Samsung Plans 2nm HBM Base Die Production in Texas From 2030

Samsung plans to manufacture HBM base dies on a 2nm process at its second Taylor, Texas fab starting in 2030, according to a November 3, 2026 report by South Korean business newspaper Maeil Business. AMD is named as the primary customer, NVIDIA as a potential buyer — and neither claim has been officially confirmed. Samsung declined to comment on individual customer relationships when asked.

What has actually been reported?

Citing unnamed industry sources, Maeil Business says Samsung has already received orders covering the entire planned capacity of the second Taylor plant. There's been no independent confirmation of those contract volumes so far. What's on the record is the site plan itself:

• The new fab covers roughly 250,000 square meters — about one and a half times the size of the first Taylor facility
• Preparatory work is reportedly already underway, with construction set to begin in November 2026
• Manufacturing equipment installation is scheduled to start in February 2029
• Mass production is expected in 2030

Why the base die matters

The base die sits at the bottom of an HBM stack, handling communication between the vertically arranged DRAM layers and the connected processor. Unlike conventional DDR memory, HBM stacks numerous DRAM dies vertically and links them through very wide internal connections — that's how AI accelerators get enormous bandwidth in a compact footprint.

With HBM4, the demands on this component jump considerably. It must process more signals and integrate additional control functions. And here's where the process node becomes interesting: Samsung's current HBM4 base dies are built on its 4nm process, while the Taylor plan reportedly targets 2nm or smaller. A two-generation jump in process technology could allow more complex logic on the base die and better energy efficiency — though no specific performance gains can be inferred until Samsung publishes technical data on future products.

The NVIDIA complication

One detail in the report deserves particular attention. NVIDIA is pursuing a concept called NVHBM, which would integrate certain control functions more extensively into the base die. According to Maeil Business, NVIDIA intends to favor TSMC for manufacturing those dies, aiming for a consistent technical foundation across different memory suppliers.

If NVIDIA follows through, that could limit Samsung's ability to produce the corresponding base dies in its own foundry. It's conceivable that conventional HBM configurations and customized variants like NVHBM could coexist during a transition period — but right now nobody can say definitively what role the Texas plant would actually play in NVIDIA's supply chain.

What it means for AMD and the supply chain

For AMD, an additional manufacturing partner is attractive in principle — with highly integrated AI accelerators, memory, packaging, and logic design are closely interlinked. A second foundry broadens the supply chain, but only if the products are qualified and economically competitive. The technical coordination required shouldn't be underestimated: an alternative partner doesn't just slot in.

The bigger picture here isn't really the 2nm process. It's the growing strategic weight of the base die itself. So far, most attention has gone to stack counts, capacities, and bandwidths — but the underlying logic is becoming a genuine differentiator in modern AI architectures. More control functions could move directly into the memory stack, changing how tasks divide between accelerator and memory, and raising the bar for development, manufacturing, and validation.

Keep the caveats in mind

This shouldn't be read as a confirmed order win. Every specific claim about AMD, NVIDIA, and full capacity utilization traces back to a single media report, with Reuters relaying the same unconfirmed customer claims. Technical requirements and customer strategies could shift substantially before mass production starts in 2030.

What the report does show is how intense the competition for the next HBM generation already is — and how Samsung could use Taylor Fab 2 to bring its foundry business and memory operations closer together.