Samsung Electronics will reportedly make a core part of Nvidia's custom memory chips on TSMC's 3-nanometer process, according to the Seoul Economic Daily, which cited industry sources.
The part in question is the base die, the logic layer of high-bandwidth memory (HBM). HBM is a stack of memory chips packaged next to an AI processor, and the base die sits underneath the stack. It is the piece that increasingly gets tailored to a specific customer. The arrangement would be the first time Samsung and TSMC have worked together on HBM. Nvidia reportedly wants all of its HBM suppliers to use the same TSMC process.
Samsung runs its own chip foundry, a business that competes with TSMC for outside customers. Sending a base die to a rival's fab, if the report holds, means Samsung is treating the part as something to buy in rather than a place to show off its own manufacturing. The source does not say whether Nvidia required this or Samsung chose it, and the outcome looks the same either way.
The broader shift is easier to read. Memory makers have long competed on DRAM, the memory chips themselves. A custom base die moves part of the product into logic-chip territory, where TSMC is the dominant name. If every supplier builds that layer on one process, the logic stops being a differentiator. What is left is how well each company stacks and packages the memory on top.
That raises a question for Nvidia's suppliers: does a shared base-die process make them easier to swap, or does it simply move the competition to the stacking step?