When AMD launched Epyc, it made it clear that the server CPU would face a slow and cautious ramp into market rather than a full-on sprint. Prior to the launch of its 7nm Epyc CPUs, AMD’s total server market share had edged upwards to ~3.4 percent. Now that 7nm hardware is in-market, there are reports that AMD is gaining market share more rapidly.
DigiTimes reports that AMD is pulling roughly 5 percent market share today and could blow past 10 percent by the end of 2020. This would represent a significant short-term share gain for the company. Prior to the launch of Epyc, AMD’s server market share had fallen to ~0 percent, with Intel estimated to have more than 99 percent of the market. It took AMD roughly two years to grow to 3.4 percent, but that made sense, given the slow pace of servers and the conservative nature of the space. Moving from 3.4 percent to 10 percent in 12-18 months would represent a much faster gain.
The chart above shows gains through Q2 of 2019. All eyes will be on AMD’s Q3 results this year. AMD had Navi and Ryzen chips in-market for almost the entire quarter and it launched its Epyc 7nm chips early in August, which means the company’s 7nm hardware has been available for the majority of the time. AMD wasn’t sure how strong console demand would be in Q3, which may limit our ability to see into server sales — the company combines Enterprise, Embedded, and Semi-Custom reporting into the same business segment.
Outside of unit sales, the past few weeks have been strong for AMD. The company has unveiled a new, HPC-focused Epyc 7H12, announced dozens of new world records in tests conducted by third parties, and been making a serious play for attention across the HPC and server markets. Even Intel has acknowledged in its quarterly calls that AMD is now competitive across a range of markets that Intel had dominated for most of the past decade. In my own opinion, AMD is doing well enough to call this the second “Golden Age” of the company.
AMD hitting 10 percent of the server market by the end of 2020 seems like a fairly reasonable goal given how well the company is executing on 7nm. The one fly in the ointment may be the overall health of the server market itself. The space boomed sharply in the back half of 2018 but sales have been slower in the first half of the year. AMD might take 10 percent of the market but wind up making significantly less money than it might have during a period of higher sales. Then again, when you’ve previously had 0 percent of a space, any sales are better than the alternative.
It’s not fair or accurate to say that companies “haven’t been” paying attention to Epyc. They have. In some cases, customers that are now publicly announcing AMD availability are companies that have likely been testing Epyc for years. Server and HPC customers install hardware for significant periods of time, and it’s not unheard of for compute clusters to be upgraded as new chips become available. Even when installations aren’t updated, it’s beneficial to companies to know that they’ll be able to count on the same CPU vendor to deliver improved parts year-on-year. This reduces the amount of validation work that has to be performed on an ongoing basis.
If you look at where AMD has historically had trouble, it’s been in the transition from one node to the next or when rolling out a major architecture update. The Zen 3 CPU core is a significant update on Zen 2 and moving to 7nm at TSMC was a significant shift from 12/14nm at GlobalFoundries. AMD has had some supply issues that required it to push back the introduction of a 16-core chip, but the company appears to be selling every chip it can make. We’ll find out more about what sales have looked like in about a month when AMD announces quarterly results.