Samsung plans to manufacture next-generation AI memory chips for Nvidia on TSMC's 3nm process, according to a report by The News International. The report says Samsung will use a TSMC fab rather than its own foundry network for the parts.

The deal, as described, covers memory chips destined for Nvidia's upcoming AI products. TSMC would supply the 3nm wafer capacity. Samsung would deliver the memory device itself.

Why would Samsung fabricate at a rival foundry?

The arrangement would be unusual. Samsung operates its own leading-edge foundry business and competes directly with TSMC at advanced nodes. Memory makers, however, often build logic-based chiplets or stacked packages on external wafers when the host processor demands tight integration.

The report did not specify which 3nm variant TSMC would use. TSMC runs two 3nm families at volume: N3B and the refined N3E. It also did not state volumes, pricing, or a production start date.

What does this mean for Nvidia's AI roadmap?

For Nvidia, the arrangement would secure a second supply path for AI memory beyond conventional HBM suppliers. Memory bandwidth remains the binding constraint on large AI accelerators, and closer coupling between memory and compute logic has driven recent packaging shifts across the industry.

The report frames the chips as "next-gen," which places them on Nvidia's future accelerator roadmap rather than current shipping products. No release window was given.

What is confirmed and what is not?

Confirmed by the report: Samsung as the memory supplier, TSMC as the wafer foundry, 3nm as the process node, and Nvidia as the end customer.

Not addressed:

• Which TSMC 3nm node (N3B or N3E) would carry the chips
• Wafer volumes or capacity commitments
• Deal value or pricing terms
• Production timeline or first shipment date
• Which Nvidia product family the memory targets

Samsung and TSMC have not publicly confirmed the arrangement. The claim rests on a single published report.

How does this fit the competitive picture?

A split of this kind would blur the usual lines in the memory market. Samsung would act as both a device maker and a TSMC customer. TSMC would gain AI memory volume on top of its accelerator logic business. Samsung's own foundry arm would sit out this particular product, a notable outcome given its push to land advanced-node AI customers.

Neither company has commented on the report. Confirmation of volumes and node details, when they arrive, will show whether this is a pilot-scale collaboration or a sustained supply commitment.