Samsung Reported to Build Nvidia Memory Chips on TSMC Process

Samsung plans to use TSMC process technology for the base dies of next-generation memory chips supplied to Nvidia, according to a report surfaced by Seeking Alpha. That single sentence, if confirmed, redraws a supply line that Samsung has fought to keep in-house.

Let us be precise about what is and is not established. The claim is attributed, not confirmed. Neither Samsung, nor TSMC, nor Nvidia has issued a statement. Seeking Alpha cites unnamed sources. The ticker reference in the original headline (SSNLF:OTCMKTS) concerns Samsung's over-the-counter listing only. No capacity figures, no deal value, no timeline appears in the report.

Why would TSMC silicon sit inside a Samsung memory stack?

High-bandwidth memory is not a single chip. It is a stack of DRAM dies mounted on a logic base die. That base die handles signal routing and increasingly performs processing near the memory itself.

The logic die is fabricated on a foundry logic process, not on the DRAM maker's memory line. This is where a split arises:

• The DRAM wafers come from the memory maker's own fabs.
• The base die can come from any foundry the customer or supplier chooses.

SK Hynix, the current HBM leader supplying Nvidia, has leaned on TSMC for advanced base-die work. If Samsung now does the same, the move would concede that TSMC's logic nodes offer something Samsung Foundry cannot match for this specific product, at least for the Nvidia qualification cycle.

What does the report not say?

Plenty. That gap matters more than the headline.

There is no stated process node. There is no wafer volume. There is no revenue estimate. There is no indication of whether TSMC would fabricate all Samsung base dies destined for Nvidia, or only those for a specific generation. The phrase "next-generation" is undefined in the source material.

Samsung has invested heavily in its own foundry business. Admitting, even implicitly, that a flagship customer's memory product needs a rival's process would be a commercial embarrassment in Seoul. That embarrassment is a reason for skepticism until parties confirm.

Is the claim plausible?

Context supports it, with caveats.

Nvidia dominates high-bandwidth memory demand for AI accelerators. The accelerator maker holds qualification power over its memory suppliers. If Nvidia's silicon requires a base-die process that Samsung Foundry cannot yet deliver at yield and volume, then routing the logic die to TSMC becomes the fastest path to shipment.

It would also be consistent with a broader pattern. Memory makers have historically partnered with logic foundries when advanced packaging and logic integration converged. SK Hynix set that precedent. Samsung following it is unremarkable from an engineering standpoint, and remarkable only from a competitive-standpoint one.

But plausibility is not confirmation. Sourcing-based reports in the semiconductor trade have been wrong before, particularly on unannounced customer relationships governed by strict non-disclosure agreements.

What should readers watch next?

Watch for three signals. First, any statement from the three companies, which is unlikely. Second, packaging and qualification timelines disclosed in earnings calls, where such arrangements sometimes surface indirectly. Third, Samsung Foundry's capacity and customer commentary in its next quarterly report, which would show whether internal logic wafer demand shifted.

Until then, treat this as a credible but unverified report about a single supply relationship. If confirmed, it strengthens TSMC's grip on the AI memory supply chain and marks another step in Samsung's struggle to convert foundry investment into flagship logic wins. If it is wrong, Samsung keeps the full stack, and the story dies quietly.