Samsung locked in AMD as its latest high-bandwidth memory customer this week, securing a partnership that comes as the chipmaker faces unprecedented demand threatening to exhaust its inventory through 2027.
The companies signed a memorandum of understanding on March 18 at Samsung's Pyeongtaek semiconductor campus, expanding their collaboration to include next-generation AI memory solutions. Under the agreement, Samsung becomes the primary supplier of seventh-generation High Bandwidth Memory (HBM4) for AMD's upcoming Instinct MI455X GPUs and will provide optimized DDR5 modules for sixth-generation EPYC server processors codenamed "Venice."
This marks the first confirmation of HBM4 integration in AMD's next-generation AI chips, positioning Samsung as an alternative supplier beyond its existing relationship with Nvidia. The Korean manufacturer began mass production shipments of HBM4 last month using its advanced 1c DRAM process and 4nm base die technology.
Samsung's HBM4 delivers processing speeds up to 13 gigabits-per-second per pin with maximum bandwidth reaching 3.3 terabytes-per-second, exceeding current industry standards. These specifications target high-end AI training and inference workloads in hyperscale data centers where memory bandwidth and power efficiency increasingly determine system-level performance.
The timing coincides with reports that Samsung may run out of its entire 2027 memory supply soon due to historic low inventory levels across the industry. KB Securities analyst Jeff Kim believes this supply constraint could persist for years as demand from AI data centers continues surging.
AMD CEO Lisa Su met with Samsung executives including Vice Chairman Jun Young-hyun during her visit to South Korea, securing priority access to HBM4 production capacity. The two companies will also discuss opportunities for foundry partnerships where Samsung would manufacture next-generation AMD products using its advanced 3nm and 4nm process technologies.
"Powering the next generation of AI infrastructure requires deep collaboration across the industry,"
said Su in a statement released alongside the announcement. The partnership extends beyond memory supply to include potential manufacturing cooperation that could diversify AMD's production capabilities beyond current foundry partners.
Samsung is reportedly considering a shift toward multi-year contracts for memory chips as demand for AI components shows no signs of slowing through 2026 and beyond. This move follows similar long-term agreements by competitors responding to sustained market pressure.
The expanded collaboration arrives weeks after AMD secured one of its largest AI hardware deals with Meta valued at over $100 billion, including up to $60 billion in AI chip sales over five years. That agreement followed a similar partnership with OpenAI, positioning AMD as a growing challenger in the AI accelerator market dominated by Nvidia.
