By Marcus Bennett

Samsung plans to build next-generation memory chips for Nvidia using TSMC process technology, according to a report from a Seoul newspaper picked up by The Business Times. The claim, if confirmed, would link three of the semiconductor industry's largest players in a single supply chain for AI memory.

The report is thin on detail. It states the deal, the parties and the technology direction. It does not state volumes, timelines, pricing or the specific memory standard involved. That matters, because the specifics determine who actually gains from the arrangement.

What does the deal involve?

Three parties appear in the report:

• Samsung — the memory manufacturer, producing the chips.
• TSMC — the foundry whose process technology would sit underneath Samsung's memory design.
• Nvidia — the customer whose next-generation products the chips would serve.

The arrangement, as described, splits the work. Samsung contributes the memory stack. TSMC contributes fabrication technology. Nvidia buys the output for future systems.

That split is unusual for Samsung. The company operates its own foundry business and has historically kept memory production in-house, end to end. Choosing a rival's process for a flagship customer's next-generation parts would mark a shift in how Samsung routes critical work.

Why would Samsung lean on a rival foundry?

Memory chips for AI accelerators increasingly depend on the logic layers that sit beneath or beside the memory itself. High-bandwidth memory and its successors pair stacked DRAM with a base die, and that base die is silicon logic. Foundries build it.

If TSMC builds that logic layer, Samsung's contribution narrows to the memory arrays themselves. TSMC's process leadership on advanced logic nodes is one possible reason a memory maker would route the base-layer work outward rather than keep it internal.

The report does not confirm this structure. It names the technology transfer, not the die architecture. But the mechanism — memory maker plus external foundry plus one large buyer — fits the pattern the AI memory market has followed elsewhere.

What does this change for Nvidia?

For Nvidia, a second qualified supply path for advanced memory would reduce dependence on a single vendor. SK Hynix has led the high-bandwidth memory segment, and Samsung has worked to win Nvidia qualification for its own stack. A Samsung memory product built on TSMC logic technology could give Nvidia a qualified alternative, or a complementary one, for next-generation accelerators.

For Samsung, the deal would keep its memory business inside Nvidia's supply chain, even if part of the value flows to a competitor's foundry.

For TSMC, it would mean revenue and process deployment outside its usual customer set, inside a product category where it does not sell the memory itself.

How firm is the report?

One Seoul newspaper carries the story. The Business Times relayed it. No company named in the report has confirmed the arrangement publicly, and none of the three has published specifications, dates or financial terms.

Readers should treat this as a sourcing lead rather than a signed contract. The practical questions — which memory generation, which node, what volume, when production starts — all remain open. Each answer would shift the market impact considerably.

A single unconfirmed report does not move silicon. But the structure it describes, with Samsung borrowing TSMC technology to serve Nvidia, would tighten an already concentrated supply chain around three companies that already control most of the world's advanced chip capacity. Confirmation from any of the parties would make this a story worth tracking closely.