Samsung Joins TSMC on Nvidia's Custom HBM, Report Says

Samsung is working with TSMC on custom high-bandwidth memory (HBM) for Nvidia, according to a report published by SammyGuru.

The report names three parties: Samsung, TSMC and Nvidia. It places Samsung's memory division inside a supply effort that runs through TSMC, the foundry that manufactures Nvidia's AI accelerators.

What does the reported arrangement cover?

The report's headline states the core fact: Samsung works with TSMC on Nvidia's custom HBM. SammyGuru published no further technical specifications, volumes or contract values alongside the headline.

HBM sits next to AI processors on the same package. Nvidia's accelerators depend on it for the memory bandwidth that large-model training and inference require. A custom HBM variant, as the headline describes, points to memory tailored to a specific customer's design rather than a standard catalogue part.

Why does the pairing matter?

The report links two companies that usually compete in different layers of the chip supply chain: Samsung, a leading memory maker, and TSMC, the dominant contract manufacturer of advanced logic. Nvidia sits downstream as the customer that specifies the memory.

For Samsung, an HBM role tied to Nvidia's supply chain would follow its established position in DRAM and HBM development. For TSMC, the report describes coordination with a memory supplier rather than a memory rival. The headline gives no dates, capacities or financial terms.

Samsung has not confirmed the arrangement in the source material. The report carries SammyGuru's byline and cites no named executives.

What remains unconfirmed?

The published report consists of the headline and outlet attribution. It includes:

• No contract value
• No production timeline
• No named Samsung, TSMC or Nvidia executives
• No specification of which Nvidia product the custom HBM targets

Die Signal could not independently verify the arrangement at publication time.

The story, as reported, is one line long: Samsung and TSMC are cooperating on custom HBM for Nvidia.