Nvidia Reportedly Pushes Samsung, SK Hynix and Micron Toward TSMC 3nm

Nvidia has asked Samsung, SK Hynix and Micron to build products on TSMC's 3nm-class process node, The Next Web reports. The request, if confirmed, would pull the world's three largest memory makers toward a single foundry for logic-heavy memory products.

The report did not specify which products the request covers. Nor did it state a timeline, volume target or pricing arrangement. Nvidia has not commented publicly, and none of the three memory companies has confirmed the approach.

Why would a GPU maker care about memory makers' foundry choice?

The three suppliers named in the report produce the HBM stacks that sit beside Nvidia's data center GPUs. Those stacks combine DRAM dies with a logic layer at the base. The foundry that builds that logic layer determines interface speed, power profile and yield.

A shift to TSMC 3nm-class nodes for that logic would unify Nvidia's supply chain around one foundry. Nvidia already builds its own GPU compute dies at TSMC on advanced nodes. A common platform across GPU and HBM base die could simplify co-design of the memory interface, according to the framing of the report.

Samsung and SK Hynix operate their own logic fabs, so the request carries commercial weight. Samsung in particular markets its foundry as a home-grown alternative to TSMC. Moving HBM logic layers to a competitor's 3nm lines would redirect advanced-node demand away from Samsung Foundry.

Micron has no leading-edge logic foundry of its own and already outsources such work. For Micron, the request mainly affects which external foundry wins the business.

What is confirmed and what is not?

Only one element is on the record: The Next Web's report that Nvidia made the approach. Everything else remains open:

• No confirmed product family tied to the request
• No confirmed capacity figures or wafer volumes
• No timeline for a possible transition
• No comment from Nvidia, Samsung, SK Hynix, Micron or TSMC

TSMC's 3nm family, branded N3, is in volume production at the company's Taiwan fabs. The node family serves customers across mobile and high-performance computing silicon.

How does this fit the supply chain picture?

The reported request lands amid tight supply of HBM, the memory type Nvidia consumes in large volumes for AI accelerators. HBM supply agreements for coming years are largely spoken for among the major GPU vendors. Any change in where the supporting logic dies are fabricated would ripple through advanced-node allocation at TSMC.

It would also sharpen the divide in the memory industry. SK Hynix and Samsung compete on HBM share while Samsung simultaneously competes with TSMC for foundry customers. A single customer steering all three memory suppliers to one foundry concentrates advanced packaging and logic capacity in fewer hands.

Geopolitically, the move would deepen reliance on Taiwan-based leading-edge manufacturing. Samsung's Korean fabs represent the main scaled alternative at the 3nm-class generation. Nvidia steering suppliers toward TSMC would narrow that hedge, a factor boards and regulators now watch closely.

What happens next?

Confirmation would require statements from any of the four companies or supply-chain evidence such as wafer orders. Until then, the report stands as a signal of intent: Nvidia wants tighter control over the silicon ecosystem around its GPUs, and foundry alignment is its next lever. If the memory makers comply, competition for TSMC 3nm-class capacity among memory vendors would intensify through the coming procurement cycles.