AMD Extends Memory Planning Horizon to Five Years Amid HBM Scarcity
AMD CEO Lisa Su is shifting the company's supply chain strategy to a three-to-five-year outlook as high-bandwidth memory demand continues to outpace production.
AMD is adjusting its supply chain strategy to account for the ongoing scarcity of high-bandwidth memory, a critical component for AI hardware. CEO Lisa Su recently identified Samsung and SK Hynix as key partners for the company's long-term needs, noting that AMD must now plan its memory requirements three to five years into the future, up from its previous one-to-two-year window.
The shift follows Su's early October 2026 trip to Taiwan and South Korea, where she focused on securing long-term supply agreements. AMD intends to significantly increase its supply in 2027 to keep pace with the rapid growth of the AI market, though the company expects HBM supply constraints to persist through 2028.
A central pillar of this strategy is a memorandum of understanding signed on March 18, 2026, which designates Samsung as the priority supplier for sixth-generation HBM4. These components are slated for use in AMD’s Instinct MI455X AI accelerators and EPYC Venice server processors. Samsung, which began shipping HBM4 in February 2026, has served as a long-term partner for nearly two decades.
While SK Hynix remains the largest player in the HBM segment with a market share of approximately 57-58%, its primary partnership with Nvidia has limited AMD's access to its capacity. This dynamic has driven AMD to deepen its reliance on Samsung, though this concentration of supply carries risks; any production or yield issues at Samsung could directly impact AMD's accelerator roadmap.
Looking ahead, market observers are monitoring whether the existing memorandum of understanding with Samsung will evolve into firm, multi-year commitments that align with Su’s new planning timeline.
