All three major HBM makers — Samsung, Micron and SK Hynix — plan to use TSMC base dies for next-generation HBM, according to Electronics Weekly. The report names the full DRAM triumvirate. That alone changes the competitive map of high-bandwidth memory supply.

The claim is narrow but significant. TSMC, not the memory makers themselves, would fabricate the logic base die that sits under the DRAM stacks. Each of the three suppliers currently builds HBM with its own approach to that layer. A common TSMC foundation would narrow the differences between their products.

What the report actually says

The Electronics Weekly headline states the fact plainly: Samsung, Micron and SK Hynix will use TSMC base dies for NVHBM. No capacity figures, deal values or timelines appear in the report. No executive was quoted. No process node was named.

That absence matters. A sourcing shift of this scale normally arrives with a joint statement, a wafer volume estimate or at least a named product generation. None is present here. Readers should treat the story as a direction of travel, not a signed contract.

Why the base die is the battleground

The base die is the logic layer of an HBM stack. It carries the interface that connects DRAM to the host processor. As HBM generations advance, that die absorbs more intelligence and moves to more advanced logic processes.

This is why TSMC keeps surfacing in HBM supply chains. TSMC has already publicized logic die work tied to major AI accelerator programs. The new report extends that pattern to the memory suppliers themselves, on all three fronts at once.

If confirmed, the shift would hand TSMC a chokepoint position in AI memory, mirroring the one it holds in AI logic. It would also pull Samsung's foundry out of a role it might otherwise have played for its own memory division. That is a notable concession if it holds.

What is not said

The report does not say when the transition starts. It does not say which HBM generation carries the TSMC base die. It does not clarify whether "NVHBM" refers to a custom variant for a specific customer or a general next-generation product family.

No dollar figures, wafer allocations or yield data appear. No party confirmed the arrangement on the record. Each of those gaps will shape whether this is a marginal sourcing tweak or a structural realignment of the HBM stack.

Competitive read

For TSMC, the report implies a widening of its HBM-related logic business from customer-specific base dies to memory-maker base dies across the board. For Samsung, Micron and SK Hynix, a shared TSMC base die would commoditize part of their differentiation and shift margin to the foundry.

It also tightens geographic concentration. Taiwan would then host both the leading AI logic and the leading HBM logic layer. That single sentence should concentrate the minds of procurement teams already worried about corridor risk.

Verification should come quickly if the story is real. Watch for TSMC capacity disclosures, HBM roadmap slides naming a foundry partner, or supply-chain filings that show wafers crossing from Hsinchu to memory packaging lines in Korea and the United States.