TrendForce reports that NVIDIA's proprietary NVHBM memory specification is reshaping how memory makers approach base die production, with Micron and Samsung moving to bring TSMC into their supply chains.

The report, distributed through TrendForce's news channel, frames NVHBM as the trigger for a strategic shift in base die sourcing across the high-bandwidth memory (HBM) segment. The specification, developed by NVIDIA for its AI accelerator roadmap, sets requirements that directly affect who fabricates the logic layer beneath the memory stacks.

What does NVHBM change?

Base dies sit at the bottom of every HBM stack. They carry the through-silicon via (TSV) interconnect structure and, increasingly, logic functions that manage the DRAM layers above them. Historically, memory makers such as Samsung and SK hynix fabricated these base dies on their own logic processes.

According to TrendForce, NVIDIA's NVHBM specification changes the economics and technical requirements of that arrangement. The report identifies Micron and Samsung as the two suppliers now adjusting their base die strategies in response, with both moving to incorporate TSMC as a fabrication partner for this layer.

Who is doing what?

The TrendForce report names three principal actors:

• NVIDIA — defines the NVHBM specification and drives demand through its AI accelerator product lines.
• Micron — adjusting its base die strategy to bring TSMC into production of the logic layer.
• Samsung — likewise reworking its base die approach to include TSMC fabrication.

TrendForce does not state in the headline material whether SK hynix, the third major HBM supplier, is making an equivalent change. The report's focus rests on Micron and Samsung.

Why does TSMC's involvement matter?

Bringing TSMC into HBM base die production marks a notable boundary shift in the memory supply chain. Foundry logic processes — the kind TSMC operates at advanced nodes — can deliver transistor density and performance characteristics that memory-optimized fabs cannot match. If NVHBM requires more sophisticated logic on the base die, memory vendors gain from outsourcing that layer to a logic specialist.

For TSMC, the development adds a new product category to its advanced-node customer base beyond CPUs, GPUs and mobile silicon. For Samsung and Micron, it splits the HBM stack across two manufacturers, with implications for supply chain coordination, wafer allocation and cost structure.

What remains open?

The publicly available headline material does not specify production volumes, timelines, process nodes or which NVIDIA products first carry NVHBM memory. TrendForce's full analysis covers the detail behind the strategic shift; readers tracking HBM supply dynamics should watch for confirmation of node assignments and qualification schedules from the parties involved.

The direction, however, is clear from the report's framing: NVIDIA's specification is pulling TSMC into a memory supply chain tier it previously served only peripherally, and Micron and Samsung are the first movers in that realignment.