With quarterly earnings approaching April 28, Advanced Micro Devices Inc. will soon offer a first look at how its performance stacks up against an ambitious new plan designed to erode Intel Corp.'s domination of the data-center market.
The fast-growing, high-margin market for data-center and cloud-platform servers is a lucrative target and one owned almost entirely by Intel, which holds between 90% and 95% share in the market for data-center processors, according to most analyst estimates. Intel's Data Center Group reported $7.2 billion in revenue for the fourth quarter of 2019, exceeding in one segment AMD's entire full-year 2019 revenue of $6.7 billion.
About $2.02 billion of AMD's 2019 revenue came from its Enterprise, Embedded and Semicustom group, which includes servers, embedded processors, semicustom system-on-chip designs, development services, game-console technology and the licensing of AMD designs to be used in other processors. Growth for the enterprise segment has lagged that of AMD's larger Computing and Graphics division in recent years. Enterprise segment revenue fell 16.2% year over year in 2019 compared to growth of 12.4% for Computing and Graphics.
The Computing and Graphics segment also reported double the operating income of the enterprise segment in 2019.
AMD's high-end ambitions are somewhat split between its two product segments, however. The enterprise product group is responsible for developing AMD's EPYC multicore central processing units and marketing them for enterprise data centers, cloud platforms and high-performance computing applications. But if graphical processing units play a significant role in expanding AMD's presence in the data center, their contribution will be counted as part of the Computing and Graphics group, which is focused on gaming and desktop markets but also includes server and desktop GPU development as part of its portfolio.
During a product-roadmap-focused Financial Analyst Day in March 2020, AMD executives laid out a three-year plan to boost revenues 20% per year and push gross margins above 50% by expanding the company's sales to data centers and cloud platforms to 30% of total revenue, compared to the 15% or so it accounts for now.
As of March 30, AMD processors made up about 8% of the servers available for users to configure on the four largest cloud platforms, according to Paul Teich, principal analyst at market-research firm Liftr Insights.
Liftr Insights' Cloud Component Tracker Service scans the user configuration applications of cloud platform providers, including Amazon.com Inc.'s Amazon Web Services, Alphabet Inc.'s Google Cloud Platform, Microsoft Corp.'s Azure and dominant Chinese cloud provider Alibaba Group Holding Ltd. to gather data on the specific hardware configurations among which cloud-platform customers can choose. The service counts the server configurations available in each data center for each cloud platform on the list to identify which are available in which specific areas.
"We don't see how many servers are behind each configuration, but we know where every configuration they offer is available on the planet," said Teich, who spent 20 years in technical marketing and strategy roles at AMD before becoming an analyst 8 years ago. "It's like looking at products on a retail shelf. We know what's available to be purchased in what styles and what configuration, how many cores, how much memory, which accelerators are being offered."
Of the AMD processors available, very few were the latest, highest-performance versions of processors from either Intel or AMD. Of the AMD chips, 7% were fifth-generation EPYC Ryzen 1000 CPUs code-named "Naples," which were shipped in 2017, and 1% were the EPYC Ryzen 2000 series code-named "Rome," first shipped in 2019.
Of the Intel processors, 13% were Xeon version 3 "Haswell' processors first shipped in 2013.
However, Liftr Insights' market measurement method may underestimate AMD's progress by not counting enterprise data centers, noncloud installations and the possibility that cloud providers primarily keep the latest, greatest and fastest processors for uses not visible to end-user configuration tools.
Using more traditional survey methods, The Linley Group estimates AMD's server market share at closer to 10%, according to Linley Gwennap, president and principal analyst of the semiconductor-industry consulting firm.
"We're getting a lot more robust competition in the server market, but AMD has been able to win business in mainstream servers just by offering lower prices," Gwennap said. "Intel has been top dog for so long that they've taken a lot of opportunities to increase prices over time to make as much money as possible. But some things that AMD can do, like offering 64-core server chips, are way beyond what Intel can offer right now."
During AMD's analyst briefing, company President and CEO Lisa Su said AMD is ahead of Intel in many performance areas and is coming up to parity with NVIDIA Corp., which will make it possible to increase prices and margins as the market becomes aware of AMD's potential as a high-performance computing provider.
During the past year, AMD announced that its processors had been chosen to power two $600 million Department of Energy supercomputers: Oak Ridge National Laboratory's Frontier and Lawrence Livermore National Laboratory's El Capitan. Each is defined as being the world's fastest supercomputer; El Capitan is expected to be faster than the world's 200 fastest supercomputers combined.
Linley Group's Gwennap said AMD's growth plan sounds reasonable, especially given the company's recent alliances with Google and other cloud providers, but Intel will fight back on pricing and will likely make up lost ground on technical and performance grounds as well.
"If you add their wins in high-performance computing, like El Capitan and Frontier, which extends their credibility, it seems reasonable they [AMD] could see their revenue from the data center grow," said IDC semiconductor analyst Shane Rau. He declined to guess whether that growth could match the pace AMD executives suggested.