Samsung to use TSMC 3nm process for Nvidia custom HBM chips

By Splitfeed EditorialLast updated Oct 19, 2026, 05:20 UTC

On October 18, 2026, Seoul Economic Daily reported that Samsung Electronics will use TSMC's 3nm fabrication process to build the base die for Nvidia's next-generation custom high-bandwidth memory chips, a product referred to in some reports as NVHBM. Samsung is bypassing its own 4nm foundry node for this component, sourcing it from TSMC at Nvidia's direction. The result is a single product that combines manufacturing from two companies that are ordinarily direct competitors.

The significance lies in what the arrangement reveals about where leverage sits in the AI hardware supply chain. Nvidia's custom HBM specification requires a logic-class base die built on an advanced process node, and Samsung's own foundry was apparently not the preferred option for that work. For Samsung, ceding a component to a rival within its own product is an unusual concession. For TSMC, landing base die work inside a Samsung HBM order extends its reach further into memory infrastructure. The arrangement also raises questions about how Nvidia's growing influence over supplier decisions shapes competition between foundries.

The pro-AI camp has posted no reactions to this story yet. Based on typical patterns, that camp would likely frame the development as a sign of healthy industry coordination, with leading companies pooling capabilities to accelerate AI hardware. The anti-AI camp is more vocal, though its concerns reach beyond this specific deal. Frank Frankly raised the tension between official rhetoric about beating China in AI and the continued sale of Nvidia chips to China. Samuel argued that Nvidia has drifted far from its origins, calling for a leaner company focused on consumer graphics rather than what he called AI garbage. Pashha Theosifon pointed to CoreWeave, described as a key part of Nvidia's financing ecosystem, being rated at 64% likelihood of default. Argonian Mage called the RTX 50 Series a waste of money. Dr. Steven Buckley was more terse, simply questioning the stability of the industry as a whole. The one middle-ground voice, Techmeme X Chatter, treated the unusual manufacturing pairing as a natural product of turbulent conditions.

The immediate question is whether Samsung's foundry division can maintain relevance in advanced AI chip components as Nvidia increasingly dictates process requirements. Any announcement from Nvidia clarifying the NVHBM specification, or from Samsung about its own next-generation node progress, would help clarify whether this arrangement is a one-time accommodation or a lasting shift in how AI memory supply chains are structured.