Samsung TSMC Nvidia Memory Deal Pairs Two Chip Rivals

The Samsung TSMC Nvidia memory team-up just got real: Samsung will reportedly use TSMC's three-nanometre process to make a key part of Nvidia's next custom memory chips, according to Seoul Economic Daily. Two rivals are suddenly working on the same chip.

Reuters picked up the story on Sunday, and it matters because memory is now the bottleneck of the whole artificial intelligence boom. Faster processors are useless if data cannot reach them quickly enough.

What is actually happening

The report says Samsung Electronics will use TSMC's three-nanometre process to produce base dies for Nvidia's next-generation custom high-bandwidth memory, known as HBM. It would be the first time Samsung has worked with the Taiwanese chipmaker at the HBM manufacturing stage, according to Reuters via The Star.

Until now, Samsung made both parts of its HBM itself: the core die, which is the memory chip, and the base die, which works like a small logic chip underneath. Seoul Economic Daily says the company decided to accept Nvidia's technical requirements and change that all-in-one approach.

Why Nvidia wants this

The custom memory is called NVHBM and was unveiled in August. It puts Nvidia's own memory controller into the base die, and the goal is to standardise how memory from several suppliers plugs into its chips.

Nvidia has asked Samsung, SK Hynix and Micron Technology to all build the base die on the TSMC process, and to use sixth-generation DRAM technology for the core dies, the newspaper said, citing industry sources. Think of it as one shared blueprint for three suppliers.

Extra circuits on the base die are meant to speed up data transfer. Seoul Economic Daily also reports that TSMC's three-nanometre yield tops ninety percent, which helped it win the order, and that HBM demand is seen rising ninety-one percent next year.

Rivals, but also partners

Samsung and TSMC fight for the same big foundry customers, so this is a strange alliance. The logic is simple, though: the demand for AI memory is outrunning supply, and TSMC adds manufacturing capacity Samsung does not have to build from scratch.

Samsung is not giving up on its own factories. It plans to keep its in-house model for standard HBM and for custom products sold to other customers. It is also assessing whether its own foundry could make NVHBM base dies if demand exceeds TSMC's capacity, an unnamed source told the paper.

What it means for you

If you care about gadgets, games or chatbots, this is the plumbing behind all of them. More memory supply could eventually ease the shortages that have pushed up prices for AI hardware, though nothing here promises cheaper devices soon.

For now, all of this rests on industry sources, and neither company had confirmed it in the reports we saw. Treat it as a strong lead rather than a signed deal. We will keep following the chip race on our tech page.

Why base dies matter

A high-bandwidth memory stack is built like a tiny skyscraper. Layers of memory chips sit on top, and the base die is the ground floor that connects them to the processor. Whoever controls that ground floor controls how quickly data moves in and out.

That is why the Samsung TSMC Nvidia memory story is bigger than a routine supplier tweak. By putting its controller logic into the base die, Nvidia pulls a piece of the design in-house and asks every memory maker to follow the same recipe, which makes suppliers easier to swap and compare.

For Samsung, the shift also carries a risk of looking dependent on a rival. Yet the company has long struggled to convince Nvidia that its memory is ready for the most demanding chips, and a shared TSMC process could help close that gap quickly.

For SK Hynix and Micron, the ask is similar, so no single supplier gets an unfair edge on the logic side. The contest then moves to who can stack memory most reliably and ship the most of it, which is exactly where capacity matters.

The bigger picture

Chipmaking used to be a clean split: some firms designed, some firms manufactured, some firms made memory. The AI rush is blurring those lines fast, and Nvidia is clearly steering the whole supply chain by setting the specs for everyone else.

Watch for official statements from Samsung, TSMC and Nvidia, and for how SK Hynix and Micron respond. If all three suppliers really do converge on one TSMC-built base die, the memory market will look very different by next year.