Nvidia’s growing influence in the AI hardware market is now reshaping how high-bandwidth memory is designed and produced. The company’s next-generation custom memory technology, reportedly called NVHBM, is expected to change the traditional relationship between memory suppliers and chip customers by giving Nvidia more control over the structure of the memory used in its advanced AI accelerators.
According to industry reports from South Korea, Samsung Electronics has decided to use TSMC’s 3nm process for the base die of Nvidia’s upcoming custom HBM. This marks a major shift for Samsung, which has historically handled the full production of its HBM stacks internally. If the plan moves forward, it would be the first known collaboration between Samsung and TSMC at this stage of HBM manufacturing.
High-bandwidth memory, or HBM, is a critical component in modern AI processors. It allows massive amounts of data to move quickly between memory and computing units, making it essential for training and running large AI models. As demand for AI chips continues to rise, companies like Nvidia are looking for ways to improve performance, power efficiency, and supply reliability.
The base die plays an important role in an HBM stack. It acts as the foundation that connects the memory layers to the processor, helping manage communication and performance. By using TSMC’s advanced 3nm manufacturing technology for this part of the design, Nvidia may be aiming to improve efficiency and better tailor memory performance for its future AI platforms.
For Samsung, the move is especially significant. The company is one of the world’s leading memory manufacturers, but Nvidia’s custom HBM approach suggests that major customers may increasingly influence not only what memory is produced, but also how and where key parts of it are made. This could signal a broader change in the semiconductor industry, where custom designs and closer customer involvement become more common.
The reported decision also highlights TSMC’s strong position in advanced chip manufacturing. Its 3nm process is widely seen as a leading technology for high-performance computing, and its involvement in Nvidia’s custom HBM could strengthen its role in the AI chip supply chain.
Nvidia’s NVHBM strategy reflects the intense competition surrounding AI hardware. As artificial intelligence workloads become more demanding, standard memory solutions may no longer be enough. Custom HBM designs could give Nvidia an advantage by allowing tighter integration between GPUs, memory, and packaging technologies.
While full details of NVHBM remain limited, the reported partnership between Samsung and TSMC shows how quickly the AI semiconductor landscape is evolving. Memory makers, foundries, and AI chip designers are becoming more interconnected as they race to deliver faster, more efficient hardware.
If confirmed, Samsung’s use of TSMC’s 3nm process for Nvidia’s custom HBM base die would represent a major milestone in advanced memory production. It would also underline Nvidia’s expanding power in the global chip industry, where its AI accelerators continue to drive demand for cutting-edge manufacturing and next-generation memory solutions.
