Nvidia Corp. has asked all three major memory manufacturers — Samsung Electronics, SK Hynix, and Micron Technology — to fabricate the base dies for its custom NVHBM memory stacks using Taiwan Semiconductor Manufacturing Co.'s 3-nanometer process1,2. Samsung Electronics has agreed, marking the first time the company has used Taiwan Semiconductor Manufacturing Co. to produce HBM base dies3.
The arrangement, reported on October 18, 2026, consolidates a critical layer of AI memory manufacturing under a single foundry process.
Why one process for three suppliers
Nvidia Corp. announced NVHBM in August 2026 as a way to standardize custom HBM production rather than letting each memory supplier take its own approach. The design moves Nvidia Corp.'s memory controller off the GPU and onto the base die, the logic layer at the bottom of the HBM stack. That shift frees GPU die area for additional compute circuits.
To fit the controller onto the base die, the physical-layer circuitry that exchanges electrical signals with external chips was shrunk by 60% relative to standard HBM. Nvidia Corp. says the result is up to 30% higher memory bandwidth and 15% lower power consumption compared with standard HBM4E4.
Routing all three suppliers through Taiwan Semiconductor Manufacturing Co.'s 3nm node is intended to yield consistent performance from the finished product and reduce quality-verification steps. The core die, meanwhile, is specified to use the sixth-generation DRAM process, known as 1c.
Samsung crosses a foundry line
Samsung Electronics has historically taken a turnkey approach to HBM, handling both core and base dies internally. Agreeing to use Taiwan Semiconductor Manufacturing Co.'s lines for the NVHBM base die breaks that pattern. Industry sources cited by Seoul Economic Daily said the decision reflects a memory bottleneck: HBM data-transfer speeds have not kept pace with GPU performance, and the plan is to lift HBM throughput using Taiwan Semiconductor Manufacturing Co.'s advanced logic process while also tapping its manufacturing capacity to expand supply.
Nvidia Corp. is co-designing the base die with Samsung Electronics and Taiwan Semiconductor Manufacturing Co., according to those sources. SK Hynix and Micron Technology have also been asked to manufacture NVHBM base dies on the same Taiwan Semiconductor Manufacturing Co. 3nm node.
NVHBM has been linked to Nvidia Corp.'s Feynman-generation GPUs. ANALYSIS By mandating a single foundry process across all three memory suppliers, Nvidia Corp. is effectively inserting Taiwan Semiconductor Manufacturing Co. as a chokepoint in the AI memory supply chain — a structural change that trades supplier manufacturing independence for performance uniformity.