Advanced Micro Devices made history this year when it surpassed Intel by market cap for the first time ever. Intel has long held the lead in the market for computer processors, but AMD's ascent results from the company branching out into entirely new sectors.
In one of the biggest semiconductor acquisitions in history, AMD purchased adaptive chip company Xilinx in February for $49 billion. Now, AMD chips are in two Tesla models, NASA’s Mars Perseverance land rover, 5G cell towers and the world’s fastest supercomputer.
“AMD is beating Intel on all the metrics that matter, and until and unless Intel can fix its manufacturing, find some new way to manufacture things, they will continue to do that,” said Jay Goldberg, semiconductor consultant at D2D Advisory.
But a decade ago, analysts had a very different outlook for AMD.
“It was almost a joke, right? Because for decades they had these incredible performance problems,” Goldberg said. “And that's changed.”
CNBC sat down with AMD CEO Lisa Su to hear about her company’s remarkable comeback, and huge bets on new types of chips in the face of a PC slump, fresh restrictions on exports to China and shifting industry trends.
AMD was founded in 1969 by eight men, chief among them Jerry Sanders. The famously colorful marketing executive had recently left Fairchild Semiconductor, which shares credit for the invention of the integrated circuit.
“He was one of the best salesmen that Silicon Valley had ever seen," said Stacy Rasgon, semiconductor analyst at Bernstein Research. "Stories of lavish parties that they would throw. And there's one story about him and his wife coming down the stairs of the turret at the party in matching fur coats.”
He also coined an infamous phrase about chip fabrication plants, or fabs.
“Jerry Sanders was very famous for saying, ‘Real men have fabs,’ which obviously is a comment that is problematic on a number of levels and has largely been disproven by history,” Goldberg said.
As technology advances, making chips has become prohibitively expensive. It now takes billions of dollars and several years to build a fab. AMD now designs and tests chips and has no fabs.
“When you think about what do you need to do to be world class and design, it's a certain set of skills," Su said. "And then what do you need to do to be world class In manufacturing? It's a different set of skills and the business model is different, the capital model is different.”
Back in the '70s, AMD was pumping out computer chips. By the '80s, it was a second-source supplier for Intel. After AMD and Intel parted ways, AMD reverse engineered Intel’s chips to make its own products that were compatible with Intel’s groundbreaking x86 software. Intel sued AMD, but a settlement in 1995 gave AMD the right to continue designing x86 chips, making personal computer pricing more competitive for end consumers.
“That's when their execution really started to take off because they no longer had to worry about the foundry side of things,” Goldberg said.
GlobalFoundries went public in 2021 and remains a top maker of the less advanced chips found in simpler components like a car’s anti-lock brakes or heads-up display. But it stopped making leading-edge chips in 2018. For those, AMD turned to Taiwan Semiconductor Manufacturing Co., which now makes all of AMD’s most advanced chips.
AMD only has major competition from two other companies when it comes to designing the most advanced microprocessors: Nvidia in graphics processing units, GPUs, and Intel in central processing units, CPUs.
While AMD controls far less GPU and CPU market share than Nvidia and Intel, respectively, it’s made remarkable strides since moving away from manufacturing and reducing capital expenditure.
Meanwhile, Intel doubled down on manufacturing last year, committing $20 billion for new fabs in Arizona and up to $100 billion in Ohio, for what it says will be the world’s largest chip-making complex. But the projects are still years away from coming online.
“Intel is just not moving forward fast enough," Goldberg said. "They've said they expect to continue to lose share in next year and I think we'll see that on the client side. And that's helped out AMD tremendously on the data center side.”
AMD’s Zen line of CPUs, first released in 2017, is often seen as the key to the company's recent success. Su told CNBC it's her favorite product. It’s also what analysts say saved AMD from near bankruptcy.
“They were like literally, like probably six months away from the edge and somehow they pulled out of it," Rasgon said. "They have this Hail Mary on this new product design that they're still selling like later generations of today, they call it Zen is their name for it. And it worked. It had a massively improved performance and enabled them to stem the share losses and ultimately turn them around.”
Among the Zen products, AMD’s EPYC family of CPUs made monumental leaps on the data center side. Its latest, Genoa, was released earlier this month. AMD’s data center customers include Amazon Web Services, Google Cloud, Oracle, IBM and Microsoft Azure.
“If you looked at our business five years ago, we were probably more than 80% - 90% in the consumer markets and very PC-centric and gaming-centric,” Su said. “As I thought about what we wanted for the strategy of the company, we believed that for high-performance computing, really the data center was the most strategic piece of the business.”
AMD's revenue more than tripled between 2017 and 2021, growing from $5.3 billion to over $16 billion. Intel's annual revenue over that stretched, meanwhile, increased about 25% from close to $63 billion in 2017 to $79 billion last year.
AMD’s success at catching up to Intel’s technological advances is something many attribute to Su, who took over as CEO in 2014. AMD has more than tripled its employee count since then. Su was Fortune’s #2 Business Person of the Year in 2020 and the recipient of three of the semiconductor industry’s top honors. She also serves on President Joe Biden’s Council of Advisors on Science on Technology, which pushed hard for the recent passage of the CHIPS Act. It sets aside $52 billion for U.S. companies to manufacture chips domestically instead of overseas.
“It's a recognition of just how important semiconductors are to both economic prosperity as well as national security in the United States,” Su said.
With all the world’s most advanced semiconductors currently made in Asia, the chip shortage highlighted the problems of overseas dependency, especially amid continued tension between China and Taiwan. Now, TSMC is building a $12 billion 5-nanometer chip fab outside Phoenix.
“We're pleased with the expansion in Arizona," Su said. "We think that's a great thing and we'd like to see it expand even more.”
Earlier this month, the Biden administration enacted big new bans on semiconductor exports to China. AMD has about 3,000 employees in China and 25% of its sales were to China last year. But Su says the revenue impact has been “very small.”
“When we look at the most recent regulations, they're not significantly impacting our business," Su said. "It does affect some of our highest-end chips that are used in sort of AI applications. And we were not selling those into China.”
What is hurting AMD’s revenue, at least for now, is the PC slump. In its third-quarter earnings report earlier this month, AMD missed expectations, shortly after Intel warned of a soft fourth quarter. PC shipments were down nearly 20% in the third quarter, the steepest decline in more than 20 years.
“It's down a bit more than perhaps we expected,” Su said. “There is a cycle of correction which happens from time to time, but we're very focused on the long-term road map.”
It’s not just PC sales that are slowing. The very core of computer chip technology advancement is changing. An industry rule called Moore’s Law has long dictated that the number of transistors on a chip should double about every two years.
“The process that we call Moore's Law still has at least another decade to go, but there's definitely, it's slowing down,” Goldberg said. “Everybody sort of used CPUs for everything, general purpose compute, but that's all slowed down. And so now it suddenly makes sense to do more customized solutions.”
That’s why AMD acquired Xilinx, known for its adaptive chips called Field-Programmable Gate Arrays, or FPGAs. Earlier this year, AMD also bought cloud startup Pensando for $1.9 billion.
“We can quibble about some of the prices they paid for some of these things and what the returns will look like,” said Goldberg, adding that the acquisitions were ultimately a good decision. “They're building a custom compute business to help their customers design their own chips. I think that's a very, it's a smart strategy.”
More and more big companies are designing their own custom chips. Amazon has its own Graviton processors for AWS. Google designs its own AI chips for the Pixel phone and a specific video chip for YouTube. Even John Deere is coming out with its own chips for autonomous tractors.
“If you really look underneath what's happening in the chip industry over the last five years, everybody needs more chips and you see them everywhere, right?" Su said. "Particularly the growth of the cloud has been such a key trend over the last five years. And what that means is when you have very high volume growth in chips, you do want to do more customization.”
Even basic chip architecture is at a transition point. AMD and Intel chips are based on the five-decade-old x86 architecture. Now ARM architecture chips are growing in popularity, with companies like Nvidia and Ampere making major promises about developing Arm CPUs, and Apple switching from Intel to self-designed ARM processors.
“My view is it's really not a debate between x86 and Arm," Su said. "You're going to see basically, these two are the most important architectures out there in the market. And what we've seen is it's really about what you do with the compute.”
For now, analysts say AMD is in a strong position as it diversifies alongside its core business of x86 computing chips.
“AMD should fare much better in 2023 as we come out of the cycle, as their performance gains versus Intel start to become apparent, and as they start to build out on some of these new businesses,” Goldberg said.
Intel did not immediately respond to a request for comment.
Correction: "And we were not selling those into China," said Lisa Su, AMD's CEO. Her quote has been updated to reflect a typo that appeared in an earlier version of this article. An industry rule called Moore’s Law has long dictated that the number of transistors on a chip should double about every two years. An earlier version misstated the rule.