Samsung plans to use TSMC process technology for Nvidia's next-generation memory chips, according to a report from a Seoul newspaper carried by The Business Times. That single sentence, if confirmed, redraws a supply line that the industry has treated as settled for a decade.

The report is thin on numbers. What it signals, though, is structural. Samsung's memory division — the world's largest NAND supplier and Hynix's chief rival in high-bandwidth memory — would, on this account, lean on its foundry arch-rival to wire the logic dies that sit beneath Nvidia's HBM stacks. Nothing in the public record yet pins the claim to a named node, a wafer size, or a volume figure, and neither Samsung, TSMC, nor Nvidia has commented through official channels.

Why would Samsung turn to TSMC?

The likely answer sits at the bottom of an HBM stack. High-bandwidth memory packages DRAM on top of a logic base die, and that base die has become the hard part. As Nvidia pushes each HBM generation, the base die carries more routing, more speed, and tighter timing budgets than the DRAM layers above it.

Foundry leadership matters here. Whoever builds the base die needs a dense, fast logic process, and that has been a two-horse race between TSMC and Samsung Foundry. A report that Samsung's own memory group would hand that die to TSMC reads as an admission that the performance gap between the two foundries still bites at the leading edge — at least where an Nvidia spec sheet is involved.

It wouldn't be unprecedented. SK Hynix, Samsung's HBM competitor, already works with TSMC on logic dies for its HBM supply to Nvidia. If the Seoul report holds, Samsung would be adopting the same pattern: keep the DRAM in-house, buy the logic outside.

What's at stake commercially?

HBM is where memory margins now live. Nvidia's accelerator roadmap consumes HBM faster than any prior demand signal in DRAM history, and winning a next-generation socket is worth billions in annual revenue at current pricing. Samsung has spent quarters fighting to qualify its HBM3E with Nvidia; a deal that pairs Samsung DRAM with TSMC-built base dies would be a pragmatic route into the next generation rather than a purist's route.

For TSMC, the upside is direct. Every HBM stack Nvidia sells would carry a TSMC-fabricated component, extending the foundry's reach deeper into the memory supply chain it has historically left to the DRAM makers. Samsung Foundry, meanwhile, loses a showcase socket it could ill afford to lose while it works to close the gap at 3nm-class nodes and below.

For Nvidia, the arrangement would mean a second qualified HBM pipeline built the way it likes: SK Hynix memory on TSMC logic, and now Samsung memory on TSMC logic too. Dual sourcing, same interface, same supplier discipline.

What remains unconfirmed?

Plenty. The claim rests on a single Seoul newspaper account relayed by The Business Times. No filing, no keynote, no analyst note in the public domain yet names:

• the process node TSMC would run the base dies on,
• the memory generation involved — whether HBM4 or beyond,
• the volumes or revenue either party expects,
• the timeline for qualification or mass production.

Until Samsung or Nvidia confirms, this sits firmly in the roadmap-and-rumor column, not the capacity-and-revenue column. Analysts covering the HBM market have flagged base-die outsourcing as a live question for the HBM4 cycle, and this report is the sharpest public signal yet that the answer may favor TSMC.

If the deal is real, expect confirmation through the usual channels — an earnings call aside, a supply-chain filing, or a silent capacity reservation at one of TSMC's Taiwanese fabs. The competitive map of the HBM market, and the foundry pecking order beneath it, may hinge on which way that reservation lands.