Nvidia Pushes Memory Makers Toward TSMC for Next-Gen HBM Base Dies

Nvidia has asked the leading memory makers to manufacture the base dies of their next-generation HBM at TSMC, according to a report carried by TipRanks. The request targets the suppliers of high-bandwidth memory that feeds Nvidia's AI accelerators. It would move a critical packaging component out of the memory makers' own fabs and into the world's largest contract foundry.

The report names no dollar value, wafer volume or timeline. It states the direction of Nvidia's demand, not a signed supply agreement.

What is an HBM base die?

High-bandwidth memory stacks DRAM dies on top of a logic die. That base die carries the through-silicon vias, the interface logic and the routing that connects the stack to the processor package. As HBM generations advance, the base die absorbs more functions and resembles a small logic chip more than a passive interposer.

This shift explains Nvidia's reported position. Memory makers such as SK hynix, Samsung and Micron build DRAM on their own process lines. Base dies with denser logic require advanced logic processes that these companies do not run at leading-edge nodes. TSMC does.

Why does the customer get a say?

Nvidia buys the overwhelming share of global HBM output for its data center GPUs. A customer with that purchasing weight can set technical specifications across its supply chain, including the process source of individual components inside a memory stack.

The report frames the request as Nvidia's preference. TipRanks does not state whether SK hynix, Samsung or Micron has agreed. It also does not state which HBM generation the request covers.

What changes for the supply chain?

A move to TSMC for base dies would deepen the AI memory supply chain's dependence on one foundry. It would also redraw the division of labor in HBM production.

The memory makers would keep DRAM wafer fabrication and stack assembly. TSMC would gain the logic-die layer, a component that grows in value with each HBM generation. For Samsung, which runs both memory and logic foundry operations, the request carries an added commercial edge, since it would route base-die work to its chief foundry rival.

No capacity figures, node assignments or pricing terms appear in the report. Any revenue impact on TSMC or the memory makers remains unquantified.

What is confirmed and what is not?

Confirmed by the report:

• Nvidia has asked top memory makers to use TSMC for next-generation HBM base dies.

Not established by the report:

• Whether any memory maker has accepted the request.
• The HBM generation, process node or wafer size involved.
• Volumes, timelines or contract terms.

Readers should treat the story as a statement of Nvidia's demand, attributed to a single aggregator report, rather than a confirmed multi-party supply arrangement.

What comes next?

The commercial signal is clear regardless of contract status: base-die logic now matters enough that Nvidia wants it on a leading-edge foundry process. Watch for confirmation from the memory makers, and for TSMC disclosures on HBM-related base-die capacity, before treating the shift as settled.