Samsung to use TSMC 3nm process for Nvidia custom HBM

By Splitfeed EditorialLast updated Oct 18, 2026, 12:50 UTC

On October 18, 2026, Seoul Economic Daily reported that Samsung Electronics plans to use TSMC's 3nm fabrication process to produce the base die for Nvidia's custom high-bandwidth memory chips. Under the reported arrangement, Samsung retains responsibility for HBM memory stacking while TSMC handles the logic base die, meaning a single Nvidia memory product would draw on manufacturing from both companies. The story was subsequently picked up by multiple outlets the same day.

The significance of this arrangement lies in how it scrambles the usual competitive dynamic in the foundry market. Samsung and TSMC compete directly for the same chip manufacturing contracts, so a setup where they contribute complementary work to one Nvidia product is unusual. It also signals that Nvidia is exerting enough demand-side influence to shape supply chain relationships that would not otherwise exist. For Samsung, the deal preserves a role in a high-profile Nvidia product but represents a concession on the base die, where TSMC's 3nm process was apparently preferred. Whether this reflects a longer-term shift in how Samsung competes for advanced logic work, or a one-off accommodation to meet Nvidia's specifications, remains an open question.

None of the Pro-AI, Anti-AI, or Middle Ground camps have published reactions to this story yet. The Pro-AI camp would typically frame this kind of advanced chip collaboration as evidence that AI hardware demand is driving innovation in semiconductor supply chains. The Anti-AI camp might point to the concentration of AI-related manufacturing power among a small number of players as a systemic risk worth scrutinizing. The Middle Ground camp would likely focus on the technical and geopolitical complexity of having rival foundries cooperate under pressure from a dominant customer like Nvidia.

The details most worth watching are any official confirmation from Samsung, TSMC, or Nvidia, none of which had commented as of the initial reporting, as well as any subsequent disclosure about the timeline for production and whether this base die arrangement carries over into future HBM generations.