Samsung to Work With TSMC on Custom HBM for Nvidia

Samsung is collaborating with TSMC to supply custom high-bandwidth memory chips for Nvidia, according to a report carried by Yahoo Finance.

The reported arrangement links the two largest memory and foundry suppliers in the semiconductor industry. Samsung would provide the HBM chips. TSMC would handle work tied to their integration with Nvidia's processors. Nvidia is the intended end customer.

What do we know so far?

The report itself is the confirmed fact at this stage. It states that Samsung and TSMC are cooperating on custom HBM for Nvidia. No capacity figures, deal values, process nodes, or shipment numbers appear in the source material. Financial terms were not disclosed.

HBM, or high-bandwidth memory, sits at the center of the AI accelerator supply chain. Nvidia's data center GPUs depend on it. Memory vendors have raced to qualify their stacks with the chip designer, and qualification decisions shape billions of dollars in annual memory revenue.

Why would Samsung bring in TSMC?

Custom HBM typically means memory tuned to one customer's packaging and interconnect requirements rather than sold as a standard catalog part. Advanced packaging is where TSMC holds a strong position. Nvidia already uses TSMC's CoWoS family of packaging technology for its AI accelerators.

A joint Samsung-TSMC effort would therefore pair Samsung's memory fabrication with the foundry's packaging and integration role. For Samsung, a qualified custom HBM win at Nvidia would mark a significant step in its push to secure AI memory orders. For TSMC, it deepens its position in the AI supply chain. For Nvidia, it adds a second major HBM supplier path alongside existing memory partners.

Any deal of this kind would carry commercial weight beyond the three companies. HBM supply is tight, and AI chip demand has made memory qualification a decisive factor in accelerator output. A broader supplier base could ease constraints across the market.

What remains unconfirmed?

Several questions stay open. The report does not specify which HBM generation is involved, whether the chips target a named Nvidia product family, or when volume shipments would begin. It does not state whether the collaboration covers logic die manufacturing, base die work, or packaging alone.

Samsung, TSMC, and Nvidia have not, per the source material, published confirmation of the arrangement or its terms. Readers should treat capacity, timing, and revenue implications as unquantified until the companies disclose them.

If the collaboration proceeds to volume supply, it would reshape the competitive balance in HBM, where qualification at Nvidia has been the industry's most closely watched commercial contest.