An excerpt from a market report indicates that AMD is purportedly raising prices of its EPYC data center processors for some customers by 10% to 30%, a new reality in the face of chip shortages that have resulted in reduced visibility for future shipments. We reached out to AMD for comment on the matter, but the company declined due to its policy of not commenting on customer pricing. The report also cites a potential delay for Intel's next-gen Sapphire Rapids chips into Q3 2022, meaning Intel could miss the re-revised timeline it announced last year.
The excerpt comes from a report by Mizuho Securities' Managing Director Jordan Klein that cites a leading server OEM as the source.
The report says that AMD has increased pricing by 10 to 30%, but the impact varies by customer, with large cloud customers given less of an upcharge. The report claims that without any guidance of when more CPUs could arrive, AMD's customers are simply accepting the higher prices.
AMD's Milan holds the peak performance, and perhaps more importantly, performance-per-watt crown in the data center. Milan's higher core counts also offer superior performance density, thus allowing data centers to cram in more performance per server and further lower their operating costs, all of which is obviously attractive enough to justify the price increases.
AMD's price hikes aren't surprising given ongoing shortages coupled with its reliance upon outside firms for both lithographic (wafer) and packaging (OSAT) capacity, with the latter being a particular pain point in the industry. Both of these factors could impact EPYC supply more acutely than AMD's other types of chips due to EPYCs multi-chip design that scales up to nine die in a single package. Manufacturing costs have risen at nearly every link in the supply chain, and AMD's price increases probably reflect the company passing that increased pricing on to its customers rather than being a means to increase margin.
The report also says that Intel has ramped Ice Lake production and projects a possible 50% year-over-year increase in supply in 2022. In addition, in an attempt to retain market share, Intel isn't raising prices for its Ice Lake processors. Both of those factors could help blunt AMD's market share gains.
However, the report claims that Intel's full-scale Sapphire Rapids volume launch will be delayed until 'maybe' Q3 2022 (Intel had previously projected the formal launch in Q2 2022). Furthermore, the firm doesn't anticipate Sapphire Rapids will help Intel much, predicting that Intel will see further market share degradation due to 'materially higher' BOM pricing for Sapphire Rapids.
Intel's Sapphire Rapids comes with a multi-chip architecture fabbed on the 'Intel 7' process and tied together via a copious number of EMIB connections that provide superior throughput and latency compared to other packaging techniques, but it also adds cost. While Intel does most of its packaging in-house, packaging is still a production bottleneck due to shortages of key materials. Sapphire Rapids will be the company's first widespread implementation of its EMIB tech in a high-volume lineup like Xeon, which could contribute to the predicted higher pricing.
Intel CEO Pat Gelsinger recently said that Sapphire Rapids should put the company close to parity with AMD's data center chips (Sapphire Rapids will win some, lose some), but that he doesn't expect to gain a clear lead over AMD until Intel fields the next generation of data center chips. Wu sees that a bit differently, projecting that Milan and Genoa will retain the performance advantage over Xeon, resulting in AMD's continuing its explosive growth in the data center.
AMD is more supply-constrained than Intel, so its supply of silicon is obviously holding the company back from gaining market share more quickly. Still, Intel faces a tough challenge as TSMC continues to ramp 7nm capacity and AMD's 5nm Genoa comes to market later this year. That means we'll see the pitched battle for server sockets intensify throughout the year.