Zhi Dong
Author | Chen Jundao
Editor | Yun Peng
Zhi Dong, August 14th. Recently, Dr. Lisa Su, Chairman and CEO of AMD, gave a deep interview to Wired magazine, sharing her latest insights on key topics such as export controls, competition with NVIDIA, the potential of AI technology, and the battle for talent in Silicon Valley. She believes that even if the United States does not export AI chips, AI technology in other parts of the world will continue to progress, and export controls are just "noise" to her.
Although AMD is facing Intel or NVIDIA in multiple areas, Su does not want the outside world to directly compare AMD with these two companies - AMD has a presence in CPU, AI accelerators, personal computing, and data centers. She believes there are over $50 billion market opportunities in these areas in the next three to four years.
Su hopes that AMD's AI accelerators will become the preferred choice for companies like OpenAI, Meta, and xAI, but she is not in a hurry. She emphasized that AMD will not change its understanding of compute power because of low-cost models like DeepSeek, as they have already bet on inference computing; although AMD has its own AI models, the purpose is to test and improve products, not to compete with large model vendors.
Regarding the criticism that AMD's ROCm ecosystem lags behind NVIDIA's CUDA, she believes the difference is not about who is better, but more about developer habits, and said AMD is "accelerating forward," making up for libraries and operators, expanding hiring and acquisitions, and continuously listening to developer feedback to improve the software ecosystem. However, they will not recruit people with nine-digit salaries like Meta, because in the semiconductor world, achievements have never been achieved by one person's single breakthrough.
When talking about the future of AI, Su believes AGI will come, but does not think AI will eventually be smarter than humans, and the current popular intelligent agents are only doing trivial tasks. In the future, AI can be used to enhance productivity, reduce repetitive work, allowing humans to focus on more interesting work; secondly, solve extremely complex problems, such as greatly compressing the development time of chip design, which originally took several years.
Su also lowered her guard with Wired magazine journalist Lauren Goode, sharing some very personal topics. Due to her mother's experience at the end of her life, she realized that current medicine is more of an art than a science. In the tech field, they integrate complex systems to work together, but in medicine, different departments have not yet achieved this.
Additionally, Su revealed that she does not like others repeatedly asking about her relationship with Jensen Huang. Although it does not annoy her, her inner thought is: "Really? Is that really the most important topic we can talk about?"
In the end of the interview, Su defined herself as a "pragmatic super-technological optimist" - she believes technology can greatly improve our life experiences, but the way to achieve it needs to be done step by step, constantly learning, listening, and adjusting.
One, Even if the US does not export chips, it will not affect AI development
CEOs of large US tech companies have recently often needed to express opinions on hot topics - this was not common before. Su said she indeed felt the "responsibility" on her shoulders has increased, when it comes to technology policy and the development of the semiconductor industry, AMD has a responsibility to express its views and help decision-makers formulate policies based on facts.
Question: What do you hope the Trump administration and the public have a better understanding of the AI accelerators your company is developing?
Su: We, as tech companies, actually benefit from having more users. Limiting the number of users in the ecosystem is bad for AMD, and also bad for the US. Because there will always be other choices in the market.
Some people think that if we don't export chips around the world, the AI development in those regions would stop - that's not true. AI progress will not stop, it will continue to advance. We would rather they develop on our technology, rather than on someone else's technology.
Question: Recently, there have been some incentives to bring chip manufacturing back to the US. What is the most complex part of rebuilding American manufacturing?
Su: I fully support bringing manufacturing back to the US. Completely support.
Question: Why?
Su: Because it is crucial for national security and economic interests. A few years ago, central Texas suffered an ice storm, and traffic was almost completely shut down within days, and several wafer plants were forced to halt production. For businesses, supply chain diversification is fundamental.
Bringing manufacturing back to the US takes time, but it is feasible. The previous idea was "the US cannot do the most advanced manufacturing," but now in TSMC's Arizona factory, they are already producing the latest generation of server processors, and the operation is very good.
This shows it can be done. Although the cost will be higher, that's fine. I think we need to change a mindset - not always pursuing the lowest cost option.
Question: When you became president and CEO of AMD in 2014, did you think you would need to make so many comments on geopolitical and social issues? Do you feel more pressure in your dialogue with the current government?
Su: It is indeed different from before. I won't say we are politicized, nor do I say I am politicized. You won't see me commenting on general social issues, because that is not where I can truly add value.
But when it comes to technology policy and the development of the semiconductor industry, we must participate. I wouldn't call it increased pressure, but increased responsibility, because we want the rules to be correctly set.
Question: You are one of the most well-known female leaders in the tech and semiconductor industries. Why do you think you have "little value" on these issues? Do you feel more pressure in your dialogue with the current government?
Su: Maybe the term "little value" is not accurate. More precisely, my personal opinion may be interesting, but what is more important is to set policies based on facts.
Two, Not wanting to compare with Intel or NVIDIA, hoping to become the preferred choice for companies like OpenAI
In 2024, Su was named the Best CEO of the Year by Fortune Magazine. From 2012 when she joined AMD until now, she has led the company to gain a significant market share in the CPU market. According to PassMark data, the latest market share of AMD in the entire CPU market (including personal and server) is 38.9%; at the same time, AMD's stock price has risen from about $1.8 in November 2012 to around $175 today. Now, AMD is also competing with NVIDIA in the AI accelerator market, and Su said she expects to become the preferred choice for manufacturers in the AI field, but not in a rush.
Question: The question that has annoyed you the most recently might be "When will AMD..." For example, when will AMD surpass NVIDIA in the AI GPU market? Ten years ago, people might have thought this idea was absurd, but now some people think it is worth discussing.
Su: When I first became CEO, people asked me, "Why did you take this job?" I was confused - I thought, "Are you joking?" This was the best opportunity for me. We are a company that has long underperformed in an important industry, and I had the chance to lead the team to do something important. If you ask me what I wanted to be when I grew up, my answer wasn't "to be a CEO," but "to do meaningful things."
At that time, people always compared us with Intel, and we had to defend ourselves. I told my team, "We know what we can do, let the world see." So, for me, these "when" questions are not something I like. Media obviously like these black-and-white comparisons, right?
Question: I don't think it's just black-and-white. AMD's situation is more complicated, because you have many clients that rely on x86 and CPUs, and your data center business has grown rapidly, from $6 billion in 2022 to $12.6 billion last year. But at present, NVIDIA is the competitor that people are most concerned about.
Su: My view is that I don't want to be compared only with Intel or NVIDIA. My vision - and also our vision - has always been "there is no one-size-fits-all solution." Data centers need the best products, CPUs need the best, AI accelerators need the best, and personal computing also needs the best. Our product portfolio is broad, and the market size is huge - over $50 billion in opportunities in the next three to four years, and we have many opportunities.
Question: At AMD's AI event in June this year, Sam Altman also attended. You have collaborations with OpenAI, Meta, Tesla and xAI under Elon Musk. But the reality is that these companies love NVIDIA GPUs, and AMD is just "an alternative." Do you hope that one day you can become their "preferred choice"?
Su: Of course. That's the position we have in the CPU market today. If you ask many of these companies, they will say that AMD is their strategic CPU partner. We also expect to do this in the AI field, but I'm not in a hurry.
Question: You don't want to set a timeline for this goal?
Su: For example, when I joined AMD in 2012, Microsoft was just an early partner in our gaming business. Over the past decade, we have built a lot of trust, and now we work together with them - Microsoft announced that not only the next generation of Xbox consoles will use AMD technology, but also the entire cloud platform will adopt AMD.
With Meta, it's the same. The first time I talked to them, I said, "Give me a chance. I don't have to tell you I'll be number one, because I know I'll prove it to you. I'll be your best partner, not only technically, but also helping you build your technical infrastructure." That's what we're doing.
Three, Training models is not for competing with AI vendors, will not offer nine-digit salaries to recruit talent
AMD has always been considered the biggest competitor of NVIDIA. In the past year, the company has launched multiple AI computing chips, narrowing the gap with NVIDIA's top GPUs. At the same time, AMD is also continuing to catch up in the software ecosystem, and Su shared their strategies in this market.
Question: At the AMD AI event, you mentioned China's DeepSeek large model, which is said to have much lower training costs and compute requirements. Will such models change your view on compute power?
Su: This is just another example of the rapid changes in AI workloads. In the past, everyone focused on large-scale training, but now with the development of inference models and fine-tuning, the industry has shifted to a faster-growing stage of inference computing. This is why hardware must be flexible enough.
So it didn't change our thinking, because we have always believed that inference is more important. In a way, we made the right bet. We have optimized aspects critical to inference computing, such as memory capacity.
Question: NVIDIA itself trains models and provides the NeMo framework for developers to build generative AI models. Has AMD seriously considered training its own large models?
Su: We are indeed training our own AI models. We have an AI model team, but the purpose is not to compete with large model vendors, but to learn. The more we use (test) our own products, the more we can learn, and then we can improve and build new features faster.
Question: Analysts and some customers say AMD is very customer-centric. But some technologists believe that AMD's software tools ROCm are still not as good as NVIDIA's CUDA. What specific measures will you take to attract more developers?
Su: I agree that software is the most critical layer, because it's where developers directly interact. The difference between ROCm and CUDA is not about who is better, but about CUDA being around for a long time, and people are used to its ecosystem, so we are teaching them a new ecosystem.
Question: What I've heard is that it's not just about habit. Some people say "the compiler is not easy to use", "the performance library is not good enough", "we hope for higher portability".
Su: Of course, but part of the reason is that they are used to the way CUDA works and hope ROCm can be exactly the same. In AI customers, I haven't met any that couldn't run their business and have good performance. Although we don't have all the libraries, many special operators are still being added. Answering your question "what have we done", my answer is - we are accelerating forward.
I have long understood that when facing criticism, it's not necessary to agree, but it's necessary to understand it as a perspective, and then decide how to respond. We have a lot to do, so we are hiring a lot, acquiring, and listening to developer voices. I think we have seen that progress can be achieved quickly.
Question: How do you see Meta's Super Intelligence Lab, and Zuckerberg's reported willingness to offer nine-digit salaries for AI talent? What impact will such salaries have on Silicon Valley recruitment?
Su: Frankly, I haven't encountered this directly. Talent competition is very fierce. I believe money is important, but it's not the only factor in attracting talent. Salaries should be in roughly the same range, but more importantly, people should believe in the mission you are doing.
At AMD, I feel that the salaries are okay, because the stock price is good. From a recruitment perspective, we always ask: "Do you want to join our mission?" Most people are attracted by this journey - if you want to do important technology and make an impact, not just a gear in the machine, but someone who truly drives the roadmap, you would want to join AMD.
Question: Would you consider offering nine-digit salaries to build the software ecosystem?
Su: I won't.
Question: Because you have to explain to shareholders or the board?
Su: Because in our world, achievements have never been achieved by one person. Of course, we have many very talented people. For example, we acquired Nod.ai, and Anush Elangovan, the CEO of Nod.ai, now oversees the entire software ecosystem. His enthusiasm is amazing - anyone with a problem with ROCm, he personally communicates to resolve it. I am looking for people like this who are passionate about their work, and there are many such people.
Four, AI will not be smarter than humans, the tasks currently performed by intelligent agents are too "trivial"
Question: How do you understand "super intelligence"?
Su: I think AI can make each of us "smarter," which is a beautiful vision, and we are just beginning.
One of the areas I am most concerned about is healthcare. I have personally experienced the shortcomings of the medical system, and I think it should be much better. We should be able to cure more diseases, not just rely on trial and error. AI is a perfect example of this - connecting drug development, therapies, hospital treatments, etc., which is a field that is very suitable for transformation. I'm not sure if this is called "super intelligence."
Question: Some people think AI will eventually become smart enough to eliminate humans. What do you think of such predictions? Do you believe in AGI?
Su: I believe in AGI, but I don't think AI will be smarter than humans. I also don't believe in apocalyptic theories. Technology is great, but its goodness depends on the people who create and guide it. So I think these discussions are a bit abstract. Our focus is: "The technology is good, but not good enough, how can we make it especially good?"
Question: How do you measure "especially good"?
Su: When AI can solve truly difficult problems, it is "especially good." We often say that agents are the next big thing, but most agents are currently dealing with relatively trivial tasks.
Question: For example, putting items into a shopping cart?
Su: Yes, something like that. I think AI has two directions for the future. One is purely to enhance productivity - how to remove the tedious, repetitive work from humans' hands, allowing them to focus on more interesting things. This is what we are already doing.
The other side is using AI to solve very, very difficult problems, such as compressing the time it takes to solve a problem that would have taken ten years into six months. Imagine, for example, that designing a chip now takes three years, and if I could complete it in six months, what would that look like?
Question: But will there come a day when humans can't keep up with this speed?
Su: I don't know. But I would bet on humans being able to cope.
Question: Technology sometimes does make people a bit overwhelmed.
Su: That's actually the point. When technology becomes mature enough, you don't need to think about it. Today, when you use it, you still have to think - which one do you use more? ChatGPT or Grok?
Question: I use ChatGPT. Not every day, but-
Su: Often?
Question: Sort of, my job is to test these things.
Su: Yeah, but you still have to check, "Is the answer correct?"
Question: Right. Especially as a journalist. We clearly draw the line between "using it to cook steak" and "using it to write news."
Su: But you do use it for research.
Question: Sometimes, but the "hallucination" issue is still very concerning.
Su: That's the problem - it's not good enough yet. But one day, it will be good enough for you to trust it directly.
Five, Today's medicine is art rather than science, wants to change this status with AI
Question: You mentioned medicine earlier. When we get old and sick, will the doctors who treat us be ChatGPT doctors?
Su: I hope that generation of doctors can have access to massive data like ChatGPT, so that they can have more solid evidence when making diagnoses.
Question: From a philosophical perspective, do you think AI is a revolution like the Internet, or more like Linux - becoming an operating system running on all devices? Or is it electricity? Fire? I remember Sundar Pichai once compared AI to fire, saying its transformative power is extremely strong.
Su: The Internet is a good analogy, but I think AI is more than that. The Internet is more about the transmission of information, while AI is a more fundamental transformation in productivity. Some people compare it to the Industrial Revolution, and I think that's also appropriate.
Question: But in previous revolutions, we didn't have to worry much about the issue of false information.
Su: You can have two perspectives. One is to find ways to limit AI because it may be risky; the other is to develop it as fast as possible, but add proper filtering to the information. I firmly believe in the latter. So I don't think there will be a situation where "few people are needed." Because ultimately, it's people who determine the truth. We are still hiring more engineers, because they are the final judges of engineering.
Question: Do you still have hope in humans?
Su: Yes. And it will be better in the future. Like the Internet today, you take it for granted. We shouldn't evaluate technology based on its current level, but on its "upward curve" - what we can achieve in the future. We will do it, although there may be some fluctuations along the way.
Question: Do you seem to have some concerns about AI, are you being a contrarian?
Su: I think those who are most likely to benefit from technology often have the conditions to be more optimistic and actively promote it. That old saying you know - "the distribution of the future is not uniform." Even if medical technology improves, we will still see bias, leading to some people being denied medical care or insurance. We have seen such things.
For me, medicine is a very personal matter, because my mother was very ill. I watched her entire medical process, and realized - no matter who you are, you can't guarantee the best medical care, because the current medical system is more of an "art" than a science. And I think it should be a science.
Question: Why do you think it's "art"?
Su: The human body is an extremely complex system. There are cardiologists, nephrologists, but there aren't many general practitioners who can integrate these together. This seems tragic to me. I think: Please, this can be solved.
In the tech field, our work is to integrate complex systems so they can work together. But in medicine, we often only focus on part of health. I firmly believe that if we can combine this professional knowledge with technology, we can better treat patients. I have seen this firsthand. So, in my next life, if I have time to do something besides my current work -
Question: Will you become a doctor?
Su: No, but I hope to be someone who can bridge this gap and let technology play its real role.
Question: Actually, you can do it in this life.
Su: I still have a lot to do at the moment.
Question: Did your mother recover?
Su: Unfortunately, no. But you understand what I mean, right? I just realized - wow, this could happen to anyone.
Question: Yes, but even with the best doctors, the key is the quality of medical care. About a year ago, my mother also had a serious health crisis and was put on a ventilator in the ICU. Doctors kept coming to look at the images, but couldn't find the cause. I was sitting in the hospital thinking - I report on AI medical breakthroughs every day, but we can't even understand the images?
Su: You completely understand what I mean. What made me most angry was that my mother stayed in the ICU for 60 days, and everyone said, "No one can survive." They said, "She can't do it." But I thought, "She can, I know she can." Although I wasn't the one making the decision, she did it. She lived for two more years after that.
Six, Others ask about relationship with Huang Renxun, after becoming CEO release through boxing
Question: You often talk about "resilience." From a personal perspective - not the company - how do you maintain resilience? Is it through Starbucks? I noticed you also brought a cup at the AI event.
Su: Yes, it's a passion fruit tea lemon drink, which helps me a lot. I don't drink much caffeine, so I rely on it and a little exercise to stay alert. I have cycles with Starbucks, sometimes I drink a lot, sometimes I stop.
Question: Me too. What is your favorite form of exercise?
Su: I like boxing. I have a trainer who comes to my house, lets me punch him - of course, it's a punching bag, not a person.
Question: How long have you been doing it?
Su: About seven or eight years.
Question: So it was after you became CEO for two or three years that you thought "I need to find a way to release some stress"?
Su: Yes.
Question: How many hours do you sleep per day?
Su: Five or six hours. Six hours is the ideal state. On weekends, I can sleep up to seven hours.
Question: As a leader, what quality do you value most?
Su: Passion for work. It can carry you through the high points and also through the lows. Things will go wrong, but if you really love your work, you will shine.
Question: What is the most annoying thing for you?
Su: The most annoying thing? Hmm... I can't say it's being asked "Is Huang Renxun your cousin?"
Question: You can say it freely.
Su: Actually, it's not really annoying, more like - "Really? Is that really the most important topic we can talk about?"
Question: What do you think people don't understand about you, but you wish they knew?
Su: I think people know me pretty well? Not really? Well... I get up every day because I believe our products can change the world and make it better. External noise - like export controls - are just noise to me.
Question: So you're a super-optimist?
Su: I don't see myself that way, but I might be a super-technological optimist. I'm actually quite pragmatic. So I'm a pragmatic super-technological optimist. How does that sound?
Question: Sounds like it was generated by ChatGPT.
Su: That's not a pre-set answer. If I were to describe myself, I would say - I do believe technology can greatly improve our life experiences, so in that sense, I'm a super-technological optimist. But in terms of implementation, I'm practical. How to achieve it? Just step by step, constantly learning, listening, adjusting, and applying what you've learned. That's our way of working.
Original: https://www.toutiao.com/article/7538293300223066633/
Statement: This article represents the views of the author, and you are welcome to express your attitude below with the 【Up/Down】 buttons.