View From The Top with Lisa Su: Chair and CEO of AMD
Stanford Graduate School of Business48:022025-02-26English
Chapters
In an interview at Stanford, AMD Chair and CEO Lisa Su used AMD’s turnaround to make a point: when technology reaches a turning point, don’t chase every hot market. Pick a large market where you have an advantage, and give the team enough time to deliver on the strategy.
Overview
View From The Top with Lisa Su: Chair and CEO…
- Lisa Su talks about her family’s influence, the part-time job she happened to land in a semiconductor lab at MIT, and why she chose to make products she could see and touch.
- She looks back at how managing projects, dealing with a chip failure just before launch, and serving as technical assistant to IBM’s CEO taught her to lead teams and give newcomers a chance.
- She explains why she joined AMD as the industry shifted to mobile devices and why, after becoming CEO, she stopped chasing mobile chips and laid out a multiyear plan for high-performance computing.
- She moves from AMD’s routine reviews to the balance between U.S. regulations and global operations, and explains why different AI applications need different computing resources and why open software lets customers choose their own components.
- She uses DeepSeek to show how open innovation leads to new uses and discusses AI’s potential to improve productivity, five-year milestones, and how outside feedback can help refine a plan.
- The audience asks about inference cost and accuracy, chip manufacturing in the U.S. and Taiwan, data center investment, model training, and whether to start a company. The session closes with hobbies and advice for students.
Key points
- Lisa Su believes leadership is developed through practice: put people with potential in roles they aren’t fully ready for yet, and give them plenty of support. Her own time as technical assistant to IBM CEO Lou Gerstner was an opportunity that changed how she saw her career.
- AMD evaluated the hot mobile chip market but decided it wasn’t playing to its strengths, so it bet on high-performance computing instead. The team rebuilt its products from scratch and allowed three to five years, or roughly three product generations, for the strategy to pay off.
- Company culture isn’t a slogan on paper; it’s what employees see people do every day. At AMD, that includes reviewing every product launch, especially looking at mistakes to see what could be improved in the next generation.
- Lisa Su doesn’t see training and inference as an either-or choice. The cloud, the industrial edge, and personal devices need different kinds of computing power, while training, inference, and reinforcement learning rely on similar underlying technology.
- AMD’s open software approach lets customers and partners choose their own components. The trade-off is more work to make those components work together, with the expectation of a larger developer ecosystem.
- She thinks the more interesting thing about DeepSeek is how its open model sparks further innovation, not the exact cost of training it. The cheaper and easier AI is to use, the more uses are likely to emerge.
- She is optimistic about long-term demand for computing power and believes shortages of chips, electricity, and data centers have held back AI progress. For individual inference queries, she wants to lower costs, shorten wait times, and improve answer accuracy.
Chapter notes
From Taking Apart Toys to Making Chips
- Lisa Su was born in Taiwan and moved to New York with her parents at age two. Her father came to the United States for graduate school. Her family placed a high value on education, and her parents always encouraged her to aim high and accomplish something. She says that encouragement shaped her whole life.
- When she was a child, her older brother’s remote-controlled car suddenly stopped working, and Lisa Su wanted to figure out why it had broken. She always liked tinkering with things. She focused on math and science in high school and went on to earn her bachelor’s, master’s, and doctoral degrees at MIT.
- She hadn’t planned a career in semiconductors. While looking for a part-time job in her first year of college, she happened to land in a semiconductor lab, where she put on a cleanroom suit for the first time and helped a graduate student run experiments. She was amazed that such a tiny chip could do so much. She fell in love with semiconductors and kept working in the field throughout college and graduate school.
- When she finished her PhD, she had to choose between going into industry and staying in academia to become a professor. She felt that staying at the forefront of research was hard and wasn’t her strength. She preferred making products she could see and touch: she could go to Best Buy and see some of her team’s products, or go to Lawrence Livermore National Labs and see large supercomputers.
- She worked at Texas Instruments for less than a year. At the time, she felt Texas wasn’t right for her, so she moved back to New York, where her family lived. She sees IBM as the place where she grew professionally. She rose to become director of emerging products there, then joined Freescale Semiconductor as CTO.
From Engineer to Team Leader
- Lisa Su worked at IBM for about 13 years. She believes career growth comes from learning something new at every step. Management appealed to her because there’s only so much one researcher can do, whereas leading a team of 10, 50, or even 100 people lets you accomplish far more than you could on your own.
- She had wanted to get an MBA but never managed to set aside two years for it. She learned most of what she knows about management on the job, by trying her hand at leading people, getting exposure to the business side, and taking responsibility for larger projects.
- The first product she worked on was a microprocessor. Just before launch, the chip wouldn’t work, and the team didn’t know why. She believes curiosity about what went wrong matters, and that a difficult, urgent problem can help a team focus, pull together resources, and get creative. A leader’s job is to bring the team together to do something nobody thought was possible.
- Lisa Su believes leaders can’t guarantee that someone will succeed, but they can spot potential and give people opportunities. Leadership isn’t something you’re born with; you develop it through different experiences. About five years into her time at IBM, someone gave her an opportunity that made her quite nervous at the time: she was asked to meet with then-Chairman and CEO Lou Gerstner.
- As Lou Gerstner’s technical assistant, she explained the latest technology to a CEO who didn’t come from a technical background. She also got to see up close how the person at the helm of a global company worked. She considers it one of the most influential experiences of her career because it showed her what being a CEO was like.
- Developing talent means being willing to put someone in a role when you’re not yet sure they can handle it, while giving them enough support to succeed. She also advocates trying a different job every few years, because each new experience is a chance to learn.
Taking Over AMD and Betting on High-Performance Computing
- Some mentors advised Lisa Su not to join AMD because its performance was up and down and its execution was inconsistent. She still wanted a chance to lead a company that could influence the industry. Processors, she believed, were the “brains” of most devices, and that was what drew her to AMD.
- Lisa Su joined AMD in 2012 to oversee its business units. At the time, the company didn’t yet have clearly defined divisions or clear accountability for profit and loss. The industry was shifting from PCs to mobile devices, and both AMD’s business and its leadership were in transition. She became COO in 2014, and just six months later, the chairman called to ask her to take over as CEO.
- Lisa Su took over as CEO in October 2014. According to the host, AMD was then on the brink of bankruptcy and its shares were hovering around $3. During her tenure, they later rose above $200 at one point. She believes the balance sheet and profit and loss matter, of course, but a technology company can’t win by cutting costs alone. What matters more is deciding where the company is going to excel.
- As smartphones rapidly caught on, AMD spent considerable time evaluating the mobile chip market and found that, while it was a good business, it wasn’t what AMD did best. In semiconductors, decisions made today often take three to five years to pay off, so Lisa Su chose to bet on high-performance computing, where she believed AMD had a chance to be the best.
- Lisa Su looks for three things when choosing a market: it has to be big enough, the industry has to be approaching a turning point, and the company has to have an advantage of its own. AMD’s strength was building high-performance computers, so she set a clear direction and gave the team time to develop products from scratch: three to five years, spanning three product generations.
- The slowdown in Moore’s Law was the turning point AMD saw at the time. Computing power could no longer roughly double every two years while costs came down, as it had before, opening the door for new technologies. Lisa Su sees the experience as more than a return to profitability: the key was seeing where the industry was headed and focusing resources and attention in that direction.
A Learning Culture, Chip Policy, and AI Strategy
- Lisa Su believes company culture isn’t about what you write down; it’s about what employees see and feel every day. At AMD, pursuing cutting-edge technology means putting in extra effort, taking some risks, and learning from everything the company does.
- AMD learns more from mistakes than from successes. After every product launch, the team reviews what it could have done better, and with every generation, it considers what to change in the next one. The point of all that learning and collaboration is to build the best technology and products and win in the market.
- AMD is a U.S. company that operates globally. It wants to serve a large global market while making sure its critical technologies comply with all U.S. regulations. Lisa Su believes it’s hard to know where to draw regulatory lines and that companies should take an active part in the discussion to help find an approach that respects the interests of both sides.
- Lisa Su thinks the pace of AI adoption over the past 18 months has been especially striking. AI had been around for a long time, but it used to be hard to integrate into business workflows. Generative AI has made it possible for more people to use it in different settings. AMD’s goal is to match the right computing power to each application, from large cloud environments to industrial AI and robotics at the edge, all the way to personal computers and phones.
- Lisa Su sees how much emphasis to put on training versus inference as a tactical decision for a given year. AMD’s long-term direction is to provide the right computing power for models of different sizes running in different environments. Large open and closed foundation models will coexist with small and midsize models, some of which can provide personal AI features directly on a phone.
- AMD chose open software so customers and partners could pick the best components for each part of their systems. That means more work to get the components to work together, but AMD believes it will ultimately result in a larger developer ecosystem. The company is also working to make AI easier to use.
From DeepSeek to Five-Year Goals
- Lisa Su thinks the point isn’t whether DeepSeek cost a few million or tens of millions of dollars to train. What matters more is that it combined existing work in a new way to create an open model, letting other people build on it to develop applications tailored to their own fields.
- Innovation leads to more innovation: the cheaper, easier to use, and more widely adopted AI becomes, the more uses will emerge. She thinks AI adoption is just getting started. Today’s capabilities are still fairly basic compared with what could be possible, and it will take several rounds of learning and improvement.
- Lisa Su prefers to set five-year milestones for herself and the company rather than ask what they should become in 20 years. Two years is too short, and too much can change in ten years. She wants AMD to play a major role in making computing power available around the world, because computing can fundamentally improve productivity in businesses and everyday life.
- More promising than making existing work more efficient is the possibility that AI could change workflows and business models and help tackle problems in areas like healthcare, drug development, and climate change. She sees cutting the time it takes to solve problems from years to months or even weeks as a milestone worth pursuing over the next five years.
- In the face of rapid change, Lisa Su recommends always being in learning mode. An MBA or PhD isn’t job training; it’s a way to learn how to think, solve problems, and anticipate what’s coming. She also wants to combine AMD’s expertise in hardware, software, and systems with its partners’ strengths in applications and understanding end-user needs, so that “one plus one is greater than three.”
- As research and new product launch cycles get shorter, you need both a clear roadmap and a deliberate effort to get feedback and check whether you’re headed in the right direction. Lisa Su talks to customers, partners, and people in the industry to refine her priorities and decide what’s most worth doing next.
Audience Q&A and Rapid-Fire Questions
- Lisa Su thinks AI still needs to bring down the cost of each inference query, shorten wait times, and improve accuracy. Longer queries can sometimes take 30 to 60 seconds, and despite clear progress over the past 18 months, the answers still can’t be trusted completely. AI agents could move from providing information to completing tasks on their own, but only if they can do those tasks correctly.
- To make supply chains more resilient and ensure critical components are available locally or regionally, Lisa Su supports bringing more semiconductor manufacturing back to the United States. But she doesn’t think it can happen overnight: Taiwan’s supply chain is more efficient and still accounts for the vast majority of advanced semiconductor production. She expects more capacity to move gradually to the United States, given the CHIPS Act and the Trump administration’s push to bring manufacturing investment back.
- Asked about the risk of building too many data centers, Lisa Su puts more weight on long-term demand. Over the past 18 months, shortages of chips, electricity, and data centers have limited further progress in AI. She believes that, under scaling laws, more computing power can produce better answers and drive wider use of the technology around the world. Overall, she is optimistic about data center construction.
- Asked whether AMD can challenge Nvidia in pretraining hardware, Lisa Su said she doesn’t think only one technology will be used. Inference, training, and reinforcement learning are connected and require similar underlying technology. She said a fair amount of training already runs on AMD hardware and more will run on it in the future, but she didn’t give a specific timeline.
- There’s no single answer to whether you should start a company after graduation or join a large one. Starting a company can give you access to funding, mentorship, and support, while a large company lets you try things that are hard to do in a startup. Lisa Su chose a large company not because she wanted to become CEO as quickly as possible, but because she thought she had a better chance of making a significant impact on the industry there.
- In the rapid-fire round, Lisa Su said she likes boxing for exercise and enjoys Texas Hold’em. If she weren’t leading AMD, she’d like to improve her golf handicap. She encouraged Stanford students to dream big and act on those dreams now, and described AMD’s future as “Phenomenal.”
What it answers
- Why didn’t AMD chase the hot mobile chip market at the time?
- Lisa Su thought mobile chips were a good business, but not the business AMD did best. High-performance computing made better use of the company’s technical strengths, and the team set aside three to five years for research and development.
- How can leaders develop talent that hasn’t been fully recognized yet?
- Don’t wait until someone has proved they can do the job before giving them a chance. Give them a challenging role and enough support, and help them build their skills by giving them experience in different roles.
- Is AMD betting only on inference and staying out of training?
- No. Lisa Su sees the balance between inference and training as a tactical choice that can change over time. She views inference, training, and reinforcement learning as a continuum that shares underlying technology, and expects more training to run on AMD platforms in the future.
- What still needs to improve for AI to be more useful?
- She wants to lower the cost of each inference query, make answers closer to instantaneous, and improve accuracy. For agents that can carry out tasks on their own, reliably getting the task right is especially important.
- How does she view increased U.S. chip manufacturing and Taiwan’s role?
- She supports increasing domestic U.S. manufacturing to make supply chains more resilient, but stresses that it can’t happen overnight. She points out that Taiwan still produces the vast majority of advanced semiconductors.
Quotes
What leaders do is they actually bring teams together to do something that nobody thought was possible.
Actually, we learn more from our mistakes than we learn from our successes, because with every product launch we can say, "Hey, we coulda done that a little bit better."
There's no one-size-fits-all in AI.
You get these degrees to learn how to think, to learn how to solve problems, to learn how to really kind of see the future.
Mentioned
- Lisa Su — AMD’s chair and CEO and the subject of the interview, she took over as CEO in October 2014.
- AMD — The semiconductor company at the center of the interview. It bet on high-performance computing during its turnaround and aims to provide the right computing power for different AI applications.
- MIT — The school where Lisa Su earned her bachelor’s, master’s, and doctoral degrees. She found her career direction through a part-time job in a semiconductor lab in her first year.
- IBM — The company where Lisa Su worked for about 13 years, moving from research into managing projects and people.
- Lou Gerstner — IBM’s chairman and CEO at the time. Lisa Su served as his technical assistant, which gave her a close look at the top management job at a global company.
- Freescale Semiconductor — The company where Lisa Su was CTO and ran a relatively large business unit before joining AMD.
- Mark Papermaster — AMD’s CTO, whom Lisa Su called her partner in driving the company’s turnaround.
- 摩尔定律 — Lisa Su saw the slowdown in Moore’s Law as an industry turning point when AMD was developing its high-performance computing strategy.
- NVIDIA — The company contrasted with AMD’s open approach in the interview. Lisa Su praised its end-to-end AI capabilities.
- DeepSeek — Lisa Su used its open model to show how a new approach can inspire other teams to keep innovating.
- CHIPS Act — The law Lisa Su mentioned when discussing increased domestic semiconductor manufacturing in the United States.