Zen 1, they said, "Well, this is good," Zen 2 is better, and Zen 3 is much better. The execution of the roadmap has put us in such a position: now we have established very in-depth partnerships with all hyperscale enterprises, and we are very grateful for this. When you think about the AI journey again, you will find that it is a similar journey.
Question: There is another question about x86. How do you view the consumer field related to all of this? You can imagine a company like Intel, they must keep the wafer factory running at full capacity, so they need to maximize the utilization of chips to meet all the demands. The problem with the wafer factory is that Intel wants to achieve integration, while AMD is in a different position, so they can meet the needs of hyperscale producers, and they are better at manufacturing excellent chips. But, do you consider the quantity, just because you want to take advantage of the design costs and IP investment? I am just curious, in a world that is not your wafer factory, not your billions of dollars in capital expenditure, how are these calculations made? I am curious about your different views from the integrators.
Lisa Su: We believe that it is about scale. In 2014-15, we were a company with a market value of 4 billion yuan, and in this case, you can invest a certain amount of R&D funds. Last year, we were a company with a market value of more than 22 billion US dollars, and you can invest more in R&D.
This is the same as the calculation of how we use leverage.
Question: But if you invest too much in the wafer factory, the risk of bankruptcy may be reduced.
Lisa Su: Well, I think the key is to leverage IP. It is the engine, the computing engine that we have. This is definitely our top priority, to put these computing engines on a very positive development roadmap, and then we build products based on this.
III. Artificial Intelligence
Question: What was your reaction when ChatGPT appeared in November 2022?
Lisa Su: Well, this is actually the crystallization of the essence of AI.
Question: Obviously, you have been in the graphics gaming industry for a long time and have been considering high-performance computing, so the idea of the importance of GPUs is not unfamiliar to you. But has it changed the views of others around you, and what happened afterwards, are you surprised?Lisa Su: We place a high priority on the development of GPUs for high-performance computing and artificial intelligence. In fact, this might be a very important arc that we can trace back to a time frame after 2017. We have always been researching GPUs, but the real focus is—
Q: What happened in 2017 that made you realize, "Wait a minute, we have these, we thought we bought ATI to play games, but suddenly, a completely different application emerged"?
Lisa Su: This is the next big opportunity, and we knew it was the next big opportunity. This is something Mark and I discussed, that by placing CPUs and GPUs in a system and designing them together, we would get better answers, and the first near-term application was supercomputing. We were very focused on these large machines that would reside in national laboratories and deep research facilities, and we knew we could build these large-scale parallel GPU machines to achieve this. In the AI part, we also always believed that it was obviously a combination of HPC plus AI.
Q: You previously said that AI is the killer application for HPC.
Lisa Su: Yes.
Q: But when you talk to people in the high-performance computing field, they will say, "Well, this is a bit different," to what extent is this the same category and adjacent category?
Lisa Su: They are adjacent but highly related categories, which completely depends on the precision you want in computing, whether you are using full precision or some other data format. But I think the real key, and what we really have vision for, is that due to our chiplet strategy, we can build a highly modular system that can be called an integrated CPU and GPU, or it may just be the incredible GPU functionality that people need.
So, for me, the emergence of ChatGPT has made it clearer that now everyone knows the use of AI. Before, only scientists and engineers would consider AI, and now everyone can use AI. These models are not perfect, but they are very good, so I think how we can get more AI computing into people's hands as quickly as possible has become very clear. Because of the way we build and design systems, we can actually have two styles. We have the HPC-only style, which is what we call MI300A, and the AI-only style, which is MI300X.
Q: Is this an uncomfortable shift, such as, "Actually, no, we want lower precision because scalability is very important"?
Lisa Su: It's not uncomfortable. The pace is very fast.Q: Things are happening so fast. AMD has been performing very well, setting a historical high a few months ago. But overall, it is clear that Nvidia has taken the lead because it has a lot of momentum and room for growth. From your perspective, during that period, AMD needed to catch up, and what advantages does Nvidia have?
Lisa Su: I think the way to think about this is, where is the focus, relatively speaking—see, I have a lot of admiration for [Nvidia CEO] Jensen [Huang] and Nvidia. They have invested in this field for a long time until the direction of things became completely clear. We are also investing, although I would say we have several arcs. We have the CPU arc, and then we have the GPU arc.
Q: Hey, you're busy crushing Intel, so I understand.
Lisa Su: I want to put it another way, we are at the beginning of AI. I find a strange phenomenon is that people always think about technology in a short time. Technology is not a short-term movement, we are in a 10-year arc, and we may have already gone through the first 18 months. From this perspective, I think we are very clear about where we need to go and what the roadmap should look like. You mentioned software before, it's very clear how we make it easy for developers to make this transition, one of the great benefits of our acquisition of Xilinx is that we have obtained an extraordinary team of 5,000 people, including a large number of software talents, who are currently working hard to make AMD AI as easy to use as possible.
Q: One of the things that really impressed me in this comparison is that one of Nvidia's really smart moves is the acquisition of Mellanox and its product portfolio in the networking field, and integrating all these chips together, especially for training. In your Computex keynote speech, you talked about the new Ultra Accelerator Link and Ultra Ethernet Link standards, and the idea of bringing many companies together, which is a bit reminiscent of the Open Compute Project in the data center field. This is very reasonable, especially considering that Nvidia's proprietary solutions have the high profit margins we know and love, just like their other products.
But I think this is my question about your long-term development—do you think, from Clayton Christensen's theory perspective, because we are in the early stage of artificial intelligence, in many ways, more proprietary integrated solutions become the focus, which may not be surprising? To some extent, openness and modularity make sense, but may not be good enough for a while.
Lisa Su: I would say this. When you look at the market five years from now, what I see is a world with a variety of solutions. I don't believe in one-size-fits-all, and the beauty of openness and modularity is that you can... I don't want to use the word "customize" here, because they may not all be customized, but you can tailor them.
Tailoring is the right word—you can tailor solutions for different workloads, and I believe no single company can provide all possible solutions for all possible workloads. So, I think we will achieve this in different ways.
By the way, I firmly believe that these large GPUs we are going to build will continue to be the center of the universe for a while, yes, you will need the entire network system and reference system to come together. The focus of what we have done is that all these parts will become the reference architecture for the future, so I think this will be very important from an architectural point of view.
The only thing I want to say is that there is no one-size-fits-all solution, so modularity and openness will allow the ecosystem to innovate where they want to innovate. The solution you want for Hyperscale Enterprise 1 may be different from the solution you want for Hyperscale Enterprise 2 or 3.Q: So, where do you think the balance lies between the standard approach and "this is the Microsoft way," "this is the Meta way"? They have some commonalities, but in reality, they have been customized to a considerable extent according to their respective use cases and requirements. Similarly, this is not about next year, but about the long term.
Lisa Su: I believe that in the next three to four or five years, you will see more customization for different workloads, and the algorithms will — currently, we are in a period where algorithms are changing very rapidly. At some point, you will feel "hey, it's more stable, clearer," and in terms of the scale we are discussing, you can gain significant benefits from it, not only from a cost perspective but also from a power perspective. People talk about chip efficiency, system efficiency, which is now as important as performance, or even more so, for all these reasons, I think you will see multiple solutions.
Q: Is this a tailwind for your undervalued x86 business? In your keynote speech, you mentioned the fact that most CPUs in the cloud are over five years old, and you have said something like: "one of our CPUs can replace five or six old CPUs." Do you really think this is the case — because I think currently both your company and Intel are worried that all the spending is going into AI, and no one is buying CPUs anymore, is this a power wall? If we can take a bunch of CPUs out of the data center, can we save electricity by placing other CPUs?
Lisa Su: I think both points are correct. I think the modernization of data centers is absolutely necessary. It will happen, and then another point is — this may not happen now.
I think we are seeing investment returning to the field of modernization, but another thing that is really important is that, although we like GPUs very much, they are a huge growth driver for our future, but not all workloads will use GPUs. You will have traditional workloads, you will have hybrid workloads, and I think this is the key point of the story. In large enterprises, you have to do a lot of things, and our goal is to ensure that we have the right solutions in all these functionalities.
Q: How much inference do you think can actually go back to the CPU?
Lisa Su: I think a lot of inference will be done on the CPU, as you think, the very large models we are talking about obviously need to be done on GPUs, but how many companies can really afford the largest models? So, you can already see that for smaller models, they have been more fine-tuned for these things, and the CPU is fully capable of doing this, especially if you go to the edge.
Q: In terms of competing with Nvidia, you pointed out in the last earnings call that the supply of MI300 is limited, which is the fastest growth rate ever, but may be different from some investors' expectations and somewhat disappointing for the year-end forecast. How much do you think this shift in demand constraint is related to the launch of the 325, and the fact that Nvidia's overall supply has increased because everyone is trying to figure this out? Is your long-term opportunity to become this kind of customized supplier - a tailored supplier? Sorry, this is the word we want to say - rather than "see, I don't want to say buy, but as long as we need GPUs, we will buy from anyone." Where do you think your demand curve is in relation to competition and the rapid development of the field?
Lisa Su: Again, let me step back and make sure we can grasp the theme of the conversation. The demand for AI computing has exceeded expectations, and I don't think anyone would have predicted this demand, so when I say supply chain tension, it is to be expected because no one would have anticipated that you would need so many GPUs in this time period. In fact, the semiconductor industry is very good at building capacity, which is what we are seeing. As we began to predict -Q: So do you think it's more because there's a large supply online?
Lisa Su: Of course, that's our job. Our job is to make sure you're not limited by manufacturing capacity.
In fact, for us, it's about ensuring that customers really increase their workload, which requires a lot of in-depth work and deep collaboration with our customers. So to be honest, I am very excited about the opportunities here. We've been through this before, it's very similar to what we saw when we first added data center server CPUs, our customers worked closely with us to optimize their software, then they added new workloads, added more capacity, and that's what I hope will happen here.
The difference with artificial intelligence is that I think customers are willing to take more risks because they want to get as much benefit as possible as quickly as possible.
Q: Is this a challenge for you? Because being willing to take more risks means they are more likely to accept high profits to get a leading GPU or anything else, or to have the GPU with the largest ecosystem, the developer ecosystem?
Lisa Su: I would say I am very pleased with the progress we have made in software.
What we see is excellent out-of-the-box performance. In fact, everything works properly, and in fact, many developer ecosystems want to lift the abstraction layer because everyone wants to choose.
Q: Do you think you will enter a stage where the lifting of the abstraction layer will become a public layer across companies, rather than letting a company lift the abstraction layer internally, so that they can buy any CPU, but this is not necessarily beneficial to you entering another company, or do you think this will be -
Lisa Su: I absolutely believe it will spread throughout the industry. Technologies like PyTorch, I think PyTorch is widely adopted, and OpenAI Triton is also the same. These are larger industry things, and frankly, part of the desire is that it takes a long time to program into hardware. Everyone wants to innovate quickly, so from a quick innovation perspective, the abstraction layer is good.
Q: You are the second wave of adopters of TSMC's new node, possibly lagging by a year or a year and a half. Do you feel the pressure to rise to the top? Obviously, for some players in this world, you are a relatively small company, $22 billion is impressive, but you still need to consider the costs in this regard. Or do you just desperately need to be at the absolute forefront?Lisa Su: Well, I believe from the fabless perspective, in terms of overall output, we are definitely one of the top five, and having cutting-edge technology is certainly helpful. We don't consider whether we should do this or not; I think what we consider is from the roadmap perspective, for example, when we talk about the annual rhythm of GPU launches.
Q: Unfortunately, for you, the situation is a bit opposite to Nvidia, is that a bit frustrating?
Lisa Su: No, not at all. Again, one of the most important things for me is that our roadmap is based on what we think is achievable and what we think customers want and need.
Q: Is there any possibility that AMD might use Intel's foundry?
Lisa Su: I would say that we are very satisfied with our current manufacturing relationships.
Q: I did think of it, Intel and AMD have been one of the biggest competitors in the history of technology from the very beginning. But when you take a step back and think, do you ever step back in these conversations and is there a bit where you are fighting side by side, because the real enemy is Arm?
Lisa Su: You speak as if Arm is the enemy, but I don't think ARM is the enemy, so let me start with that. We use Arm throughout our product portfolio. I think x86 is an extraordinary architecture with capabilities, but please don't view AMD as an x86 company; we are a computing company, and we will use the right computing engine for the right workload.
This is related to my thoughts—-if you look at the semiconductor industry today, you will find that we have both competitive and cooperative areas. So, regarding what you mentioned about Intel, we do compete in some areas, but we also cooperate in some areas. Intel is part of the UALink consortium, and they are part of the Super Ethernet consortium.
Q: They are also very interested in this modularity and standardization.
Lisa Su: We agree with the idea that establishing a link that can span different accelerators is actually a good thing. So, I think the whole industry is like this. We are in a place where there are both competitive areas and areas where we can cooperate.Q: Over the past 10 years, you have achieved astonishing success in the x86 field, and your achievements in the server and data center fields are self-evident. Now, it's as if a new champion has emerged, are you ready to face the new round of challenges?
Lisa Su: This is the next arc. I can tell you, the achievements we have made in the field of high-performance computing today are astonishing. Who could have imagined this? It's like a new world. It's incredibly exciting.
Q: Do you feel energetic and ready to set off?
Lisa Su: Absolutely ready. Very ready.
Comments