Samsung to Use TSMC's 3nm for NVIDIA's NVHBM Base Die, Skipping Its Own 4nm, Report Says - XenoSpectrum

Samsung's decision to source TSMC's 3nm for NVIDIA's NVHBM base die—bypassing its own 4nm—is not a procurement choice; it is a credibility collapse for Samsung Foundry at the leading edge. Technically, HBM4-generation base dies have evolved into complex SoCs integrating PHY, ECC engines, and scheduling logic. The 3nm node delivers roughly 15-20% better power efficiency than 4nm. By outsourcing rather than manufacturing in-house, Samsung concedes its 4nm PPA metrics fail NVIDIA's spec threshold. The deeper consequence: once NVHBM's reference design locks to TSMC 3nm, the HBM4/4E roadmap naturally converges toward 2nm, permanently marginalizing Samsung's 4nm in the HBM ecosystem. On supply-chain security, routing a core product's front-end fabrication through a direct competitor creates a vulnerability far exceeding conventional foundry outsourcing. NVIDIA's HBM roadmap now depends on a single geographic node's capacity, handing export-control policymakers a new chokepoint variable. Competitively, TSMC locks in long-term HBM logic-die orders. If Samsung cannot present credible 3nm GAA yield data within 12 months, expect another one or two of its top-five customers to defect. Within 24 months, the HBM base die becomes the second "TSMC-dependent" compute-critical component after the GPU itself. Samsung's "memory-plus-foundry" vertical-integration narrative is being structurally dismantled.