Nvidia Taps TSMC 3nm Base Die for Custom NVHBM, Samsung Agrees

The notable signal here is architectural rather than commercial: Nvidia is reportedly steering the logic layer of its custom HBM to a foundry outside the memory makers' own process roadmaps, which would deepen TSMC's position in the AI memory stack even as Samsung, SK Hynix and Micron supply the DRAM. Samsung's willingness to accept those terms suggests memory vendors are conceding some design control to secure Nvidia's custom HBM business. Whether the other two suppliers follow, and how this reshapes foundry-versus-memory-maker leverage, is the open question worth watching.