AMD seeks 'broader partnership' with Samsung as it looks to secure memory supply

Samsung's move is not about supply security—it's about weaponizing HBM scarcity to extract foundry commitments from AMD. The playbook mirrors Intel's 2000s strategy of bundling chipsets with CPUs to lock in OEMs. Technically, MI300X already depends on SK Hynix HBM3E, while MI350/MI400 will demand roughly double the HBM4 bandwidth. Samsung packaging memory with 2nm GAA logic is a direct probe into TSMC's territory. For AMD, this isn't mere supplier diversification—it's building a second channel in the 30-40% HBM slice of AI accelerator BOM cost, diluting any single vendor's pricing leverage. Strategically, the NVIDIA–SK Hynix axis is near-exclusive. AMD anchoring to Samsung creates a parallel memory-plus-logic vertical chain. Samsung's calculus is clear: use HBM capacity as leverage to pull AMD's 2nm/3nm logic workloads off TSMC. The risk is lock-in. If Samsung's HBM4 yield remains below the industry-estimated 70%, AMD's shipping cadence becomes hostage to a single fab's ramp curve. Over the next 18 months, expect memory-fab bundling to become the industry default. HBM is no longer a commodity—it's the hardest strategic asset in the AI compute stack. Whoever controls HBM allocation controls the pricing power of next-generation AI silicon.