Sources: Nvidia asked Samsung, SK Hynix, and Micron to make base dies for its custom NVHBM memory on TSMC's 3nm process; Samsung agreed, its first use of TSMC

Nvidia's custom HBM now uses TSMC's 3nm process, meaning Samsung's foundries are no longer the sole producer of this critical AI component.

Samsung Electronics has partnered with its rival TSMC to produce custom High Bandwidth Memory (HBM) for Nvidia, addressing memory bottlenecks impacting AI chip performance and supply. Nvidia is co-designing its NVHBM base die with Samsung and TSMC, integrating the memory controller directly onto the base die. This integration requires TSMC's advanced 3-nanometer process due to its ability to pack more circuits into a smaller area and achieve a high yield of over 90%. The move is driven by projected HBM demand increasing by 91% next year, with supply expected to fall short. Samsung's existing base die capacity is limited, prompting Nvidia to request TSMC's manufacturing capacity. While Samsung is also preparing to produce the base die on its own foundry, TSMC's involvement ensures a stable supply to meet Nvidia's demands.

This collaboration signifies a strategic shift for Samsung, forcing it to rely on a direct foundry competitor, TSMC, to meet the escalating demands of a major AI chip designer like Nvidia. It highlights the intense competition and interdependence within the semiconductor industry, where custom solutions and cutting-edge processes are paramount for AI acceleration. The reliance on TSMC's 3nm process for the base die underscores the technological hurdles in memory integration and the urgent need for increased HBM production capacity, which Samsung's own facilities are currently struggling to fulfill.

The decision to leverage TSMC's foundry capacity for Nvidia's custom HBM base die has significant implications for the broader AI chip market. It indicates that the demand for high-performance AI memory is outstripping the specialized manufacturing capabilities of individual memory suppliers, creating opportunities for foundries to play a more integrated role in chip design and production. This move could set a precedent for future custom AI memory solutions, potentially increasing TSMC's influence and prompting other memory manufacturers to explore similar cross-company collaborations to secure necessary production capacity and meet market demand.