AMD now offers competitive products at every price point in both the server and consumer markets. However, Intel has one major advantage: It is one of the few chipmakers who fab their own chips. This simplifies the various intricacies of the supply chain, especially through rough patches. According to Mizuho Securities, Intel’s Xeon server processors are presently available with a lead time of just 1-2 weeks. In comparison, the same for AMD’s Epyc Rome and Milan offerings range between 20 to 26 weeks. This info was obtained in July 2021 during a call with a global server OEM.
NVIDIA isn’t doing any better with the lead time for the A100 accelerators (also manufactured on TSMC’s N7 node) being around 26 weeks. This largely boils down to just how packed TSMC’s advanced process nodes are. You’ve got AMD, NVIDIA, Qualcomm, MediaTek, and soon even Intel relying on the same process technology with growing demand from all segments (amid rising costs). The result is…well…this.
Both NVIDIA and AMD have repeatedly affirmed to customers and investors that they’re making investments to reduce pressure on the supply chain. But creating new foundry capacity takes time, resources, and more time, even more so in today’s semi landscape. The semiconductor shortages are expected to stretch well into 2022, so don’t expect things to change anytime soon.
Add to this Intel’s ability to absorb slimmer profit margins (even negative) due to a more diverse mix of products, and it makes things even harder for AMD. The latter has still been able to penetrate the server market with its share just crossing 10% for the first time in forever. However, Intel’s closely knit ecosystem partnerships and ability to offer large-scale discounts (known as the meet comp program) to customers means that bringing new clients on board is a slow and drawn-out process.