AMD’s Lisa Su has already beaten Intel. Now comes Nvidia

When Lisa Su became CEO of Advanced Micro Devices in late 2014, the company was in dire straits and on the brink of potential bankruptcy.

In its core market, computer processors, AMD wasn’t competitive with rival Intel. The company had billions of dollars of debt, and had committed to manufacturing chips for customers that may or may not exist. By the end of the year, AMD was worth a mere $2 billion in market value, while Intel was valued at about $180 billion.

“I actually had mentors in my career saying, you know, I don’t think that that’s a good move,” Su, 55, said in a speech at Stanford University in March, reflecting on her decision to take the gig a little over a decade earlier.

Su is in a very different spot today.

AMD passed rival Intel in market value in 2022 and is now worth $172 billion, a roughly 85-fold increase during Su’s tenure. Millions of gamers rely on AMD processors every day as they power up their Microsoft Xbox and Sony PlayStation consoles. AMD chips are so important that the U.S. government sees them as critical to national security.

Yet AMD still views itself as an underdog. That’s because it’s a distant second in artificial intelligence, behind Nvidia, the almost $3 trillion behemoth that dominates the market for graphics processing units, or GPUs. To have a serious role in the future of technology, AMD knows it needs a bigger chunk of the AI GPU market, where the biggest tech companies in the world are spending many billions of dollars a year on advanced infrastructure.

Underpinning AMD’s strategy is Su’s belief, rooted in her engineering background, that success in this industry is driven by making the right technical decisions that lead to the highest-performance chips. It can take a long time to see the results show up in products.

“One of the things that I like to say about the semiconductor industry, or technology in general, is the decisions that we make today, you will really see the impact three to five years down the road,” Su said at Stanford. “It is all about making the right bets.”

Su stands out in Silicon Valley, especially during Women’s History Month, because she’s the only female CEO among the top 10 semiconductor companies by market cap. Chips remain a male-dominant field, with women accounting for just 30% of the industry’s headcount, according to Accenture.

She’s also been the highest-paid female CEO for five years running, according to an Associated Press survey.

The stats aren’t favorable across the corporate world. About 10% of Fortune 500 CEOs are women, with 52 in the group in 2024, up from 41 in 2021, according to the Women Business Collaborative. 

Su, who’s known to make bets in her personal life — she’s a fan of Texas Hold ’em, and has been spotted playing poker with her sales staff — is making a long-term professional wager on AI.

AMD has committed to announcing new AI chips on an annual basis and building out a software division with thousands of employees to create open-source tools for its chips that can compete with Nvidia’s dominant CUDA language. 

Su is deeply involved with the company’s technology road map, frequently visiting the company’s chip labs, said AMD technology chief Mark Papermaster.

“Lisa is so admired by the engineering team because she knows what they do, and they know she knows it,” Papermaster said.

AMD declined to make Su available for an interview.

Su was born in Taiwan in 1969 and came to the U.S. as a child so her dad could go to graduate school. She’s described her childhood as a “typical Asian upbringing” centered around school. She attended The Bronx High School of Science in New York.

She first encountered the semiconductor industry as a freshman at the Massachusetts Institute of Technology, where she did her undergraduate work before pursuing a Ph.D. in electrical engineering. Her research centered on the manufacturing side of chips. As a graduate student in 1992, she wrote an award-winning paper about testing silicon-on-insulator technology. That ended up being one of the core production methods used to improve chips during the 1990s and 2000s.

Su went to work at Texas Instruments after graduating and then IBM, overseeing research and design teams focused on chip manufacturing. There, she was tapped to be the technical assistant for Lou Gerstner, IBM’s chair and CEO at the time.

After IBM, Su became chief technology officer at Freescale Semiconductor, a particularly rare feat as even today only about 8% of tech companies have women in the CTO role.

In 2012, she was hired by AMD’s then-CEO Rory Read to run the company’s business units.

Su’s ascension to the top role in 2014 kicked off one of the biggest turnarounds in the history of the technology industry. When she joined AMD, it was a company at a low point and many people believed it might not survive much longer.

“AMD was having some serious cash issues,” said Patrick Moorhead, founder of Moor Insights and a former AMD executive who left the year before Su joined. “AMD had to completely press the reset button.”

On her first day as CEO, AMD’s stock was trading at just over $3 per share. It’s now at more than $106 per share. AMD is the fourth-most valuable U.S. chip company, ahead of Intel, and behind Nvidia, Broadcom and Qualcomm.

The rebound started with tough decisions nearly as soon as Su took over. That included cutting 7% of AMD’s staff.

She also turned to Sony and Microsoft to build AMD’s game console chips, a move designed to help the company’s cash flow and to fill the chip factories it had already committed to using.

But Su said, during her talk at Stanford, that it wasn’t the cuts that saved AMD. Rather, it was the bet on developing new technology.

“It’s fairly clear in tech, there’s no such thing as cutting yourself to be a winner,” she said. 

Su turned AMD’s attention to products, and she encouraged staff to prioritize building the highest-performance chips. She also made the decision to stop selling lucrative server chips until AMD had products that could compete with Intel on performance, Papermaster said.

One big problem for AMD was that its processor design was aging and quickly.

Su approved a “blank sheet” approach that emphasized a redesigned compute core — AMD’s most valuable intellectual property — from the ground up. It took years, and AMD didn’t start selling products based on its “Zen” core until 2017. The technology is now on its fifth generation of improvements.

On Su’s watch, AMD was the first major company to embrace a technology called “chiplets.” Instead of manufacturing one big chip with all the elements needed — the compute cores as well as an input and output block — AMD could make smaller chips and then assemble them together.

Chiplets allowed AMD to be more flexible and more efficient in manufacturing. The change also allowed AMD to use different foundries for different parts of the chip, and reduce the risk associated with manufacturing, a critical consideration due to the company’s tight cash position. It was a good fit for AMD, especially as it transitioned in 2019 to use Taiwan Semiconductor Manufacturing to build its chips.

The focus on chiplets allowed AMD to grow at a time when Intel was suffering from its investments in the wrong manufacturing technologies, said Scott Thompson, professor of electrical engineering at University of Florida, who has followed Su’s career since her doctorate.

“Lisa Su knows fundamentally how chips are made,” Thompson said. “She understands the risk with different technology options, and she can assess that risk and make the right choice.”

Intel declined to comment.

Chiplets are now used by nearly every major processor company, and the technology was a core part of AMD’s first “big GPU.”

AMD’s experience in chiplets enabled it to take an announced data center GPU — originally designed for supercomputing, which is different than AI applications — and simply swap out the CPU chiplets to make it one large GPU better suited for AI. AMD’s first big GPU, the Instinct MI300X, launched in 2023, just over a year after OpenAI launched ChatGPT and kicked off the generative AI boom.

AMD’s revenue in 2024 jumped 14% to about $26 billion, nearly five times more than when Su took over as CEO. The company spent $6.5 billion on research last year, over six times its outlay in 2014.

But its current relationship to the GPU market leader, Nvidia, resembles its standing against Intel in CPUs a decade ago.

AMD has AI accelerators on the market, has told customers and investors when to expect future chips, and recorded $5 billion in AI chip sales last year, up from $100 million in 2023, according to Papermaster.

Still, it lags far behind Nvidia, which reported $115 billion of data center chip and networking sales in its latest fiscal year.

AMD shares are down more than 10% so far this year after dropping 18% in 2024. That followed a 128% pop the prior year.

The biggest challenge for Su may lie in software.

AMD’s hardware is comparable to Nvidia’s chips and can even win in some benchmarks. The company’s AI chips are viable, and Su is building their credibility.

While AMD is the only real alternative to Nvidia, most AI developers already know how to use Nvidia’s proprietary CUDA software, which allows programmers to access core hardware parts of a GPU for AI development.

For Su, the answer is ROCm.

That’s AMD version of CUDA, and it’s free to use. Su is counting on ROCm to help AMD win over developers who want more than just what Nvidia offers.

Nvidia did not respond to a request for comment.

Last year, AMD reorganized some of its software groups, including the teams working on ROCm, into a new AI software division. AMD also began to open source many of its important components and hardware details, allowing programmers to more easily access the raw power of the company’s AI chips.

To achieve widespread adoption of ROCm, Su will need her team to hit up AI developers and make efforts to shake off AMD’s reputation as a hardware-only company.

In December, Su personally reached out to one public critic of AMD’s software, posting on social media after the meeting that “we have put a ton of work into customer and workload optimizations but there is lots more we can do.”

Su generally plays down the market share rivalry with Nvidia. She sees the AI chip market growing to $500 billion per year by 2028 — the same size of the entire semiconductor market before the AI boom.

She’s counting on AMD to get a big chunk of it.