AUSTIN, Texas—At a board meeting in late 2022, Lisa Su, chief executive of chip designer Advanced Micro Devices, announced that she was radically changing course.
“I’m going to pivot the entire company,” she told the directors gathered around a boardroom table at the company’s Austin campus. The rise of artificial intelligence was a “once-in-a-lifetime opportunity,” she said, and the company had to put AI at the center of its entire product line.
Three years later, the Santa Clara, Calif.-based company has nearly quadrupled in size, its market value rising from $90 billion to more than $335 billion despite a recent pullback. AMD’s strategy of positioning itself at the center of the global AI race has paid off handsomely, making Su into a billionaire and her company into one of the only viable designers of the powerful chips needed to power advanced AI models.
In a market that has in recent years been completely dominated by Nvidia, Su is showing that there may be a place for a strong No. 2 that can compete on price and keep AI developers from having to rely on a single chip company.
Su has a Ph.D. in electrical engineering from the Massachusetts Institute of Technology and a deep understanding of the physics behind her company’s products. She developed a reputation as a giant slayer by outcompeting market-leader Intel a decade ago to take the lead in producing central processing units, or CPUs, for personal computers and data centers.
Now Su is staring down the ultimate goliath in Nvidia, the world’s most valuable company and the foremost maker of the chips that power AI data centers. AMD will have to deliver on its promise to produce chips that are comparable to Nvidia’s—some of the most sophisticated precision electronics ever designed. And Su will need to prove her mettle in a less familiar set of roles: as a saleswoman, dealmaker and lobbyist.
Investors bid up AMD’s stock price in October after the company announced marquee deals with Oracle and OpenAI—both of which have agreed to buy tens of thousands of AMD’s newest generation of AI chips, known as the MI450.
The OpenAI deal “turbocharges our road map,” Su said, and represents a huge vote of confidence in the MI450, which is set to launch next year. The growing AI market is “a huge opportunity, and we want to make sure that we have deep partnerships that enable us to get a big piece of that, and, you know, the rest will handle itself,” the CEO said in an interview in Austin, the city where she’s lived for decades.
Su’s dealmaking was on display again this week, when the company said it would work with Cisco Systems and a Saudi Arabian AI venture to build a large cluster of data centers in the kingdom. Su attended a black-tie dinner for Saudi Crown Prince Mohammed bin Salman alongside other executives at the White House Tuesday, and AMD sent a representative to a U.S.-Saudi Arabia investment forum Wednesday.
“Up until now, data centers have been almost entirely Nvidia’s market,” said Gil Luria, a tech analyst with D.A. Davidson. “AMD is in a position to take a more meaningful piece of that.”
$1 trillion a year
At the heart of Su’s strategy is her belief that there is “insatiable demand” for computing power, and that as the market for AI grows, the companies offering the best and most reliable AI infrastructure will thrive. She has said she believes AI is “not a zero-sum game” and that recent concerns about an overheated market for chips and data centers are exaggerated.
In recent weeks, the massive spending on data centers has intensified concerns that an AI bubble is building. AMD’s stock, which had jumped nearly 60% in October, has fallen about 20% this month.
“I am not concerned about an AI bubble,” Su said in the interview. “I do think that those who are thinking that way are a bit too shortsighted. They don’t really see the power of the technology.”
Su said people who are willing to take “big, bold bets” are reaping the rewards—including skyrocketing share prices and increased market share—and believes that tech companies have barely scratched the surface of how AI can be used.
For many of the largest tech firms, including AMD, taking big, bold bets has meant entering into circular funding arrangements. To secure its deal to sell chips to OpenAI, for example, Su promised the ChatGPT maker warrants that would allow it to buy up to 10% of AMD at a deep discount. Such deals have added to concerns that the AI boom is being propped up by unsustainable financing.
“This is not the time to stay on the sidelines and worry, ‘Hey, am I over-investing?’” she said. “It’s much more dangerous if you underinvest than if you over-invest, in my opinion.”
A year ago, Sue predicted that the market for AI chips would hit $500 billion in sales annually by 2028. At an investor day presentation last week in New York, she was even more optimistic: The market for AI and data-center computing, she said, will reach $1 trillion a year by 2030.
AMD’s revenue in that segment will grow by 80% a year and the company will achieve double-digit percentage market share in AI chips over the next three to five years, Su predicted. Most analysts peg the company’s AI share at just 3% to 5% today.
The CEO said that the company is currently in talks with multiple customers about doing deals similar in scale to the OpenAI agreement, under which AMD is providing enough MI450 chips for OpenAI to deploy 6 gigawatts of data-center capacity—meaning the facilities would use enough electricity to power 4.5 million homes for a year.
Stealing significant market share from Nvidia is far from a foregone conclusion, investors and analysts say. AMD first has to execute smoothly on next year’s launch of the MI450 and demonstrate to the market that it can deliver OpenAI all the computing power it needs on a tight schedule.
Su is also hoping to capitalize on an inflection point in AI—many executives in the chips industry, including Nvidia chief Jensen Huang, expect demand to shift from clusters designed to assist in training large language models to what’s known as inferencing, or the billions of computations required to run those models and allow them to respond to user queries. Inferencing doesn’t require chips with as much computing muscle as training does.
As AI models like ChatGPT and Gemini become increasingly integrated into daily life, and as companies design thousands of enterprise-software tools that rely on AI models, demand for inference functions is “about to go up by a billion times,” Huang said last month. (Su and Huang are distant cousins, but they didn’t meet until both were established executives.)
AMD has a strong line of inference computing chips but has struggled to design chips powerful enough to compete with Nvidia in training.
After the investor-day presentation in New York, Su gathered with investors in a reception area on the 10th floor of Nasdaq’s headquarters, overlooking Times Square. As guests sipped wine and nibbled mini crabcakes, she admitted to a small group of investors and analysts that it’s been difficult to compete with Nvidia on training, but said she was optimistic that MI450 would help with both training and inference, said Hendi Susanto, a portfolio manager at Gabelli Funds, which owns more than $11 million worth of AMD shares.
“I have a high level of confidence that they’ll be able to win deals in inferencing and boost their market share meaningfully,” said Susanto. “What’s important now is strategic partnerships.”
Taking down Intel
Early in her career, the Taiwan-born, Queens, N.Y.-raised Su worked for International Business Machines, where she focused on designing products according to the needs of big customers. When she took over as CEO of AMD in 2014, the company had a market value of less than $3 billion.
Su did deals with Chinese partners that helped stabilize the company’s finances and moved the chip designer deeper into the processors that powered PCs and data centers, capitalizing on Intel’s weaknesses. Today AMD sells an estimated 41% of data-center CPUs, up from essentially zero about five years ago.
It was that success that allowed AMD’s board of directors to trust Su when she proposed, in that fateful board meeting in 2022, to pivot the company to go head-to-head with Nvidia, said Abhi Talwalkar, one of the board members who attended the meeting.
“She had vision around our ability to be AI from end point to data center and everything in between,” Talwalkar, a veteran chip executive, said. “She had the conviction of the need, and the need to move fast.”
With the directors backing her, Su embarked on an ambitious plan to ramp up design work on several generations of AMD’s Instinct line of data-center chips, culminating with its most powerful AI processor yet, the MI450.
Su was able to dispense with Intel as a competitor because Intel had “self-inflicted wounds,” said Daniel Newman, CEO of research and advisory firm Futurum Group. That’s in sharp contrast to Nvidia, a company that’s expected to generate more than $200 billion in revenue next year and that has gotten thousands of AI developers hooked on the proprietary software that Nvidia chips rely on, Newman said.
But Nvidia can’t make enough chips to keep up with demand, and the OpenAI deal proved that large AI companies are eager to diversify their supplier base, said Newman.
Darrick Horton, founder and CEO of TensorWave, a Las Vegas-based cloud services company that is backed by AMD and uses only AMD’s hardware for its data-center clusters, said AMD has another factor working in its favor: Its chips and related products are less expensive than Nvidia’s, with prices for comparable hardware sometimes as much as 20% lower.
Global dealmaker
As AMD has become a bigger player in AI, Su has begun to embrace the spotlight. Earlier this year at an industry summit in Paris, Su was treated like a celebrity, with many young women asking her to take selfies and sign autographs.
Su is also spending more time in Washington and on the global stage, recognizing that decisions by the Trump administration and governments globally could make or break her plans to attract more customers. The Trump administration approved chip exports directly to state-backed companies in Saudi Arabia and the United Arab Emirates this week, a boon for AMD and Nvidia.
While Huang has gotten most of the attention for lobbying the administration to allow chip exports to China and other countries, Su has also been relentless in pushing for AMD-friendly policies in discussions with senior administration officials such as Commerce Secretary Howard Lutnick and White House AI Czar David Sacks, people familiar with the matter said.
Like Huang, Su has argued that restricting exports of American chips would boost Chinese competitors like Huawei Technologies, an argument that administration officials have found compelling. Su has also said that strict export controls could disproportionately hurt AMD while it tries to get a foothold in overseas markets and chip away at Nvidia’s lead. Those in favor of such controls argue that allowing the sale of the most advanced chips would boost China’s AI data-center capabilities and threaten national security.
In July, Su attended an AI summit where Trump rolled out his AI strategy. When she addressed the audience, Su pulled one of the company’s chips out of her pocket and said it had 185 billion transistors and took nine months to make. Such chips can cost tens of thousands of dollars. Trump later mentioned her in his speech as one of the notable executives in attendance.
She was still in the shadow of Huang, however, who Trump asked to stand while the crowd applauded. “What a job you’ve done,” the president said to the Nvidia CEO. Huang said at the event that Trump was the biggest advantage the U.S. has in the AI race.
Su returned for a September dinner in which she and other CEOs took turns praising Trump’s policies.
On Thursday afternoon, the chip industry’s trade group, the Semiconductor Industry Association, said that it had elected Su as its new chair.
“I do seek to be part of the conversation where decisions are made,” Su said. “In this administration, I think what’s different is there’s such a sort of open door to be part of the conversation.”
Write to Robbie Whelan at robbie.whelan@wsj.com and Amrith Ramkumar at amrith.ramkumar@wsj.com