Samsung Uses TSMC 3nm for Nvidia NVHBM Base Die; All Three Memory Makers Follow

Samsung is sourcing base dies for Nvidia's custom NVHBM memory chip from TSMC's 3nm process, with Samsung's own 4nm production lines at Hwaseong Line 17 and Pyeongtaek P3 lacking capacity to absorb the required volumes. The arrangement, reported by Seoul Economic Daily citing unnamed industry sources, means all three major HBM suppliers to Nvidia, Samsung, SK Hynix, and Micron, become buyers of TSMC 3nm logic wafers for NVHBM stacks. Nvidia's NVHBM specification, announced by Nvidia, delivers 30% higher bandwidth and 15% lower power than commodity HBM4e, and will be available to NVLink Fusion partners via a custom base die and PHY. The base die houses Nvidia's memory controller, which Nvidia moved out of its GPU compute die. Samsung Foundry is simultaneously validating an in-house production plan as a hedge against TSMC capacity constraints; Samsung is also considering repurposing its NRD-K Line 2 facility from R&D to foundry production to target 2nm base dies for HBM5. TSMC plans to add two new 3nm lines at its Fab 18 in Tainan and is equipping its second Arizona and second Kumamoto fabs with 3nm capability. An 8-layer NVHBM variant targeting 18 Gbps is described as a near-term drop-in-compatible product, while standard HBM4E is tied to Nvidia's Vera Rubin Ultra in H2 2027. TrendForce projects HBM will account for 30% of the top three DRAM makers' wafer input by end of 2027, up from 18% at end of 2025.